xref: /openbmc/linux/drivers/block/rnbd/rnbd-srv.c (revision bef7a78d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * RDMA Network Block Driver
4  *
5  * Copyright (c) 2014 - 2018 ProfitBricks GmbH. All rights reserved.
6  * Copyright (c) 2018 - 2019 1&1 IONOS Cloud GmbH. All rights reserved.
7  * Copyright (c) 2019 - 2020 1&1 IONOS SE. All rights reserved.
8  */
9 #undef pr_fmt
10 #define pr_fmt(fmt) KBUILD_MODNAME " L" __stringify(__LINE__) ": " fmt
11 
12 #include <linux/module.h>
13 #include <linux/blkdev.h>
14 
15 #include "rnbd-srv.h"
16 #include "rnbd-srv-dev.h"
17 
18 MODULE_DESCRIPTION("RDMA Network Block Device Server");
19 MODULE_LICENSE("GPL");
20 
21 static u16 port_nr = RTRS_PORT;
22 
23 module_param_named(port_nr, port_nr, ushort, 0444);
24 MODULE_PARM_DESC(port_nr,
25 		 "The port number the server is listening on (default: "
26 		 __stringify(RTRS_PORT)")");
27 
28 #define DEFAULT_DEV_SEARCH_PATH "/"
29 
30 static char dev_search_path[PATH_MAX] = DEFAULT_DEV_SEARCH_PATH;
31 
32 static int dev_search_path_set(const char *val, const struct kernel_param *kp)
33 {
34 	const char *p = strrchr(val, '\n') ? : val + strlen(val);
35 
36 	if (strlen(val) >= sizeof(dev_search_path))
37 		return -EINVAL;
38 
39 	snprintf(dev_search_path, sizeof(dev_search_path), "%.*s",
40 		 (int)(p - val), val);
41 
42 	pr_info("dev_search_path changed to '%s'\n", dev_search_path);
43 
44 	return 0;
45 }
46 
47 static struct kparam_string dev_search_path_kparam_str = {
48 	.maxlen	= sizeof(dev_search_path),
49 	.string	= dev_search_path
50 };
51 
52 static const struct kernel_param_ops dev_search_path_ops = {
53 	.set	= dev_search_path_set,
54 	.get	= param_get_string,
55 };
56 
57 module_param_cb(dev_search_path, &dev_search_path_ops,
58 		&dev_search_path_kparam_str, 0444);
59 MODULE_PARM_DESC(dev_search_path,
60 		 "Sets the dev_search_path. When a device is mapped this path is prepended to the device path from the map device operation.  If %SESSNAME% is specified in a path, then device will be searched in a session namespace. (default: "
61 		 DEFAULT_DEV_SEARCH_PATH ")");
62 
63 static DEFINE_MUTEX(sess_lock);
64 static DEFINE_SPINLOCK(dev_lock);
65 
66 static LIST_HEAD(sess_list);
67 static LIST_HEAD(dev_list);
68 
69 struct rnbd_io_private {
70 	struct rtrs_srv_op		*id;
71 	struct rnbd_srv_sess_dev	*sess_dev;
72 };
73 
74 static void rnbd_sess_dev_release(struct kref *kref)
75 {
76 	struct rnbd_srv_sess_dev *sess_dev;
77 
78 	sess_dev = container_of(kref, struct rnbd_srv_sess_dev, kref);
79 	complete(sess_dev->destroy_comp);
80 }
81 
82 static inline void rnbd_put_sess_dev(struct rnbd_srv_sess_dev *sess_dev)
83 {
84 	kref_put(&sess_dev->kref, rnbd_sess_dev_release);
85 }
86 
87 void rnbd_endio(void *priv, int error)
88 {
89 	struct rnbd_io_private *rnbd_priv = priv;
90 	struct rnbd_srv_sess_dev *sess_dev = rnbd_priv->sess_dev;
91 
