xref: /linux/drivers/block/rnbd/rnbd-srv.c (revision 26bb0d3f38a764b743a3ad5c8b6e5b5044d7ceb4)
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 
dev_search_path_set(const char * val,const struct kernel_param * kp)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 
rnbd_sess_dev_release(struct kref * kref)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 
rnbd_put_sess_dev(struct rnbd_srv_sess_dev * sess_dev)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 *
rnbd_get_sess_dev(int dev_id,struct rnbd_srv_session * srv_sess)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 (!ret)
100 		return ERR_PTR(-ENXIO);
101 
102 	return sess_dev;
103 }
104 
rnbd_dev_bi_end_io(struct bio * bio)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 
process_rdma(struct rnbd_srv_session * srv_sess,struct rtrs_srv_op * id,void * data,u32 datalen,const void * usr,size_t usrlen)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: %pe\n",
140 				   srv_sess->sessname, dev_id, sess_dev);
141 		err = -ENOTCONN;
142 		goto err;
143 	}
144 
145 	priv->sess_dev = sess_dev;
146 	priv->id = id;
147 
148 	bio = bio_alloc(file_bdev(sess_dev->bdev_file), 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_rl(sess_dev, "Failed to map data to bio\n");
153 		err = -EINVAL;
154 		goto bio_put;
155 	}
156 
157 	bio->bi_opf = rnbd_to_bio_flags(le32_to_cpu(msg->rw));
158 	if (bio_has_data(bio) &&
159 	    bio->bi_iter.bi_size != le32_to_cpu(msg->bi_size)) {
160 		rnbd_srv_err_rl(sess_dev, "Datalen mismatch:  bio bi_size (%u), bi_size (%u)\n",
161 				bio->bi_iter.bi_size, msg->bi_size);
162 		err = -EINVAL;
163 		goto bio_put;
164 	}
165 	bio->bi_end_io = rnbd_dev_bi_end_io;
166 	bio->bi_private = priv;
167 	bio->bi_iter.bi_sector = le64_to_cpu(msg->sector);
168 	prio = srv_sess->ver < RNBD_PROTO_VER_MAJOR ||
169 	       usrlen < sizeof(*msg) ? 0 : le16_to_cpu(msg->prio);
170 	bio_set_prio(bio, prio);
171 
172 	submit_bio(bio);
173 
174 	return 0;
175 
176 bio_put:
177 	bio_put(bio);
178 	rnbd_put_sess_dev(sess_dev);
179 err:
180 	kfree(priv);
181 	return err;
182 }
183 
destroy_device(struct kref * kref)184 static void destroy_device(struct kref *kref)
185 {
186 	struct rnbd_srv_dev *dev = container_of(kref, struct rnbd_srv_dev, kref);
187 
188 	WARN_ONCE(!list_empty(&dev->sess_dev_list),
189 		  "Device %s is being destroyed but still in use!\n",
190 		  dev->name);
191 
192 	spin_lock(&dev_lock);
193 	list_del(&dev->list);
194 	spin_unlock(&dev_lock);
195 
196 	mutex_destroy(&dev->lock);
197 	if (dev->dev_kobj.state_in_sysfs)
198 		/*
199 		 * Destroy kobj only if it was really created.
