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/init.h>
59 
60 #include "xenbus_comms.h"
61 
62 #include <xen/xenbus.h>
63 #include <xen/xen.h>
64 #include <asm/xen/hypervisor.h>
65 
66 /*
67  * An element of a list of outstanding transactions, for which we're
68  * still waiting a reply.
69  */
70 struct xenbus_transaction_holder {
71 	struct list_head list;
72 	struct xenbus_transaction handle;
73 };
74 
75 /*
76  * A buffer of data on the queue.
77  */
78 struct read_buffer {
79 	struct list_head list;
80 	unsigned int cons;
81 	unsigned int len;
82 	char msg[];
83 };
84 
85 struct xenbus_file_priv {
86 	/*
87 	 * msgbuffer_mutex is held while partial requests are built up
88 	 * and complete requests are acted on.  It therefore protects
89 	 * the "transactions" and "watches" lists, and the partial
90 	 * request length and buffer.
91 	 *
92 	 * reply_mutex protects the reply being built up to return to
93 	 * usermode.  It nests inside msgbuffer_mutex but may be held
94 	 * alone during a watch callback.
95 	 */
96 	struct mutex msgbuffer_mutex;
97 
98 	/* In-progress transactions */
99 	struct list_head transactions;
100 
101 	/* Active watches. */
102 	struct list_head watches;
103 
104 	/* Partial request. */
105 	unsigned int len;
106 	union {
107 		struct xsd_sockmsg msg;
108 		char buffer[XENSTORE_PAYLOAD_MAX];
109 	} u;
110 
111 	/* Response queue. */
112 	struct mutex reply_mutex;
113 	struct list_head read_buffers;
114 	wait_queue_head_t read_waitq;
115 
116 };
117 
118 /* Read out any raw xenbus messages queued up. */
119 static ssize_t xenbus_file_read(struct file *filp,
120 			       char __user *ubuf,
121 			       size_t len, loff_t *ppos)
122 {
123 	struct xenbus_file_priv *u = filp->private_data;
124 	struct read_buffer *rb;
125 	unsigned i;
126 	int ret;
127 
128 	mutex_lock(&u->reply_mutex);
129 again:
130 	while (list_empty(&u->read_buffers)) {
131 		mutex_unlock(&u->reply_mutex);
132 		if (filp->f_flags & O_NONBLOCK)
133 			return -EAGAIN;
134 
135 		ret = wait_event_interruptible(u->read_waitq,
136 					       !list_empty(&u->read_buffers));
137 		if (ret)
138 			return ret;
139 		mutex_lock(&u->reply_mutex);
140 	}
141 
142 	rb = list_entry(u->read_buffers.next, struct read_buffer, list);
143 	i = 0;
144 	while (i < len) {
145 		unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
146 
147 		ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
148 
149 		i += sz - ret;
150 		rb->cons += sz - ret;
151 
152 		if (ret != 0) {
153 			if (i == 0)
154 				i = -EFAULT;
155 			goto out;
156 		}
157 
158 		/* Clear out buffer if it has been consumed */
159 		if (rb->cons == rb->len) {
160 			list_del(&rb->list);
161 			kfree(rb);
162 			if (list_empty(&u->read_buffers))
163 				break;
164 			rb = list_entry(u->read_buffers.next,
165 					struct read_buffer, list);
166 		}
167 	}
168 	if (i == 0)
169 		goto again;
170 
171 out:
172 	mutex_unlock(&u->reply_mutex);
173 	return i;
174 }
175 
176 /*
177  * Add a buffer to the queue.  Caller must hold the appropriate lock
178  * if the queue is not local.  (Commonly the caller will build up
179  * multiple queued buffers on a temporary local list, and then add it
180  * to the appropriate list under lock once all the buffers have een
181  * successfully allocated.)
182  */
183 static int queue_reply(struct list_head *queue, const void *data, size_t len)
184 {
185 	struct read_buffer *rb;
186 
187 	if (len == 0)
188 		return 0;
189 	if (len > XENSTORE_PAYLOAD_MAX)
190 		return -EINVAL;
191 
192 	rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
193 	if (rb == NULL)
194 		return -ENOMEM;
195 
196 	rb->cons = 0;
197 	rb->len = len;
198 
199 	memcpy(rb->msg, data, len);
200 
201 	list_add_tail(&rb->list, queue);
202 	return 0;
203 }
204 
205 /*
206  * Free all the read_buffer s on a list.
