xref: /linux/drivers/nvme/target/core.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
1 /*
2  * Common code for the NVMe target.
3  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  */
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/module.h>
16 #include <linux/random.h>
17 #include "nvmet.h"
18 
19 static struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX];
20 
21 /*
22  * This read/write semaphore is used to synchronize access to configuration
23  * information on a target system that will result in discovery log page
24  * information change for at least one host.
25  * The full list of resources to protected by this semaphore is:
26  *
27  *  - subsystems list
28  *  - per-subsystem allowed hosts list
29  *  - allow_any_host subsystem attribute
30  *  - nvmet_genctr
31  *  - the nvmet_transports array
32  *
33  * When updating any of those lists/structures write lock should be obtained,
34  * while when reading (popolating discovery log page or checking host-subsystem
35  * link) read lock is obtained to allow concurrent reads.
36  */
37 DECLARE_RWSEM(nvmet_config_sem);
38 
39 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
40 		const char *subsysnqn);
41 
42 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
43 		size_t len)
44 {
45 	if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len)
46 		return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
47 	return 0;
48 }
49 
50 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len)
51 {
52 	if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len)
53 		return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
54 	return 0;
55 }
56 
57 static u32 nvmet_async_event_result(struct nvmet_async_event *aen)
58 {
59 	return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16);
60 }
61 
62 static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
63 {
64 	struct nvmet_req *req;
65 
66 	while (1) {
67 		mutex_lock(&ctrl->lock);
68 		if (!ctrl->nr_async_event_cmds) {
69 			mutex_unlock(&ctrl->lock);
70 			return;
71 		}
72 
73 		req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
74 		mutex_unlock(&ctrl->lock);
75 		nvmet_req_complete(req, NVME_SC_INTERNAL | NVME_SC_DNR);
76 	}
77 }
78 
79 static void nvmet_async_event_work(struct work_struct *work)
80 {
81 	struct nvmet_ctrl *ctrl =
82 		container_of(work, struct nvmet_ctrl, async_event_work);
83 	struct nvmet_async_event *aen;
84 	struct nvmet_req *req;
85 
86 	while (1) {
87 		mutex_lock(&ctrl->lock);
88 		aen = list_first_entry_or_null(&ctrl->async_events,
89 				struct nvmet_async_event, entry);
90 		if (!aen || !ctrl->nr_async_event_cmds) {
91 			mutex_unlock(&ctrl->lock);
92 			return;
93 		}
94 
95 		req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
96 		nvmet_set_result(req, nvmet_async_event_result(aen));
97 
98 		list_del(&aen->entry);
99 		kfree(aen);
100 
101 		mutex_unlock(&ctrl->lock);
102 		nvmet_req_complete(req, 0);
103 	}
104 }
105 
106 static void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
107 		u8 event_info, u8 log_page)
108 {
109 	struct nvmet_async_event *aen;
110 
111 	aen = kmalloc(sizeof(*aen), GFP_KERNEL);
112 	if (!aen)
113 		return;
114 
115 	aen->event_type = event_type;
116 	aen->event_info = event_info;
117 	aen->log_page = log_page;
118 
119 	mutex_lock(&ctrl->lock);
120 	list_add_tail(&aen->entry, &ctrl->async_events);
121 	mutex_unlock(&ctrl->lock);
122 
123 	schedule_work(&ctrl->async_event_work);
124 }
125 
126 int nvmet_register_transport(struct nvmet_fabrics_ops *ops)
127 {
128 	int ret = 0;
129 
130 	down_write(&nvmet_config_sem);
131 	if (nvmet_transports[ops->type])
132 		ret = -EINVAL;
133 	else
134 		nvmet_transports[ops->type] = ops;
135 	up_write(&nvmet_config_sem);
136 
137 	return ret;
138 }
139 EXPORT_SYMBOL_GPL(nvmet_register_transport);
140 
141 void nvmet_unregister_transport(struct nvmet_fabrics_ops *ops)
142 {
143 	down_write(&nvmet_config_sem);
