1 /*
2  * Driver giving user-space access to the kernel's xenbus connection
3  * to xenstore.
4  *
5  * Copyright (c) 2005, Christian Limpach
6  * Copyright (c) 2005, Rusty Russell, IBM Corporation
7  *
8  * This program is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU General Public License version 2
10  * as published by the Free Software Foundation; or, when distributed
11  * separately from the Linux kernel or incorporated into other
12  * software packages, subject to the following license:
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this source file (the "Software"), to deal in the Software without
16  * restriction, including without limitation the rights to use, copy, modify,
17  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
18  * and to permit persons to whom the Software is furnished to do so, subject to
19  * the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
27  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
29  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
30  * IN THE SOFTWARE.
31  *
32  * Changes:
33  * 2008-10-07  Alex Zeffertt    Replaced /proc/xen/xenbus with xenfs filesystem
34  *                              and /proc/xen compatibility mount point.
35  *                              Turned xenfs into a loadable module.
36  */
37 
38 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 
40 #include <linux/kernel.h>
41 #include <linux/errno.h>
42 #include <linux/uio.h>
43 #include <linux/notifier.h>
44 #include <linux/wait.h>
45 #include <linux/fs.h>
46 #include <linux/poll.h>
47 #include <linux/mutex.h>
48 #include <linux/sched.h>
49 #include <linux/spinlock.h>
50 #include <linux/mount.h>
51 #include <linux/pagemap.h>
52 #include <linux/uaccess.h>
53 #include <linux/init.h>
54 #include <linux/namei.h>
55 #include <linux/string.h>
56 #include <linux/slab.h>
57 #include <linux/miscdevice.h>
58 #include <linux/workqueue.h>
59 
60 #include <xen/xenbus.h>
61 #include <xen/xen.h>
62 #include <asm/xen/hypervisor.h>
63 
64 #include "xenbus.h"
65 
66 unsigned int xb_dev_generation_id;
67 
68 /*
69  * An element of a list of outstanding transactions, for which we're
70  * still waiting a reply.
71  */
72 struct xenbus_transaction_holder {
73 	struct list_head list;
74 	struct xenbus_transaction handle;
75 	unsigned int generation_id;
76 };
77 
78 /*
79  * A buffer of data on the queue.
80  */
81 struct read_buffer {
82 	struct list_head list;
83 	unsigned int cons;
84 	unsigned int len;
85 	char msg[];
86 };
87 
88 struct xenbus_file_priv {
89 	/*
90 	 * msgbuffer_mutex is held while partial requests are built up
91 	 * and complete requests are acted on.  It therefore protects
92 	 * the "transactions" and "watches" lists, and the partial
93 	 * request length and buffer.
94 	 *
95 	 * reply_mutex protects the reply being built up to return to
96 	 * usermode.  It nests inside msgbuffer_mutex but may be held
97 	 * alone during a watch callback.
98 	 */
99 	struct mutex msgbuffer_mutex;
100 
101 	/* In-progress transactions */
102 	struct list_head transactions;
103 
104 	/* Active watches. */
105 	struct list_head watches;
106 
107 	/* Partial request. */
108 	unsigned int len;
109 	union {
110 		struct xsd_sockmsg msg;
111 		char buffer[XENSTORE_PAYLOAD_MAX];
112 	} u;
113 
114 	/* Response queue. */
115 	struct mutex reply_mutex;
116 	struct list_head read_buffers;
117 	wait_queue_head_t read_waitq;
118 
119 	struct kref kref;
120 
121 	struct work_struct wq;
122 };
123 
124 /* Read out any raw xenbus messages queued up. */
125 static ssize_t xenbus_file_read(struct file *filp,
126 			       char __user *ubuf,
127 			       size_t len, loff_t *ppos)
128 {
129 	struct xenbus_file_priv *u = filp->private_data;
130 	struct read_buffer *rb;
131 	ssize_t i;
132 	int ret;
133 
134 	mutex_lock(&u->reply_mutex);
135 again:
136 	while (list_empty(&u->read_buffers)) {
137 		mutex_unlock(&u->reply_mutex);
138 		if (filp->f_flags & O_NONBLOCK)
139 			return -EAGAIN;
140 
141 		ret = wait_event_interruptible(u->read_waitq,
142 					       !list_empty(&u->read_buffers));
143 		if (ret)
144 			return ret;
145 		mutex_lock(&u->reply_mutex);
146 	}
147 
148 	rb = list_entry(u->read_buffers.next, struct read_buffer, list);
149 	i = 0;
150 	while (i < len) {
151 		size_t sz = min_t(size_t, len - i, rb->len - rb->cons);
152 
153 		ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
154 
155 		i += sz - ret;
156 		rb->cons += sz - ret;
157 
158 		if (ret != 0) {
159 			if (i == 0)
160 				i = -EFAULT;
161 			goto out;
162 		}
163 
164 		/* Clear out buffer if it has been consumed */
165 		if (rb->cons == rb->len) {
166 			list_del(&rb->list);
167 			kfree(rb);
168 			if (list_empty(&u->read_buffers))
169 				break;
170 			rb = list_entry(u->read_buffers.next,
171 					struct read_buffer, list);
172 		}
173 	}
174 	if (i == 0)
175 		goto again;
176 
177 out:
178 	mutex_unlock(&u->reply_mutex);
179 	return i;
180 }
181 
182 /*
183  * Add a buffer to the queue.  Caller must hold the appropriate lock
184  * if the queue is not local.  (Commonly the caller will build up
185  * multiple queued buffers on a temporary local list, and then add it
186  * to the appropriate list under lock once all the buffers have een
187  * successfully allocated.)