200 		 */
201 		rnbd_srv_destroy_dev_sysfs(dev);
202 	else
203 		kfree(dev);
204 }
205 
rnbd_put_srv_dev(struct rnbd_srv_dev * dev)206 static void rnbd_put_srv_dev(struct rnbd_srv_dev *dev)
207 {
208 	kref_put(&dev->kref, destroy_device);
209 }
210 
rnbd_destroy_sess_dev(struct rnbd_srv_sess_dev * sess_dev,bool keep_id)211 void rnbd_destroy_sess_dev(struct rnbd_srv_sess_dev *sess_dev, bool keep_id)
212 {
213 	DECLARE_COMPLETION_ONSTACK(dc);
214 
215 	if (keep_id)
216 		/* free the resources for the id but don't  */
217 		/* allow to re-use the id itself because it */
218 		/* is still used by the client              */
219 		xa_cmpxchg(&sess_dev->sess->index_idr, sess_dev->device_id,
220 			   sess_dev, NULL, 0);
221 	else
222 		xa_erase(&sess_dev->sess->index_idr, sess_dev->device_id);
223 	synchronize_rcu();
224 
225 	sess_dev->destroy_comp = &dc;
226 	rnbd_put_sess_dev(sess_dev);
227 	wait_for_completion(&dc); /* wait for inflights to drop to zero */
228 
229 	fput(sess_dev->bdev_file);
230 	mutex_lock(&sess_dev->dev->lock);
231 	list_del(&sess_dev->dev_list);
232 	if (!sess_dev->readonly)
233 		sess_dev->dev->open_write_cnt--;
234 	mutex_unlock(&sess_dev->dev->lock);
235 
236 	rnbd_put_srv_dev(sess_dev->dev);
237 
238 	rnbd_srv_info(sess_dev, "Device closed\n");
239 	kfree(sess_dev);
240 }
241 
destroy_sess(struct rnbd_srv_session * srv_sess)242 static void destroy_sess(struct rnbd_srv_session *srv_sess)
243 {
244 	struct rnbd_srv_sess_dev *sess_dev;
245 	unsigned long index;
246 
247 	if (xa_empty(&srv_sess->index_idr))
248 		goto out;
249 
250 	trace_destroy_sess(srv_sess);
251 
252 	mutex_lock(&srv_sess->lock);
253 	xa_for_each(&srv_sess->index_idr, index, sess_dev)
254 		rnbd_srv_destroy_dev_session_sysfs(sess_dev);
255 	mutex_unlock(&srv_sess->lock);
256 
257 out:
258 	xa_destroy(&srv_sess->index_idr);
259 
260 	pr_info("RTRS Session %s disconnected\n", srv_sess->sessname);
261 
262 	mutex_lock(&sess_lock);
263 	list_del(&srv_sess->list);
264 	mutex_unlock(&sess_lock);
265 
266 	mutex_destroy(&srv_sess->lock);
267 	kfree(srv_sess);
268 }
269 
create_sess(struct rtrs_srv_sess * rtrs)270 static int create_sess(struct rtrs_srv_sess *rtrs)
271 {
272 	struct rnbd_srv_session *srv_sess;
273 	char pathname[NAME_MAX];
274 	int err;
275 
276 	err = rtrs_srv_get_path_name(rtrs, pathname, sizeof(pathname));
277 	if (err) {
278 		pr_err("rtrs_srv_get_path_name(%s): %d\n", pathname, err);
279 
280 		return err;
281 	}
282 	srv_sess = kzalloc(sizeof(*srv_sess), GFP_KERNEL);
283 	if (!srv_sess)
284 		return -ENOMEM;
285 
286 	srv_sess->queue_depth = rtrs_srv_get_queue_depth(rtrs);
287 	xa_init_flags(&srv_sess->index_idr, XA_FLAGS_ALLOC);
288 	mutex_init(&srv_sess->lock);
289 	mutex_lock(&sess_lock);
290 	list_add(&srv_sess->list, &sess_list);
291 	mutex_unlock(&sess_lock);
292 
293 	srv_sess->rtrs = rtrs;
294 	strscpy(srv_sess->sessname, pathname, sizeof(srv_sess->sessname));
295 
296 	rtrs_srv_set_sess_priv(rtrs, srv_sess);
297 
298 	trace_create_sess(srv_sess);
299 
300 	return 0;
301 }
302 
rnbd_srv_link_ev(struct rtrs_srv_sess * rtrs,enum rtrs_srv_link_ev ev,void * priv)303 static int rnbd_srv_link_ev(struct rtrs_srv_sess *rtrs,
304 			     enum rtrs_srv_link_ev ev, void *priv)
305 {
306 	struct rnbd_srv_session *srv_sess = priv;
307 
308 	switch (ev) {
309 	case RTRS_SRV_LINK_EV_CONNECTED:
