xref: /openbmc/linux/drivers/block/rnbd/rnbd-srv.c (revision faffb083)
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-trace.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 static struct rnbd_srv_sess_dev *
88 rnbd_get_sess_dev(int dev_id, struct rnbd_srv_session *srv_sess)
89 {
90 	struct rnbd_srv_sess_dev *sess_dev;
91 	int ret = 0;
92 
93 	rcu_read_lock();
94 	sess_dev = xa_load(&srv_sess->index_idr, dev_id);
95 	if (sess_dev)
96 		ret = kref_get_unless_zero(&sess_dev->kref);
97 	rcu_read_unlock();
98 
99 	if (!sess_dev || !ret)
100 		return ERR_PTR(-ENXIO);
101 
102 	return sess_dev;
103 }
104 
105 static void rnbd_dev_bi_end_io(struct bio *bio)
106 {
107 	struct rnbd_io_private *rnbd_priv = bio->bi_private;
108 	struct rnbd_srv_sess_dev *sess_dev = rnbd_priv->sess_dev;
109 
110 	rnbd_put_sess_dev(sess_dev);
111 	rtrs_srv_resp_rdma(rnbd_priv->id, blk_status_to_errno(bio->bi_status));
112 
113 	kfree(rnbd_priv);
114 	bio_put(bio);
115 }
116 
117 static int process_rdma(struct rnbd_srv_session *srv_sess,
118 			struct rtrs_srv_op *id, void *data, u32 datalen,
119 			const void *usr, size_t usrlen)
120 {
121 	const struct rnbd_msg_io *msg = usr;
122 	struct rnbd_io_private *priv;
123 	struct rnbd_srv_sess_dev *sess_dev;
124 	u32 dev_id;
125 	int err;
126 	struct bio *bio;
127 	short prio;
128 
129 	trace_process_rdma(srv_sess, msg, id, datalen, usrlen);
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 	bio = bio_alloc(sess_dev->bdev, 1,
149 			rnbd_to_bio_flags(le32_to_cpu(msg->rw)), GFP_KERNEL);
150 	if (bio_add_page(bio, virt_to_page(data), datalen,
151 			offset_in_page(data)) != datalen) {
152 		rnbd_srv_err(sess_dev, "Failed to map data to bio\n");
153 		err = -EINVAL;
154 		goto bio_put;
155 	}
156 
157 	bio->bi_end_io = rnbd_dev_bi_end_io;
158 	bio->bi_private = priv;
159 	bio->bi_iter.bi_sector = le64_to_cpu(msg->sector);
160 	bio->bi_iter.bi_size = le32_to_cpu(msg->bi_size);
161 	prio = srv_sess->ver < RNBD_PROTO_VER_MAJOR ||
162 	       usrlen < sizeof(*msg) ? 0 : le16_to_cpu(msg->prio);
163 	bio_set_prio(bio, prio);
164 
165 	submit_bio(bio);
166 
167 	return 0;
168 
169 bio_put:
170 	bio_put(bio);
171 	rnbd_put_sess_dev(sess_dev);
172 err:
173 	kfree(priv);
174 	return err;
175 }
176 
177 static void destroy_device(struct kref *kref)
178 {
179 	struct rnbd_srv_dev *dev = container_of(kref, struct rnbd_srv_dev, kref);
180 
181 	WARN_ONCE(!list_empty(&dev->sess_dev_list),
182 		  "Device %s is being destroyed but still in use!\n",
183 		  dev->id);
184 
185 	spin_lock(&dev_lock);
186 	list_del(&dev->list);
187 	spin_unlock(&dev_lock);
188 
189 	mutex_destroy(&dev->lock);
190 	if (dev->dev_kobj.state_in_sysfs)
191 		/*
192 		 * Destroy kobj only if it was really created.