92 	rnbd_put_sess_dev(sess_dev);
93 
94 	rtrs_srv_resp_rdma(rnbd_priv->id, error);
95 
96 	kfree(priv);
97 }
98 
99 static struct rnbd_srv_sess_dev *
100 rnbd_get_sess_dev(int dev_id, struct rnbd_srv_session *srv_sess)
101 {
102 	struct rnbd_srv_sess_dev *sess_dev;
103 	int ret = 0;
104 
105 	rcu_read_lock();
106 	sess_dev = xa_load(&srv_sess->index_idr, dev_id);
107 	if (likely(sess_dev))
108 		ret = kref_get_unless_zero(&sess_dev->kref);
109 	rcu_read_unlock();
110 
111 	if (!sess_dev || !ret)
112 		return ERR_PTR(-ENXIO);
113 
114 	return sess_dev;
115 }
116 
117 static int process_rdma(struct rtrs_srv *sess,
118 			struct rnbd_srv_session *srv_sess,
119 			struct rtrs_srv_op *id, void *data, u32 datalen,
120 			const void *usr, size_t usrlen)
121 {
122 	const struct rnbd_msg_io *msg = usr;
123 	struct rnbd_io_private *priv;
124 	struct rnbd_srv_sess_dev *sess_dev;
125 	u32 dev_id;
126 	int err;
127 	struct rnbd_dev_blk_io *io;
128 	struct bio *bio;
129 	short prio;
130 
131 	priv = kmalloc(sizeof(*priv), GFP_KERNEL);
132 	if (!priv)
133 		return -ENOMEM;
134 
135 	dev_id = le32_to_cpu(msg->device_id);
136 
137 	sess_dev = rnbd_get_sess_dev(dev_id, srv_sess);
138 	if (IS_ERR(sess_dev)) {
139 		pr_err_ratelimited("Got I/O request on session %s for unknown device id %d\n",
140 				   srv_sess->sessname, dev_id);
141 		err = -ENOTCONN;
142 		goto err;
143 	}
144 
145 	priv->sess_dev = sess_dev;
146 	priv->id = id;
147 
148 	/* Generate bio with pages pointing to the rdma buffer */
149 	bio = rnbd_bio_map_kern(data, sess_dev->rnbd_dev->ibd_bio_set, datalen, GFP_KERNEL);
150 	if (IS_ERR(bio)) {
151 		err = PTR_ERR(bio);
152 		rnbd_srv_err(sess_dev, "Failed to generate bio, err: %d\n", err);
153 		goto sess_dev_put;
154 	}
155 
156 	io = container_of(bio, struct rnbd_dev_blk_io, bio);
157 	io->dev = sess_dev->rnbd_dev;
158 	io->priv = priv;
159 
160 	bio->bi_end_io = rnbd_dev_bi_end_io;
161 	bio->bi_private = io;
162 	bio->bi_opf = rnbd_to_bio_flags(le32_to_cpu(msg->rw));
163 	bio->bi_iter.bi_sector = le64_to_cpu(msg->sector);
164 	bio->bi_iter.bi_size = le32_to_cpu(msg->bi_size);
165 	prio = srv_sess->ver < RNBD_PROTO_VER_MAJOR ||
166 	       usrlen < sizeof(*msg) ? 0 : le16_to_cpu(msg->prio);
167 	bio_set_prio(bio, prio);
168 	bio_set_dev(bio, sess_dev->rnbd_dev->bdev);
169 
170 	submit_bio(bio);
171 
172 	return 0;
173 
174 sess_dev_put:
175 	rnbd_put_sess_dev(sess_dev);
176 err:
177 	kfree(priv);
178 	return err;
179 }
180 
181 static void destroy_device(struct rnbd_srv_dev *dev)
182 {
183 	WARN_ONCE(!list_empty(&dev->sess_dev_list),
184 		  "Device %s is being destroyed but still in use!\n",
185 		  dev->id);
186 
187 	spin_lock(&dev_lock);
188 	list_del(&dev->list);
189 	spin_unlock(&dev_lock);
190 
191 	mutex_destroy(&dev->lock);
192 	if (dev->dev_kobj.state_in_sysfs)
193 		/*
194 		 * Destroy kobj only if it was really created.
