xref: /linux/drivers/nvme/host/fabrics.c (revision 6d8854216ebb60959ddb6f4ea4123bd449ba6cf6)
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 		/* Check for secure concatenation */
476 		if ((result & NVME_CONNECT_AUTHREQ_ASCR) &&
477 		    !ctrl->opts->concat) {
478 			dev_warn(ctrl->device,
479 				 "qid 0: secure concatenation is not supported\n");
480 			ret = -EOPNOTSUPP;
481 			goto out_free_data;
482 		}
483 		/* Authentication required */
484 		ret = nvme_auth_negotiate(ctrl, 0);
485 		if (ret) {
486 			dev_warn(ctrl->device,
487 				 "qid 0: authentication setup failed\n");
488 			goto out_free_data;
489 		}
490 		ret = nvme_auth_wait(ctrl, 0);
491 		if (ret) {
492 			dev_warn(ctrl->device,
493 				 "qid 0: authentication failed, error %d\n",
494 				 ret);
495 		} else
496 			dev_info(ctrl->device,
497 				 "qid 0: authenticated\n");
498 	}
499 out_free_data:
500 	kfree(data);
501 	return ret;
502 }
503 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
504 
505 /**
506  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
507  *			     API function.
508  * @ctrl:	Host nvme controller instance used to establish an
509  *		NVMe I/O queue connection to the already allocated NVMe
510  *		controller on the target system.
511  * @qid:	NVMe I/O queue number for the new I/O connection between
512  *		host and target (note qid == 0 is illegal as this is
513  *		the Admin queue, per NVMe standard).
514  *
515  * This function issues a fabrics-protocol connection
516  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
517  * between the host system device and the allocated NVMe controller
518  * on the target system.
519  *
520  * Return:
521  *	0: success
522  *	> 0: NVMe error status code
523  *	< 0: Linux errno error code
524  */
nvmf_connect_io_queue(struct nvme_ctrl * ctrl,u16 qid)525 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
526 {
527 	struct nvme_command cmd = { };
528 	struct nvmf_connect_data *data;
529 	union nvme_result res;
530 	int ret;
531 	u32 result;
532 
533 	nvmf_connect_cmd_prep(ctrl, qid, &cmd);
534 
535 	data = nvmf_connect_data_prep(ctrl, ctrl->cntlid);
536 	if (!data)
537 		return -ENOMEM;
538 
539 	ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
540 			data, sizeof(*data), qid,
541 			NVME_SUBMIT_AT_HEAD |
542 			NVME_SUBMIT_RESERVED |
543 			NVME_SUBMIT_NOWAIT);
544 	if (ret) {
545 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
546 				       &cmd, data);
547 		goto out_free_data;
548 	}
549 	result = le32_to_cpu(res.u32);
550 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
551 		/* Secure concatenation is not implemented */
552 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
553 			dev_warn(ctrl->device,
554 				 "qid %d: secure concatenation is not supported\n", qid);
555 			ret = -EOPNOTSUPP;
556 			goto out_free_data;
557 		}
558 		/* Authentication required */
559 		ret = nvme_auth_negotiate(ctrl, qid);
560 		if (ret) {
561 			dev_warn(ctrl->device,
562 				 "qid %d: authentication setup failed\n", qid);
563 			goto out_free_data;
564 		}
565 		ret = nvme_auth_wait(ctrl, qid);
566 		if (ret) {
567 			dev_warn(ctrl->device,
568 				 "qid %u: authentication failed, error %d\n",
569 				 qid, ret);
570 		}
571 	}
572 out_free_data:
573 	kfree(data);
574 	return ret;
575 }
576 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
577 
578 /*
579  * Evaluate the status information returned by the transport in order to decided
580  * if a reconnect attempt should be scheduled.
581  *
582  * Do not retry when:
583  *
584  * - the DNR bit is set and the specification states no further connect
585  *   attempts with the same set of parameters should be attempted.
586  *
587  * - when the authentication attempt fails, because the key was invalid.
588  *   This error code is set on the host side.
589  */
nvmf_should_reconnect(struct nvme_ctrl * ctrl,int status)590 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl, int status)
591 {
592 	if (status > 0 && (status & NVME_STATUS_DNR))
593 		return false;
594 
595 	if (status == -EKEYREJECTED)
596 		return false;
597 
598 	if (ctrl->opts->max_reconnects == -1 ||
599 	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
600 		return true;
601 
602 	return false;
603 }
604 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
605 
606 /**
607  * nvmf_register_transport() - NVMe Fabrics Library registration function.
608  * @ops:	Transport ops instance to be registered to the
609  *		common fabrics library.
610  *
611  * API function that registers the type of specific transport fabric
612  * being implemented to the common NVMe fabrics library. Part of
613  * the overall init sequence of starting up a fabrics driver.