207  * Caller must have sole reference to list.
208  */
209 static void queue_cleanup(struct list_head *list)
210 {
211 	struct read_buffer *rb;
212 
213 	while (!list_empty(list)) {
214 		rb = list_entry(list->next, struct read_buffer, list);
215 		list_del(list->next);
216 		kfree(rb);
217 	}
218 }
219 
220 struct watch_adapter {
221 	struct list_head list;
222 	struct xenbus_watch watch;
223 	struct xenbus_file_priv *dev_data;
224 	char *token;
225 };
226 
227 static void free_watch_adapter(struct watch_adapter *watch)
228 {
229 	kfree(watch->watch.node);
230 	kfree(watch->token);
231 	kfree(watch);
232 }
233 
234 static struct watch_adapter *alloc_watch_adapter(const char *path,
235 						 const char *token)
236 {
237 	struct watch_adapter *watch;
238 
239 	watch = kzalloc(sizeof(*watch), GFP_KERNEL);
240 	if (watch == NULL)
241 		goto out_fail;
242 
243 	watch->watch.node = kstrdup(path, GFP_KERNEL);
244 	if (watch->watch.node == NULL)
245 		goto out_free;
246 
247 	watch->token = kstrdup(token, GFP_KERNEL);
248 	if (watch->token == NULL)
249 		goto out_free;
250 
251 	return watch;
252 
253 out_free:
254 	free_watch_adapter(watch);
255 
256 out_fail:
257 	return NULL;
258 }
259 
260 static void watch_fired(struct xenbus_watch *watch,
261 			const char **vec,
262 			unsigned int len)
263 {
264 	struct watch_adapter *adap;
265 	struct xsd_sockmsg hdr;
266 	const char *path, *token;
267 	int path_len, tok_len, body_len, data_len = 0;
268 	int ret;
269 	LIST_HEAD(staging_q);
270 
271 	adap = container_of(watch, struct watch_adapter, watch);
272 
273 	path = vec[XS_WATCH_PATH];
274 	token = adap->token;
275 
276 	path_len = strlen(path) + 1;
277 	tok_len = strlen(token) + 1;
278 	if (len > 2)
279 		data_len = vec[len] - vec[2] + 1;
280 	body_len = path_len + tok_len + data_len;
281 
282 	hdr.type = XS_WATCH_EVENT;
283 	hdr.len = body_len;
284 
285 	mutex_lock(&adap->dev_data->reply_mutex);
286 
287 	ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
288 	if (!ret)
289 		ret = queue_reply(&staging_q, path, path_len);
290 	if (!ret)
291 		ret = queue_reply(&staging_q, token, tok_len);
292 	if (!ret && len > 2)
293 		ret = queue_reply(&staging_q, vec[2], data_len);
294 
295 	if (!ret) {
296 		/* success: pass reply list onto watcher */
297 		list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
298 		wake_up(&adap->dev_data->read_waitq);
299 	} else
300 		queue_cleanup(&staging_q);
301 
302 	mutex_unlock(&adap->dev_data->reply_mutex);
303 }
304 
305 static int xenbus_write_transaction(unsigned msg_type,
306 				    struct xenbus_file_priv *u)
307 {
308 	int rc;
309 	void *reply;
310 	struct xenbus_transaction_holder *trans = NULL;
311 	LIST_HEAD(staging_q);
312 
313 	if (msg_type == XS_TRANSACTION_START) {
314 		trans = kmalloc(sizeof(*trans), GFP_KERNEL);
315 		if (!trans) {
316 			rc = -ENOMEM;
317 			goto out;
318 		}
319 	} else if (msg_type == XS_TRANSACTION_END) {
320 		list_for_each_entry(trans, &u->transactions, list)
321 			if (trans->handle.id == u->u.msg.tx_id)
322 				break;
323 		if (&trans->list == &u->transactions)
324 			return -ESRCH;
325 	}
326 
327 	reply = xenbus_dev_request_and_reply(&u->u.msg);
328 	if (IS_ERR(reply)) {
329 		if (msg_type == XS_TRANSACTION_START)
330 			kfree(trans);
331 		rc = PTR_ERR(reply);
332 		goto out;
333 	}
334 