193 		 */
194 		rnbd_srv_destroy_dev_sysfs(dev);
195 	else
196 		kfree(dev);
197 }
198 
199 static void rnbd_put_srv_dev(struct rnbd_srv_dev *dev)
200 {
201 	kref_put(&dev->kref, destroy_device);
202 }
203 
204 void rnbd_destroy_sess_dev(struct rnbd_srv_sess_dev *sess_dev, bool keep_id)
205 {
206 	DECLARE_COMPLETION_ONSTACK(dc);
207 
208 	if (keep_id)
209 		/* free the resources for the id but don't  */
210 		/* allow to re-use the id itself because it */
211 		/* is still used by the client              */
212 		xa_cmpxchg(&sess_dev->sess->index_idr, sess_dev->device_id,
213 			   sess_dev, NULL, 0);
214 	else
215 		xa_erase(&sess_dev->sess->index_idr, sess_dev->device_id);
216 	synchronize_rcu();
217 
218 	sess_dev->destroy_comp = &dc;
219 	rnbd_put_sess_dev(sess_dev);
220 	wait_for_completion(&dc); /* wait for inflights to drop to zero */
221 
222 	blkdev_put(sess_dev->bdev, sess_dev->open_flags);
223 	mutex_lock(&sess_dev->dev->lock);
224 	list_del(&sess_dev->dev_list);
225 	if (sess_dev->open_flags & FMODE_WRITE)
226 		sess_dev->dev->open_write_cnt--;
227 	mutex_unlock(&sess_dev->dev->lock);
228 
229 	rnbd_put_srv_dev(sess_dev->dev);
230 
231 	rnbd_srv_info(sess_dev, "Device closed\n");
232 	kfree(sess_dev);
233 }
234 
235 static void destroy_sess(struct rnbd_srv_session *srv_sess)
236 {
237 	struct rnbd_srv_sess_dev *sess_dev;
238 	unsigned long index;
239 
240 	if (xa_empty(&srv_sess->index_idr))
241 		goto out;
242 
243 	trace_destroy_sess(srv_sess);
244 
245 	mutex_lock(&srv_sess->lock);
246 	xa_for_each(&srv_sess->index_idr, index, sess_dev)
247 		rnbd_srv_destroy_dev_session_sysfs(sess_dev);
248 	mutex_unlock(&srv_sess->lock);
249 
250 out:
251 	xa_destroy(&srv_sess->index_idr);
252 
253 	pr_info("RTRS Session %s disconnected\n", srv_sess->sessname);
254 
255 	mutex_lock(&sess_lock);
256 	list_del(&srv_sess->list);
257 	mutex_unlock(&sess_lock);
258 
259 	mutex_destroy(&srv_sess->lock);
260 	kfree(srv_sess);
261 }
262 
263 static int create_sess(struct rtrs_srv_sess *rtrs)
264 {
265 	struct rnbd_srv_session *srv_sess;
266 	char pathname[NAME_MAX];
267 	int err;
268 
269 	err = rtrs_srv_get_path_name(rtrs, pathname, sizeof(pathname));
270 	if (err) {
271 		pr_err("rtrs_srv_get_path_name(%s): %d\n", pathname, err);
272 
273 		return err;
274 	}
275 	srv_sess = kzalloc(sizeof(*srv_sess), GFP_KERNEL);
276 	if (!srv_sess)
277 		return -ENOMEM;
278 
279 	srv_sess->queue_depth = rtrs_srv_get_queue_depth(rtrs);
280 	xa_init_flags(&srv_sess->index_idr, XA_FLAGS_ALLOC);
281 	mutex_init(&srv_sess->lock);
282 	mutex_lock(&sess_lock);
283 	list_add(&srv_sess->list, &sess_list);
284 	mutex_unlock(&sess_lock);
285 
286 	srv_sess->rtrs = rtrs;
287 	strscpy(srv_sess->sessname, pathname, sizeof(srv_sess->sessname));
288 
289 	rtrs_srv_set_sess_priv(rtrs, srv_sess);
290 
291 	trace_create_sess(srv_sess);
292 
293 	return 0;
294 }
295 
296 static int rnbd_srv_link_ev(struct rtrs_srv_sess *rtrs,
297 			     enum rtrs_srv_link_ev ev, void *priv)
298 {
299 	struct rnbd_srv_session *srv_sess = priv;
300 
301 	switch (ev) {
302 	case RTRS_SRV_LINK_EV_CONNECTED:
303 		return create_sess(rtrs);
304 
305 	case RTRS_SRV_LINK_EV_DISCONNECTED:
306 		if (WARN_ON_ONCE(!srv_sess))
307 			return -EINVAL;
308 
309 		destroy_sess(srv_sess);
310 		return 0;
311 
312 	default:
313 		pr_warn("Received unknown RTRS session event %d from session %s\n",
314 			ev, srv_sess->sessname);
315 		return -EINVAL;
316 	}
317 }
318 
319 void rnbd_srv_sess_dev_force_close(struct rnbd_srv_sess_dev *sess_dev,
320 				   struct kobj_attribute *attr)
321 {
322 	struct rnbd_srv_session	*sess = sess_dev->sess;
323 
324 	/* It is already started to close by client's close message. */
325 	if (!mutex_trylock(&sess->lock))
326 		return;
327 
328 	sess_dev->keep_id = true;