310 		return create_sess(rtrs);
311 
312 	case RTRS_SRV_LINK_EV_DISCONNECTED:
313 		if (WARN_ON_ONCE(!srv_sess))
314 			return -EINVAL;
315 
316 		destroy_sess(srv_sess);
317 		return 0;
318 
319 	default:
320 		pr_warn("Received unknown RTRS session event %d from session %s\n",
321 			ev, srv_sess->sessname);
322 		return -EINVAL;
323 	}
324 }
325 
rnbd_srv_sess_dev_force_close(struct rnbd_srv_sess_dev * sess_dev,struct kobj_attribute * attr)326 void rnbd_srv_sess_dev_force_close(struct rnbd_srv_sess_dev *sess_dev,
327 				   struct kobj_attribute *attr)
328 {
329 	struct rnbd_srv_session	*sess = sess_dev->sess;
330 
331 	/* It is already started to close by client's close message. */
332 	if (!mutex_trylock(&sess->lock))
333 		return;
334 
335 	sess_dev->keep_id = true;
336 	/* first remove sysfs itself to avoid deadlock */
337 	sysfs_remove_file_self(&sess_dev->kobj, &attr->attr);
338 	rnbd_srv_destroy_dev_session_sysfs(sess_dev);
339 	mutex_unlock(&sess->lock);
340 }
341 
process_msg_close(struct rnbd_srv_session * srv_sess,void * data,size_t datalen,const void * usr,size_t usrlen)342 static void process_msg_close(struct rnbd_srv_session *srv_sess,
343 			     void *data, size_t datalen, const void *usr,
344 			     size_t usrlen)
345 {
346 	const struct rnbd_msg_close *close_msg = usr;
347 	struct rnbd_srv_sess_dev *sess_dev;
348 
349 	trace_process_msg_close(srv_sess, close_msg);
350 
351 	sess_dev = rnbd_get_sess_dev(le32_to_cpu(close_msg->device_id),
352 				      srv_sess);
353 	if (IS_ERR(sess_dev))
354 		return;
355 
356 	rnbd_put_sess_dev(sess_dev);
357 	mutex_lock(&srv_sess->lock);
358 	rnbd_srv_destroy_dev_session_sysfs(sess_dev);
359 	mutex_unlock(&srv_sess->lock);
360 }
361 
362 static int process_msg_open(struct rnbd_srv_session *srv_sess,
363 			    const void *msg, size_t len,
364 			    void *data, size_t datalen);
365 
366 static void process_msg_sess_info(struct rnbd_srv_session *srv_sess,
367 				 const void *msg, size_t len,
368 				 void *data, size_t datalen);
369 
rnbd_srv_rdma_ev(void * priv,struct rtrs_srv_op * id,void * data,size_t datalen,const void * usr,size_t usrlen)370 static int rnbd_srv_rdma_ev(void *priv, struct rtrs_srv_op *id,
371 			    void *data, size_t datalen,
372 			    const void *usr, size_t usrlen)
373 {
374 	struct rnbd_srv_session *srv_sess = priv;
375 	const struct rnbd_msg_hdr *hdr = usr;
376 	int ret = 0;
377 	u16 type;
378 
379 	if (WARN_ON_ONCE(!srv_sess))
380 		return -ENODEV;
381 
382 	type = le16_to_cpu(hdr->type);
383 
384 	switch (type) {
385 	case RNBD_MSG_IO:
386 		return process_rdma(srv_sess, id, data, datalen, usr, usrlen);
387 	case RNBD_MSG_CLOSE:
388 		process_msg_close(srv_sess, data, datalen, usr, usrlen);
389 		break;
390 	case RNBD_MSG_OPEN:
391 		ret = process_msg_open(srv_sess, usr, usrlen, data, datalen);
392 		break;
393 	case RNBD_MSG_SESS_INFO:
394 		process_msg_sess_info(srv_sess, usr, usrlen, data, datalen);
395 		break;
396 	default:
397 		pr_warn("Received unexpected message type %d from session %s\n",
398 			type, srv_sess->sessname);
399 		return -EINVAL;
400 	}
401 
402 	/*
403 	 * Since ret is passed to rtrs to handle the failure case, we
404 	 * just return 0 at the end otherwise callers in rtrs would call
405 	 * send_io_resp_imm again to print redundant err message.