335 	if (msg_type == XS_TRANSACTION_START) {
336 		if (u->u.msg.type == XS_ERROR)
337 			kfree(trans);
338 		else {
339 			trans->handle.id = simple_strtoul(reply, NULL, 0);
340 			list_add(&trans->list, &u->transactions);
341 		}
342 	} else if (u->u.msg.type == XS_TRANSACTION_END) {
343 		list_del(&trans->list);
344 		kfree(trans);
345 	}
346 
347 	mutex_lock(&u->reply_mutex);
348 	rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
349 	if (!rc)
350 		rc = queue_reply(&staging_q, reply, u->u.msg.len);
351 	if (!rc) {
352 		list_splice_tail(&staging_q, &u->read_buffers);
353 		wake_up(&u->read_waitq);
354 	} else {
355 		queue_cleanup(&staging_q);
356 	}
357 	mutex_unlock(&u->reply_mutex);
358 
359 	kfree(reply);
360 
361 out:
362 	return rc;
363 }
364 
365 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
366 {
367 	struct watch_adapter *watch, *tmp_watch;
368 	char *path, *token;
369 	int err, rc;
370 	LIST_HEAD(staging_q);
371 
372 	path = u->u.buffer + sizeof(u->u.msg);
373 	token = memchr(path, 0, u->u.msg.len);
374 	if (token == NULL) {
375 		rc = -EILSEQ;
376 		goto out;
377 	}
378 	token++;
379 	if (memchr(token, 0, u->u.msg.len - (token - path)) == NULL) {
380 		rc = -EILSEQ;
381 		goto out;
382 	}
383 
384 	if (msg_type == XS_WATCH) {
385 		watch = alloc_watch_adapter(path, token);
386 		if (watch == NULL) {
387 			rc = -ENOMEM;
388 			goto out;
389 		}
390 
391 		watch->watch.callback = watch_fired;
392 		watch->dev_data = u;
393 
394 		err = register_xenbus_watch(&watch->watch);
395 		if (err) {
396 			free_watch_adapter(watch);
397 			rc = err;
398 			goto out;
399 		}
400 		list_add(&watch->list, &u->watches);
401 	} else {
402 		list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
403 			if (!strcmp(watch->token, token) &&
404 			    !strcmp(watch->watch.node, path)) {
405 				unregister_xenbus_watch(&watch->watch);
406 				list_del(&watch->list);
407 				free_watch_adapter(watch);
408 				break;
409 			}
410 		}
411 	}
412 
413 	/* Success.  Synthesize a reply to say all is OK. */
414 	{
415 		struct {
416 			struct xsd_sockmsg hdr;
417 			char body[3];
418 		} __packed reply = {
419 			{
420 				.type = msg_type,
421 				.len = sizeof(reply.body)
422 			},
423 			"OK"
424 		};
425 
426 		mutex_lock(&u->reply_mutex);
427 		rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
428 		wake_up(&u->read_waitq);
429 		mutex_unlock(&u->reply_mutex);
430 	}
431 
432 out:
433 	return rc;
434 }
435 
436 static ssize_t xenbus_file_write(struct file *filp,
437 				const char __user *ubuf,
438 				size_t len, loff_t *ppos)
439 {
440 	struct xenbus_file_priv *u = filp->private_data;
441 	uint32_t msg_type;
442 	int rc = len;
443 	int ret;
444 	LIST_HEAD(staging_q);
445 
446 	/*
447 	 * We're expecting usermode to be writing properly formed
448 	 * xenbus messages.  If they write an incomplete message we
449 	 * buffer it up.  Once it is complete, we act on it.
450 	 */
451 
452 	/*
453 	 * Make sure concurrent writers can't stomp all over each
454 	 * other's messages and make a mess of our partial message
455 	 * buffer.  We don't make any attemppt to stop multiple
456 	 * writers from making a mess of each other's incomplete
457 	 * messages; we're just trying to guarantee our own internal
458 	 * consistency and make sure that single writes are handled
459 	 * atomically.