329 	/* first remove sysfs itself to avoid deadlock */
330 	sysfs_remove_file_self(&sess_dev->kobj, &attr->attr);
331 	rnbd_srv_destroy_dev_session_sysfs(sess_dev);
332 	mutex_unlock(&sess->lock);
333 }
334 
335 static void process_msg_close(struct rnbd_srv_session *srv_sess,
336 			     void *data, size_t datalen, const void *usr,
337 			     size_t usrlen)
338 {
339 	const struct rnbd_msg_close *close_msg = usr;
340 	struct rnbd_srv_sess_dev *sess_dev;
341 
342 	trace_process_msg_close(srv_sess, close_msg);
343 
344 	sess_dev = rnbd_get_sess_dev(le32_to_cpu(close_msg->device_id),
345 				      srv_sess);
346 	if (IS_ERR(sess_dev))
347 		return;
348 
349 	rnbd_put_sess_dev(sess_dev);
350 	mutex_lock(&srv_sess->lock);
351 	rnbd_srv_destroy_dev_session_sysfs(sess_dev);
352 	mutex_unlock(&srv_sess->lock);
353 }
354 
355 static int process_msg_open(struct rnbd_srv_session *srv_sess,
356 			    const void *msg, size_t len,
357 			    void *data, size_t datalen);
358 
359 static int process_msg_sess_info(struct rnbd_srv_session *srv_sess,
360 				 const void *msg, size_t len,
361 				 void *data, size_t datalen);
362 
363 static int rnbd_srv_rdma_ev(void *priv, struct rtrs_srv_op *id,
364 			    void *data, size_t datalen,
365 			    const void *usr, size_t usrlen)
366 {
367 	struct rnbd_srv_session *srv_sess = priv;
368 	const struct rnbd_msg_hdr *hdr = usr;
369 	int ret = 0;
370 	u16 type;
371 
372 	if (WARN_ON_ONCE(!srv_sess))
373 		return -ENODEV;
374 
375 	type = le16_to_cpu(hdr->type);
376 
377 	switch (type) {
378 	case RNBD_MSG_IO:
379 		return process_rdma(srv_sess, id, data, datalen, usr, usrlen);
380 	case RNBD_MSG_CLOSE:
381 		process_msg_close(srv_sess, data, datalen, usr, usrlen);
382 		break;
383 	case RNBD_MSG_OPEN:
384 		ret = process_msg_open(srv_sess, usr, usrlen, data, datalen);
385 		break;
386 	case RNBD_MSG_SESS_INFO:
387 		ret = process_msg_sess_info(srv_sess, usr, usrlen, data,
388 					    datalen);
389 		break;
390 	default:
391 		pr_warn("Received unexpected message type %d from session %s\n",
392 			type, srv_sess->sessname);
393 		return -EINVAL;
394 	}
395 
396 	/*
397 	 * Since ret is passed to rtrs to handle the failure case, we
398 	 * just return 0 at the end otherwise callers in rtrs would call
399 	 * send_io_resp_imm again to print redundant err message.
400 	 */
401 	rtrs_srv_resp_rdma(id, ret);
402 	return 0;
403 }
404 
405 static struct rnbd_srv_sess_dev
406 *rnbd_sess_dev_alloc(struct rnbd_srv_session *srv_sess)
407 {
408 	struct rnbd_srv_sess_dev *sess_dev;
409 	int error;
410 
411 	sess_dev = kzalloc(sizeof(*sess_dev), GFP_KERNEL);
412 	if (!sess_dev)
413 		return ERR_PTR(-ENOMEM);
414 
415 	error = xa_alloc(&srv_sess->index_idr, &sess_dev->device_id, sess_dev,
416 			 xa_limit_32b, GFP_NOWAIT);
417 	if (error < 0) {
418 		pr_warn("Allocating idr failed, err: %d\n", error);
419 		kfree(sess_dev);
420 		return ERR_PTR(error);
421 	}
422 
423 	return sess_dev;
424 }
425 
426 static struct rnbd_srv_dev *rnbd_srv_init_srv_dev(struct block_device *bdev)
427 {
428 	struct rnbd_srv_dev *dev;
429 
430 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
431 	if (!dev)
432 		return ERR_PTR(-ENOMEM);
433 
434 	snprintf(dev->id, sizeof(dev->id), "%pg", bdev);
435 	kref_init(&dev->kref);
436 	INIT_LIST_HEAD(&dev->sess_dev_list);
437 	mutex_init(&dev->lock);
438 
439 	return dev;
440 }
441 
442 static struct rnbd_srv_dev *
443 rnbd_srv_find_or_add_srv_dev(struct rnbd_srv_dev *new_dev)
444 {
445 	struct rnbd_srv_dev *dev;
446 
447 	spin_lock(&dev_lock);
448 	list_for_each_entry(dev, &dev_list, list) {
449 		if (!strncmp(dev->id, new_dev->id, sizeof(dev->id))) {
450 			if (!kref_get_unless_zero(&dev->kref))
451 				/*
452 				 * We lost the race, device is almost dead.