188  */
189 static int queue_reply(struct list_head *queue, const void *data, size_t len)
190 {
191 	struct read_buffer *rb;
192 
193 	if (len == 0)
194 		return 0;
195 	if (len > XENSTORE_PAYLOAD_MAX)
196 		return -EINVAL;
197 
198 	rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
199 	if (rb == NULL)
200 		return -ENOMEM;
201 
202 	rb->cons = 0;
203 	rb->len = len;
204 
205 	memcpy(rb->msg, data, len);
206 
207 	list_add_tail(&rb->list, queue);
208 	return 0;
209 }
210 
211 /*
212  * Free all the read_buffer s on a list.
213  * Caller must have sole reference to list.
214  */
215 static void queue_cleanup(struct list_head *list)
216 {
217 	struct read_buffer *rb;
218 
219 	while (!list_empty(list)) {
220 		rb = list_entry(list->next, struct read_buffer, list);
221 		list_del(list->next);
222 		kfree(rb);
223 	}
224 }
225 
226 struct watch_adapter {
227 	struct list_head list;
228 	struct xenbus_watch watch;
229 	struct xenbus_file_priv *dev_data;
230 	char *token;
231 };
232 
233 static void free_watch_adapter(struct watch_adapter *watch)
234 {
235 	kfree(watch->watch.node);
236 	kfree(watch->token);
237 	kfree(watch);
238 }
239 
240 static struct watch_adapter *alloc_watch_adapter(const char *path,
241 						 const char *token)
242 {
243 	struct watch_adapter *watch;
244 
245 	watch = kzalloc(sizeof(*watch), GFP_KERNEL);
246 	if (watch == NULL)
247 		goto out_fail;
248 
249 	watch->watch.node = kstrdup(path, GFP_KERNEL);
250 	if (watch->watch.node == NULL)
251 		goto out_free;
252 
253 	watch->token = kstrdup(token, GFP_KERNEL);
254 	if (watch->token == NULL)
255 		goto out_free;
256 
257 	return watch;
258 
259 out_free:
260 	free_watch_adapter(watch);
261 
262 out_fail:
263 	return NULL;
264 }
265 
266 static void watch_fired(struct xenbus_watch *watch,
267 			const char *path,
268 			const char *token)
269 {
270 	struct watch_adapter *adap;
271 	struct xsd_sockmsg hdr;
272 	const char *token_caller;
273 	int path_len, tok_len, body_len;
274 	int ret;
275 	LIST_HEAD(staging_q);
276 
277 	adap = container_of(watch, struct watch_adapter, watch);
278 
279 	token_caller = adap->token;
280 
281 	path_len = strlen(path) + 1;
282 	tok_len = strlen(token_caller) + 1;
283 	body_len = path_len + tok_len;
284 
285 	hdr.type = XS_WATCH_EVENT;
286 	hdr.len = body_len;
287 
288 	mutex_lock(&adap->dev_data->reply_mutex);
289 
290 	ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
291 	if (!ret)
292 		ret = queue_reply(&staging_q, path, path_len);
293 	if (!ret)
294 		ret = queue_reply(&staging_q, token_caller, tok_len);
295 
296 	if (!ret) {
297 		/* success: pass reply list onto watcher */
298 		list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
299 		wake_up(&adap->dev_data->read_waitq);
300 	} else
301 		queue_cleanup(&staging_q);
302 
303 	mutex_unlock(&adap->dev_data->reply_mutex);
304 }
305 
306 static void xenbus_worker(struct work_struct *wq)
307 {
308 	struct xenbus_file_priv *u;
309 	struct xenbus_transaction_holder *trans, *tmp;
310 	struct watch_adapter *watch, *tmp_watch;
311 	struct read_buffer *rb, *tmp_rb;
312 
313 	u = container_of(wq, struct xenbus_file_priv, wq);
314 
315 	/*
316 	 * No need for locking here because there are no other users,
317 	 * by definition.
