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 
59 #include <xen/xenbus.h>
60 #include <xen/xen.h>
61 #include <asm/xen/hypervisor.h>
62 
63 #include "xenbus.h"
64 
65 /*
66  * An element of a list of outstanding transactions, for which we're
67  * still waiting a reply.
68  */
69 struct xenbus_transaction_holder {
70 	struct list_head list;
71 	struct xenbus_transaction handle;
72 };
73 
74 /*
75  * A buffer of data on the queue.
76  */
77 struct read_buffer {
78 	struct list_head list;
79 	unsigned int cons;
80 	unsigned int len;
81 	char msg[];
82 };
83 
84 struct xenbus_file_priv {
85 	/*
86 	 * msgbuffer_mutex is held while partial requests are built up
87 	 * and complete requests are acted on.  It therefore protects
88 	 * the "transactions" and "watches" lists, and the partial
89 	 * request length and buffer.
90 	 *
91 	 * reply_mutex protects the reply being built up to return to
92 	 * usermode.  It nests inside msgbuffer_mutex but may be held
93 	 * alone during a watch callback.
94 	 */
95 	struct mutex msgbuffer_mutex;
96 
97 	/* In-progress transactions */
98 	struct list_head transactions;
99 
100 	/* Active watches. */
101 	struct list_head watches;
102 
103 	/* Partial request. */
104 	unsigned int len;
105 	union {
106 		struct xsd_sockmsg msg;
107 		char buffer[XENSTORE_PAYLOAD_MAX];
108 	} u;
109 
110 	/* Response queue. */
111 	struct mutex reply_mutex;
112 	struct list_head read_buffers;
113 	wait_queue_head_t read_waitq;
114 
115 	struct kref kref;
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 *path,
262 			const char *token)
263 {
264 	struct watch_adapter *adap;
265 	struct xsd_sockmsg hdr;
266 	const char *token_caller;
267 	int path_len, tok_len, body_len;
268 	int ret;
269 	LIST_HEAD(staging_q);
270 
271 	adap = container_of(watch, struct watch_adapter, watch);
272 
273 	token_caller = adap->token;
274 
275 	path_len = strlen(path) + 1;
276 	tok_len = strlen(token_caller) + 1;
277 	body_len = path_len + tok_len;
278 
279 	hdr.type = XS_WATCH_EVENT;
280 	hdr.len = body_len;
281 
282 	mutex_lock(&adap->dev_data->reply_mutex);
283 
284 	ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
285 	if (!ret)
286 		ret = queue_reply(&staging_q, path, path_len);
287 	if (!ret)
288 		ret = queue_reply(&staging_q, token_caller, tok_len);
289 
290 	if (!ret) {
291 		/* success: pass reply list onto watcher */
292 		list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
293 		wake_up(&adap->dev_data->read_waitq);
294 	} else
295 		queue_cleanup(&staging_q);
296 
297 	mutex_unlock(&adap->dev_data->reply_mutex);
298 }
299 
300 static void xenbus_file_free(struct kref *kref)
301 {
302 	struct xenbus_file_priv *u;
303 	struct xenbus_transaction_holder *trans, *tmp;
304 	struct watch_adapter *watch, *tmp_watch;
305 	struct read_buffer *rb, *tmp_rb;
306 
307 	u = container_of(kref, struct xenbus_file_priv, kref);
308 
309 	/*
310 	 * No need for locking here because there are no other users,
311 	 * by definition.