453 				 *  Continue traversing to find a valid one.
454 				 */
455 				continue;
456 			spin_unlock(&dev_lock);
457 			return dev;
458 		}
459 	}
460 	list_add(&new_dev->list, &dev_list);
461 	spin_unlock(&dev_lock);
462 
463 	return new_dev;
464 }
465 
466 static int rnbd_srv_check_update_open_perm(struct rnbd_srv_dev *srv_dev,
467 					    struct rnbd_srv_session *srv_sess,
468 					    enum rnbd_access_mode access_mode)
469 {
470 	int ret = -EPERM;
471 
472 	mutex_lock(&srv_dev->lock);
473 
474 	switch (access_mode) {
475 	case RNBD_ACCESS_RO:
476 		ret = 0;
477 		break;
478 	case RNBD_ACCESS_RW:
479 		if (srv_dev->open_write_cnt == 0)  {
480 			srv_dev->open_write_cnt++;
481 			ret = 0;
482 		} else {
483 			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",
484 			       srv_dev->id, srv_sess->sessname,
485 			       srv_dev->open_write_cnt,
486 			       rnbd_access_mode_str(access_mode));
487 		}
488 		break;
489 	case RNBD_ACCESS_MIGRATION:
490 		if (srv_dev->open_write_cnt < 2) {
491 			srv_dev->open_write_cnt++;
492 			ret = 0;
493 		} else {
494 			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",
495 			       srv_dev->id, srv_sess->sessname,
496 			       srv_dev->open_write_cnt,
497 			       rnbd_access_mode_str(access_mode));
498 		}
499 		break;
500 	default:
501 		pr_err("Received mapping request for device '%s' on session %s with invalid access mode: %d\n",
502 		       srv_dev->id, srv_sess->sessname, access_mode);
503 		ret = -EINVAL;
504 	}
505 
506 	mutex_unlock(&srv_dev->lock);
507 
508 	return ret;
509 }
510 
511 static struct rnbd_srv_dev *
512 rnbd_srv_get_or_create_srv_dev(struct block_device *bdev,
513 				struct rnbd_srv_session *srv_sess,
514 				enum rnbd_access_mode access_mode)
515 {
516 	int ret;
517 	struct rnbd_srv_dev *new_dev, *dev;
518 
519 	new_dev = rnbd_srv_init_srv_dev(bdev);
520 	if (IS_ERR(new_dev))
521 		return new_dev;
522 
523 	dev = rnbd_srv_find_or_add_srv_dev(new_dev);
524 	if (dev != new_dev)
525 		kfree(new_dev);
526 
527 	ret = rnbd_srv_check_update_open_perm(dev, srv_sess, access_mode);
528 	if (ret) {
529 		rnbd_put_srv_dev(dev);
530 		return ERR_PTR(ret);
531 	}
532 
533 	return dev;
534 }
535 
536 static void rnbd_srv_fill_msg_open_rsp(struct rnbd_msg_open_rsp *rsp,
537 					struct rnbd_srv_sess_dev *sess_dev)
538 {
539 	struct block_device *bdev = sess_dev->bdev;
540 
541 	rsp->hdr.type = cpu_to_le16(RNBD_MSG_OPEN_RSP);
542 	rsp->device_id = cpu_to_le32(sess_dev->device_id);
543 	rsp->nsectors = cpu_to_le64(bdev_nr_sectors(bdev));
544 	rsp->logical_block_size	= cpu_to_le16(bdev_logical_block_size(bdev));
545 	rsp->physical_block_size = cpu_to_le16(bdev_physical_block_size(bdev));
546 	rsp->max_segments = cpu_to_le16(bdev_max_segments(bdev));
547 	rsp->max_hw_sectors =
548 		cpu_to_le32(queue_max_hw_sectors(bdev_get_queue(bdev)));
549 	rsp->max_write_same_sectors = 0;
550 	rsp->max_discard_sectors = cpu_to_le32(bdev_max_discard_sectors(bdev));