195 		 */
196 		rnbd_srv_destroy_dev_sysfs(dev);
197 	else
198 		kfree(dev);
199 }
200 
201 static void destroy_device_cb(struct kref *kref)
202 {
203 	struct rnbd_srv_dev *dev;
204 
205 	dev = container_of(kref, struct rnbd_srv_dev, kref);
206 
207 	destroy_device(dev);
208 }
209 
210 static void rnbd_put_srv_dev(struct rnbd_srv_dev *dev)
211 {
212 	kref_put(&dev->kref, destroy_device_cb);
213 }
214 
215 void rnbd_destroy_sess_dev(struct rnbd_srv_sess_dev *sess_dev, bool keep_id)
216 {
217 	DECLARE_COMPLETION_ONSTACK(dc);
218 
219 	if (keep_id)
220 		/* free the resources for the id but don't  */
221 		/* allow to re-use the id itself because it */
222 		/* is still used by the client              */
223 		xa_cmpxchg(&sess_dev->sess->index_idr, sess_dev->device_id,
224 			   sess_dev, NULL, 0);
225 	else
226 		xa_erase(&sess_dev->sess->index_idr, sess_dev->device_id);
227 	synchronize_rcu();
228 
229 	sess_dev->destroy_comp = &dc;
230 	rnbd_put_sess_dev(sess_dev);
231 	wait_for_completion(&dc); /* wait for inflights to drop to zero */
232 
233 	rnbd_dev_close(sess_dev->rnbd_dev);
234 	list_del(&sess_dev->sess_list);
235 	mutex_lock(&sess_dev->dev->lock);
236 	list_del(&sess_dev->dev_list);
237 	if (sess_dev->open_flags & FMODE_WRITE)
238 		sess_dev->dev->open_write_cnt--;
239 	mutex_unlock(&sess_dev->dev->lock);
240 
241 	rnbd_put_srv_dev(sess_dev->dev);
242 
243 	rnbd_srv_info(sess_dev, "Device closed\n");
244 	kfree(sess_dev);
245 }
246 
247 static void destroy_sess(struct rnbd_srv_session *srv_sess)
248 {
249 	struct rnbd_srv_sess_dev *sess_dev, *tmp;
250 
251 	if (list_empty(&srv_sess->sess_dev_list))
252 		goto out;
253 
254 	mutex_lock(&srv_sess->lock);
255 	list_for_each_entry_safe(sess_dev, tmp, &srv_sess->sess_dev_list,
256 				 sess_list)
257 		rnbd_srv_destroy_dev_session_sysfs(sess_dev);
258 	mutex_unlock(&srv_sess->lock);
259 
260 out:
261 	xa_destroy(&srv_sess->index_idr);
262 	bioset_exit(&srv_sess->sess_bio_set);
263 
264 	pr_info("RTRS Session %s disconnected\n", srv_sess->sessname);
265 
266 	mutex_lock(&sess_lock);
267 	list_del(&srv_sess->list);
268 	mutex_unlock(&sess_lock);
269 
270 	mutex_destroy(&srv_sess->lock);
271 	kfree(srv_sess);
272 }
273 
274 static int create_sess(struct rtrs_srv *rtrs)
275 {
276 	struct rnbd_srv_session *srv_sess;
277 	char sessname[NAME_MAX];
278 	int err;
279 
280 	err = rtrs_srv_get_sess_name(rtrs, sessname, sizeof(sessname));
281 	if (err) {
282 		pr_err("rtrs_srv_get_sess_name(%s): %d\n", sessname, err);
283 
284 		return err;
285 	}
286 	srv_sess = kzalloc(sizeof(*srv_sess), GFP_KERNEL);
287 	if (!srv_sess)
288 		return -ENOMEM;
289 
290 	srv_sess->queue_depth = rtrs_srv_get_queue_depth(rtrs);
291 	err = bioset_init(&srv_sess->sess_bio_set, srv_sess->queue_depth,
292 			  offsetof(struct rnbd_dev_blk_io, bio),
293 			  BIOSET_NEED_BVECS);
294 	if (err) {
295 		pr_err("Allocating srv_session for session %s failed\n",
296 		       sessname);
297 		kfree(srv_sess);
298 		return err;
299 	}
300 
301 	xa_init_flags(&srv_sess->index_idr, XA_FLAGS_ALLOC);
302 	INIT_LIST_HEAD(&srv_sess->sess_dev_list);
303 	mutex_init(&srv_sess->lock);
304 	mutex_lock(&sess_lock);
305 	list_add(&srv_sess->list, &sess_list);
306 	mutex_unlock(&sess_lock);
307 
308 	srv_sess->rtrs = rtrs;
309 	strlcpy(srv_sess->sessname, sessname, sizeof(srv_sess->sessname));
310 
311 	rtrs_srv_set_sess_priv(rtrs, srv_sess);
312 
313 	return 0;
314 }
315 
316 static int rnbd_srv_link_ev(struct rtrs_srv *rtrs,
317 			     enum rtrs_srv_link_ev ev, void *priv)
318 {
319 	struct rnbd_srv_session *srv_sess = priv;
320 
321 	switch (ev) {
322 	case RTRS_SRV_LINK_EV_CONNECTED:
323 		return create_sess(rtrs);
324 
325 	case RTRS_SRV_LINK_EV_DISCONNECTED:
326 		if (WARN_ON_ONCE(!srv_sess))
327 			return -EINVAL;
328 
329 		destroy_sess(srv_sess);
330 		return 0;
331 
332 	default:
