1// SPDX-License-Identifier: GPL-2.0
2/*
3 * NVMe over Fabrics loopback device.
4 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
5 */
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7#include <linux/scatterlist.h>
8#include <linux/blk-mq.h>
9#include <linux/nvme.h>
10#include <linux/module.h>
11#include <linux/parser.h>
12#include "nvmet.h"
13#include "../host/nvme.h"
14#include "../host/fabrics.h"
15
16#define NVME_LOOP_MAX_SEGMENTS		256
17
18struct nvme_loop_iod {
19	struct nvme_request	nvme_req;
20	struct nvme_command	cmd;
21	struct nvme_completion	cqe;
22	struct nvmet_req	req;
23	struct nvme_loop_queue	*queue;
24	struct work_struct	work;
25	struct sg_table		sg_table;
26	struct scatterlist	first_sgl[];
27};
28
29struct nvme_loop_ctrl {
30	struct nvme_loop_queue	*queues;
31
32	struct blk_mq_tag_set	admin_tag_set;
33
34	struct list_head	list;
35	struct blk_mq_tag_set	tag_set;
36	struct nvme_loop_iod	async_event_iod;
37	struct nvme_ctrl	ctrl;
38
39	struct nvmet_port	*port;
40};
41
42static inline struct nvme_loop_ctrl *to_loop_ctrl(struct nvme_ctrl *ctrl)
43{
44	return container_of(ctrl, struct nvme_loop_ctrl, ctrl);
45}
46
47enum nvme_loop_queue_flags {
48	NVME_LOOP_Q_LIVE	= 0,
49};
50
51struct nvme_loop_queue {
52	struct nvmet_cq		nvme_cq;
53	struct nvmet_sq		nvme_sq;
54	struct nvme_loop_ctrl	*ctrl;
55	unsigned long		flags;
56};
57
58static LIST_HEAD(nvme_loop_ports);
59static DEFINE_MUTEX(nvme_loop_ports_mutex);
60
61static LIST_HEAD(nvme_loop_ctrl_list);
62static DEFINE_MUTEX(nvme_loop_ctrl_mutex);
63
64static void nvme_loop_queue_response(struct nvmet_req *nvme_req);
65static void nvme_loop_delete_ctrl(struct nvmet_ctrl *ctrl);
66
67static const struct nvmet_fabrics_ops nvme_loop_ops;
68
69static inline int nvme_loop_queue_idx(struct nvme_loop_queue *queue)
70{
71	return queue - queue->ctrl->queues;
72}
73
74static void nvme_loop_complete_rq(struct request *req)
75{
76	struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req);
77
78	sg_free_table_chained(&iod->sg_table, NVME_INLINE_SG_CNT);
79	nvme_complete_rq(req);
80}
81
82static struct blk_mq_tags *nvme_loop_tagset(struct nvme_loop_queue *queue)
83{
84	u32 queue_idx = nvme_loop_queue_idx(queue);
85
86	if (queue_idx == 0)
87		return queue->ctrl->admin_tag_set.tags[queue_idx];
88	return queue->ctrl->tag_set.tags[queue_idx - 1];
89}
90
91static void nvme_loop_queue_response(struct nvmet_req *req)
92{
93	struct nvme_loop_queue *queue =
94		container_of(req->sq, struct nvme_loop_queue, nvme_sq);
95	struct nvme_completion *cqe = req->cqe;
96
97	/*
98	 * AEN requests are special as they don't time out and can
99	 * survive any kind of queue freeze and often don't respond to
100	 * aborts.  We don't even bother to allocate a struct request
101	 * for them but rather special case them here.
