xref: /openbmc/linux/drivers/xen/xenbus/xenbus_xs.c (revision 63dc02bd)
1 /******************************************************************************
2  * xenbus_xs.c
3  *
4  * This is the kernel equivalent of the "xs" library.  We don't need everything
5  * and we use xenbus_comms for communication.
6  *
7  * Copyright (C) 2005 Rusty Russell, IBM Corporation
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License version 2
11  * as published by the Free Software Foundation; or, when distributed
12  * separately from the Linux kernel or incorporated into other
13  * software packages, subject to the following license:
14  *
15  * Permission is hereby granted, free of charge, to any person obtaining a copy
16  * of this source file (the "Software"), to deal in the Software without
17  * restriction, including without limitation the rights to use, copy, modify,
18  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
19  * and to permit persons to whom the Software is furnished to do so, subject to
20  * the following conditions:
21  *
22  * The above copyright notice and this permission notice shall be included in
23  * all copies or substantial portions of the Software.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
26  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
27  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
28  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
29  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
30  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
31  * IN THE SOFTWARE.
32  */
33 
34 #include <linux/unistd.h>
35 #include <linux/errno.h>
36 #include <linux/types.h>
37 #include <linux/uio.h>
38 #include <linux/kernel.h>
39 #include <linux/string.h>
40 #include <linux/err.h>
41 #include <linux/slab.h>
42 #include <linux/fcntl.h>
43 #include <linux/kthread.h>
44 #include <linux/rwsem.h>
45 #include <linux/module.h>
46 #include <linux/mutex.h>
47 #include <xen/xenbus.h>
48 #include <xen/xen.h>
49 #include "xenbus_comms.h"
50 
51 struct xs_stored_msg {
52 	struct list_head list;
53 
54 	struct xsd_sockmsg hdr;
55 
56 	union {
57 		/* Queued replies. */
58 		struct {
59 			char *body;
60 		} reply;
61 
62 		/* Queued watch events. */
63 		struct {
64 			struct xenbus_watch *handle;
65 			char **vec;
66 			unsigned int vec_size;
67 		} watch;
68 	} u;
69 };
70 
71 struct xs_handle {
72 	/* A list of replies. Currently only one will ever be outstanding. */
73 	struct list_head reply_list;
74 	spinlock_t reply_lock;
75 	wait_queue_head_t reply_waitq;
76 
77 	/*
78 	 * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex.
79 	 * response_mutex is never taken simultaneously with the other three.
80 	 *
81 	 * transaction_mutex must be held before incrementing
82 	 * transaction_count. The mutex is held when a suspend is in
83 	 * progress to prevent new transactions starting.
84 	 *
85 	 * When decrementing transaction_count to zero the wait queue
86 	 * should be woken up, the suspend code waits for count to
87 	 * reach zero.
88 	 */
89 
90 	/* One request at a time. */
91 	struct mutex request_mutex;
92 
93 	/* Protect xenbus reader thread against save/restore. */
94 	struct mutex response_mutex;
95 
96 	/* Protect transactions against save/restore. */
97 	struct mutex transaction_mutex;
98 	atomic_t transaction_count;
99 	wait_queue_head_t transaction_wq;
100 
101 	/* Protect watch (de)register against save/restore. */
102 	struct rw_semaphore watch_mutex;
103 };
104 
105 static struct xs_handle xs_state;
106 
107 /* List of registered watches, and a lock to protect it. */
108 static LIST_HEAD(watches);
109 static DEFINE_SPINLOCK(watches_lock);
110 
111 /* List of pending watch callback events, and a lock to protect it. */
112 static LIST_HEAD(watch_events);
113 static DEFINE_SPINLOCK(watch_events_lock);
114 
115 /*
116  * Details of the xenwatch callback kernel thread. The thread waits on the
117  * watch_events_waitq for work to do (queued on watch_events list). When it
118  * wakes up it acquires the xenwatch_mutex before reading the list and
119  * carrying out work.