318 	 */
319 
320 	list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
321 		xenbus_transaction_end(trans->handle, 1);
322 		list_del(&trans->list);
323 		kfree(trans);
324 	}
325 
326 	list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
327 		unregister_xenbus_watch(&watch->watch);
328 		list_del(&watch->list);
329 		free_watch_adapter(watch);
330 	}
331 
332 	list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
333 		list_del(&rb->list);
334 		kfree(rb);
335 	}
336 	kfree(u);
337 }
338 
339 static void xenbus_file_free(struct kref *kref)
340 {
341 	struct xenbus_file_priv *u;
342 
343 	/*
344 	 * We might be called in xenbus_thread().
345 	 * Use workqueue to avoid deadlock.
346 	 */
347 	u = container_of(kref, struct xenbus_file_priv, kref);
348 	schedule_work(&u->wq);
349 }
350 
351 static struct xenbus_transaction_holder *xenbus_get_transaction(
352 	struct xenbus_file_priv *u, uint32_t tx_id)
353 {
354 	struct xenbus_transaction_holder *trans;
355 
356 	list_for_each_entry(trans, &u->transactions, list)
357 		if (trans->handle.id == tx_id)
358 			return trans;
359 
360 	return NULL;
361 }
362 
363 void xenbus_dev_queue_reply(struct xb_req_data *req)
364 {
365 	struct xenbus_file_priv *u = req->par;
366 	struct xenbus_transaction_holder *trans = NULL;
367 	int rc;
368 	LIST_HEAD(staging_q);
369 
370 	xs_request_exit(req);
371 
372 	mutex_lock(&u->msgbuffer_mutex);
373 
374 	if (req->type == XS_TRANSACTION_START) {
375 		trans = xenbus_get_transaction(u, 0);
376 		if (WARN_ON(!trans))
377 			goto out;
378 		if (req->msg.type == XS_ERROR) {
379 			list_del(&trans->list);
380 			kfree(trans);
381 		} else {
382 			rc = kstrtou32(req->body, 10, &trans->handle.id);
383 			if (WARN_ON(rc))
384 				goto out;
385 		}
386 	} else if (req->type == XS_TRANSACTION_END) {
387 		trans = xenbus_get_transaction(u, req->msg.tx_id);
388 		if (WARN_ON(!trans))
389 			goto out;
390 		list_del(&trans->list);
391 		kfree(trans);
392 	}
393 
394 	mutex_unlock(&u->msgbuffer_mutex);
395 
396 	mutex_lock(&u->reply_mutex);
397 	rc = queue_reply(&staging_q, &req->msg, sizeof(req->msg));
398 	if (!rc)
399 		rc = queue_reply(&staging_q, req->body, req->msg.len);
400 	if (!rc) {
401 		list_splice_tail(&staging_q, &u->read_buffers);
402 		wake_up(&u->read_waitq);
403 	} else {
404 		queue_cleanup(&staging_q);
405 	}
406 	mutex_unlock(&u->reply_mutex);
407 
408 	kfree(req->body);
409 	kfree(req);
410 
411 	kref_put(&u->kref, xenbus_file_free);
412 
413 	return;
414 
415  out:
416 	mutex_unlock(&u->msgbuffer_mutex);
417 }
418 
419 static int xenbus_command_reply(struct xenbus_file_priv *u,
420 				unsigned int msg_type, const char *reply)
421 {
422 	struct {
423 		struct xsd_sockmsg hdr;
424 		char body[16];
425 	} msg;
426 	int rc;
427 
428 	msg.hdr = u->u.msg;
429 	msg.hdr.type = msg_type;
430 	msg.hdr.len = strlen(reply) + 1;
431 	if (msg.hdr.len > sizeof(msg.body))
432 		return -E2BIG;
433 	memcpy(&msg.body, reply, msg.hdr.len);
434 
435 	mutex_lock(&u->reply_mutex);
436 	rc = queue_reply(&u->read_buffers, &msg, sizeof(msg.hdr) + msg.hdr.len);
437 	wake_up(&u->read_waitq);
438 	mutex_unlock(&u->reply_mutex);
439 
440 	if (!rc)
441 		kref_put(&u->kref, xenbus_file_free);
442 
443 	return rc;
444 }
445 
446 static int xenbus_write_transaction(unsigned msg_type,
447 				    struct xenbus_file_priv *u)
448 {
449 	int rc;
450 	struct xenbus_transaction_holder *trans = NULL;
451 	struct {
452 		struct xsd_sockmsg hdr;
453 		char body[];
454 	} *msg = (void *)u->u.buffer;