102	 */
103	if (unlikely(nvme_is_aen_req(nvme_loop_queue_idx(queue),
104				     cqe->command_id))) {
105		nvme_complete_async_event(&queue->ctrl->ctrl, cqe->status,
106				&cqe->result);
107	} else {
108		struct request *rq;
109
110		rq = nvme_find_rq(nvme_loop_tagset(queue), cqe->command_id);
111		if (!rq) {
112			dev_err(queue->ctrl->ctrl.device,
113				"got bad command_id %#x on queue %d\n",
114				cqe->command_id, nvme_loop_queue_idx(queue));
115			return;
116		}
117
118		if (!nvme_try_complete_req(rq, cqe->status, cqe->result))
119			nvme_loop_complete_rq(rq);
120	}
121}
122
123static void nvme_loop_execute_work(struct work_struct *work)
124{
125	struct nvme_loop_iod *iod =
126		container_of(work, struct nvme_loop_iod, work);
127
128	iod->req.execute(&iod->req);
129}
130
131static blk_status_t nvme_loop_queue_rq(struct blk_mq_hw_ctx *hctx,
132		const struct blk_mq_queue_data *bd)
133{
134	struct nvme_ns *ns = hctx->queue->queuedata;
135	struct nvme_loop_queue *queue = hctx->driver_data;
136	struct request *req = bd->rq;
137	struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req);
138	bool queue_ready = test_bit(NVME_LOOP_Q_LIVE, &queue->flags);
139	blk_status_t ret;
140
141	if (!nvmf_check_ready(&queue->ctrl->ctrl, req, queue_ready))
142		return nvmf_fail_nonready_command(&queue->ctrl->ctrl, req);
143
144	ret = nvme_setup_cmd(ns, req, &iod->cmd);
145	if (ret)
146		return ret;
147
148	blk_mq_start_request(req);
149	iod->cmd.common.flags |= NVME_CMD_SGL_METABUF;
150	iod->req.port = queue->ctrl->port;
151	if (!nvmet_req_init(&iod->req, &queue->nvme_cq,
152			&queue->nvme_sq, &nvme_loop_ops))
153		return BLK_STS_OK;
154
155	if (blk_rq_nr_phys_segments(req)) {
156		iod->sg_table.sgl = iod->first_sgl;
157		if (sg_alloc_table_chained(&iod->sg_table,
158				blk_rq_nr_phys_segments(req),
159				iod->sg_table.sgl, NVME_INLINE_SG_CNT)) {
160			nvme_cleanup_cmd(req);
161			return BLK_STS_RESOURCE;
162		}
163
164		iod->req.sg = iod->sg_table.sgl;
165		iod->req.sg_cnt = blk_rq_map_sg(req->q, req, iod->sg_table.sgl);
166		iod->req.transfer_len = blk_rq_payload_bytes(req);
167	}
168
169	schedule_work(&iod->work);
170	return BLK_STS_OK;
171}
172
173static void nvme_loop_submit_async_event(struct nvme_ctrl *arg)
174{
175	struct nvme_loop_ctrl *ctrl = to_loop_ctrl(arg);
176	struct nvme_loop_queue *queue = &ctrl->queues[0];
177	struct nvme_loop_iod *iod = &ctrl->async_event_iod;
178
179	memset(&iod->cmd, 0, sizeof(iod->cmd));
180	iod->cmd.common.opcode = nvme_admin_async_event;
181	iod->cmd.common.command_id = NVME_AQ_BLK_MQ_DEPTH;
182	iod->cmd.common.flags |= NVME_CMD_SGL_METABUF;
183
184	if (!nvmet_req_init(&iod->req, &queue->nvme_cq, &queue->nvme_sq,
185			&nvme_loop_ops)) {
186		dev_err(ctrl->ctrl.device, "failed async event work\n");
187		return;
188	}
189
190	schedule_work(&iod->work);
191}
192
193static int nvme_loop_init_iod(struct nvme_loop_ctrl *ctrl,
194		struct nvme_loop_iod *iod, unsigned int queue_idx)
195{
196	iod->req.cmd = &iod->cmd;
197	iod->req.cqe = &iod->cqe;
198	iod->queue = &ctrl->queues[queue_idx];
199	INIT_WORK(&iod->work, nvme_loop_execute_work);
200	return 0;
201}
202
203static int nvme_loop_init_request(struct blk_mq_tag_set *set,
204		struct request *req, unsigned int hctx_idx,
205		unsigned int numa_node)
206{
207	struct nvme_loop_ctrl *ctrl = set->driver_data;
208
209	nvme_req(req)->ctrl = &ctrl->ctrl;
210	return nvme_loop_init_iod(ctrl, blk_mq_rq_to_pdu(req),
211			(set == &ctrl->tag_set) ? hctx_idx + 1 : 0);
212}
213
214static struct lock_class_key loop_hctx_fq_lock_key;
215
216static int nvme_loop_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
217		unsigned int hctx_idx)
218{
219	struct nvme_loop_ctrl *ctrl = data;
220	struct nvme_loop_queue *queue = &ctrl->queues[hctx_idx + 1];
221
222	BUG_ON(hctx_idx >= ctrl->ctrl.queue_count);
223
224	/*
225	 * flush_end_io() can be called recursively for us, so use our own
226	 * lock class key for avoiding lockdep possible recursive locking,
227	 * then we can remove the dynamically allocated lock class for each
228	 * flush queue, that way may cause horrible boot delay.