614  */
nvmf_register_transport(struct nvmf_transport_ops * ops)615 int nvmf_register_transport(struct nvmf_transport_ops *ops)
616 {
617 	if (!ops->create_ctrl)
618 		return -EINVAL;
619 
620 	down_write(&nvmf_transports_rwsem);
621 	list_add_tail(&ops->entry, &nvmf_transports);
622 	up_write(&nvmf_transports_rwsem);
623 
624 	return 0;
625 }
626 EXPORT_SYMBOL_GPL(nvmf_register_transport);
627 
628 /**
629  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
630  * @ops:	Transport ops instance to be unregistered from the
631  *		common fabrics library.
632  *
633  * Fabrics API function that unregisters the type of specific transport
634  * fabric being implemented from the common NVMe fabrics library.
635  * Part of the overall exit sequence of unloading the implemented driver.
636  */
nvmf_unregister_transport(struct nvmf_transport_ops * ops)637 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
638 {
639 	down_write(&nvmf_transports_rwsem);
640 	list_del(&ops->entry);
641 	up_write(&nvmf_transports_rwsem);
642 }
643 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
644 
nvmf_lookup_transport(struct nvmf_ctrl_options * opts)645 static struct nvmf_transport_ops *nvmf_lookup_transport(
646 		struct nvmf_ctrl_options *opts)
647 {
648 	struct nvmf_transport_ops *ops;
649 
650 	lockdep_assert_held(&nvmf_transports_rwsem);
651 
652 	list_for_each_entry(ops, &nvmf_transports, entry) {
653 		if (strcmp(ops->name, opts->transport) == 0)
654 			return ops;
655 	}
656 
657 	return NULL;
658 }
659 
nvmf_parse_key(int key_id)660 static struct key *nvmf_parse_key(int key_id)
661 {
662 	struct key *key;
663 
664 	if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
665 		pr_err("TLS is not supported\n");
666 		return ERR_PTR(-EINVAL);
667 	}
668 
669 	key = nvme_tls_key_lookup(key_id);
670 	if (IS_ERR(key))
671 		pr_err("key id %08x not found\n", key_id);
672 	else
673 		pr_debug("Using key id %08x\n", key_id);
674 	return key;
675 }
676 
677 static const match_table_t opt_tokens = {
678 	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
679 	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
680 	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
681 	{ NVMF_OPT_NQN,			"nqn=%s"		},
682 	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
683 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
684 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
685 	{ NVMF_OPT_CTRL_LOSS_TMO,	"ctrl_loss_tmo=%d"	},
686 	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
687 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
688 	{ NVMF_OPT_HOST_TRADDR,		"host_traddr=%s"	},
689 	{ NVMF_OPT_HOST_IFACE,		"host_iface=%s"		},
690 	{ NVMF_OPT_HOST_ID,		"hostid=%s"		},
691 	{ NVMF_OPT_DUP_CONNECT,		"duplicate_connect"	},
692 	{ NVMF_OPT_DISABLE_SQFLOW,	"disable_sqflow"	},
693 	{ NVMF_OPT_HDR_DIGEST,		"hdr_digest"		},
694 	{ NVMF_OPT_DATA_DIGEST,		"data_digest"		},
695 	{ NVMF_OPT_NR_WRITE_QUEUES,	"nr_write_queues=%d"	},
696 	{ NVMF_OPT_NR_POLL_QUEUES,	"nr_poll_queues=%d"	},
697 	{ NVMF_OPT_TOS,			"tos=%d"		},
698 #ifdef CONFIG_NVME_TCP_TLS
699 	{ NVMF_OPT_KEYRING,		"keyring=%d"		},
700 	{ NVMF_OPT_TLS_KEY,		"tls_key=%d"		},
701 #endif
702 	{ NVMF_OPT_FAIL_FAST_TMO,	"fast_io_fail_tmo=%d"	},
703 	{ NVMF_OPT_DISCOVERY,		"discovery"		},
704 #ifdef CONFIG_NVME_HOST_AUTH
705 	{ NVMF_OPT_DHCHAP_SECRET,	"dhchap_secret=%s"	},
706 	{ NVMF_OPT_DHCHAP_CTRL_SECRET,	"dhchap_ctrl_secret=%s"	},
707 #endif
708 #ifdef CONFIG_NVME_TCP_TLS
709 	{ NVMF_OPT_TLS,			"tls"			},
710 	{ NVMF_OPT_CONCAT,		"concat"		},
711 #endif
712 	{ NVMF_OPT_ERR,			NULL			}
713 };
714 
nvmf_parse_options(struct nvmf_ctrl_options * opts,const char * buf)715 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
716 		const char *buf)
717 {
718 	substring_t args[MAX_OPT_ARGS];
719 	char *options, *o, *p;