333 		pr_warn("Received unknown RTRS session event %d from session %s\n",
334 			ev, srv_sess->sessname);
335 		return -EINVAL;
336 	}
337 }
338 
339 void rnbd_srv_sess_dev_force_close(struct rnbd_srv_sess_dev *sess_dev)
340 {
341 	mutex_lock(&sess_dev->sess->lock);
342 	rnbd_srv_destroy_dev_session_sysfs(sess_dev);
343 	mutex_unlock(&sess_dev->sess->lock);
344 	sess_dev->keep_id = true;
345 }
346 
347 static int process_msg_close(struct rtrs_srv *rtrs,
348 			     struct rnbd_srv_session *srv_sess,
349 			     void *data, size_t datalen, const void *usr,
350 			     size_t usrlen)
351 {
352 	const struct rnbd_msg_close *close_msg = usr;
353 	struct rnbd_srv_sess_dev *sess_dev;
354 
355 	sess_dev = rnbd_get_sess_dev(le32_to_cpu(close_msg->device_id),
356 				      srv_sess);
357 	if (IS_ERR(sess_dev))
358 		return 0;
359 
360 	rnbd_put_sess_dev(sess_dev);
361 	mutex_lock(&srv_sess->lock);
362 	rnbd_srv_destroy_dev_session_sysfs(sess_dev);
363 	mutex_unlock(&srv_sess->lock);
364 	return 0;
365 }
366 
367 static int process_msg_open(struct rtrs_srv *rtrs,
368 			    struct rnbd_srv_session *srv_sess,
369 			    const void *msg, size_t len,
370 			    void *data, size_t datalen);
371 
372 static int process_msg_sess_info(struct rtrs_srv *rtrs,
373 				 struct rnbd_srv_session *srv_sess,
374 				 const void *msg, size_t len,
375 				 void *data, size_t datalen);
376 
377 static int rnbd_srv_rdma_ev(struct rtrs_srv *rtrs, void *priv,
378 			     struct rtrs_srv_op *id, int dir,
379 			     void *data, size_t datalen, const void *usr,
380 			     size_t usrlen)
381 {
382 	struct rnbd_srv_session *srv_sess = priv;
383 	const struct rnbd_msg_hdr *hdr = usr;
384 	int ret = 0;
385 	u16 type;
386 
387 	if (WARN_ON_ONCE(!srv_sess))
388 		return -ENODEV;
389 
390 	type = le16_to_cpu(hdr->type);
391 
392 	switch (type) {
393 	case RNBD_MSG_IO:
394 		return process_rdma(rtrs, srv_sess, id, data, datalen, usr,
395 				    usrlen);
396 	case RNBD_MSG_CLOSE:
397 		ret = process_msg_close(rtrs, srv_sess, data, datalen,
398 					usr, usrlen);
399 		break;
400 	case RNBD_MSG_OPEN:
401 		ret = process_msg_open(rtrs, srv_sess, usr, usrlen,
402 				       data, datalen);
403 		break;
404 	case RNBD_MSG_SESS_INFO:
405 		ret = process_msg_sess_info(rtrs, srv_sess, usr, usrlen,
406 					    data, datalen);
407 		break;
408 	default:
409 		pr_warn("Received unexpected message type %d with dir %d from session %s\n",
410 			type, dir, srv_sess->sessname);
411 		return -EINVAL;
412 	}
413 
414 	rtrs_srv_resp_rdma(id, ret);
415 	return 0;
416 }
417 
418 static struct rnbd_srv_sess_dev
419 *rnbd_sess_dev_alloc(struct rnbd_srv_session *srv_sess)
420 {
421 	struct rnbd_srv_sess_dev *sess_dev;
422 	int error;
423 
424 	sess_dev = kzalloc(sizeof(*sess_dev), GFP_KERNEL);
425 	if (!sess_dev)
426 		return ERR_PTR(-ENOMEM);
427 
428 	error = xa_alloc(&srv_sess->index_idr, &sess_dev->device_id, sess_dev,
429 			 xa_limit_32b, GFP_NOWAIT);
430 	if (error < 0) {
431 		pr_warn("Allocating idr failed, err: %d\n", error);
432 		kfree(sess_dev);
433 		return ERR_PTR(error);
434 	}
435 
436 	return sess_dev;
437 }
438 
439 static struct rnbd_srv_dev *rnbd_srv_init_srv_dev(const char *id)
440 {
441 	struct rnbd_srv_dev *dev;
442 
443 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
444 	if (!dev)
445 		return ERR_PTR(-ENOMEM);
446 
447 	strlcpy(dev->id, id, sizeof(dev->id));
448 	kref_init(&dev->kref);
449 	INIT_LIST_HEAD(&dev->sess_dev_list);
450 	mutex_init(&dev->lock);
451 
452 	return dev;
453 }
454 
455 static struct rnbd_srv_dev *
456 rnbd_srv_find_or_add_srv_dev(struct rnbd_srv_dev *new_dev)
457 {
458 	struct rnbd_srv_dev *dev;
459 
460 	spin_lock(&dev_lock);
461 	list_for_each_entry(dev, &dev_list, list) {
462 		if (!strncmp(dev->id, new_dev->id, sizeof(dev->id))) {
463 			if (!kref_get_unless_zero(&dev->kref))
464 				/*
465 				 * We lost the race, device is almost dead.