144 	nvmet_transports[ops->type] = NULL;
145 	up_write(&nvmet_config_sem);
146 }
147 EXPORT_SYMBOL_GPL(nvmet_unregister_transport);
148 
149 int nvmet_enable_port(struct nvmet_port *port)
150 {
151 	struct nvmet_fabrics_ops *ops;
152 	int ret;
153 
154 	lockdep_assert_held(&nvmet_config_sem);
155 
156 	ops = nvmet_transports[port->disc_addr.trtype];
157 	if (!ops) {
158 		up_write(&nvmet_config_sem);
159 		request_module("nvmet-transport-%d", port->disc_addr.trtype);
160 		down_write(&nvmet_config_sem);
161 		ops = nvmet_transports[port->disc_addr.trtype];
162 		if (!ops) {
163 			pr_err("transport type %d not supported\n",
164 				port->disc_addr.trtype);
165 			return -EINVAL;
166 		}
167 	}
168 
169 	if (!try_module_get(ops->owner))
170 		return -EINVAL;
171 
172 	ret = ops->add_port(port);
173 	if (ret) {
174 		module_put(ops->owner);
175 		return ret;
176 	}
177 
178 	port->enabled = true;
179 	return 0;
180 }
181 
182 void nvmet_disable_port(struct nvmet_port *port)
183 {
184 	struct nvmet_fabrics_ops *ops;
185 
186 	lockdep_assert_held(&nvmet_config_sem);
187 
188 	port->enabled = false;
189 
190 	ops = nvmet_transports[port->disc_addr.trtype];
191 	ops->remove_port(port);
192 	module_put(ops->owner);
193 }
194 
195 static void nvmet_keep_alive_timer(struct work_struct *work)
196 {
197 	struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
198 			struct nvmet_ctrl, ka_work);
199 
200 	pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
201 		ctrl->cntlid, ctrl->kato);
202 
203 	ctrl->ops->delete_ctrl(ctrl);
204 }
205 
206 static void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
207 {
208 	pr_debug("ctrl %d start keep-alive timer for %d secs\n",
209 		ctrl->cntlid, ctrl->kato);
210 
211 	INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer);
212 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
213 }
214 
215 static void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
216 {
217 	pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid);
218 
219 	cancel_delayed_work_sync(&ctrl->ka_work);
220 }
221 
222 static struct nvmet_ns *__nvmet_find_namespace(struct nvmet_ctrl *ctrl,
223 		__le32 nsid)
224 {
225 	struct nvmet_ns *ns;
226 
227 	list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
228 		if (ns->nsid == le32_to_cpu(nsid))
229 			return ns;
230 	}
231 
232 	return NULL;
233 }
234 
235 struct nvmet_ns *nvmet_find_namespace(struct nvmet_ctrl *ctrl, __le32 nsid)
236 {
237 	struct nvmet_ns *ns;
238 
239 	rcu_read_lock();
240 	ns = __nvmet_find_namespace(ctrl, nsid);
241 	if (ns)
242 		percpu_ref_get(&ns->ref);
243 	rcu_read_unlock();
244 
245 	return ns;
246 }
247 
248 static void nvmet_destroy_namespace(struct percpu_ref *ref)
249 {
250 	struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref);
251 
252 	complete(&ns->disable_done);
253 }
254 
255 void nvmet_put_namespace(struct nvmet_ns *ns)
256 {
257 	percpu_ref_put(&ns->ref);
258 }
259 
260 int nvmet_ns_enable(struct nvmet_ns *ns)
261 {
262 	struct nvmet_subsys *subsys = ns->subsys;
263 	struct nvmet_ctrl *ctrl;
264 	int ret = 0;
265 
266 	mutex_lock(&subsys->lock);
267 	if (!list_empty(&ns->dev_link))
268 		goto out_unlock;
269 
270 	ns->bdev = blkdev_get_by_path(ns->device_path, FMODE_READ | FMODE_WRITE,
271 			NULL);
272 	if (IS_ERR(ns->bdev)) {
273 		pr_err("nvmet: failed to open block device %s: (%ld)\n",
274 			ns->device_path, PTR_ERR(ns->bdev));
275 		ret = PTR_ERR(ns->bdev);
276 		ns->bdev = NULL;
277 		goto out_unlock;
278 	}
279 
280 	ns->size = i_size_read(ns->bdev->bd_inode);
281 	ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
282 
283 	ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace,
284 				0, GFP_KERNEL);
285 	if (ret)
286 		goto out_blkdev_put;
287 
288 	if (ns->nsid > subsys->max_nsid)
289 		subsys->max_nsid = ns->nsid;
290 
291 	/*
292 	 * The namespaces list needs to be sorted to simplify the implementation
293 	 * of the Identify Namepace List subcommand.