229	 */
230	blk_mq_hctx_set_fq_lock_class(hctx, &loop_hctx_fq_lock_key);
231
232	hctx->driver_data = queue;
233	return 0;
234}
235
236static int nvme_loop_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data,
237		unsigned int hctx_idx)
238{
239	struct nvme_loop_ctrl *ctrl = data;
240	struct nvme_loop_queue *queue = &ctrl->queues[0];
241
242	BUG_ON(hctx_idx != 0);
243
244	hctx->driver_data = queue;
245	return 0;
246}
247
248static const struct blk_mq_ops nvme_loop_mq_ops = {
249	.queue_rq	= nvme_loop_queue_rq,
250	.complete	= nvme_loop_complete_rq,
251	.init_request	= nvme_loop_init_request,
252	.init_hctx	= nvme_loop_init_hctx,
253};
254
255static const struct blk_mq_ops nvme_loop_admin_mq_ops = {
256	.queue_rq	= nvme_loop_queue_rq,
257	.complete	= nvme_loop_complete_rq,
258	.init_request	= nvme_loop_init_request,
259	.init_hctx	= nvme_loop_init_admin_hctx,
260};
261
262static void nvme_loop_destroy_admin_queue(struct nvme_loop_ctrl *ctrl)
263{
264	if (!test_and_clear_bit(NVME_LOOP_Q_LIVE, &ctrl->queues[0].flags))
265		return;
266	nvmet_sq_destroy(&ctrl->queues[0].nvme_sq);
267	blk_cleanup_queue(ctrl->ctrl.admin_q);
268	blk_cleanup_queue(ctrl->ctrl.fabrics_q);
269	blk_mq_free_tag_set(&ctrl->admin_tag_set);
270}
271
272static void nvme_loop_free_ctrl(struct nvme_ctrl *nctrl)
273{
274	struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl);
275
276	if (list_empty(&ctrl->list))
277		goto free_ctrl;
278
279	mutex_lock(&nvme_loop_ctrl_mutex);
280	list_del(&ctrl->list);
281	mutex_unlock(&nvme_loop_ctrl_mutex);
282
283	if (nctrl->tagset) {
284		blk_cleanup_queue(ctrl->ctrl.connect_q);
285		blk_mq_free_tag_set(&ctrl->tag_set);
286	}
287	kfree(ctrl->queues);
288	nvmf_free_options(nctrl->opts);
289free_ctrl:
290	kfree(ctrl);
291}
292
293static void nvme_loop_destroy_io_queues(struct nvme_loop_ctrl *ctrl)
294{
295	int i;
296
297	for (i = 1; i < ctrl->ctrl.queue_count; i++) {
298		clear_bit(NVME_LOOP_Q_LIVE, &ctrl->queues[i].flags);
299		nvmet_sq_destroy(&ctrl->queues[i].nvme_sq);
300	}
301	ctrl->ctrl.queue_count = 1;
302}
303
304static int nvme_loop_init_io_queues(struct nvme_loop_ctrl *ctrl)
305{
306	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
307	unsigned int nr_io_queues;
308	int ret, i;
309
310	nr_io_queues = min(opts->nr_io_queues, num_online_cpus());
311	ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
312	if (ret || !nr_io_queues)
313		return ret;
314
315	dev_info(ctrl->ctrl.device, "creating %d I/O queues.\n", nr_io_queues);
316
317	for (i = 1; i <= nr_io_queues; i++) {
318		ctrl->queues[i].ctrl = ctrl;
319		ret = nvmet_sq_init(&ctrl->queues[i].nvme_sq);
320		if (ret)
321			goto out_destroy_queues;
322
323		ctrl->ctrl.queue_count++;
324	}
325
326	return 0;
327
328out_destroy_queues:
329	nvme_loop_destroy_io_queues(ctrl);
330	return ret;
331}
332
333static int nvme_loop_connect_io_queues(struct nvme_loop_ctrl *ctrl)
334{
335	int i, ret;
336