406 	 */
407 	rtrs_srv_resp_rdma(id, ret);
408 	return 0;
409 }
410 
411 static struct rnbd_srv_sess_dev
rnbd_sess_dev_alloc(struct rnbd_srv_session * srv_sess)412 *rnbd_sess_dev_alloc(struct rnbd_srv_session *srv_sess)
413 {
414 	struct rnbd_srv_sess_dev *sess_dev;
415 	int error;
416 
417 	sess_dev = kzalloc(sizeof(*sess_dev), GFP_KERNEL);
418 	if (!sess_dev)
419 		return ERR_PTR(-ENOMEM);
420 
421 	error = xa_alloc(&srv_sess->index_idr, &sess_dev->device_id, sess_dev,
422 			 xa_limit_32b, GFP_NOWAIT);
423 	if (error < 0) {
424 		pr_warn("Allocating idr failed, err: %d\n", error);
425 		kfree(sess_dev);
426 		return ERR_PTR(error);
427 	}
428 
429 	return sess_dev;
430 }
431 
rnbd_srv_init_srv_dev(struct block_device * bdev)432 static struct rnbd_srv_dev *rnbd_srv_init_srv_dev(struct block_device *bdev)
433 {
434 	struct rnbd_srv_dev *dev;
435 
436 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
437 	if (!dev)
438 		return ERR_PTR(-ENOMEM);
439 
440 	snprintf(dev->name, sizeof(dev->name), "%pg", bdev);
441 	kref_init(&dev->kref);
442 	INIT_LIST_HEAD(&dev->sess_dev_list);
443 	mutex_init(&dev->lock);
444 
445 	return dev;
446 }
447 
448 static struct rnbd_srv_dev *
rnbd_srv_find_or_add_srv_dev(struct rnbd_srv_dev * new_dev)449 rnbd_srv_find_or_add_srv_dev(struct rnbd_srv_dev *new_dev)
450 {
451 	struct rnbd_srv_dev *dev;
452 
453 	spin_lock(&dev_lock);
454 	list_for_each_entry(dev, &dev_list, list) {
455 		if (!strncmp(dev->name, new_dev->name, sizeof(dev->name))) {
456 			if (!kref_get_unless_zero(&dev->kref))
457 				/*
458 				 * We lost the race, device is almost dead.
459 				 *  Continue traversing to find a valid one.
460 				 */
461 				continue;
462 			spin_unlock(&dev_lock);
463 			return dev;
464 		}
465 	}
466 	list_add(&new_dev->list, &dev_list);
467 	spin_unlock(&dev_lock);
468 
469 	return new_dev;
470 }
471 
rnbd_srv_check_update_open_perm(struct rnbd_srv_dev * srv_dev,struct rnbd_srv_session * srv_sess,enum rnbd_access_mode access_mode)472 static int rnbd_srv_check_update_open_perm(struct rnbd_srv_dev *srv_dev,
473 					    struct rnbd_srv_session *srv_sess,
474 					    enum rnbd_access_mode access_mode)
475 {
476 	int ret = 0;
477 
478 	mutex_lock(&srv_dev->lock);
479 
480 	switch (access_mode) {
481 	case RNBD_ACCESS_RO:
482 		break;
483 	case RNBD_ACCESS_RW:
484 		if (srv_dev->open_write_cnt == 0)  {
485 			srv_dev->open_write_cnt++;
486 		} else {
487 			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",
488 			       srv_dev->name, srv_sess->sessname,
489 			       srv_dev->open_write_cnt,
490 			       rnbd_access_modes[access_mode].str);
491 			ret = -EPERM;
492 		}
493 		break;
494 	case RNBD_ACCESS_MIGRATION:
495 		if (srv_dev->open_write_cnt < 2) {
496 			srv_dev->open_write_cnt++;
497 		} else {
498 			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",
499 			       srv_dev->name, srv_sess->sessname,
500 			       srv_dev->open_write_cnt,
501 			       rnbd_access_modes[access_mode].str);
502 			ret = -EPERM;
503 		}
504 		break;
505 	default:
506 		pr_err("Received mapping request for device '%s' on session %s with invalid access mode: %d\n",
507 		       srv_dev->name, srv_sess->sessname, access_mode);
508 		ret = -EINVAL;
509 	}
510 
511 	mutex_unlock(&srv_dev->lock);
512 
513 	return ret;
514 }
515 
516 static struct rnbd_srv_dev *