294 	 */
295 	if (list_empty(&subsys->namespaces)) {
296 		list_add_tail_rcu(&ns->dev_link, &subsys->namespaces);
297 	} else {
298 		struct nvmet_ns *old;
299 
300 		list_for_each_entry_rcu(old, &subsys->namespaces, dev_link) {
301 			BUG_ON(ns->nsid == old->nsid);
302 			if (ns->nsid < old->nsid)
303 				break;
304 		}
305 
306 		list_add_tail_rcu(&ns->dev_link, &old->dev_link);
307 	}
308 
309 	list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
310 		nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE, 0, 0);
311 
312 	ret = 0;
313 out_unlock:
314 	mutex_unlock(&subsys->lock);
315 	return ret;
316 out_blkdev_put:
317 	blkdev_put(ns->bdev, FMODE_WRITE|FMODE_READ);
318 	ns->bdev = NULL;
319 	goto out_unlock;
320 }
321 
322 void nvmet_ns_disable(struct nvmet_ns *ns)
323 {
324 	struct nvmet_subsys *subsys = ns->subsys;
325 	struct nvmet_ctrl *ctrl;
326 
327 	mutex_lock(&subsys->lock);
328 	if (list_empty(&ns->dev_link)) {
329 		mutex_unlock(&subsys->lock);
330 		return;
331 	}
332 	list_del_init(&ns->dev_link);
333 	mutex_unlock(&subsys->lock);
334 
335 	/*
336 	 * Now that we removed the namespaces from the lookup list, we
337 	 * can kill the per_cpu ref and wait for any remaining references
338 	 * to be dropped, as well as a RCU grace period for anyone only
339 	 * using the namepace under rcu_read_lock().  Note that we can't
340 	 * use call_rcu here as we need to ensure the namespaces have
341 	 * been fully destroyed before unloading the module.