337	for (i = 1; i < ctrl->ctrl.queue_count; i++) {
338		ret = nvmf_connect_io_queue(&ctrl->ctrl, i, false);
339		if (ret)
340			return ret;
341		set_bit(NVME_LOOP_Q_LIVE, &ctrl->queues[i].flags);
342	}
343
344	return 0;
345}
346
347static int nvme_loop_configure_admin_queue(struct nvme_loop_ctrl *ctrl)
348{
349	int error;
350
351	memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set));
352	ctrl->admin_tag_set.ops = &nvme_loop_admin_mq_ops;
353	ctrl->admin_tag_set.queue_depth = NVME_AQ_MQ_TAG_DEPTH;
354	ctrl->admin_tag_set.reserved_tags = 2; /* connect + keep-alive */
355	ctrl->admin_tag_set.numa_node = ctrl->ctrl.numa_node;
356	ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_loop_iod) +
357		NVME_INLINE_SG_CNT * sizeof(struct scatterlist);
358	ctrl->admin_tag_set.driver_data = ctrl;
359	ctrl->admin_tag_set.nr_hw_queues = 1;
360	ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT;
361	ctrl->admin_tag_set.flags = BLK_MQ_F_NO_SCHED;
362
363	ctrl->queues[0].ctrl = ctrl;
364	error = nvmet_sq_init(&ctrl->queues[0].nvme_sq);
365	if (error)
366		return error;
367	ctrl->ctrl.queue_count = 1;
368
369	error = blk_mq_alloc_tag_set(&ctrl->admin_tag_set);
370	if (error)
371		goto out_free_sq;
372	ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set;
373
374	ctrl->ctrl.fabrics_q = blk_mq_init_queue(&ctrl->admin_tag_set);
375	if (IS_ERR(ctrl->ctrl.fabrics_q)) {
376		error = PTR_ERR(ctrl->ctrl.fabrics_q);
377		goto out_free_tagset;
378	}
379
380	ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);
381	if (IS_ERR(ctrl->ctrl.admin_q)) {
382		error = PTR_ERR(ctrl->ctrl.admin_q);
383		goto out_cleanup_fabrics_q;
384	}
385	/* reset stopped state for the fresh admin queue */
386	clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->ctrl.flags);
387
388	error = nvmf_connect_admin_queue(&ctrl->ctrl);
389	if (error)
390		goto out_cleanup_queue;
391
392	set_bit(NVME_LOOP_Q_LIVE, &ctrl->queues[0].flags);
393
394	error = nvme_enable_ctrl(&ctrl->ctrl);
395	if (error)
396		goto out_cleanup_queue;
397
398	ctrl->ctrl.max_hw_sectors =
399		(NVME_LOOP_MAX_SEGMENTS - 1) << (PAGE_SHIFT - 9);
400
401	nvme_start_admin_queue(&ctrl->ctrl);
402
403	error = nvme_init_identify(&ctrl->ctrl);
404	if (error)
405		goto out_cleanup_queue;
406
407	return 0;
408
409out_cleanup_queue:
410	clear_bit(NVME_LOOP_Q_LIVE, &ctrl->queues[0].flags);
411	blk_cleanup_queue(ctrl->ctrl.admin_q);
412out_cleanup_fabrics_q:
413	blk_cleanup_queue(ctrl->ctrl.fabrics_q);
414out_free_tagset:
415	blk_mq_free_tag_set(&ctrl->admin_tag_set);
416out_free_sq:
417	nvmet_sq_destroy(&ctrl->queues[0].nvme_sq);
418	return error;
419}
420
421static void nvme_loop_shutdown_ctrl(struct nvme_loop_ctrl *ctrl)
422{
423	if (ctrl->ctrl.queue_count > 1) {
424		nvme_stop_queues(&ctrl->ctrl);
425		blk_mq_tagset_busy_iter(&ctrl->tag_set,
426					nvme_cancel_request, &ctrl->ctrl);
427		blk_mq_tagset_wait_completed_request(&ctrl->tag_set);
428		nvme_loop_destroy_io_queues(ctrl);