rnbd_srv_get_or_create_srv_dev(struct block_device * bdev,struct rnbd_srv_session * srv_sess,enum rnbd_access_mode access_mode)517 rnbd_srv_get_or_create_srv_dev(struct block_device *bdev,
518 				struct rnbd_srv_session *srv_sess,
519 				enum rnbd_access_mode access_mode)
520 {
521 	int ret;
522 	struct rnbd_srv_dev *new_dev, *dev;
523 
524 	new_dev = rnbd_srv_init_srv_dev(bdev);
525 	if (IS_ERR(new_dev))
526 		return new_dev;
527 
528 	dev = rnbd_srv_find_or_add_srv_dev(new_dev);
529 	if (dev != new_dev)
530 		kfree(new_dev);
531 
532 	ret = rnbd_srv_check_update_open_perm(dev, srv_sess, access_mode);
533 	if (ret) {
534 		rnbd_put_srv_dev(dev);
535 		return ERR_PTR(ret);
536 	}
537 
538 	return dev;
539 }
540 
rnbd_srv_fill_msg_open_rsp(struct rnbd_msg_open_rsp * rsp,struct rnbd_srv_sess_dev * sess_dev)541 static void rnbd_srv_fill_msg_open_rsp(struct rnbd_msg_open_rsp *rsp,
542 					struct rnbd_srv_sess_dev *sess_dev)
543 {
544 	struct block_device *bdev = file_bdev(sess_dev->bdev_file);
545 
546 	rsp->hdr.type = cpu_to_le16(RNBD_MSG_OPEN_RSP);
547 	rsp->device_id = cpu_to_le32(sess_dev->device_id);
548 	rsp->nsectors = cpu_to_le64(bdev_nr_sectors(bdev));
549 	rsp->logical_block_size	= cpu_to_le16(bdev_logical_block_size(bdev));
550 	rsp->physical_block_size = cpu_to_le16(bdev_physical_block_size(bdev));
551 	rsp->max_segments = cpu_to_le16(bdev_max_segments(bdev));
552 	rsp->max_hw_sectors =
553 		cpu_to_le32(queue_max_hw_sectors(bdev_get_queue(bdev)));
554 	rsp->max_write_zeroes_sectors =
555 		cpu_to_le32(bdev_write_zeroes_sectors(bdev));
556 	rsp->max_discard_sectors = cpu_to_le32(bdev_max_discard_sectors(bdev));
557 	rsp->discard_granularity = cpu_to_le32(bdev_discard_granularity(bdev));
558 	rsp->discard_alignment = cpu_to_le32(bdev_discard_alignment(bdev));
559 	rsp->secure_discard = cpu_to_le16(bdev_max_secure_erase_sectors(bdev));
560 	rsp->cache_policy = 0;
561 	if (bdev_write_cache(bdev))
562 		rsp->cache_policy |= RNBD_WRITEBACK;
563 	if (bdev_fua(bdev))
564 		rsp->cache_policy |= RNBD_FUA;
565 }
566 
567 static struct rnbd_srv_sess_dev *
rnbd_srv_create_set_sess_dev(struct rnbd_srv_session * srv_sess,const struct rnbd_msg_open * open_msg,struct file * bdev_file,bool readonly,struct rnbd_srv_dev * srv_dev)568 rnbd_srv_create_set_sess_dev(struct rnbd_srv_session *srv_sess,
569 			      const struct rnbd_msg_open *open_msg,
570 			      struct file *bdev_file, bool readonly,
571 			      struct rnbd_srv_dev *srv_dev)
572 {
573 	struct rnbd_srv_sess_dev *sdev = rnbd_sess_dev_alloc(srv_sess);
574 
575 	if (IS_ERR(sdev))
576 		return sdev;
577 
578 	kref_init(&sdev->kref);
579 
580 	strscpy(sdev->pathname, open_msg->dev_name, sizeof(sdev->pathname));
581 
582 	sdev->bdev_file		= bdev_file;
583 	sdev->sess		= srv_sess;
584 	sdev->dev		= srv_dev;
585 	sdev->readonly		= readonly;
586 	sdev->access_mode	= open_msg->access_mode;
587 
588 	return sdev;
589 }
590 
rnbd_srv_get_full_path(struct rnbd_srv_session * srv_sess,const char * dev_name)591 static char *rnbd_srv_get_full_path(struct rnbd_srv_session *srv_sess,
592 				     const char *dev_name)
593 {
594 	char *full_path;
595 	char *a, *b;
596 	int len;
597 
598 	full_path = kmalloc(PATH_MAX, GFP_KERNEL);
599 	if (!full_path)
600 		return ERR_PTR(-ENOMEM);
601 
602 	/*
603 	 * Replace %SESSNAME% with a real session name in order to
604 	 * create device namespace.