342 	 */
343 	percpu_ref_kill(&ns->ref);
344 	synchronize_rcu();
345 	wait_for_completion(&ns->disable_done);
346 	percpu_ref_exit(&ns->ref);
347 
348 	mutex_lock(&subsys->lock);
349 	list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
350 		nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE, 0, 0);
351 
352 	if (ns->bdev)
353 		blkdev_put(ns->bdev, FMODE_WRITE|FMODE_READ);
354 	mutex_unlock(&subsys->lock);
355 }
356 
357 void nvmet_ns_free(struct nvmet_ns *ns)
358 {
359 	nvmet_ns_disable(ns);
360 
361 	kfree(ns->device_path);
362 	kfree(ns);
363 }
364 
365 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
366 {
367 	struct nvmet_ns *ns;
368 
369 	ns = kzalloc(sizeof(*ns), GFP_KERNEL);
370 	if (!ns)
371 		return NULL;
372 
373 	INIT_LIST_HEAD(&ns->dev_link);
374 	init_completion(&ns->disable_done);
375 
376 	ns->nsid = nsid;
377 	ns->subsys = subsys;
378 
379 	return ns;
380 }
381 
382 static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
383 {
384 	if (status)
385 		nvmet_set_status(req, status);
386 
387 	/* XXX: need to fill in something useful for sq_head */
388 	req->rsp->sq_head = 0;
389 	if (likely(req->sq)) /* may happen during early failure */
390 		req->rsp->sq_id = cpu_to_le16(req->sq->qid);
391 	req->rsp->command_id = req->cmd->common.command_id;
392 
393 	if (req->ns)
394 		nvmet_put_namespace(req->ns);
395 	req->ops->queue_response(req);
396 }
397 
398 void nvmet_req_complete(struct nvmet_req *req, u16 status)
399 {
400 	__nvmet_req_complete(req, status);
401 	percpu_ref_put(&req->sq->ref);
402 }
403 EXPORT_SYMBOL_GPL(nvmet_req_complete);
404 
405 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq,
406 		u16 qid, u16 size)
407 {
408 	cq->qid = qid;
409 	cq->size = size;
410 
411 	ctrl->cqs[qid] = cq;
412 }
413 
414 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq,
415 		u16 qid, u16 size)
416 {
417 	sq->qid = qid;
418 	sq->size = size;
419 
420 	ctrl->sqs[qid] = sq;
421 }
422 
423 void nvmet_sq_destroy(struct nvmet_sq *sq)
424 {
425 	/*
426 	 * If this is the admin queue, complete all AERs so that our
427 	 * queue doesn't have outstanding requests on it.
428 	 */
429 	if (sq->ctrl && sq->ctrl->sqs && sq->ctrl->sqs[0] == sq)
430 		nvmet_async_events_free(sq->ctrl);
431 	percpu_ref_kill(&sq->ref);
432 	wait_for_completion(&sq->free_done);
433 	percpu_ref_exit(&sq->ref);
434 
435 	if (sq->ctrl) {
436 		nvmet_ctrl_put(sq->ctrl);
437 		sq->ctrl = NULL; /* allows reusing the queue later */
438 	}
439 }
440 EXPORT_SYMBOL_GPL(nvmet_sq_destroy);
441 
442 static void nvmet_sq_free(struct percpu_ref *ref)
443 {
444 	struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
445 
446 	complete(&sq->free_done);
447 }
448 
449 int nvmet_sq_init(struct nvmet_sq *sq)
450 {
451 	int ret;
452 
453 	ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL);
454 	if (ret) {
455 		pr_err("percpu_ref init failed!\n");
456 		return ret;
457 	}
458 	init_completion(&sq->free_done);
459 
460 	return 0;
461 }
462 EXPORT_SYMBOL_GPL(nvmet_sq_init);
463 
464 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
465 		struct nvmet_sq *sq, struct nvmet_fabrics_ops *ops)
466 {
467 	u8 flags = req->cmd->common.flags;
468 	u16 status;
469 
470 	req->cq = cq;
471 	req->sq = sq;
472 	req->ops = ops;
473 	req->sg = NULL;
474 	req->sg_cnt = 0;
475 	req->rsp->status = 0;
476 