312 	 */
313 
314 	list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
315 		xenbus_transaction_end(trans->handle, 1);
316 		list_del(&trans->list);
317 		kfree(trans);
318 	}
319 
320 	list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
321 		unregister_xenbus_watch(&watch->watch);
322 		list_del(&watch->list);
323 		free_watch_adapter(watch);
324 	}
325 
326 	list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
327 		list_del(&rb->list);
328 		kfree(rb);
329 	}
330 	kfree(u);
331 }
332 
333 static struct xenbus_transaction_holder *xenbus_get_transaction(
334 	struct xenbus_file_priv *u, uint32_t tx_id)
335 {
336 	struct xenbus_transaction_holder *trans;
337 
338 	list_for_each_entry(trans, &u->transactions, list)
339 		if (trans->handle.id == tx_id)
340 			return trans;
341 
342 	return NULL;
343 }
344 
345 void xenbus_dev_queue_reply(struct xb_req_data *req)
346 {
347 	struct xenbus_file_priv *u = req->par;
348 	struct xenbus_transaction_holder *trans = NULL;
349 	int rc;
350 	LIST_HEAD(staging_q);
351 
352 	xs_request_exit(req);
353 
354 	mutex_lock(&u->msgbuffer_mutex);
355 
356 	if (req->type == XS_TRANSACTION_START) {
357 		trans = xenbus_get_transaction(u, 0);
358 		if (WARN_ON(!trans))
359 			goto out;
360 		if (req->msg.type == XS_ERROR) {
361 			list_del(&trans->list);
362 			kfree(trans);
363 		} else {
364 			rc = kstrtou32(req->body, 10, &trans->handle.id);
365 			if (WARN_ON(rc))
366 				goto out;
367 		}
368 	} else if (req->msg.type == XS_TRANSACTION_END) {
369 		trans = xenbus_get_transaction(u, req->msg.tx_id);
370 		if (WARN_ON(!trans))
371 			goto out;
372 		list_del(&trans->list);
373 		kfree(trans);
374 	}
375 
376 	mutex_unlock(&u->msgbuffer_mutex);
377 
378 	mutex_lock(&u->reply_mutex);
379 	rc = queue_reply(&staging_q, &req->msg, sizeof(req->msg));
380 	if (!rc)
381 		rc = queue_reply(&staging_q, req->body, req->msg.len);
382 	if (!rc) {
383 		list_splice_tail(&staging_q, &u->read_buffers);
384 		wake_up(&u->read_waitq);
385 	} else {
386 		queue_cleanup(&staging_q);
387 	}
388 	mutex_unlock(&u->reply_mutex);
389 
390 	kfree(req->body);
391 	kfree(req);
392 
393 	kref_put(&u->kref, xenbus_file_free);
394 
395 	return;
396 
397  out:
398 	mutex_unlock(&u->msgbuffer_mutex);
399 }
400 
401 static int xenbus_command_reply(struct xenbus_file_priv *u,
402 				unsigned int msg_type, const char *reply)
403 {
404 	struct {
405 		struct xsd_sockmsg hdr;
406 		const char body[16];
407 	} msg;
408 	int rc;
409 
410 	msg.hdr = u->u.msg;
411 	msg.hdr.type = msg_type;
412 	msg.hdr.len = strlen(reply) + 1;
413 	if (msg.hdr.len > sizeof(msg.body))
414 		return -E2BIG;
415 
416 	mutex_lock(&u->reply_mutex);
417 	rc = queue_reply(&u->read_buffers, &msg, sizeof(msg.hdr) + msg.hdr.len);
418 	wake_up(&u->read_waitq);
419 	mutex_unlock(&u->reply_mutex);
420 
421 	if (!rc)
422 		kref_put(&u->kref, xenbus_file_free);
423 
424 	return rc;
425 }
426 
427 static int xenbus_write_transaction(unsigned msg_type,
428 				    struct xenbus_file_priv *u)
429 {
430 	int rc;
431 	struct xenbus_transaction_holder *trans = NULL;
432 
433 	if (msg_type == XS_TRANSACTION_START) {
434 		trans = kzalloc(sizeof(*trans), GFP_KERNEL);
435 		if (!trans) {
436 			rc = -ENOMEM;
437 			goto out;
438 		}
439 		list_add(&trans->list, &u->transactions);
440 	} else if (u->u.msg.tx_id != 0 &&
441 		   !xenbus_get_transaction(u, u->u.msg.tx_id))
442 		return xenbus_command_reply(u, XS_ERROR, "ENOENT");
443 
444 	rc = xenbus_dev_request_and_reply(&u->u.msg, u);
445 	if (rc && trans) {
446 		list_del(&trans->list);