720 	int token, ret = 0;
721 	size_t nqnlen  = 0;
722 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO, key_id;
723 	uuid_t hostid;
724 	char hostnqn[NVMF_NQN_SIZE];
725 	struct key *key;
726 
727 	/* Set defaults */
728 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
729 	opts->nr_io_queues = num_online_cpus();
730 	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
731 	opts->kato = 0;
732 	opts->duplicate_connect = false;
733 	opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
734 	opts->hdr_digest = false;
735 	opts->data_digest = false;
736 	opts->tos = -1; /* < 0 == use transport default */
737 	opts->tls = false;
738 	opts->tls_key = NULL;
739 	opts->keyring = NULL;
740 	opts->concat = false;
741 
742 	options = o = kstrdup(buf, GFP_KERNEL);
743 	if (!options)
744 		return -ENOMEM;
745 
746 	/* use default host if not given by user space */
747 	uuid_copy(&hostid, &nvmf_default_host->id);
748 	strscpy(hostnqn, nvmf_default_host->nqn, NVMF_NQN_SIZE);
749 
750 	while ((p = strsep(&o, ",\n")) != NULL) {
751 		if (!*p)
752 			continue;
753 
754 		token = match_token(p, opt_tokens, args);
755 		opts->mask |= token;
756 		switch (token) {
757 		case NVMF_OPT_TRANSPORT:
758 			p = match_strdup(args);
759 			if (!p) {
760 				ret = -ENOMEM;
761 				goto out;
762 			}
763 			kfree(opts->transport);
764 			opts->transport = p;
765 			break;
766 		case NVMF_OPT_NQN:
767 			p = match_strdup(args);
768 			if (!p) {
769 				ret = -ENOMEM;
770 				goto out;
771 			}
772 			kfree(opts->subsysnqn);
773 			opts->subsysnqn = p;
774 			nqnlen = strlen(opts->subsysnqn);
775 			if (nqnlen >= NVMF_NQN_SIZE) {
776 				pr_err("%s needs to be < %d bytes\n",
777 					opts->subsysnqn, NVMF_NQN_SIZE);
778 				ret = -EINVAL;
779 				goto out;
780 			}
781 			opts->discovery_nqn =
782 				!(strcmp(opts->subsysnqn,
783 					 NVME_DISC_SUBSYS_NAME));
784 			break;
785 		case NVMF_OPT_TRADDR:
786 			p = match_strdup(args);
787 			if (!p) {
788 				ret = -ENOMEM;
789 				goto out;
790 			}
791 			kfree(opts->traddr);
792 			opts->traddr = p;
793 			break;
794 		case NVMF_OPT_TRSVCID:
795 			p = match_strdup(args);
796 			if (!p) {
797 				ret = -ENOMEM;
798 				goto out;
799 			}
800 			kfree(opts->trsvcid);
801 			opts->trsvcid = p;
802 			break;
803 		case NVMF_OPT_QUEUE_SIZE:
804 			if (match_int(args, &token)) {
805 				ret = -EINVAL;
806 				goto out;
807 			}
808 			if (token < NVMF_MIN_QUEUE_SIZE ||
809 			    token > NVMF_MAX_QUEUE_SIZE) {
810 				pr_err("Invalid queue_size %d\n", token);
811 				ret = -EINVAL;
812 				goto out;
813 			}
814 			opts->queue_size = token;
815 			break;
816 		case NVMF_OPT_NR_IO_QUEUES:
817 			if (match_int(args, &token)) {
818 				ret = -EINVAL;
819 				goto out;
820 			}
821 			if (token <= 0) {
822 				pr_err("Invalid number of IOQs %d\n", token);
823 				ret = -EINVAL;
824 				goto out;
825 			}
826 			if (opts->discovery_nqn) {
827 				pr_debug("Ignoring nr_io_queues value for discovery controller\n");
828 				break;
829 			}
830 
831 			opts->nr_io_queues = min_t(unsigned int,
832 					num_online_cpus(), token);
833 			break;
834 		case NVMF_OPT_KATO:
835 			if (match_int(args, &token)) {
836 				ret = -EINVAL;
837 				goto out;
838 			}
839 
840 			if (token < 0) {
841 				pr_err("Invalid keep_alive_tmo %d\n", token);
842 				ret = -EINVAL;
843 				goto out;
844 			} else if (token == 0 && !opts->discovery_nqn) {
845 				/* Allowed for debug */
846 				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
847 			}
848 			opts->kato = token;
849 			break;
850 		case NVMF_OPT_CTRL_LOSS_TMO:
851 			if (match_int(args, &token)) {
852 				ret = -EINVAL;
853 				goto out;
854 			}
855 
856 			if (token < 0)
857 				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
858 			ctrl_loss_tmo = token;
859 			break;