477 	/* no support for fused commands yet */
478 	if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) {
479 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
480 		goto fail;
481 	}
482 
483 	/* either variant of SGLs is fine, as we don't support metadata */
484 	if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF &&
485 		     (flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METASEG)) {
486 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
487 		goto fail;
488 	}
489 
490 	if (unlikely(!req->sq->ctrl))
491 		/* will return an error for any Non-connect command: */
492 		status = nvmet_parse_connect_cmd(req);
493 	else if (likely(req->sq->qid != 0))
494 		status = nvmet_parse_io_cmd(req);
495 	else if (req->cmd->common.opcode == nvme_fabrics_command)
496 		status = nvmet_parse_fabrics_cmd(req);
497 	else if (req->sq->ctrl->subsys->type == NVME_NQN_DISC)
498 		status = nvmet_parse_discovery_cmd(req);
499 	else
500 		status = nvmet_parse_admin_cmd(req);
501 
502 	if (status)
503 		goto fail;
504 
505 	if (unlikely(!percpu_ref_tryget_live(&sq->ref))) {
506 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
507 		goto fail;
508 	}
509 
510 	return true;
511 
512 fail:
513 	__nvmet_req_complete(req, status);
514 	return false;
515 }
516 EXPORT_SYMBOL_GPL(nvmet_req_init);
517 
518 static inline bool nvmet_cc_en(u32 cc)
519 {
520 	return cc & 0x1;
521 }
522 
523 static inline u8 nvmet_cc_css(u32 cc)
524 {
525 	return (cc >> 4) & 0x7;
526 }
527 
528 static inline u8 nvmet_cc_mps(u32 cc)
529 {
530 	return (cc >> 7) & 0xf;
531 }
532 
533 static inline u8 nvmet_cc_ams(u32 cc)
534 {
535 	return (cc >> 11) & 0x7;
536 }
537 
538 static inline u8 nvmet_cc_shn(u32 cc)
539 {
540 	return (cc >> 14) & 0x3;
541 }
542 
543 static inline u8 nvmet_cc_iosqes(u32 cc)
544 {
545 	return (cc >> 16) & 0xf;
546 }
547 
548 static inline u8 nvmet_cc_iocqes(u32 cc)
549 {
550 	return (cc >> 20) & 0xf;
551 }
552 
553 static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl)
554 {
555 	lockdep_assert_held(&ctrl->lock);
556 
557 	if (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES ||
558 	    nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES ||
559 	    nvmet_cc_mps(ctrl->cc) != 0 ||
560 	    nvmet_cc_ams(ctrl->cc) != 0 ||
561 	    nvmet_cc_css(ctrl->cc) != 0) {
562 		ctrl->csts = NVME_CSTS_CFS;
563 		return;
564 	}
565 
566 	ctrl->csts = NVME_CSTS_RDY;
567 }
568 
569 static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl)
570 {
571 	lockdep_assert_held(&ctrl->lock);
572 
573 	/* XXX: tear down queues? */
574 	ctrl->csts &= ~NVME_CSTS_RDY;
575 	ctrl->cc = 0;
576 }
577 
578 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new)
579 {
580 	u32 old;
581 
582 	mutex_lock(&ctrl->lock);
583 	old = ctrl->cc;
584 	ctrl->cc = new;
585 
586 	if (nvmet_cc_en(new) && !nvmet_cc_en(old))
587 		nvmet_start_ctrl(ctrl);
588 	if (!nvmet_cc_en(new) && nvmet_cc_en(old))
589 		nvmet_clear_ctrl(ctrl);
590 	if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) {
591 		nvmet_clear_ctrl(ctrl);
592 		ctrl->csts |= NVME_CSTS_SHST_CMPLT;
593 	}
594 	if (!nvmet_cc_shn(new) && nvmet_cc_shn(old))
595 		ctrl->csts &= ~NVME_CSTS_SHST_CMPLT;
596 	mutex_unlock(&ctrl->lock);
597 }
598 
599 static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
600 {
601 	/* command sets supported: NVMe command set: */
602 	ctrl->cap = (1ULL << 37);