460 	 */
461 	mutex_lock(&u->msgbuffer_mutex);
462 
463 	/* Get this out of the way early to avoid confusion */
464 	if (len == 0)
465 		goto out;
466 
467 	/* Can't write a xenbus message larger we can buffer */
468 	if (len > sizeof(u->u.buffer) - u->len) {
469 		/* On error, dump existing buffer */
470 		u->len = 0;
471 		rc = -EINVAL;
472 		goto out;
473 	}
474 
475 	ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
476 
477 	if (ret != 0) {
478 		rc = -EFAULT;
479 		goto out;
480 	}
481 
482 	/* Deal with a partial copy. */
483 	len -= ret;
484 	rc = len;
485 
486 	u->len += len;
487 
488 	/* Return if we haven't got a full message yet */
489 	if (u->len < sizeof(u->u.msg))
490 		goto out;	/* not even the header yet */
491 
492 	/* If we're expecting a message that's larger than we can
493 	   possibly send, dump what we have and return an error. */
494 	if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
495 		rc = -E2BIG;
496 		u->len = 0;
497 		goto out;
498 	}
499 
500 	if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
501 		goto out;	/* incomplete data portion */
502 
503 	/*
504 	 * OK, now we have a complete message.  Do something with it.
505 	 */
506 
507 	msg_type = u->u.msg.type;
508 
509 	switch (msg_type) {
510 	case XS_WATCH:
511 	case XS_UNWATCH:
512 		/* (Un)Ask for some path to be watched for changes */
513 		ret = xenbus_write_watch(msg_type, u);
514 		break;
515 
516 	default:
517 		/* Send out a transaction */
518 		ret = xenbus_write_transaction(msg_type, u);
519 		break;
520 	}
521 	if (ret != 0)
522 		rc = ret;
523 
524 	/* Buffered message consumed */
525 	u->len = 0;
526 
527  out:
528 	mutex_unlock(&u->msgbuffer_mutex);
529 	return rc;
530 }
531 
532 static int xenbus_file_open(struct inode *inode, struct file *filp)
533 {
534 	struct xenbus_file_priv *u;
535 
536 	if (xen_store_evtchn == 0)
537 		return -ENOENT;
538 
539 	nonseekable_open(inode, filp);
540 
541 	u = kzalloc(sizeof(*u), GFP_KERNEL);
542 	if (u == NULL)
543 		return -ENOMEM;
544 
545 	INIT_LIST_HEAD(&u->transactions);
546 	INIT_LIST_HEAD(&u->watches);
547 	INIT_LIST_HEAD(&u->read_buffers);
548 	init_waitqueue_head(&u->read_waitq);
549 
550 	mutex_init(&u->reply_mutex);
551 	mutex_init(&u->msgbuffer_mutex);
552 
553 	filp->private_data = u;
554 
555 	return 0;
556 }
557 
558 static int xenbus_file_release(struct inode *inode, struct file *filp)
559 {
560 	struct xenbus_file_priv *u = filp->private_data;
561 	struct xenbus_transaction_holder *trans, *tmp;
562 	struct watch_adapter *watch, *tmp_watch;
563 	struct read_buffer *rb, *tmp_rb;
564 
565 	/*
566 	 * No need for locking here because there are no other users,
567 	 * by definition.
568 	 */
569 
570 	list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
571 		xenbus_transaction_end(trans->handle, 1);
572 		list_del(&trans->list);
573 		kfree(trans);
574 	}
575 
576 	list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
577 		unregister_xenbus_watch(&watch->watch);
578 		list_del(&watch->list);
579 		free_watch_adapter(watch);
580 	}
581 
582 	list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
583 		list_del(&rb->list);
584 		kfree(rb);
585 	}
586 	kfree(u);
587 
588 	return 0;
589 }
590 
591 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
592 {
593 	struct xenbus_file_priv *u = file->private_data;
594 
595 	poll_wait(file, &u->read_waitq, wait);
596 	if (!list_empty(&u->read_buffers))
597 		return POLLIN | POLLRDNORM;
598 	return 0;
599 }
600 
601 const struct file_operations xen_xenbus_fops = {
602 	.read = xenbus_file_read,
603 	.write = xenbus_file_write,
604 	.open = xenbus_file_open,
605 	.release = xenbus_file_release,
606 	.poll = xenbus_file_poll,
607 	.llseek = no_llseek,
608 };
609 EXPORT_SYMBOL_GPL(xen_xenbus_fops);
610 
611 static struct miscdevice xenbus_dev = {
612 	.minor = MISC_DYNAMIC_MINOR,
613 	.name = "xen/xenbus",
614 	.fops = &xen_xenbus_fops,
615 };
616 
617 static int __init xenbus_init(void)
618 {
619 	int err;
620 
621 	if (!xen_domain())
622 		return -ENODEV;
623 
624 	err = misc_register(&xenbus_dev);
625 	if (err)
626 		pr_err("Could not register xenbus frontend device\n");
627 	return err;
628 }
629 device_initcall(xenbus_init);
630