466 				 *  Continue traversing to find a valid one.
467 				 */
468 				continue;
469 			spin_unlock(&dev_lock);
470 			return dev;
471 		}
472 	}
473 	list_add(&new_dev->list, &dev_list);
474 	spin_unlock(&dev_lock);
475 
476 	return new_dev;
477 }
478 
479 static int rnbd_srv_check_update_open_perm(struct rnbd_srv_dev *srv_dev,
480 					    struct rnbd_srv_session *srv_sess,
481 					    enum rnbd_access_mode access_mode)
482 {
483 	int ret = -EPERM;
484 
485 	mutex_lock(&srv_dev->lock);
486 
487 	switch (access_mode) {
488 	case RNBD_ACCESS_RO:
489 		ret = 0;
490 		break;
491 	case RNBD_ACCESS_RW:
492 		if (srv_dev->open_write_cnt == 0)  {
493 			srv_dev->open_write_cnt++;
494 			ret = 0;
495 		} else {
496 			pr_err("Mapping device '%s' for session %s with RW permissions failed. Device already opened as 'RW' by %d client(s), access mode %s.\n",
497 			       srv_dev->id, srv_sess->sessname,
498 			       srv_dev->open_write_cnt,
499 			       rnbd_access_mode_str(access_mode));
500 		}
501 		break;
502 	case RNBD_ACCESS_MIGRATION:
503 		if (srv_dev->open_write_cnt < 2) {
504 			srv_dev->open_write_cnt++;
505 			ret = 0;
506 		} else {
507 			pr_err("Mapping device '%s' for session %s with migration permissions failed. Device already opened as 'RW' by %d client(s), access mode %s.\n",
508 			       srv_dev->id, srv_sess->sessname,
509 			       srv_dev->open_write_cnt,
510 			       rnbd_access_mode_str(access_mode));
511 		}
512 		break;
513 	default:
514 		pr_err("Received mapping request for device '%s' on session %s with invalid access mode: %d\n",
515 		       srv_dev->id, srv_sess->sessname, access_mode);
516 		ret = -EINVAL;
517 	}
518 
519 	mutex_unlock(&srv_dev->lock);
520 
521 	return ret;
522 }
523 
524 static struct rnbd_srv_dev *
525 rnbd_srv_get_or_create_srv_dev(struct rnbd_dev *rnbd_dev,
526 				struct rnbd_srv_session *srv_sess,
527 				enum rnbd_access_mode access_mode)
528 {
529 	int ret;
530 	struct rnbd_srv_dev *new_dev, *dev;
531 
532 	new_dev = rnbd_srv_init_srv_dev(rnbd_dev->name);
533 	if (IS_ERR(new_dev))
534 		return new_dev;
535 
536 	dev = rnbd_srv_find_or_add_srv_dev(new_dev);
537 	if (dev != new_dev)
538 		kfree(new_dev);
539 
540 	ret = rnbd_srv_check_update_open_perm(dev, srv_sess, access_mode);
541 	if (ret) {
542 		rnbd_put_srv_dev(dev);
543 		return ERR_PTR(ret);
544 	}
545 
546 	return dev;
547 }
548 
549 static void rnbd_srv_fill_msg_open_rsp(struct rnbd_msg_open_rsp *rsp,
550 					struct rnbd_srv_sess_dev *sess_dev)
551 {
552 	struct rnbd_dev *rnbd_dev = sess_dev->rnbd_dev;
553 	struct request_queue *q = bdev_get_queue(rnbd_dev->bdev);
554 
555 	rsp->hdr.type = cpu_to_le16(RNBD_MSG_OPEN_RSP);
556 	rsp->device_id =
557 		cpu_to_le32(sess_dev->device_id);
558 	rsp->nsectors =
559 		cpu_to_le64(get_capacity(rnbd_dev->bdev->bd_disk));
560 	rsp->logical_block_size	=
561 		cpu_to_le16(bdev_logical_block_size(rnbd_dev->bdev));
562 	rsp->physical_block_size =
563 		cpu_to_le16(bdev_physical_block_size(rnbd_dev->bdev));
564 	rsp->max_segments =
565 		cpu_to_le16(rnbd_dev_get_max_segs(rnbd_dev));
566 	rsp->max_hw_sectors =
567 		cpu_to_le32(rnbd_dev_get_max_hw_sects(rnbd_dev));
568 	rsp->max_write_same_sectors =
569 		cpu_to_le32(bdev_write_same(rnbd_dev->bdev));
570 	rsp->max_discard_sectors =