603 	/* CC.EN timeout in 500msec units: */
604 	ctrl->cap |= (15ULL << 24);
605 	/* maximum queue entries supported: */
606 	ctrl->cap |= NVMET_QUEUE_SIZE - 1;
607 }
608 
609 u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid,
610 		struct nvmet_req *req, struct nvmet_ctrl **ret)
611 {
612 	struct nvmet_subsys *subsys;
613 	struct nvmet_ctrl *ctrl;
614 	u16 status = 0;
615 
616 	subsys = nvmet_find_get_subsys(req->port, subsysnqn);
617 	if (!subsys) {
618 		pr_warn("connect request for invalid subsystem %s!\n",
619 			subsysnqn);
620 		req->rsp->result = IPO_IATTR_CONNECT_DATA(subsysnqn);
621 		return NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
622 	}
623 
624 	mutex_lock(&subsys->lock);
625 	list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
626 		if (ctrl->cntlid == cntlid) {
627 			if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) {
628 				pr_warn("hostnqn mismatch.\n");
629 				continue;
630 			}
631 			if (!kref_get_unless_zero(&ctrl->ref))
632 				continue;
633 
634 			*ret = ctrl;
635 			goto out;
636 		}
637 	}
638 
639 	pr_warn("could not find controller %d for subsys %s / host %s\n",
640 		cntlid, subsysnqn, hostnqn);
641 	req->rsp->result = IPO_IATTR_CONNECT_DATA(cntlid);
642 	status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
643 
644 out:
645 	mutex_unlock(&subsys->lock);
646 	nvmet_subsys_put(subsys);
647 	return status;
648 }
649 
650 static bool __nvmet_host_allowed(struct nvmet_subsys *subsys,
651 		const char *hostnqn)
652 {
653 	struct nvmet_host_link *p;
654 
655 	if (subsys->allow_any_host)
656 		return true;
657 
658 	list_for_each_entry(p, &subsys->hosts, entry) {
659 		if (!strcmp(nvmet_host_name(p->host), hostnqn))
660 			return true;
661 	}
662 
663 	return false;
664 }
665 
666 static bool nvmet_host_discovery_allowed(struct nvmet_req *req,
667 		const char *hostnqn)
668 {
669 	struct nvmet_subsys_link *s;
670 
671 	list_for_each_entry(s, &req->port->subsystems, entry) {
672 		if (__nvmet_host_allowed(s->subsys, hostnqn))
673 			return true;
674 	}
675 
676 	return false;
677 }
678 
679 bool nvmet_host_allowed(struct nvmet_req *req, struct nvmet_subsys *subsys,
680 		const char *hostnqn)
681 {
682 	lockdep_assert_held(&nvmet_config_sem);
683 
684 	if (subsys->type == NVME_NQN_DISC)
685 		return nvmet_host_discovery_allowed(req, hostnqn);
686 	else
687 		return __nvmet_host_allowed(subsys, hostnqn);
688 }
689 
690 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
691 		struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp)
692 {
693 	struct nvmet_subsys *subsys;
694 	struct nvmet_ctrl *ctrl;
695 	int ret;
696 	u16 status;
697 
698 	status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
699 	subsys = nvmet_find_get_subsys(req->port, subsysnqn);
700 	if (!subsys) {
701 		pr_warn("connect request for invalid subsystem %s!\n",
702 			subsysnqn);
703 		req->rsp->result = IPO_IATTR_CONNECT_DATA(subsysnqn);
704 		goto out;
705 	}
706 
707 	status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
708 	down_read(&nvmet_config_sem);
709 	if (!nvmet_host_allowed(req, subsys, hostnqn)) {
710 		pr_info("connect by host %s for subsystem %s not allowed\n",
711 			hostnqn, subsysnqn);
712 		req->rsp->result = IPO_IATTR_CONNECT_DATA(hostnqn);
713 		up_read(&nvmet_config_sem);
714 		goto out_put_subsystem;
715 	}
716 	up_read(&nvmet_config_sem);