551 	rsp->discard_granularity = cpu_to_le32(bdev_discard_granularity(bdev));
552 	rsp->discard_alignment = cpu_to_le32(bdev_discard_alignment(bdev));
553 	rsp->secure_discard = cpu_to_le16(bdev_max_secure_erase_sectors(bdev));
554 	rsp->cache_policy = 0;
555 	if (bdev_write_cache(bdev))
556 		rsp->cache_policy |= RNBD_WRITEBACK;
557 	if (bdev_fua(bdev))
558 		rsp->cache_policy |= RNBD_FUA;
559 }
560 
561 static struct rnbd_srv_sess_dev *
562 rnbd_srv_create_set_sess_dev(struct rnbd_srv_session *srv_sess,
563 			      const struct rnbd_msg_open *open_msg,
564 			      struct block_device *bdev, fmode_t open_flags,
565 			      struct rnbd_srv_dev *srv_dev)
566 {
567 	struct rnbd_srv_sess_dev *sdev = rnbd_sess_dev_alloc(srv_sess);
568 
569 	if (IS_ERR(sdev))
570 		return sdev;
571 
572 	kref_init(&sdev->kref);
573 
574 	strscpy(sdev->pathname, open_msg->dev_name, sizeof(sdev->pathname));
575 
576 	sdev->bdev		= bdev;
577 	sdev->sess		= srv_sess;
578 	sdev->dev		= srv_dev;
579 	sdev->open_flags	= open_flags;
580 	sdev->access_mode	= open_msg->access_mode;
581 
582 	return sdev;
583 }
584 
585 static char *rnbd_srv_get_full_path(struct rnbd_srv_session *srv_sess,
586 				     const char *dev_name)
587 {
588 	char *full_path;
589 	char *a, *b;
590 
591 	full_path = kmalloc(PATH_MAX, GFP_KERNEL);
592 	if (!full_path)
593 		return ERR_PTR(-ENOMEM);
594 
595 	/*
596 	 * Replace %SESSNAME% with a real session name in order to
597 	 * create device namespace.
598 	 */
599 	a = strnstr(dev_search_path, "%SESSNAME%", sizeof(dev_search_path));
600 	if (a) {
601 		int len = a - dev_search_path;
602 
603 		len = snprintf(full_path, PATH_MAX, "%.*s/%s/%s", len,
604 			       dev_search_path, srv_sess->sessname, dev_name);
605 		if (len >= PATH_MAX) {
606 			pr_err("Too long path: %s, %s, %s\n",
607 			       dev_search_path, srv_sess->sessname, dev_name);
608 			kfree(full_path);
609 			return ERR_PTR(-EINVAL);
610 		}
611 	} else {
612 		snprintf(full_path, PATH_MAX, "%s/%s",
613 			 dev_search_path, dev_name);
614 	}
615 
616 	/* eliminitate duplicated slashes */
617 	a = strchr(full_path, '/');
618 	b = a;
619 	while (*b != '\0') {
620 		if (*b == '/' && *a == '/') {
621 			b++;
622 		} else {
623 			a++;
624 			*a = *b;
625 			b++;
626 		}
627 	}
628 	a++;
629 	*a = '\0';
630 
631 	return full_path;
632 }
633 
634 static int process_msg_sess_info(struct rnbd_srv_session *srv_sess,
635 				 const void *msg, size_t len,
636 				 void *data, size_t datalen)
637 {
638 	const struct rnbd_msg_sess_info *sess_info_msg = msg;
639 	struct rnbd_msg_sess_info_rsp *rsp = data;
640 
641 	srv_sess->ver = min_t(u8, sess_info_msg->ver, RNBD_PROTO_VER_MAJOR);
642 
643 	trace_process_msg_sess_info(srv_sess, sess_info_msg);
644 
645 	rsp->hdr.type = cpu_to_le16(RNBD_MSG_SESS_INFO_RSP);
646 	rsp->ver = srv_sess->ver;
647 
648 	return 0;
649 }
650 
651 /**
652  * find_srv_sess_dev() - a dev is already opened by this name
653  * @srv_sess:	the session to search.