860 		case NVMF_OPT_FAIL_FAST_TMO:
861 			if (match_int(args, &token)) {
862 				ret = -EINVAL;
863 				goto out;
864 			}
865 
866 			if (token >= 0)
867 				pr_warn("I/O fail on reconnect controller after %d sec\n",
868 					token);
869 			else
870 				token = -1;
871 
872 			opts->fast_io_fail_tmo = token;
873 			break;
874 		case NVMF_OPT_HOSTNQN:
875 			if (opts->host) {
876 				pr_err("hostnqn already user-assigned: %s\n",
877 				       opts->host->nqn);
878 				ret = -EADDRINUSE;
879 				goto out;
880 			}
881 			p = match_strdup(args);
882 			if (!p) {
883 				ret = -ENOMEM;
884 				goto out;
885 			}
886 			nqnlen = strlen(p);
887 			if (nqnlen >= NVMF_NQN_SIZE) {
888 				pr_err("%s needs to be < %d bytes\n",
889 					p, NVMF_NQN_SIZE);
890 				kfree(p);
891 				ret = -EINVAL;
892 				goto out;
893 			}
894 			strscpy(hostnqn, p, NVMF_NQN_SIZE);
895 			kfree(p);
896 			break;
897 		case NVMF_OPT_RECONNECT_DELAY:
898 			if (match_int(args, &token)) {
899 				ret = -EINVAL;
900 				goto out;
901 			}
902 			if (token <= 0) {
903 				pr_err("Invalid reconnect_delay %d\n", token);
904 				ret = -EINVAL;
905 				goto out;
906 			}
907 			opts->reconnect_delay = token;
908 			break;
909 		case NVMF_OPT_HOST_TRADDR:
910 			p = match_strdup(args);
911 			if (!p) {
912 				ret = -ENOMEM;
913 				goto out;
914 			}
915 			kfree(opts->host_traddr);
916 			opts->host_traddr = p;
917 			break;
918 		case NVMF_OPT_HOST_IFACE:
919 			p = match_strdup(args);
920 			if (!p) {
921 				ret = -ENOMEM;
922 				goto out;
923 			}
924 			kfree(opts->host_iface);
925 			opts->host_iface = p;
926 			break;
927 		case NVMF_OPT_HOST_ID:
928 			p = match_strdup(args);
929 			if (!p) {
930 				ret = -ENOMEM;
931 				goto out;
932 			}
933 			ret = uuid_parse(p, &hostid);
934 			if (ret) {
935 				pr_err("Invalid hostid %s\n", p);
936 				ret = -EINVAL;
937 				kfree(p);
938 				goto out;
939 			}
940 			kfree(p);
941 			break;
942 		case NVMF_OPT_DUP_CONNECT:
943 			opts->duplicate_connect = true;
944 			break;
945 		case NVMF_OPT_DISABLE_SQFLOW:
946 			opts->disable_sqflow = true;
947 			break;
948 		case NVMF_OPT_HDR_DIGEST:
949 			opts->hdr_digest = true;
950 			break;
951 		case NVMF_OPT_DATA_DIGEST:
952 			opts->data_digest = true;
953 			break;
954 		case NVMF_OPT_NR_WRITE_QUEUES:
955 			if (match_int(args, &token)) {
956 				ret = -EINVAL;
957 				goto out;
958 			}
959 			if (token <= 0) {
960 				pr_err("Invalid nr_write_queues %d\n", token);
961 				ret = -EINVAL;
962 				goto out;
963 			}
964 			opts->nr_write_queues = token;
965 			break;
966 		case NVMF_OPT_NR_POLL_QUEUES:
967 			if (match_int(args, &token)) {
968 				ret = -EINVAL;
969 				goto out;
970 			}
971 			if (token <= 0) {
972 				pr_err("Invalid nr_poll_queues %d\n", token);
973 				ret = -EINVAL;
974 				goto out;
975 			}
976 			opts->nr_poll_queues = token;
977 			break;
978 		case NVMF_OPT_TOS:
979 			if (match_int(args, &token)) {
980 				ret = -EINVAL;
981 				goto out;
982 			}
983 			if (token < 0) {
984 				pr_err("Invalid type of service %d\n", token);
985 				ret = -EINVAL;
986 				goto out;
987 			}
988 			if (token > 255) {
989 				pr_warn("Clamping type of service to 255\n");
990 				token = 255;
991 			}
992 			opts->tos = token;
993 			break;
994 		case NVMF_OPT_KEYRING:
995 			if (match_int(args, &key_id) || key_id <= 0) {
996 				ret = -EINVAL;
997 				goto out;
998 			}
999 			key = nvmf_parse_key(key_id);
1000 			if (IS_ERR(key)) {
1001 				ret = PTR_ERR(key);
1002 				goto out;
1003 			}
1004 			key_put(opts->keyring);
1005 			opts->keyring = key;
1006 			break;
1007 		case NVMF_OPT_TLS_KEY:
1008 			if (match_int(args, &key_id) || key_id <= 0) {
1009 				ret = -EINVAL;
1010 				goto out;
1011 			}
1012 			key = nvmf_parse_key(key_id);
1013 			if (IS_ERR(key)) {
1014 				ret = PTR_ERR(key);