605 	 */
606 	a = strnstr(dev_search_path, "%SESSNAME%", sizeof(dev_search_path));
607 	if (a) {
608 		len = a - dev_search_path;
609 
610 		len = snprintf(full_path, PATH_MAX, "%.*s/%s/%s", len,
611 			       dev_search_path, srv_sess->sessname, dev_name);
612 	} else {
613 		len = snprintf(full_path, PATH_MAX, "%s/%s",
614 			       dev_search_path, dev_name);
615 	}
616 	if (len >= PATH_MAX) {
617 		pr_err("Too long path: %s, %s, %s\n",
618 		       dev_search_path, srv_sess->sessname, dev_name);
619 		kfree(full_path);
620 		return ERR_PTR(-EINVAL);
621 	}
622 
623 	/* eliminitate duplicated slashes */
624 	a = strchr(full_path, '/');
625 	b = a;
626 	while (*b != '\0') {
627 		if (*b == '/' && *a == '/') {
628 			b++;
629 		} else {
630 			a++;
631 			*a = *b;
632 			b++;
633 		}
634 	}
635 	a++;
636 	*a = '\0';
637 
638 	return full_path;
639 }
640 
process_msg_sess_info(struct rnbd_srv_session * srv_sess,const void * msg,size_t len,void * data,size_t datalen)641 static void process_msg_sess_info(struct rnbd_srv_session *srv_sess,
642 				 const void *msg, size_t len,
643 				 void *data, size_t datalen)
644 {
645 	const struct rnbd_msg_sess_info *sess_info_msg = msg;
646 	struct rnbd_msg_sess_info_rsp *rsp = data;
647 
648 	srv_sess->ver = min_t(u8, sess_info_msg->ver, RNBD_PROTO_VER_MAJOR);
649 
650 	trace_process_msg_sess_info(srv_sess, sess_info_msg);
651 
652 	rsp->hdr.type = cpu_to_le16(RNBD_MSG_SESS_INFO_RSP);
653 	rsp->ver = srv_sess->ver;
654 }
655 
656 /**
657  * find_srv_sess_dev() - a dev is already opened by this name
658  * @srv_sess:	the session to search.
659  * @dev_name:	string containing the name of the device.
660  *
661  * Return struct rnbd_srv_sess_dev if srv_sess already opened the dev_name
662  * NULL if the session didn't open the device yet.