717 
718 	status = NVME_SC_INTERNAL;
719 	ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
720 	if (!ctrl)
721 		goto out_put_subsystem;
722 	mutex_init(&ctrl->lock);
723 
724 	nvmet_init_cap(ctrl);
725 
726 	INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work);
727 	INIT_LIST_HEAD(&ctrl->async_events);
728 
729 	memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE);
730 	memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE);
731 
732 	/* generate a random serial number as our controllers are ephemeral: */
733 	get_random_bytes(&ctrl->serial, sizeof(ctrl->serial));
734 
735 	kref_init(&ctrl->ref);
736 	ctrl->subsys = subsys;
737 
738 	ctrl->cqs = kcalloc(subsys->max_qid + 1,
739 			sizeof(struct nvmet_cq *),
740 			GFP_KERNEL);
741 	if (!ctrl->cqs)
742 		goto out_free_ctrl;
743 
744 	ctrl->sqs = kcalloc(subsys->max_qid + 1,
745 			sizeof(struct nvmet_sq *),
746 			GFP_KERNEL);
747 	if (!ctrl->sqs)
748 		goto out_free_cqs;
749 
750 	ret = ida_simple_get(&subsys->cntlid_ida,
751 			     NVME_CNTLID_MIN, NVME_CNTLID_MAX,
752 			     GFP_KERNEL);
753 	if (ret < 0) {
754 		status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR;
755 		goto out_free_sqs;
756 	}
757 	ctrl->cntlid = ret;
758 
759 	ctrl->ops = req->ops;
760 	if (ctrl->subsys->type == NVME_NQN_DISC) {
761 		/* Don't accept keep-alive timeout for discovery controllers */
762 		if (kato) {
763 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
764 			goto out_free_sqs;
765 		}
766 
767 		/*
768 		 * Discovery controllers use some arbitrary high value in order
769 		 * to cleanup stale discovery sessions
770 		 *
771 		 * From the latest base diff RC:
772 		 * "The Keep Alive command is not supported by
773 		 * Discovery controllers. A transport may specify a
774 		 * fixed Discovery controller activity timeout value
775 		 * (e.g., 2 minutes).  If no commands are received
776 		 * by a Discovery controller within that time
777 		 * period, the controller may perform the
778 		 * actions for Keep Alive Timer expiration".
779 		 */
780 		ctrl->kato = NVMET_DISC_KATO;
781 	} else {
782 		/* keep-alive timeout in seconds */
783 		ctrl->kato = DIV_ROUND_UP(kato, 1000);
784 	}
785 	nvmet_start_keep_alive_timer(ctrl);
786 
787 	mutex_lock(&subsys->lock);
788 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
789 	mutex_unlock(&subsys->lock);
790 
791 	*ctrlp = ctrl;
792 	return 0;
793 
794 out_free_sqs:
795 	kfree(ctrl->sqs);
796 out_free_cqs:
797 	kfree(ctrl->cqs);
798 out_free_ctrl:
799 	kfree(ctrl);
800 out_put_subsystem:
801 	nvmet_subsys_put(subsys);
802 out:
803 	return status;
804 }
805 
806 static void nvmet_ctrl_free(struct kref *ref)
807 {
808 	struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref);
809 	struct nvmet_subsys *subsys = ctrl->subsys;
810 
811 	nvmet_stop_keep_alive_timer(ctrl);
812 
813 	mutex_lock(&subsys->lock);
814 	list_del(&ctrl->subsys_entry);
815 	mutex_unlock(&subsys->lock);
816 
817 	ida_simple_remove(&subsys->cntlid_ida, ctrl->cntlid);
818 	nvmet_subsys_put(subsys);
819 
820 	kfree(ctrl->sqs);
821 	kfree(ctrl->cqs);
822 	kfree(ctrl);
823 }
824 
825 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl)
826 {
827 	kref_put(&ctrl->ref, nvmet_ctrl_free);
828 }
829 
830 static void nvmet_fatal_error_handler(struct work_struct *work)
831 {
832 	struct nvmet_ctrl *ctrl =