1015 				goto out;
1016 			}
1017 			key_put(opts->tls_key);
1018 			opts->tls_key = key;
1019 			break;
1020 		case NVMF_OPT_DISCOVERY:
1021 			opts->discovery_nqn = true;
1022 			break;
1023 		case NVMF_OPT_DHCHAP_SECRET:
1024 			p = match_strdup(args);
1025 			if (!p) {
1026 				ret = -ENOMEM;
1027 				goto out;
1028 			}
1029 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1030 				pr_err("Invalid DH-CHAP secret %s\n", p);
1031 				ret = -EINVAL;
1032 				goto out;
1033 			}
1034 			kfree(opts->dhchap_secret);
1035 			opts->dhchap_secret = p;
1036 			break;
1037 		case NVMF_OPT_DHCHAP_CTRL_SECRET:
1038 			p = match_strdup(args);
1039 			if (!p) {
1040 				ret = -ENOMEM;
1041 				goto out;
1042 			}
1043 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1044 				pr_err("Invalid DH-CHAP secret %s\n", p);
1045 				ret = -EINVAL;
1046 				goto out;
1047 			}
1048 			kfree(opts->dhchap_ctrl_secret);
1049 			opts->dhchap_ctrl_secret = p;
1050 			break;
1051 		case NVMF_OPT_TLS:
1052 			if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
1053 				pr_err("TLS is not supported\n");
1054 				ret = -EINVAL;
1055 				goto out;
1056 			}
1057 			opts->tls = true;
1058 			break;
1059 		case NVMF_OPT_CONCAT:
1060 			if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
1061 				pr_err("TLS is not supported\n");
1062 				ret = -EINVAL;
1063 				goto out;
1064 			}
1065 			opts->concat = true;
1066 			break;
1067 		default:
1068 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
1069 				p);
1070 			ret = -EINVAL;
1071 			goto out;
1072 		}
1073 	}
1074 
1075 	if (opts->discovery_nqn) {
1076 		opts->nr_io_queues = 0;
1077 		opts->nr_write_queues = 0;
1078 		opts->nr_poll_queues = 0;
1079 		opts->duplicate_connect = true;
1080 	} else {
1081 		if (!opts->kato)
1082 			opts->kato = NVME_DEFAULT_KATO;
1083 	}
1084 	if (ctrl_loss_tmo < 0) {
1085 		opts->max_reconnects = -1;
1086 	} else {
1087 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
1088 						opts->reconnect_delay);
1089 		if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
1090 			pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
1091 				opts->fast_io_fail_tmo, ctrl_loss_tmo);
1092 	}
1093 	if (opts->concat) {
1094 		if (opts->tls) {
1095 			pr_err("Secure concatenation over TLS is not supported\n");
1096 			ret = -EINVAL;
1097 			goto out;
1098 		}
1099 		if (opts->tls_key) {
1100 			pr_err("Cannot specify a TLS key for secure concatenation\n");
1101 			ret = -EINVAL;
1102 			goto out;
1103 		}
1104 		if (!opts->dhchap_secret) {
1105 			pr_err("Need to enable DH-CHAP for secure concatenation\n");
1106 			ret = -EINVAL;
1107 			goto out;
1108 		}
1109 	}
1110 
1111 	opts->host = nvmf_host_add(hostnqn, &hostid);
1112 	if (IS_ERR(opts->host)) {
1113 		ret = PTR_ERR(opts->host);
1114 		opts->host = NULL;
1115 		goto out;
1116 	}
1117 
1118 out:
1119 	kfree(options);
1120 	return ret;
1121 }
1122 
nvmf_set_io_queues(struct nvmf_ctrl_options * opts,u32 nr_io_queues,u32 io_queues[HCTX_MAX_TYPES])1123 void nvmf_set_io_queues(struct nvmf_ctrl_options *opts, u32 nr_io_queues,
1124 			u32 io_queues[HCTX_MAX_TYPES])
1125 {
1126 	if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
1127 		/*
1128 		 * separate read/write queues
1129 		 * hand out dedicated default queues only after we have
1130 		 * sufficient read queues.
1131 		 */
1132 		io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
1133 		nr_io_queues -= io_queues[HCTX_TYPE_READ];
1134 		io_queues[HCTX_TYPE_DEFAULT] =
1135 			min(opts->nr_write_queues, nr_io_queues);
1136 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1137 	} else {
1138 		/*
1139 		 * shared read/write queues
1140 		 * either no write queues were requested, or we don't have
1141 		 * sufficient queue count to have dedicated default queues.