571 		cpu_to_le32(rnbd_dev_get_max_discard_sects(rnbd_dev));
572 	rsp->discard_granularity =
573 		cpu_to_le32(rnbd_dev_get_discard_granularity(rnbd_dev));
574 	rsp->discard_alignment =
575 		cpu_to_le32(rnbd_dev_get_discard_alignment(rnbd_dev));
576 	rsp->secure_discard =
577 		cpu_to_le16(rnbd_dev_get_secure_discard(rnbd_dev));
578 	rsp->rotational = !blk_queue_nonrot(q);
579 	rsp->cache_policy = 0;
580 	if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
581 		rsp->cache_policy |= RNBD_WRITEBACK;
582 	if (blk_queue_fua(q))
583 		rsp->cache_policy |= RNBD_FUA;
584 }
585 
586 static struct rnbd_srv_sess_dev *
587 rnbd_srv_create_set_sess_dev(struct rnbd_srv_session *srv_sess,
588 			      const struct rnbd_msg_open *open_msg,
589 			      struct rnbd_dev *rnbd_dev, fmode_t open_flags,
590 			      struct rnbd_srv_dev *srv_dev)
591 {
592 	struct rnbd_srv_sess_dev *sdev = rnbd_sess_dev_alloc(srv_sess);
593 
594 	if (IS_ERR(sdev))
595 		return sdev;
596 
597 	kref_init(&sdev->kref);
598 
599 	strlcpy(sdev->pathname, open_msg->dev_name, sizeof(sdev->pathname));
600 
601 	sdev->rnbd_dev		= rnbd_dev;
602 	sdev->sess		= srv_sess;
603 	sdev->dev		= srv_dev;
604 	sdev->open_flags	= open_flags;
605 	sdev->access_mode	= open_msg->access_mode;
606 
607 	return sdev;
608 }
609 
610 static char *rnbd_srv_get_full_path(struct rnbd_srv_session *srv_sess,
611 				     const char *dev_name)
612 {
613 	char *full_path;
614 	char *a, *b;
615 
616 	full_path = kmalloc(PATH_MAX, GFP_KERNEL);
617 	if (!full_path)
618 		return ERR_PTR(-ENOMEM);
619 
620 	/*
621 	 * Replace %SESSNAME% with a real session name in order to
622 	 * create device namespace.
623 	 */
624 	a = strnstr(dev_search_path, "%SESSNAME%", sizeof(dev_search_path));
625 	if (a) {
626 		int len = a - dev_search_path;
627 
628 		len = snprintf(full_path, PATH_MAX, "%.*s/%s/%s", len,
629 			       dev_search_path, srv_sess->sessname, dev_name);
630 		if (len >= PATH_MAX) {
631 			pr_err("Too long path: %s, %s, %s\n",
632 			       dev_search_path, srv_sess->sessname, dev_name);
633 			kfree(full_path);
634 			return ERR_PTR(-EINVAL);
635 		}
636 	} else {
637 		snprintf(full_path, PATH_MAX, "%s/%s",
638 			 dev_search_path, dev_name);
639 	}
640 
641 	/* eliminitate duplicated slashes */
642 	a = strchr(full_path, '/');
643 	b = a;
644 	while (*b != '\0') {
645 		if (*b == '/' && *a == '/') {
646 			b++;
647 		} else {
648 			a++;
649 			*a = *b;
650 			b++;
651 		}
652 	}
653 	a++;
654 	*a = '\0';
655 
656 	return full_path;
657 }
658 
659 static int process_msg_sess_info(struct rtrs_srv *rtrs,
660 				 struct rnbd_srv_session *srv_sess,
661 				 const void *msg, size_t len,
662 				 void *data, size_t datalen)
663 {
664 	const struct rnbd_msg_sess_info *sess_info_msg = msg;
665 	struct rnbd_msg_sess_info_rsp *rsp = data;
666 
667 	srv_sess->ver = min_t(u8, sess_info_msg->ver, RNBD_PROTO_VER_MAJOR);
668 	pr_debug("Session %s using protocol version %d (client version: %d, server version: %d)\n",
669 		 srv_sess->sessname, srv_sess->ver,
670 		 sess_info_msg->ver, RNBD_PROTO_VER_MAJOR);
671 
672 	rsp->hdr.type = cpu_to_le16(RNBD_MSG_SESS_INFO_RSP);
673 	rsp->ver = srv_sess->ver;
674 
675 	return 0;
676 }
677 
678 /**
679  * find_srv_sess_dev() - a dev is already opened by this name
680  * @srv_sess:	the session to search.