455 
456 	if (msg_type == XS_TRANSACTION_START) {
457 		trans = kzalloc(sizeof(*trans), GFP_KERNEL);
458 		if (!trans) {
459 			rc = -ENOMEM;
460 			goto out;
461 		}
462 		trans->generation_id = xb_dev_generation_id;
463 		list_add(&trans->list, &u->transactions);
464 	} else if (msg->hdr.tx_id != 0 &&
465 		   !xenbus_get_transaction(u, msg->hdr.tx_id))
466 		return xenbus_command_reply(u, XS_ERROR, "ENOENT");
467 	else if (msg_type == XS_TRANSACTION_END &&
468 		 !(msg->hdr.len == 2 &&
469 		   (!strcmp(msg->body, "T") || !strcmp(msg->body, "F"))))
470 		return xenbus_command_reply(u, XS_ERROR, "EINVAL");
471 	else if (msg_type == XS_TRANSACTION_END) {
472 		trans = xenbus_get_transaction(u, msg->hdr.tx_id);
473 		if (trans && trans->generation_id != xb_dev_generation_id) {
474 			list_del(&trans->list);
475 			kfree(trans);
476 			if (!strcmp(msg->body, "T"))
477 				return xenbus_command_reply(u, XS_ERROR,
478 							    "EAGAIN");
479 			else
480 				return xenbus_command_reply(u,
481 							    XS_TRANSACTION_END,
482 							    "OK");
483 		}
484 	}
485 
486 	rc = xenbus_dev_request_and_reply(&msg->hdr, u);
487 	if (rc && trans) {
488 		list_del(&trans->list);
489 		kfree(trans);
490 	}
491 
492 out:
493 	return rc;
494 }
495 
496 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
497 {
498 	struct watch_adapter *watch;
499 	char *path, *token;
500 	int err, rc;
501 
502 	path = u->u.buffer + sizeof(u->u.msg);
503 	token = memchr(path, 0, u->u.msg.len);
504 	if (token == NULL) {
505 		rc = xenbus_command_reply(u, XS_ERROR, "EINVAL");
506 		goto out;
507 	}
508 	token++;
509 	if (memchr(token, 0, u->u.msg.len - (token - path)) == NULL) {
510 		rc = xenbus_command_reply(u, XS_ERROR, "EINVAL");
511 		goto out;
512 	}
513 
514 	if (msg_type == XS_WATCH) {
515 		watch = alloc_watch_adapter(path, token);
516 		if (watch == NULL) {
517 			rc = -ENOMEM;
518 			goto out;
519 		}
520 
521 		watch->watch.callback = watch_fired;
522 		watch->dev_data = u;
523 
524 		err = register_xenbus_watch(&watch->watch);
525 		if (err) {
526 			free_watch_adapter(watch);
527 			rc = err;
528 			goto out;
529 		}
530 		list_add(&watch->list, &u->watches);
531 	} else {
532 		list_for_each_entry(watch, &u->watches, list) {
533 			if (!strcmp(watch->token, token) &&
534 			    !strcmp(watch->watch.node, path)) {
535 				unregister_xenbus_watch(&watch->watch);
536 				list_del(&watch->list);
537 				free_watch_adapter(watch);
538 				break;
539 			}
540 		}
541 	}
542 
543 	/* Success.  Synthesize a reply to say all is OK. */
544 	rc = xenbus_command_reply(u, msg_type, "OK");
545 
546 out:
547 	return rc;
548 }
549 
550 static ssize_t xenbus_file_write(struct file *filp,
551 				const char __user *ubuf,
552 				size_t len, loff_t *ppos)
553 {
554 	struct xenbus_file_priv *u = filp->private_data;
555 	uint32_t msg_type;
556 	int rc = len;
557 	int ret;
558 
559 	/*
560 	 * We're expecting usermode to be writing properly formed
561 	 * xenbus messages.  If they write an incomplete message we
562 	 * buffer it up.  Once it is complete, we act on it.
563 	 */
564 
565 	/*
566 	 * Make sure concurrent writers can't stomp all over each
567 	 * other's messages and make a mess of our partial message
568 	 * buffer.  We don't make any attemppt to stop multiple
569 	 * writers from making a mess of each other's incomplete
570 	 * messages; we're just trying to guarantee our own internal
571 	 * consistency and make sure that single writes are handled
572 	 * atomically.