1142 		 */
1143 		io_queues[HCTX_TYPE_DEFAULT] =
1144 			min(opts->nr_io_queues, nr_io_queues);
1145 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1146 	}
1147 
1148 	if (opts->nr_poll_queues && nr_io_queues) {
1149 		/* map dedicated poll queues only if we have queues left */
1150 		io_queues[HCTX_TYPE_POLL] =
1151 			min(opts->nr_poll_queues, nr_io_queues);
1152 	}
1153 }
1154 EXPORT_SYMBOL_GPL(nvmf_set_io_queues);
1155 
nvmf_map_queues(struct blk_mq_tag_set * set,struct nvme_ctrl * ctrl,u32 io_queues[HCTX_MAX_TYPES])1156 void nvmf_map_queues(struct blk_mq_tag_set *set, struct nvme_ctrl *ctrl,
1157 		     u32 io_queues[HCTX_MAX_TYPES])
1158 {
1159 	struct nvmf_ctrl_options *opts = ctrl->opts;
1160 
1161 	if (opts->nr_write_queues && io_queues[HCTX_TYPE_READ]) {
1162 		/* separate read/write queues */
1163 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1164 			io_queues[HCTX_TYPE_DEFAULT];
1165 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1166 		set->map[HCTX_TYPE_READ].nr_queues =
1167 			io_queues[HCTX_TYPE_READ];
1168 		set->map[HCTX_TYPE_READ].queue_offset =
1169 			io_queues[HCTX_TYPE_DEFAULT];
1170 	} else {
1171 		/* shared read/write queues */
1172 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1173 			io_queues[HCTX_TYPE_DEFAULT];
1174 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1175 		set->map[HCTX_TYPE_READ].nr_queues =
1176 			io_queues[HCTX_TYPE_DEFAULT];
1177 		set->map[HCTX_TYPE_READ].queue_offset = 0;
1178 	}
1179 
1180 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1181 	blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
1182 	if (opts->nr_poll_queues && io_queues[HCTX_TYPE_POLL]) {
1183 		/* map dedicated poll queues only if we have queues left */
1184 		set->map[HCTX_TYPE_POLL].nr_queues = io_queues[HCTX_TYPE_POLL];
1185 		set->map[HCTX_TYPE_POLL].queue_offset =
1186 			io_queues[HCTX_TYPE_DEFAULT] +
1187 			io_queues[HCTX_TYPE_READ];
1188 		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
1189 	}
1190 
1191 	dev_info(ctrl->device,
1192 		"mapped %d/%d/%d default/read/poll queues.\n",
1193 		io_queues[HCTX_TYPE_DEFAULT],
1194 		io_queues[HCTX_TYPE_READ],
1195 		io_queues[HCTX_TYPE_POLL]);
1196 }
1197 EXPORT_SYMBOL_GPL(nvmf_map_queues);
1198 
nvmf_check_required_opts(struct nvmf_ctrl_options * opts,unsigned int required_opts)1199 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
1200 		unsigned int required_opts)
1201 {
1202 	if ((opts->mask & required_opts) != required_opts) {
1203 		unsigned int i;
1204 
1205 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1206 			if ((opt_tokens[i].token & required_opts) &&
1207 			    !(opt_tokens[i].token & opts->mask)) {
1208 				pr_warn("missing parameter '%s'\n",
1209 					opt_tokens[i].pattern);
1210 			}
1211 		}
1212 
1213 		return -EINVAL;
1214 	}
1215 
1216 	return 0;
1217 }
1218 
nvmf_ip_options_match(struct nvme_ctrl * ctrl,struct nvmf_ctrl_options * opts)1219 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
1220 		struct nvmf_ctrl_options *opts)
1221 {
1222 	if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
1223 	    strcmp(opts->traddr, ctrl->opts->traddr) ||
1224 	    strcmp(opts->trsvcid, ctrl->opts->trsvcid))
1225 		return false;
1226 
1227 	/*
1228 	 * Checking the local address or host interfaces is rough.
1229 	 *
1230 	 * In most cases, none is specified and the host port or
1231 	 * host interface is selected by the stack.
1232 	 *
1233 	 * Assume no match if:
1234 	 * -  local address or host interface is specified and address
1235 	 *    or host interface is not the same
1236 	 * -  local address or host interface is not specified but
1237 	 *    remote is, or vice versa (admin using specific
1238 	 *    host_traddr/host_iface when it matters).