681  * @dev_name:	string containing the name of the device.
682  *
683  * Return struct rnbd_srv_sess_dev if srv_sess already opened the dev_name
684  * NULL if the session didn't open the device yet.
685  */
686 static struct rnbd_srv_sess_dev *
687 find_srv_sess_dev(struct rnbd_srv_session *srv_sess, const char *dev_name)
688 {
689 	struct rnbd_srv_sess_dev *sess_dev;
690 
691 	if (list_empty(&srv_sess->sess_dev_list))
692 		return NULL;
693 
694 	list_for_each_entry(sess_dev, &srv_sess->sess_dev_list, sess_list)
695 		if (!strcmp(sess_dev->pathname, dev_name))
696 			return sess_dev;
697 
698 	return NULL;
699 }
700 
701 static int process_msg_open(struct rtrs_srv *rtrs,
702 			    struct rnbd_srv_session *srv_sess,
703 			    const void *msg, size_t len,
704 			    void *data, size_t datalen)
705 {
706 	int ret;
707 	struct rnbd_srv_dev *srv_dev;
708 	struct rnbd_srv_sess_dev *srv_sess_dev;
709 	const struct rnbd_msg_open *open_msg = msg;
710 	fmode_t open_flags;
711 	char *full_path;
712 	struct rnbd_dev *rnbd_dev;
713 	struct rnbd_msg_open_rsp *rsp = data;
714 
715 	pr_debug("Open message received: session='%s' path='%s' access_mode=%d\n",
716 		 srv_sess->sessname, open_msg->dev_name,
717 		 open_msg->access_mode);
718 	open_flags = FMODE_READ;
719 	if (open_msg->access_mode != RNBD_ACCESS_RO)
720 		open_flags |= FMODE_WRITE;
721 
722 	mutex_lock(&srv_sess->lock);
723 
724 	srv_sess_dev = find_srv_sess_dev(srv_sess, open_msg->dev_name);
725 	if (srv_sess_dev)
726 		goto fill_response;
727 
728 	if ((strlen(dev_search_path) + strlen(open_msg->dev_name))
729 	    >= PATH_MAX) {
730 		pr_err("Opening device for session %s failed, device path too long. '%s/%s' is longer than PATH_MAX (%d)\n",
731 		       srv_sess->sessname, dev_search_path, open_msg->dev_name,
732 		       PATH_MAX);
733 		ret = -EINVAL;
734 		goto reject;
735 	}
736 	if (strstr(open_msg->dev_name, "..")) {
737 		pr_err("Opening device for session %s failed, device path %s contains relative path ..\n",
738 		       srv_sess->sessname, open_msg->dev_name);
739 		ret = -EINVAL;
740 		goto reject;
741 	}
742 	full_path = rnbd_srv_get_full_path(srv_sess, open_msg->dev_name);
743 	if (IS_ERR(full_path)) {
744 		ret = PTR_ERR(full_path);
745 		pr_err("Opening device '%s' for client %s failed, failed to get device full path, err: %d\n",
746 		       open_msg->dev_name, srv_sess->sessname, ret);
747 		goto reject;
748 	}
749 
750 	rnbd_dev = rnbd_dev_open(full_path, open_flags,
751 				 &srv_sess->sess_bio_set);
752 	if (IS_ERR(rnbd_dev)) {
753 		pr_err("Opening device '%s' on session %s failed, failed to open the block device, err: %ld\n",
754 		       full_path, srv_sess->sessname, PTR_ERR(rnbd_dev));
755 		ret = PTR_ERR(rnbd_dev);
756 		goto free_path;
757 	}
758 
759 	srv_dev = rnbd_srv_get_or_create_srv_dev(rnbd_dev, srv_sess,
760 						  open_msg->access_mode);
761 	if (IS_ERR(srv_dev)) {
762 		pr_err("Opening device '%s' on session %s failed, creating srv_dev failed, err: %ld\n",
763 		       full_path, srv_sess->sessname, PTR_ERR(srv_dev));
764 		ret = PTR_ERR(srv_dev);
765 		goto rnbd_dev_close;
766 	}
767 
768 	srv_sess_dev = rnbd_srv_create_set_sess_dev(srv_sess, open_msg,
769 						     rnbd_dev, open_flags,
770 						     srv_dev);
771 	if (IS_ERR(srv_sess_dev)) {
772 		pr_err("Opening device '%s' on session %s failed, creating sess_dev failed, err: %ld\n",
773 		       full_path, srv_sess->sessname, PTR_ERR(srv_sess_dev));
774 		ret = PTR_ERR(srv_sess_dev);
775 		goto srv_dev_put;
776 	}
777 
778 	/* Create the srv_dev sysfs files if they haven't been created yet. The
779 	 * reason to delay the creation is not to create the sysfs files before
780 	 * we are sure the device can be opened.