429	}
430
431	nvme_stop_admin_queue(&ctrl->ctrl);
432	if (ctrl->ctrl.state == NVME_CTRL_LIVE)
433		nvme_shutdown_ctrl(&ctrl->ctrl);
434
435	blk_mq_tagset_busy_iter(&ctrl->admin_tag_set,
436				nvme_cancel_request, &ctrl->ctrl);
437	blk_mq_tagset_wait_completed_request(&ctrl->admin_tag_set);
438	nvme_loop_destroy_admin_queue(ctrl);
439}
440
441static void nvme_loop_delete_ctrl_host(struct nvme_ctrl *ctrl)
442{
443	nvme_loop_shutdown_ctrl(to_loop_ctrl(ctrl));
444}
445
446static void nvme_loop_delete_ctrl(struct nvmet_ctrl *nctrl)
447{
448	struct nvme_loop_ctrl *ctrl;
449
450	mutex_lock(&nvme_loop_ctrl_mutex);
451	list_for_each_entry(ctrl, &nvme_loop_ctrl_list, list) {
452		if (ctrl->ctrl.cntlid == nctrl->cntlid)
453			nvme_delete_ctrl(&ctrl->ctrl);
454	}
455	mutex_unlock(&nvme_loop_ctrl_mutex);
456}
457
458static void nvme_loop_reset_ctrl_work(struct work_struct *work)
459{
460	struct nvme_loop_ctrl *ctrl =
461		container_of(work, struct nvme_loop_ctrl, ctrl.reset_work);
462	int ret;
463
464	nvme_stop_ctrl(&ctrl->ctrl);
465	nvme_loop_shutdown_ctrl(ctrl);
466
467	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_CONNECTING)) {
468		if (ctrl->ctrl.state != NVME_CTRL_DELETING &&
469		    ctrl->ctrl.state != NVME_CTRL_DELETING_NOIO)
470			/* state change failure for non-deleted ctrl? */
471			WARN_ON_ONCE(1);
472		return;
473	}
474
475	ret = nvme_loop_configure_admin_queue(ctrl);
476	if (ret)
477		goto out_disable;
478
479	ret = nvme_loop_init_io_queues(ctrl);
480	if (ret)
481		goto out_destroy_admin;
482
483	ret = nvme_loop_connect_io_queues(ctrl);
484	if (ret)
485		goto out_destroy_io;
486
487	blk_mq_update_nr_hw_queues(&ctrl->tag_set,
488			ctrl->ctrl.queue_count - 1);
489
490	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE))
491		WARN_ON_ONCE(1);
492
493	nvme_start_ctrl(&ctrl->ctrl);
494
495	return;
496
497out_destroy_io:
498	nvme_loop_destroy_io_queues(ctrl);
499out_destroy_admin:
500	nvme_loop_destroy_admin_queue(ctrl);
501out_disable:
502	dev_warn(ctrl->ctrl.device, "Removing after reset failure\n");
503	nvme_uninit_ctrl(&ctrl->ctrl);
504}
505
506static const struct nvme_ctrl_ops nvme_loop_ctrl_ops = {
507	.name			= "loop",
508	.module			= THIS_MODULE,
509	.flags			= NVME_F_FABRICS,
510	.reg_read32		= nvmf_reg_read32,
511	.reg_read64		= nvmf_reg_read64,
512	.reg_write32		= nvmf_reg_write32,
513	.free_ctrl		= nvme_loop_free_ctrl,
514	.submit_async_event	= nvme_loop_submit_async_event,
515	.delete_ctrl		= nvme_loop_delete_ctrl_host,
516	.get_address		= nvmf_get_address,
517};
518
519static int nvme_loop_create_io_queues(struct nvme_loop_ctrl *ctrl)
520{
521	int ret;
522
523	ret = nvme_loop_init_io_queues(ctrl);
524	if (ret)
525		return ret;
526
527	memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set));
528	ctrl->tag_set.ops = &nvme_loop_mq_ops;
529	ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size;
530	ctrl->tag_set.reserved_tags = 1; /* fabric connect */