1239 	 */
1240 	if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
1241 	    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1242 		if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
1243 			return false;
1244 	} else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
1245 		   (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1246 		return false;
1247 	}
1248 
1249 	if ((opts->mask & NVMF_OPT_HOST_IFACE) &&
1250 	    (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1251 		if (strcmp(opts->host_iface, ctrl->opts->host_iface))
1252 			return false;
1253 	} else if ((opts->mask & NVMF_OPT_HOST_IFACE) ||
1254 		   (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1255 		return false;
1256 	}
1257 
1258 	return true;
1259 }
1260 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
1261 
nvmf_check_allowed_opts(struct nvmf_ctrl_options * opts,unsigned int allowed_opts)1262 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
1263 		unsigned int allowed_opts)
1264 {
1265 	if (opts->mask & ~allowed_opts) {
1266 		unsigned int i;
1267 
1268 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1269 			if ((opt_tokens[i].token & opts->mask) &&
1270 			    (opt_tokens[i].token & ~allowed_opts)) {
1271 				pr_warn("invalid parameter '%s'\n",
1272 					opt_tokens[i].pattern);
1273 			}
1274 		}
1275 
1276 		return -EINVAL;
1277 	}
1278 
1279 	return 0;
1280 }
1281 
nvmf_free_options(struct nvmf_ctrl_options * opts)1282 void nvmf_free_options(struct nvmf_ctrl_options *opts)
1283 {
1284 	nvmf_host_put(opts->host);
1285 	key_put(opts->keyring);
1286 	key_put(opts->tls_key);
1287 	kfree(opts->transport);
1288 	kfree(opts->traddr);
1289 	kfree(opts->trsvcid);
1290 	kfree(opts->subsysnqn);
1291 	kfree(opts->host_traddr);
1292 	kfree(opts->host_iface);
1293 	kfree(opts->dhchap_secret);
1294 	kfree(opts->dhchap_ctrl_secret);
1295 	kfree(opts);
1296 }
1297 EXPORT_SYMBOL_GPL(nvmf_free_options);
1298 
1299 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
1300 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
1301 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
1302 				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
1303 				 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\
1304 				 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\
1305 				 NVMF_OPT_DHCHAP_CTRL_SECRET)
1306 
1307 static struct nvme_ctrl *
nvmf_create_ctrl(struct device * dev,const char * buf)1308 nvmf_create_ctrl(struct device *dev, const char *buf)
1309 {
1310 	struct nvmf_ctrl_options *opts;
1311 	struct nvmf_transport_ops *ops;
1312 	struct nvme_ctrl *ctrl;
1313 	int ret;
1314 
1315 	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1316 	if (!opts)
1317 		return ERR_PTR(-ENOMEM);
1318 
1319 	ret = nvmf_parse_options(opts, buf);
1320 	if (ret)
1321 		goto out_free_opts;
1322 
1323 
1324 	request_module("nvme-%s", opts->transport);
1325 
1326 	/*
1327 	 * Check the generic options first as we need a valid transport for
1328 	 * the lookup below.  Then clear the generic flags so that transport
1329 	 * drivers don't have to care about them.
1330 	 */
1331 	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1332 	if (ret)
1333 		goto out_free_opts;
1334 	opts->mask &= ~NVMF_REQUIRED_OPTS;
1335 
1336 	down_read(&nvmf_transports_rwsem);
1337 	ops = nvmf_lookup_transport(opts);
1338 	if (!ops) {
1339 		pr_info("no handler found for transport %s.\n",
1340 			opts->transport);
1341 		ret = -EINVAL;
1342 		goto out_unlock;
1343 	}
1344 
1345 	if (!try_module_get(ops->module)) {
1346 		ret = -EBUSY;
1347 		goto out_unlock;
1348 	}
1349 	up_read(&nvmf_transports_rwsem);
1350 
1351 	ret = nvmf_check_required_opts(opts, ops->required_opts);
1352 	if (ret)
1353 		goto out_module_put;
1354 	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1355 				ops->allowed_opts | ops->required_opts);
1356 	if (ret)
1357 		goto out_module_put;
1358 
1359 	ctrl = ops->create_ctrl(dev, opts);
1360 	if (IS_ERR(ctrl)) {
1361 		ret = PTR_ERR(ctrl);
1362 		goto out_module_put;
1363 	}
1364 
1365 	module_put(ops->module);
1366 	return ctrl;
1367 
1368 out_module_put:
1369 	module_put(ops->module);
1370 	goto out_free_opts;
1371 out_unlock:
1372 	up_read(&nvmf_transports_rwsem);
1373 out_free_opts:
1374 	nvmf_free_options(opts);
1375 	return ERR_PTR(ret);
1376 }
1377 
1378 static const struct class nvmf_class = {
1379 	.name = "nvme-fabrics",
1380 };
1381 
1382 static struct device *nvmf_device;
1383 static DEFINE_MUTEX(nvmf_dev_mutex);
1384 
nvmf_dev_write(struct file * file,const char __user * ubuf,size_t count,loff_t * pos)1385 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1386 		size_t count, loff_t *pos)
1387 {
1388 	struct seq_file *seq_file = file->private_data;
1389 	struct nvme_ctrl *ctrl;
1390 	const char *buf;
1391 	int ret = 0;
1392 
1393 	if (count > PAGE_SIZE)
1394 		return -ENOMEM;
1395 