833 			container_of(work, struct nvmet_ctrl, fatal_err_work);
834 
835 	pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid);
836 	ctrl->ops->delete_ctrl(ctrl);
837 }
838 
839 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl)
840 {
841 	ctrl->csts |= NVME_CSTS_CFS;
842 	INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler);
843 	schedule_work(&ctrl->fatal_err_work);
844 }
845 EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error);
846 
847 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
848 		const char *subsysnqn)
849 {
850 	struct nvmet_subsys_link *p;
851 
852 	if (!port)
853 		return NULL;
854 
855 	if (!strncmp(NVME_DISC_SUBSYS_NAME, subsysnqn,
856 			NVMF_NQN_SIZE)) {
857 		if (!kref_get_unless_zero(&nvmet_disc_subsys->ref))
858 			return NULL;
859 		return nvmet_disc_subsys;
860 	}
861 
862 	down_read(&nvmet_config_sem);
863 	list_for_each_entry(p, &port->subsystems, entry) {
864 		if (!strncmp(p->subsys->subsysnqn, subsysnqn,
865 				NVMF_NQN_SIZE)) {
866 			if (!kref_get_unless_zero(&p->subsys->ref))
867 				break;
868 			up_read(&nvmet_config_sem);
869 			return p->subsys;
870 		}
871 	}
872 	up_read(&nvmet_config_sem);
873 	return NULL;
874 }
875 
876 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
877 		enum nvme_subsys_type type)
878 {
879 	struct nvmet_subsys *subsys;
880 
881 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
882 	if (!subsys)
883 		return NULL;
884 
885 	subsys->ver = NVME_VS(1, 2, 1); /* NVMe 1.2.1 */
886 
887 	switch (type) {
888 	case NVME_NQN_NVME:
889 		subsys->max_qid = NVMET_NR_QUEUES;
890 		break;
891 	case NVME_NQN_DISC:
892 		subsys->max_qid = 0;
893 		break;
894 	default:
895 		pr_err("%s: Unknown Subsystem type - %d\n", __func__, type);
896 		kfree(subsys);
897 		return NULL;
898 	}
899 	subsys->type = type;
900 	subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE,
901 			GFP_KERNEL);
902 	if (!subsys->subsysnqn) {
903 		kfree(subsys);
904 		return NULL;
905 	}
906 
907 	kref_init(&subsys->ref);
908 
909 	mutex_init(&subsys->lock);
910 	INIT_LIST_HEAD(&subsys->namespaces);
911 	INIT_LIST_HEAD(&subsys->ctrls);
912 
913 	ida_init(&subsys->cntlid_ida);
914 
915 	INIT_LIST_HEAD(&subsys->hosts);
916 
917 	return subsys;
918 }
919 
920 static void nvmet_subsys_free(struct kref *ref)
921 {
922 	struct nvmet_subsys *subsys =
923 		container_of(ref, struct nvmet_subsys, ref);
924 
925 	WARN_ON_ONCE(!list_empty(&subsys->namespaces));
926 
927 	ida_destroy(&subsys->cntlid_ida);
928 	kfree(subsys->subsysnqn);
929 	kfree(subsys);
930 }
931 
932 void nvmet_subsys_put(struct nvmet_subsys *subsys)
933 {
934 	kref_put(&subsys->ref, nvmet_subsys_free);
935 }
936 
937 static int __init nvmet_init(void)
938 {
939 	int error;
940 
941 	error = nvmet_init_discovery();
942 	if (error)
943 		goto out;
944 
945 	error = nvmet_init_configfs();
946 	if (error)
947 		goto out_exit_discovery;
948 	return 0;
949 
950 out_exit_discovery:
951 	nvmet_exit_discovery();
952 out:
953 	return error;
954 }
955 
956 static void __exit nvmet_exit(void)
957 {
958 	nvmet_exit_configfs();
959 	nvmet_exit_discovery();
960 
961 	BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024);
962 	BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024);
963 }
964 
965 module_init(nvmet_init);
966 module_exit(nvmet_exit);
967 
968 MODULE_LICENSE("GPL v2");
969