654  * @dev_name:	string containing the name of the device.
655  *
656  * Return struct rnbd_srv_sess_dev if srv_sess already opened the dev_name
657  * NULL if the session didn't open the device yet.
658  */
659 static struct rnbd_srv_sess_dev *
660 find_srv_sess_dev(struct rnbd_srv_session *srv_sess, const char *dev_name)
661 {
662 	struct rnbd_srv_sess_dev *sess_dev;
663 	unsigned long index;
664 
665 	if (xa_empty(&srv_sess->index_idr))
666 		return NULL;
667 
668 	xa_for_each(&srv_sess->index_idr, index, sess_dev)
669 		if (!strcmp(sess_dev->pathname, dev_name))
670 			return sess_dev;
671 
672 	return NULL;
673 }
674 
675 static int process_msg_open(struct rnbd_srv_session *srv_sess,
676 			    const void *msg, size_t len,
677 			    void *data, size_t datalen)
678 {
679 	int ret;
680 	struct rnbd_srv_dev *srv_dev;
681 	struct rnbd_srv_sess_dev *srv_sess_dev;
682 	const struct rnbd_msg_open *open_msg = msg;
683 	struct block_device *bdev;
684 	fmode_t open_flags;
685 	char *full_path;
686 	struct rnbd_msg_open_rsp *rsp = data;
687 
688 	trace_process_msg_open(srv_sess, open_msg);
689 
690 	open_flags = FMODE_READ;
691 	if (open_msg->access_mode != RNBD_ACCESS_RO)
692 		open_flags |= FMODE_WRITE;
693 
694 	mutex_lock(&srv_sess->lock);
695 
696 	srv_sess_dev = find_srv_sess_dev(srv_sess, open_msg->dev_name);
697 	if (srv_sess_dev)
698 		goto fill_response;
699 
700 	if ((strlen(dev_search_path) + strlen(open_msg->dev_name))
701 	    >= PATH_MAX) {
702 		pr_err("Opening device for session %s failed, device path too long. '%s/%s' is longer than PATH_MAX (%d)\n",
703 		       srv_sess->sessname, dev_search_path, open_msg->dev_name,
704 		       PATH_MAX);
705 		ret = -EINVAL;
706 		goto reject;
707 	}
708 	if (strstr(open_msg->dev_name, "..")) {
709 		pr_err("Opening device for session %s failed, device path %s contains relative path ..\n",
710 		       srv_sess->sessname, open_msg->dev_name);
711 		ret = -EINVAL;
712 		goto reject;
713 	}
714 	full_path = rnbd_srv_get_full_path(srv_sess, open_msg->dev_name);
715 	if (IS_ERR(full_path)) {
716 		ret = PTR_ERR(full_path);
717 		pr_err("Opening device '%s' for client %s failed, failed to get device full path, err: %d\n",
718 		       open_msg->dev_name, srv_sess->sessname, ret);
719 		goto reject;
720 	}
721 
722 	bdev = blkdev_get_by_path(full_path, open_flags, THIS_MODULE);
723 	if (IS_ERR(bdev)) {
724 		ret = PTR_ERR(bdev);
725 		pr_err("Opening device '%s' on session %s failed, failed to open the block device, err: %d\n",
726 		       full_path, srv_sess->sessname, ret);
727 		goto free_path;
728 	}
729 
730 	srv_dev = rnbd_srv_get_or_create_srv_dev(bdev, srv_sess,
731 						  open_msg->access_mode);
732 	if (IS_ERR(srv_dev)) {
733 		pr_err("Opening device '%s' on session %s failed, creating srv_dev failed, err: %ld\n",
734 		       full_path, srv_sess->sessname, PTR_ERR(srv_dev));
735 		ret = PTR_ERR(srv_dev);
736 		goto blkdev_put;
737 	}
738 
739 	srv_sess_dev = rnbd_srv_create_set_sess_dev(srv_sess, open_msg,
740 						     bdev, open_flags,
741 						     srv_dev);
742 	if (IS_ERR(srv_sess_dev)) {
743 		pr_err("Opening device '%s' on session %s failed, creating sess_dev failed, err: %ld\n",
744 		       full_path, srv_sess->sessname, PTR_ERR(srv_sess_dev));
745 		ret = PTR_ERR(srv_sess_dev);
746 		goto srv_dev_put;
747 	}
748 
749 	/* Create the srv_dev sysfs files if they haven't been created yet. The
750 	 * reason to delay the creation is not to create the sysfs files before
751 	 * we are sure the device can be opened.