1396 	buf = memdup_user_nul(ubuf, count);
1397 	if (IS_ERR(buf))
1398 		return PTR_ERR(buf);
1399 
1400 	mutex_lock(&nvmf_dev_mutex);
1401 	if (seq_file->private) {
1402 		ret = -EINVAL;
1403 		goto out_unlock;
1404 	}
1405 
1406 	ctrl = nvmf_create_ctrl(nvmf_device, buf);
1407 	if (IS_ERR(ctrl)) {
1408 		ret = PTR_ERR(ctrl);
1409 		goto out_unlock;
1410 	}
1411 
1412 	seq_file->private = ctrl;
1413 
1414 out_unlock:
1415 	mutex_unlock(&nvmf_dev_mutex);
1416 	kfree(buf);
1417 	return ret ? ret : count;
1418 }
1419 
__nvmf_concat_opt_tokens(struct seq_file * seq_file)1420 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file)
1421 {
1422 	const struct match_token *tok;
1423 	int idx;
1424 
1425 	/*
1426 	 * Add dummy entries for instance and cntlid to
1427 	 * signal an invalid/non-existing controller
1428 	 */
1429 	seq_puts(seq_file, "instance=-1,cntlid=-1");
1430 	for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) {
1431 		tok = &opt_tokens[idx];
1432 		if (tok->token == NVMF_OPT_ERR)
1433 			continue;
1434 		seq_putc(seq_file, ',');
1435 		seq_puts(seq_file, tok->pattern);
1436 	}
1437 	seq_putc(seq_file, '\n');
1438 }
1439 
nvmf_dev_show(struct seq_file * seq_file,void * private)1440 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1441 {
1442 	struct nvme_ctrl *ctrl;
1443 
1444 	mutex_lock(&nvmf_dev_mutex);
1445 	ctrl = seq_file->private;
1446 	if (!ctrl) {
1447 		__nvmf_concat_opt_tokens(seq_file);
1448 		goto out_unlock;
1449 	}
1450 
1451 	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1452 			ctrl->instance, ctrl->cntlid);
1453 
1454 out_unlock:
1455 	mutex_unlock(&nvmf_dev_mutex);
1456 	return 0;
1457 }
1458 
nvmf_dev_open(struct inode * inode,struct file * file)1459 static int nvmf_dev_open(struct inode *inode, struct file *file)
1460 {
1461 	/*
1462 	 * The miscdevice code initializes file->private_data, but doesn't
1463 	 * make use of it later.
1464 	 */
1465 	file->private_data = NULL;
1466 	return single_open(file, nvmf_dev_show, NULL);
1467 }
1468 
nvmf_dev_release(struct inode * inode,struct file * file)1469 static int nvmf_dev_release(struct inode *inode, struct file *file)
1470 {
1471 	struct seq_file *seq_file = file->private_data;
1472 	struct nvme_ctrl *ctrl = seq_file->private;
1473 
1474 	if (ctrl)
1475 		nvme_put_ctrl(ctrl);
1476 	return single_release(inode, file);
1477 }
1478 
1479 static const struct file_operations nvmf_dev_fops = {
1480 	.owner		= THIS_MODULE,
1481 	.write		= nvmf_dev_write,
1482 	.read		= seq_read,
1483 	.open		= nvmf_dev_open,
1484 	.release	= nvmf_dev_release,
1485 };
1486 
1487 static struct miscdevice nvmf_misc = {
1488 	.minor		= MISC_DYNAMIC_MINOR,
1489 	.name           = "nvme-fabrics",
1490 	.fops		= &nvmf_dev_fops,
1491 };
1492 
nvmf_init(void)1493 static int __init nvmf_init(void)
1494 {
1495 	int ret;
1496 
1497 	nvmf_default_host = nvmf_host_default();
1498 	if (!nvmf_default_host)
1499 		return -ENOMEM;
1500 
1501 	ret = class_register(&nvmf_class);
1502 	if (ret) {
1503 		pr_err("couldn't register class nvme-fabrics\n");
1504 		goto out_free_host;
1505 	}
1506 
1507 	nvmf_device =
1508 		device_create(&nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1509 	if (IS_ERR(nvmf_device)) {
1510 		pr_err("couldn't create nvme-fabrics device!\n");
1511 		ret = PTR_ERR(nvmf_device);
1512 		goto out_destroy_class;
1513 	}
1514 
1515 	ret = misc_register(&nvmf_misc);
1516 	if (ret) {
1517 		pr_err("couldn't register misc device: %d\n", ret);
1518 		goto out_destroy_device;
1519 	}
1520 
1521 	return 0;
1522 
1523 out_destroy_device:
1524 	device_destroy(&nvmf_class, MKDEV(0, 0));
1525 out_destroy_class:
1526 	class_unregister(&nvmf_class);
1527 out_free_host:
1528 	nvmf_host_put(nvmf_default_host);
1529 	return ret;
1530 }
1531 
nvmf_exit(void)1532 static void __exit nvmf_exit(void)
1533 {
1534 	misc_deregister(&nvmf_misc);
1535 	device_destroy(&nvmf_class, MKDEV(0, 0));
1536 	class_unregister(&nvmf_class);
1537 	nvmf_host_put(nvmf_default_host);
1538 
1539 	BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1540 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1541 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1542 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1543 	BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64);
1544 	BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64);
1545 	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1546 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8);
1547 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16);
1548 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16);
1549 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16);
1550 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16);
1551 }
1552 
1553 MODULE_LICENSE("GPL v2");
1554 MODULE_DESCRIPTION("NVMe host fabrics library");
1555 
1556 module_init(nvmf_init);
1557 module_exit(nvmf_exit);
1558