663  */
664 static struct rnbd_srv_sess_dev *
find_srv_sess_dev(struct rnbd_srv_session * srv_sess,const char * dev_name)665 find_srv_sess_dev(struct rnbd_srv_session *srv_sess, const char *dev_name)
666 {
667 	struct rnbd_srv_sess_dev *sess_dev;
668 	unsigned long index;
669 
670 	if (xa_empty(&srv_sess->index_idr))
671 		return NULL;
672 
673 	xa_for_each(&srv_sess->index_idr, index, sess_dev)
674 		if (!strcmp(sess_dev->pathname, dev_name))
675 			return sess_dev;
676 
677 	return NULL;
678 }
679 
process_msg_open(struct rnbd_srv_session * srv_sess,const void * msg,size_t len,void * data,size_t datalen)680 static int process_msg_open(struct rnbd_srv_session *srv_sess,
681 			    const void *msg, size_t len,
682 			    void *data, size_t datalen)
683 {
684 	int ret;
685 	struct rnbd_srv_dev *srv_dev;
686 	struct rnbd_srv_sess_dev *srv_sess_dev;
687 	const struct rnbd_msg_open *open_msg = msg;
688 	struct file *bdev_file;
689 	blk_mode_t open_flags = BLK_OPEN_READ;
690 	char *full_path;
691 	struct rnbd_msg_open_rsp *rsp = data;
692 
693 	trace_process_msg_open(srv_sess, open_msg);
694 
695 	if (open_msg->access_mode != RNBD_ACCESS_RO)
696 		open_flags |= BLK_OPEN_WRITE;
697 
698 	mutex_lock(&srv_sess->lock);
699 
700 	srv_sess_dev = find_srv_sess_dev(srv_sess, open_msg->dev_name);
701 	if (srv_sess_dev)
702 		goto fill_response;
703 
704 	if ((strlen(dev_search_path) + strlen(open_msg->dev_name))
705 	    >= PATH_MAX) {
706 		pr_err("Opening device for session %s failed, device path too long. '%s/%s' is longer than PATH_MAX (%d)\n",
707 		       srv_sess->sessname, dev_search_path, open_msg->dev_name,
708 		       PATH_MAX);
709 		ret = -EINVAL;
710 		goto reject;
711 	}
712 	if (strstr(open_msg->dev_name, "..")) {
713 		pr_err("Opening device for session %s failed, device path %s contains relative path ..\n",
714 		       srv_sess->sessname, open_msg->dev_name);
715 		ret = -EINVAL;
716 		goto reject;
717 	}
718 	full_path = rnbd_srv_get_full_path(srv_sess, open_msg->dev_name);
719 	if (IS_ERR(full_path)) {
720 		ret = PTR_ERR(full_path);
721 		pr_err("Opening device '%s' for client %s failed, failed to get device full path, err: %pe\n",
722 		       open_msg->dev_name, srv_sess->sessname, full_path);
723 		goto reject;
724 	}
725 
726 	bdev_file = bdev_file_open_by_path(full_path, open_flags, NULL, NULL);
727 	if (IS_ERR(bdev_file)) {
728 		ret = PTR_ERR(bdev_file);
729 		pr_err("Opening device '%s' on session %s failed, failed to open the block device, err: %pe\n",
730 		       full_path, srv_sess->sessname, bdev_file);
731 		goto free_path;
732 	}
733 
734 	srv_dev = rnbd_srv_get_or_create_srv_dev(file_bdev(bdev_file), srv_sess,
735 						  open_msg->access_mode);
736 	if (IS_ERR(srv_dev)) {
737 		pr_err("Opening device '%s' on session %s failed, creating srv_dev failed, err: %pe\n",
738 		       full_path, srv_sess->sessname, srv_dev);
739 		ret = PTR_ERR(srv_dev);
740 		goto blkdev_put;
741 	}
742 
743 	srv_sess_dev = rnbd_srv_create_set_sess_dev(srv_sess, open_msg,
744 				bdev_file,
745 				open_msg->access_mode == RNBD_ACCESS_RO,
746 				srv_dev);
747 	if (IS_ERR(srv_sess_dev)) {
748 		pr_err("Opening device '%s' on session %s failed, creating sess_dev failed, err: %pe\n",
749 		       full_path, srv_sess->sessname, srv_sess_dev);
750 		ret = PTR_ERR(srv_sess_dev);
751 		goto srv_dev_put;
752 	}
753 
754 	/* Create the srv_dev sysfs files if they haven't been created yet. The
755 	 * reason to delay the creation is not to create the sysfs files before
756 	 * we are sure the device can be opened.