447 		kfree(trans);
448 	}
449 
450 out:
451 	return rc;
452 }
453 
454 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
455 {
456 	struct watch_adapter *watch;
457 	char *path, *token;
458 	int err, rc;
459 	LIST_HEAD(staging_q);
460 
461 	path = u->u.buffer + sizeof(u->u.msg);
462 	token = memchr(path, 0, u->u.msg.len);
463 	if (token == NULL) {
464 		rc = xenbus_command_reply(u, XS_ERROR, "EINVAL");
465 		goto out;
466 	}
467 	token++;
468 	if (memchr(token, 0, u->u.msg.len - (token - path)) == NULL) {
469 		rc = xenbus_command_reply(u, XS_ERROR, "EINVAL");
470 		goto out;
471 	}
472 
473 	if (msg_type == XS_WATCH) {
474 		watch = alloc_watch_adapter(path, token);
475 		if (watch == NULL) {
476 			rc = -ENOMEM;
477 			goto out;
478 		}
479 
480 		watch->watch.callback = watch_fired;
481 		watch->dev_data = u;
482 
483 		err = register_xenbus_watch(&watch->watch);
484 		if (err) {
485 			free_watch_adapter(watch);
486 			rc = err;
487 			goto out;
488 		}
489 		list_add(&watch->list, &u->watches);
490 	} else {
491 		list_for_each_entry(watch, &u->watches, list) {
492 			if (!strcmp(watch->token, token) &&
493 			    !strcmp(watch->watch.node, path)) {
494 				unregister_xenbus_watch(&watch->watch);
495 				list_del(&watch->list);
496 				free_watch_adapter(watch);
497 				break;
498 			}
499 		}
500 	}
501 
502 	/* Success.  Synthesize a reply to say all is OK. */
503 	rc = xenbus_command_reply(u, msg_type, "OK");
504 
505 out:
506 	return rc;
507 }
508 
509 static ssize_t xenbus_file_write(struct file *filp,
510 				const char __user *ubuf,
511 				size_t len, loff_t *ppos)
512 {
513 	struct xenbus_file_priv *u = filp->private_data;
514 	uint32_t msg_type;
515 	int rc = len;
516 	int ret;
517 	LIST_HEAD(staging_q);
518 
519 	/*
520 	 * We're expecting usermode to be writing properly formed
521 	 * xenbus messages.  If they write an incomplete message we
522 	 * buffer it up.  Once it is complete, we act on it.
523 	 */
524 
525 	/*
526 	 * Make sure concurrent writers can't stomp all over each
527 	 * other's messages and make a mess of our partial message
528 	 * buffer.  We don't make any attemppt to stop multiple
529 	 * writers from making a mess of each other's incomplete
530 	 * messages; we're just trying to guarantee our own internal
531 	 * consistency and make sure that single writes are handled
532 	 * atomically.
533 	 */
534 	mutex_lock(&u->msgbuffer_mutex);
535 
536 	/* Get this out of the way early to avoid confusion */
537 	if (len == 0)
538 		goto out;
539 
540 	/* Can't write a xenbus message larger we can buffer */
541 	if (len > sizeof(u->u.buffer) - u->len) {
542 		/* On error, dump existing buffer */
543 		u->len = 0;
544 		rc = -EINVAL;
545 		goto out;
546 	}
547 
548 	ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
549 
550 	if (ret != 0) {
551 		rc = -EFAULT;
552 		goto out;
553 	}
554 
555 	/* Deal with a partial copy. */
556 	len -= ret;
557 	rc = len;
558 
559 	u->len += len;
560 
561 	/* Return if we haven't got a full message yet */
562 	if (u->len < sizeof(u->u.msg))
563 		goto out;	/* not even the header yet */
564 
565 	/* If we're expecting a message that's larger than we can
566 	   possibly send, dump what we have and return an error. */
567 	if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
568 		rc = -E2BIG;
569 		u->len = 0;
570 		goto out;
571 	}
572 
573 	if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
574 		goto out;	/* incomplete data portion */
575 
576 	/*
577 	 * OK, now we have a complete message.  Do something with it.