120  */
121 static pid_t xenwatch_pid;
122 static DEFINE_MUTEX(xenwatch_mutex);
123 static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq);
124 
125 static int get_error(const char *errorstring)
126 {
127 	unsigned int i;
128 
129 	for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) {
130 		if (i == ARRAY_SIZE(xsd_errors) - 1) {
131 			printk(KERN_WARNING
132 			       "XENBUS xen store gave: unknown error %s",
133 			       errorstring);
134 			return EINVAL;
135 		}
136 	}
137 	return xsd_errors[i].errnum;
138 }
139 
140 static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len)
141 {
142 	struct xs_stored_msg *msg;
143 	char *body;
144 
145 	spin_lock(&xs_state.reply_lock);
146 
147 	while (list_empty(&xs_state.reply_list)) {
148 		spin_unlock(&xs_state.reply_lock);
149 		/* XXX FIXME: Avoid synchronous wait for response here. */
150 		wait_event(xs_state.reply_waitq,
151 			   !list_empty(&xs_state.reply_list));
152 		spin_lock(&xs_state.reply_lock);
153 	}
154 
155 	msg = list_entry(xs_state.reply_list.next,
156 			 struct xs_stored_msg, list);
157 	list_del(&msg->list);
158 
159 	spin_unlock(&xs_state.reply_lock);
160 
161 	*type = msg->hdr.type;
162 	if (len)
163 		*len = msg->hdr.len;
164 	body = msg->u.reply.body;
165 
166 	kfree(msg);
167 
168 	return body;
169 }
170 
171 static void transaction_start(void)
172 {
173 	mutex_lock(&xs_state.transaction_mutex);
174 	atomic_inc(&xs_state.transaction_count);
175 	mutex_unlock(&xs_state.transaction_mutex);
176 }
177 
178 static void transaction_end(void)
179 {
180 	if (atomic_dec_and_test(&xs_state.transaction_count))
181 		wake_up(&xs_state.transaction_wq);
182 }
183 
184 static void transaction_suspend(void)
185 {
186 	mutex_lock(&xs_state.transaction_mutex);
187 	wait_event(xs_state.transaction_wq,
188 		   atomic_read(&xs_state.transaction_count) == 0);
189 }
190 
191 static void transaction_resume(void)
192 {
193 	mutex_unlock(&xs_state.transaction_mutex);
194 }
195 
196 void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg)
197 {
198 	void *ret;
199 	struct xsd_sockmsg req_msg = *msg;
200 	int err;
201 
202 	if (req_msg.type == XS_TRANSACTION_START)
203 		transaction_start();
204 
205 	mutex_lock(&xs_state.request_mutex);
206 
207 	err = xb_write(msg, sizeof(*msg) + msg->len);
208 	if (err) {
209 		msg->type = XS_ERROR;
210 		ret = ERR_PTR(err);
211 	} else
212 		ret = read_reply(&msg->type, &msg->len);
213 
214 	mutex_unlock(&xs_state.request_mutex);
215 
216 	if ((msg->type == XS_TRANSACTION_END) ||
217 	    ((req_msg.type == XS_TRANSACTION_START) &&
218 	     (msg->type == XS_ERROR)))
219 		transaction_end();
220 
221 	return ret;
222 }
223 EXPORT_SYMBOL(xenbus_dev_request_and_reply);
224 
225 /* Send message to xs, get kmalloc'ed reply.  ERR_PTR() on error. */
226 static void *xs_talkv(struct xenbus_transaction t,
227 		      enum xsd_sockmsg_type type,
228 		      const struct kvec *iovec,
229 		      unsigned int num_vecs,
230 		      unsigned int *len)
231 {
232 	struct xsd_sockmsg msg;
233 	void *ret = NULL;
234 	unsigned int i;
235 	int err;
236 
237 	msg.tx_id = t.id;
238 	msg.req_id = 0;
239 	msg.type = type;
240 	msg.len = 0;
241 	for (i = 0; i < num_vecs; i++)
242 		msg.len += iovec[i].iov_len;
243 
244 	mutex_lock(&xs_state.request_mutex);
245 
246 	err = xb_write(&msg, sizeof(msg));
247 	if (err) {
248 		mutex_unlock(&xs_state.request_mutex);
249 		return ERR_PTR(err);
250 	}
251 
252 	for (i = 0; i < num_vecs; i++) {
253 		err = xb_write(iovec[i].iov_base, iovec[i].iov_len);
254 		if (err) {
255 			mutex_unlock(&xs_state.request_mutex);
256 			return ERR_PTR(err);
257 		}
258 	}
259 
260 	ret = read_reply(&msg.type, len);
261 
262 	mutex_unlock(&xs_state.request_mutex);
263 
264 	if (IS_ERR(ret))
265 		return ret;
266 
267 	if (msg.type == XS_ERROR) {
268 		err = get_error(ret);
269 		kfree(ret);
270 		return ERR_PTR(-err);
271 	}
272 
273 	if (msg.type != type) {
274 		if (printk_ratelimit())
275 			printk(KERN_WARNING
276 			       "XENBUS unexpected type [%d], expected [%d]\n",
277 			       msg.type, type);
278 		kfree(ret);
279 		return ERR_PTR(-EINVAL);
280 	}
281 	return ret;
282 }
283 
284 /* Simplified version of xs_talkv: single message. */
285 static void *xs_single(struct xenbus_transaction t,
286 		       enum xsd_sockmsg_type type,
287 		       const char *string,
288 		       unsigned int *len)
289 {
290 	struct kvec iovec;
291 
292 	iovec.iov_base = (void *)string;
293 	iovec.iov_len = strlen(string) + 1;
294 	return xs_talkv(t, type, &iovec, 1, len);
295 }
296 
297 /* Many commands only need an ack, don't care what it says. */
298 static int xs_error(char *reply)
299 {