757 	 */
758 	mutex_lock(&srv_dev->lock);
759 	if (!srv_dev->dev_kobj.state_in_sysfs) {
760 		ret = rnbd_srv_create_dev_sysfs(srv_dev, file_bdev(bdev_file));
761 		if (ret) {
762 			mutex_unlock(&srv_dev->lock);
763 			rnbd_srv_err(srv_sess_dev,
764 				      "Opening device failed, failed to create device sysfs files, err: %d\n",
765 				      ret);
766 			goto free_srv_sess_dev;
767 		}
768 	}
769 
770 	ret = rnbd_srv_create_dev_session_sysfs(srv_sess_dev);
771 	if (ret) {
772 		mutex_unlock(&srv_dev->lock);
773 		rnbd_srv_err(srv_sess_dev,
774 			      "Opening device failed, failed to create dev client sysfs files, err: %d\n",
775 			      ret);
776 		goto free_srv_sess_dev;
777 	}
778 
779 	list_add(&srv_sess_dev->dev_list, &srv_dev->sess_dev_list);
780 	mutex_unlock(&srv_dev->lock);
781 
782 	rnbd_srv_info(srv_sess_dev, "Opened device '%s'\n", srv_dev->name);
783 
784 	kfree(full_path);
785 
786 fill_response:
787 	rnbd_srv_fill_msg_open_rsp(rsp, srv_sess_dev);
788 	mutex_unlock(&srv_sess->lock);
789 	return 0;
790 
791 free_srv_sess_dev:
792 	xa_erase(&srv_sess->index_idr, srv_sess_dev->device_id);
793 	synchronize_rcu();
794 	kfree(srv_sess_dev);
795 srv_dev_put:
796 	if (open_msg->access_mode != RNBD_ACCESS_RO) {
797 		mutex_lock(&srv_dev->lock);
798 		srv_dev->open_write_cnt--;
799 		mutex_unlock(&srv_dev->lock);
800 	}
801 	rnbd_put_srv_dev(srv_dev);
802 blkdev_put:
803 	fput(bdev_file);
804 free_path:
805 	kfree(full_path);
806 reject:
807 	mutex_unlock(&srv_sess->lock);
808 	return ret;
809 }
810 
811 static struct rtrs_srv_ctx *rtrs_ctx;
812 
813 static struct rtrs_srv_ops rtrs_ops;
rnbd_srv_init_module(void)814 static int __init rnbd_srv_init_module(void)
815 {
816 	int err = 0;
817 
818 	BUILD_BUG_ON(sizeof(struct rnbd_msg_hdr) != 4);
819 	BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info) != 36);
820 	BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info_rsp) != 36);
821 	BUILD_BUG_ON(sizeof(struct rnbd_msg_open) != 264);
822 	BUILD_BUG_ON(sizeof(struct rnbd_msg_close) != 8);
823 	BUILD_BUG_ON(sizeof(struct rnbd_msg_open_rsp) != 56);
824 	rtrs_ops = (struct rtrs_srv_ops) {
825 		.rdma_ev = rnbd_srv_rdma_ev,
826 		.link_ev = rnbd_srv_link_ev,
827 	};
828 	rtrs_ctx = rtrs_srv_open(&rtrs_ops, port_nr);
829 	if (IS_ERR(rtrs_ctx)) {
830 		pr_err("rtrs_srv_open(), err: %pe\n", rtrs_ctx);
831 		return PTR_ERR(rtrs_ctx);
832 	}
833 
834 	err = rnbd_srv_create_sysfs_files();
835 	if (err) {
836 		pr_err("rnbd_srv_create_sysfs_files(), err: %d\n", err);
837 		rtrs_srv_close(rtrs_ctx);
838 	}
839 
840 	return err;
841 }
842 
rnbd_srv_cleanup_module(void)843 static void __exit rnbd_srv_cleanup_module(void)
844 {
845 	rtrs_srv_close(rtrs_ctx);
846 	WARN_ON(!list_empty(&sess_list));
847 	rnbd_srv_destroy_sysfs_files();
848 }
849 
850 module_init(rnbd_srv_init_module);
851 module_exit(rnbd_srv_cleanup_module);
852