531	ctrl->tag_set.numa_node = ctrl->ctrl.numa_node;
532	ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
533	ctrl->tag_set.cmd_size = sizeof(struct nvme_loop_iod) +
534		NVME_INLINE_SG_CNT * sizeof(struct scatterlist);
535	ctrl->tag_set.driver_data = ctrl;
536	ctrl->tag_set.nr_hw_queues = ctrl->ctrl.queue_count - 1;
537	ctrl->tag_set.timeout = NVME_IO_TIMEOUT;
538	ctrl->ctrl.tagset = &ctrl->tag_set;
539
540	ret = blk_mq_alloc_tag_set(&ctrl->tag_set);
541	if (ret)
542		goto out_destroy_queues;
543
544	ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set);
545	if (IS_ERR(ctrl->ctrl.connect_q)) {
546		ret = PTR_ERR(ctrl->ctrl.connect_q);
547		goto out_free_tagset;
548	}
549
550	ret = nvme_loop_connect_io_queues(ctrl);
551	if (ret)
552		goto out_cleanup_connect_q;
553
554	return 0;
555
556out_cleanup_connect_q:
557	blk_cleanup_queue(ctrl->ctrl.connect_q);
558out_free_tagset:
559	blk_mq_free_tag_set(&ctrl->tag_set);
560out_destroy_queues:
561	nvme_loop_destroy_io_queues(ctrl);
562	return ret;
563}
564
565static struct nvmet_port *nvme_loop_find_port(struct nvme_ctrl *ctrl)
566{
567	struct nvmet_port *p, *found = NULL;
568
569	mutex_lock(&nvme_loop_ports_mutex);
570	list_for_each_entry(p, &nvme_loop_ports, entry) {
571		/* if no transport address is specified use the first port */
572		if ((ctrl->opts->mask & NVMF_OPT_TRADDR) &&
573		    strcmp(ctrl->opts->traddr, p->disc_addr.traddr))
574			continue;
575		found = p;
576		break;
577	}
578	mutex_unlock(&nvme_loop_ports_mutex);
579	return found;
580}
581
582static struct nvme_ctrl *nvme_loop_create_ctrl(struct device *dev,
583		struct nvmf_ctrl_options *opts)
584{
585	struct nvme_loop_ctrl *ctrl;
586	int ret;
587
588	ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
589	if (!ctrl)
590		return ERR_PTR(-ENOMEM);
591	ctrl->ctrl.opts = opts;
592	INIT_LIST_HEAD(&ctrl->list);
593
594	INIT_WORK(&ctrl->ctrl.reset_work, nvme_loop_reset_ctrl_work);
595
596	ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_loop_ctrl_ops,
597				0 /* no quirks, we're perfect! */);
598	if (ret) {
599		kfree(ctrl);
600		goto out;
601	}
602
603	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_CONNECTING))
604		WARN_ON_ONCE(1);
605
606	ret = -ENOMEM;
607
608	ctrl->ctrl.sqsize = opts->queue_size - 1;
609	ctrl->ctrl.kato = opts->kato;
610	ctrl->port = nvme_loop_find_port(&ctrl->ctrl);
611
612	ctrl->queues = kcalloc(opts->nr_io_queues + 1, sizeof(*ctrl->queues),
613			GFP_KERNEL);
614	if (!ctrl->queues)
615		goto out_uninit_ctrl;
616
617	ret = nvme_loop_configure_admin_queue(ctrl);
618	if (ret)
619		goto out_free_queues;
620
621	if (opts->queue_size > ctrl->ctrl.maxcmd) {
622		/* warn if maxcmd is lower than queue_size */
623		dev_warn(ctrl->ctrl.device,
624			"queue_size %zu > ctrl maxcmd %u, clamping down\n",
625			opts->queue_size, ctrl->ctrl.maxcmd);
626		opts->queue_size = ctrl->ctrl.maxcmd;
627	}
628
629	if (opts->nr_io_queues) {