752 	 */
753 	mutex_lock(&srv_dev->lock);
754 	if (!srv_dev->dev_kobj.state_in_sysfs) {
755 		ret = rnbd_srv_create_dev_sysfs(srv_dev, bdev);
756 		if (ret) {
757 			mutex_unlock(&srv_dev->lock);
758 			rnbd_srv_err(srv_sess_dev,
759 				      "Opening device failed, failed to create device sysfs files, err: %d\n",
760 				      ret);
761 			goto free_srv_sess_dev;
762 		}
763 	}
764 
765 	ret = rnbd_srv_create_dev_session_sysfs(srv_sess_dev);
766 	if (ret) {
767 		mutex_unlock(&srv_dev->lock);
768 		rnbd_srv_err(srv_sess_dev,
769 			      "Opening device failed, failed to create dev client sysfs files, err: %d\n",
770 			      ret);
771 		goto free_srv_sess_dev;
772 	}
773 
774 	list_add(&srv_sess_dev->dev_list, &srv_dev->sess_dev_list);
775 	mutex_unlock(&srv_dev->lock);
776 
777 	rnbd_srv_info(srv_sess_dev, "Opened device '%s'\n", srv_dev->id);
778 
779 	kfree(full_path);
780 
781 fill_response:
782 	rnbd_srv_fill_msg_open_rsp(rsp, srv_sess_dev);
783 	mutex_unlock(&srv_sess->lock);
784 	return 0;
785 
786 free_srv_sess_dev:
787 	xa_erase(&srv_sess->index_idr, srv_sess_dev->device_id);
788 	synchronize_rcu();
789 	kfree(srv_sess_dev);
790 srv_dev_put:
791 	if (open_msg->access_mode != RNBD_ACCESS_RO) {
792 		mutex_lock(&srv_dev->lock);
793 		srv_dev->open_write_cnt--;
794 		mutex_unlock(&srv_dev->lock);
795 	}
796 	rnbd_put_srv_dev(srv_dev);
797 blkdev_put:
798 	blkdev_put(bdev, open_flags);
799 free_path:
800 	kfree(full_path);
801 reject:
802 	mutex_unlock(&srv_sess->lock);
803 	return ret;
804 }
805 
806 static struct rtrs_srv_ctx *rtrs_ctx;
807 
808 static struct rtrs_srv_ops rtrs_ops;
809 static int __init rnbd_srv_init_module(void)
810 {
811 	int err;
812 
813 	BUILD_BUG_ON(sizeof(struct rnbd_msg_hdr) != 4);
814 	BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info) != 36);
815 	BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info_rsp) != 36);
816 	BUILD_BUG_ON(sizeof(struct rnbd_msg_open) != 264);
817 	BUILD_BUG_ON(sizeof(struct rnbd_msg_close) != 8);
818 	BUILD_BUG_ON(sizeof(struct rnbd_msg_open_rsp) != 56);
819 	rtrs_ops = (struct rtrs_srv_ops) {
820 		.rdma_ev = rnbd_srv_rdma_ev,
821 		.link_ev = rnbd_srv_link_ev,
822 	};
823 	rtrs_ctx = rtrs_srv_open(&rtrs_ops, port_nr);
824 	if (IS_ERR(rtrs_ctx)) {
825 		err = PTR_ERR(rtrs_ctx);
826 		pr_err("rtrs_srv_open(), err: %d\n", err);
827 		return err;
828 	}
829 
830 	err = rnbd_srv_create_sysfs_files();
831 	if (err) {
832 		pr_err("rnbd_srv_create_sysfs_files(), err: %d\n", err);
833 		rtrs_srv_close(rtrs_ctx);
834 		return err;
835 	}
836 
837 	return 0;
838 }
839 
840 static void __exit rnbd_srv_cleanup_module(void)
841 {
842 	rtrs_srv_close(rtrs_ctx);
843 	WARN_ON(!list_empty(&sess_list));
844 	rnbd_srv_destroy_sysfs_files();
845 }
846 
847 module_init(rnbd_srv_init_module);
848 module_exit(rnbd_srv_cleanup_module);
849