781 	 */
782 	mutex_lock(&srv_dev->lock);
783 	if (!srv_dev->dev_kobj.state_in_sysfs) {
784 		ret = rnbd_srv_create_dev_sysfs(srv_dev, rnbd_dev->bdev,
785 						 rnbd_dev->name);
786 		if (ret) {
787 			mutex_unlock(&srv_dev->lock);
788 			rnbd_srv_err(srv_sess_dev,
789 				      "Opening device failed, failed to create device sysfs files, err: %d\n",
790 				      ret);
791 			goto free_srv_sess_dev;
792 		}
793 	}
794 
795 	ret = rnbd_srv_create_dev_session_sysfs(srv_sess_dev);
796 	if (ret) {
797 		mutex_unlock(&srv_dev->lock);
798 		rnbd_srv_err(srv_sess_dev,
799 			      "Opening device failed, failed to create dev client sysfs files, err: %d\n",
800 			      ret);
801 		goto free_srv_sess_dev;
802 	}
803 
804 	list_add(&srv_sess_dev->dev_list, &srv_dev->sess_dev_list);
805 	mutex_unlock(&srv_dev->lock);
806 
807 	list_add(&srv_sess_dev->sess_list, &srv_sess->sess_dev_list);
808 
809 	rnbd_srv_info(srv_sess_dev, "Opened device '%s'\n", srv_dev->id);
810 
811 	kfree(full_path);
812 
813 fill_response:
814 	rnbd_srv_fill_msg_open_rsp(rsp, srv_sess_dev);
815 	mutex_unlock(&srv_sess->lock);
816 	return 0;
817 
818 free_srv_sess_dev:
819 	xa_erase(&srv_sess->index_idr, srv_sess_dev->device_id);
820 	synchronize_rcu();
821 	kfree(srv_sess_dev);
822 srv_dev_put:
823 	if (open_msg->access_mode != RNBD_ACCESS_RO) {
824 		mutex_lock(&srv_dev->lock);
825 		srv_dev->open_write_cnt--;
826 		mutex_unlock(&srv_dev->lock);
827 	}
828 	rnbd_put_srv_dev(srv_dev);
829 rnbd_dev_close:
830 	rnbd_dev_close(rnbd_dev);
831 free_path:
832 	kfree(full_path);
833 reject:
834 	mutex_unlock(&srv_sess->lock);
835 	return ret;
836 }
837 
838 static struct rtrs_srv_ctx *rtrs_ctx;
839 
840 static struct rtrs_srv_ops rtrs_ops;
841 static int __init rnbd_srv_init_module(void)
842 {
843 	int err;
844 
845 	BUILD_BUG_ON(sizeof(struct rnbd_msg_hdr) != 4);
846 	BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info) != 36);
847 	BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info_rsp) != 36);
848 	BUILD_BUG_ON(sizeof(struct rnbd_msg_open) != 264);
849 	BUILD_BUG_ON(sizeof(struct rnbd_msg_close) != 8);
850 	BUILD_BUG_ON(sizeof(struct rnbd_msg_open_rsp) != 56);
851 	rtrs_ops = (struct rtrs_srv_ops) {
852 		.rdma_ev = rnbd_srv_rdma_ev,
853 		.link_ev = rnbd_srv_link_ev,
854 	};
855 	rtrs_ctx = rtrs_srv_open(&rtrs_ops, port_nr);
856 	if (IS_ERR(rtrs_ctx)) {
857 		err = PTR_ERR(rtrs_ctx);
858 		pr_err("rtrs_srv_open(), err: %d\n", err);
859 		return err;
860 	}
861 
862 	err = rnbd_srv_create_sysfs_files();
863 	if (err) {
864 		pr_err("rnbd_srv_create_sysfs_files(), err: %d\n", err);
865 		rtrs_srv_close(rtrs_ctx);
866 		return err;
867 	}
868 
869 	return 0;
870 }
871 
872 static void __exit rnbd_srv_cleanup_module(void)
873 {
874 	rtrs_srv_close(rtrs_ctx);
875 	WARN_ON(!list_empty(&sess_list));
876 	rnbd_srv_destroy_sysfs_files();
877 }
878 
879 module_init(rnbd_srv_init_module);
880 module_exit(rnbd_srv_cleanup_module);
881