578 	 */
579 
580 	kref_get(&u->kref);
581 
582 	msg_type = u->u.msg.type;
583 
584 	switch (msg_type) {
585 	case XS_WATCH:
586 	case XS_UNWATCH:
587 		/* (Un)Ask for some path to be watched for changes */
588 		ret = xenbus_write_watch(msg_type, u);
589 		break;
590 
591 	default:
592 		/* Send out a transaction */
593 		ret = xenbus_write_transaction(msg_type, u);
594 		break;
595 	}
596 	if (ret != 0) {
597 		rc = ret;
598 		kref_put(&u->kref, xenbus_file_free);
599 	}
600 
601 	/* Buffered message consumed */
602 	u->len = 0;
603 
604  out:
605 	mutex_unlock(&u->msgbuffer_mutex);
606 	return rc;
607 }
608 
609 static int xenbus_file_open(struct inode *inode, struct file *filp)
610 {
611 	struct xenbus_file_priv *u;
612 
613 	if (xen_store_evtchn == 0)
614 		return -ENOENT;
615 
616 	nonseekable_open(inode, filp);
617 
618 	filp->f_mode &= ~FMODE_ATOMIC_POS; /* cdev-style semantics */
619 
620 	u = kzalloc(sizeof(*u), GFP_KERNEL);
621 	if (u == NULL)
622 		return -ENOMEM;
623 
624 	kref_init(&u->kref);
625 
626 	INIT_LIST_HEAD(&u->transactions);
627 	INIT_LIST_HEAD(&u->watches);
628 	INIT_LIST_HEAD(&u->read_buffers);
629 	init_waitqueue_head(&u->read_waitq);
630 
631 	mutex_init(&u->reply_mutex);
632 	mutex_init(&u->msgbuffer_mutex);
633 
634 	filp->private_data = u;
635 
636 	return 0;
637 }
638 
639 static int xenbus_file_release(struct inode *inode, struct file *filp)
640 {
641 	struct xenbus_file_priv *u = filp->private_data;
642 
643 	kref_put(&u->kref, xenbus_file_free);
644 
645 	return 0;
646 }
647 
648 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
649 {
650 	struct xenbus_file_priv *u = file->private_data;
651 
652 	poll_wait(file, &u->read_waitq, wait);
653 	if (!list_empty(&u->read_buffers))
654 		return POLLIN | POLLRDNORM;
655 	return 0;
656 }
657 
658 const struct file_operations xen_xenbus_fops = {
659 	.read = xenbus_file_read,
660 	.write = xenbus_file_write,
661 	.open = xenbus_file_open,
662 	.release = xenbus_file_release,
663 	.poll = xenbus_file_poll,
664 	.llseek = no_llseek,
665 };
666 EXPORT_SYMBOL_GPL(xen_xenbus_fops);
667 
668 static struct miscdevice xenbus_dev = {
669 	.minor = MISC_DYNAMIC_MINOR,
670 	.name = "xen/xenbus",
671 	.fops = &xen_xenbus_fops,
672 };
673 
674 static int __init xenbus_init(void)
675 {
676 	int err;
677 
678 	if (!xen_domain())
679 		return -ENODEV;
680 
681 	err = misc_register(&xenbus_dev);
682 	if (err)
683 		pr_err("Could not register xenbus frontend device\n");
684 	return err;
685 }
686 device_initcall(xenbus_init);
687