630		ret = nvme_loop_create_io_queues(ctrl);
631		if (ret)
632			goto out_remove_admin_queue;
633	}
634
635	nvme_loop_init_iod(ctrl, &ctrl->async_event_iod, 0);
636
637	dev_info(ctrl->ctrl.device,
638		 "new ctrl: \"%s\"\n", ctrl->ctrl.opts->subsysnqn);
639
640	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE))
641		WARN_ON_ONCE(1);
642
643	mutex_lock(&nvme_loop_ctrl_mutex);
644	list_add_tail(&ctrl->list, &nvme_loop_ctrl_list);
645	mutex_unlock(&nvme_loop_ctrl_mutex);
646
647	nvme_start_ctrl(&ctrl->ctrl);
648
649	return &ctrl->ctrl;
650
651out_remove_admin_queue:
652	nvme_loop_destroy_admin_queue(ctrl);
653out_free_queues:
654	kfree(ctrl->queues);
655out_uninit_ctrl:
656	nvme_uninit_ctrl(&ctrl->ctrl);
657	nvme_put_ctrl(&ctrl->ctrl);
658out:
659	if (ret > 0)
660		ret = -EIO;
661	return ERR_PTR(ret);
662}
663
664static int nvme_loop_add_port(struct nvmet_port *port)
665{
666	mutex_lock(&nvme_loop_ports_mutex);
667	list_add_tail(&port->entry, &nvme_loop_ports);
668	mutex_unlock(&nvme_loop_ports_mutex);
669	return 0;
670}
671
672static void nvme_loop_remove_port(struct nvmet_port *port)
673{
674	mutex_lock(&nvme_loop_ports_mutex);
675	list_del_init(&port->entry);
676	mutex_unlock(&nvme_loop_ports_mutex);
677
678	/*
679	 * Ensure any ctrls that are in the process of being
680	 * deleted are in fact deleted before we return
681	 * and free the port. This is to prevent active
682	 * ctrls from using a port after it's freed.
683	 */
684	flush_workqueue(nvme_delete_wq);
685}
686
687static const struct nvmet_fabrics_ops nvme_loop_ops = {
688	.owner		= THIS_MODULE,
689	.type		= NVMF_TRTYPE_LOOP,
690	.add_port	= nvme_loop_add_port,
691	.remove_port	= nvme_loop_remove_port,
692	.queue_response = nvme_loop_queue_response,
693	.delete_ctrl	= nvme_loop_delete_ctrl,
694};
695
696static struct nvmf_transport_ops nvme_loop_transport = {
697	.name		= "loop",
698	.module		= THIS_MODULE,
699	.create_ctrl	= nvme_loop_create_ctrl,
700	.allowed_opts	= NVMF_OPT_TRADDR,
701};
702
703static int __init nvme_loop_init_module(void)
704{
705	int ret;
706
707	ret = nvmet_register_transport(&nvme_loop_ops);
708	if (ret)
709		return ret;
710
711	ret = nvmf_register_transport(&nvme_loop_transport);
712	if (ret)
713		nvmet_unregister_transport(&nvme_loop_ops);
714
715	return ret;
716}
717
718static void __exit nvme_loop_cleanup_module(void)
719{
720	struct nvme_loop_ctrl *ctrl, *next;
721
722	nvmf_unregister_transport(&nvme_loop_transport);
723	nvmet_unregister_transport(&nvme_loop_ops);
724
725	mutex_lock(&nvme_loop_ctrl_mutex);
726	list_for_each_entry_safe(ctrl, next, &nvme_loop_ctrl_list, list)
727		nvme_delete_ctrl(&ctrl->ctrl);
728	mutex_unlock(&nvme_loop_ctrl_mutex);
729
730	flush_workqueue(nvme_delete_wq);
731}
732
733module_init(nvme_loop_init_module);
734module_exit(nvme_loop_cleanup_module);
735
736MODULE_LICENSE("GPL v2");
737MODULE_ALIAS("nvmet-transport-254"); /* 254 == NVMF_TRTYPE_LOOP */
738