573 	 */
574 	mutex_lock(&u->msgbuffer_mutex);
575 
576 	/* Get this out of the way early to avoid confusion */
577 	if (len == 0)
578 		goto out;
579 
580 	/* Can't write a xenbus message larger we can buffer */
581 	if (len > sizeof(u->u.buffer) - u->len) {
582 		/* On error, dump existing buffer */
583 		u->len = 0;
584 		rc = -EINVAL;
585 		goto out;
586 	}
587 
588 	ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
589 
590 	if (ret != 0) {
591 		rc = -EFAULT;
592 		goto out;
593 	}
594 
595 	/* Deal with a partial copy. */
596 	len -= ret;
597 	rc = len;
598 
599 	u->len += len;
600 
601 	/* Return if we haven't got a full message yet */
602 	if (u->len < sizeof(u->u.msg))
603 		goto out;	/* not even the header yet */
604 
605 	/* If we're expecting a message that's larger than we can
606 	   possibly send, dump what we have and return an error. */
607 	if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
608 		rc = -E2BIG;
609 		u->len = 0;
610 		goto out;
611 	}
612 
613 	if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
614 		goto out;	/* incomplete data portion */
615 
616 	/*
617 	 * OK, now we have a complete message.  Do something with it.
618 	 */
619 
620 	kref_get(&u->kref);
621 
622 	msg_type = u->u.msg.type;
623 
624 	switch (msg_type) {
625 	case XS_WATCH:
626 	case XS_UNWATCH:
627 		/* (Un)Ask for some path to be watched for changes */
628 		ret = xenbus_write_watch(msg_type, u);
629 		break;
630 
631 	default:
632 		/* Send out a transaction */
633 		ret = xenbus_write_transaction(msg_type, u);
634 		break;
635 	}
636 	if (ret != 0) {
637 		rc = ret;
638 		kref_put(&u->kref, xenbus_file_free);
639 	}
640 
641 	/* Buffered message consumed */
642 	u->len = 0;
643 
644  out:
645 	mutex_unlock(&u->msgbuffer_mutex);
646 	return rc;
647 }
648 
649 static int xenbus_file_open(struct inode *inode, struct file *filp)
650 {
651 	struct xenbus_file_priv *u;
652 
653 	if (xen_store_evtchn == 0)
654 		return -ENOENT;
655 
656 	stream_open(inode, filp);
657 
658 	u = kzalloc(sizeof(*u), GFP_KERNEL);
659 	if (u == NULL)
660 		return -ENOMEM;
661 
662 	kref_init(&u->kref);
663 
664 	INIT_LIST_HEAD(&u->transactions);
665 	INIT_LIST_HEAD(&u->watches);
666 	INIT_LIST_HEAD(&u->read_buffers);
667 	init_waitqueue_head(&u->read_waitq);
668 	INIT_WORK(&u->wq, xenbus_worker);
669 
670 	mutex_init(&u->reply_mutex);
671 	mutex_init(&u->msgbuffer_mutex);
672 
673 	filp->private_data = u;
674 
675 	return 0;
676 }
677 
678 static int xenbus_file_release(struct inode *inode, struct file *filp)
679 {
680 	struct xenbus_file_priv *u = filp->private_data;
681 
682 	kref_put(&u->kref, xenbus_file_free);
683 
684 	return 0;
685 }
686 
687 static __poll_t xenbus_file_poll(struct file *file, poll_table *wait)
688 {
689 	struct xenbus_file_priv *u = file->private_data;
690 
691 	poll_wait(file, &u->read_waitq, wait);
692 	if (!list_empty(&u->read_buffers))
693 		return EPOLLIN | EPOLLRDNORM;
694 	return 0;
695 }
696 
697 const struct file_operations xen_xenbus_fops = {
698 	.read = xenbus_file_read,
699 	.write = xenbus_file_write,
700 	.open = xenbus_file_open,
701 	.release = xenbus_file_release,
702 	.poll = xenbus_file_poll,
703 	.llseek = no_llseek,
704 };
705 EXPORT_SYMBOL_GPL(xen_xenbus_fops);
706 
707 static struct miscdevice xenbus_dev = {
708 	.minor = MISC_DYNAMIC_MINOR,
709 	.name = "xen/xenbus",
710 	.fops = &xen_xenbus_fops,
711 };
712 
713 static int __init xenbus_init(void)
714 {
715 	int err;
716 
717 	if (!xen_domain())
718 		return -ENODEV;
719 
720 	err = misc_register(&xenbus_dev);
721 	if (err)
722 		pr_err("Could not register xenbus frontend device\n");
723 	return err;
724 }
725 device_initcall(xenbus_init);
726