300 	if (IS_ERR(reply))
301 		return PTR_ERR(reply);
302 	kfree(reply);
303 	return 0;
304 }
305 
306 static unsigned int count_strings(const char *strings, unsigned int len)
307 {
308 	unsigned int num;
309 	const char *p;
310 
311 	for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
312 		num++;
313 
314 	return num;
315 }
316 
317 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
318 static char *join(const char *dir, const char *name)
319 {
320 	char *buffer;
321 
322 	if (strlen(name) == 0)
323 		buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
324 	else
325 		buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
326 	return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
327 }
328 
329 static char **split(char *strings, unsigned int len, unsigned int *num)
330 {
331 	char *p, **ret;
332 
333 	/* Count the strings. */
334 	*num = count_strings(strings, len);
335 
336 	/* Transfer to one big alloc for easy freeing. */
337 	ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
338 	if (!ret) {
339 		kfree(strings);
340 		return ERR_PTR(-ENOMEM);
341 	}
342 	memcpy(&ret[*num], strings, len);
343 	kfree(strings);
344 
345 	strings = (char *)&ret[*num];
346 	for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
347 		ret[(*num)++] = p;
348 
349 	return ret;
350 }
351 
352 char **xenbus_directory(struct xenbus_transaction t,
353 			const char *dir, const char *node, unsigned int *num)
354 {
355 	char *strings, *path;
356 	unsigned int len;
357 
358 	path = join(dir, node);
359 	if (IS_ERR(path))
360 		return (char **)path;
361 
362 	strings = xs_single(t, XS_DIRECTORY, path, &len);
363 	kfree(path);
364 	if (IS_ERR(strings))
365 		return (char **)strings;
366 
367 	return split(strings, len, num);
368 }
369 EXPORT_SYMBOL_GPL(xenbus_directory);
370 
371 /* Check if a path exists. Return 1 if it does. */
372 int xenbus_exists(struct xenbus_transaction t,
373 		  const char *dir, const char *node)
374 {
375 	char **d;
376 	int dir_n;
377 
378 	d = xenbus_directory(t, dir, node, &dir_n);
379 	if (IS_ERR(d))
380 		return 0;
381 	kfree(d);
382 	return 1;
383 }
384 EXPORT_SYMBOL_GPL(xenbus_exists);
385 
386 /* Get the value of a single file.
387  * Returns a kmalloced value: call free() on it after use.
388  * len indicates length in bytes.
389  */
390 void *xenbus_read(struct xenbus_transaction t,
391 		  const char *dir, const char *node, unsigned int *len)
392 {
393 	char *path;
394 	void *ret;
395 
396 	path = join(dir, node);
397 	if (IS_ERR(path))
398 		return (void *)path;
399 
400 	ret = xs_single(t, XS_READ, path, len);
401 	kfree(path);
402 	return ret;
403 }
404 EXPORT_SYMBOL_GPL(xenbus_read);
405 
406 /* Write the value of a single file.
407  * Returns -err on failure.
408  */
409 int xenbus_write(struct xenbus_transaction t,
410 		 const char *dir, const char *node, const char *string)
411 {
412 	const char *path;
413 	struct kvec iovec[2];
414 	int ret;
415 
416 	path = join(dir, node);
417 	if (IS_ERR(path))
418 		return PTR_ERR(path);
419 
420 	iovec[0].iov_base = (void *)path;
421 	iovec[0].iov_len = strlen(path) + 1;
422 	iovec[1].iov_base = (void *)string;
423 	iovec[1].iov_len = strlen(string);
424 
425 	ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
426 	kfree(path);
427 	return ret;
428 }
429 EXPORT_SYMBOL_GPL(xenbus_write);
430 
431 /* Create a new directory. */
432 int xenbus_mkdir(struct xenbus_transaction t,
433 		 const char *dir, const char *node)
434 {
435 	char *path;
436 	int ret;
437 
438 	path = join(dir, node);
439 	if (IS_ERR(path))
440 		return PTR_ERR(path);
441 
442 	ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
443 	kfree(path);
444 	return ret;
445 }
446 EXPORT_SYMBOL_GPL(xenbus_mkdir);
447 
448 /* Destroy a file or directory (directories must be empty). */
449 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
450 {
451 	char *path;
452 	int ret;
453 
454 	path = join(dir, node);
455 	if (IS_ERR(path))
456 		return PTR_ERR(path);
457 
458 	ret = xs_error(xs_single(t, XS_RM, path, NULL));
459 	kfree(path);
460 	return ret;
461 }
462 EXPORT_SYMBOL_GPL(xenbus_rm);
463 
464 /* Start a transaction: changes by others will not be seen during this
465  * transaction, and changes will not be visible to others until end.
466  */
467 int xenbus_transaction_start(struct xenbus_transaction *t)
468 {
469 	char *id_str;
470 
471 	transaction_start();
472 
473 	id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
474 	if (IS_ERR(id_str)) {
475 		transaction_end();
476 		return PTR_ERR(id_str);
477 	}
478 
479 	t->id = simple_strtoul(id_str, NULL, 0);
480 	kfree(id_str);
481 	return 0;
482 }
483 EXPORT_SYMBOL_GPL(xenbus_transaction_start);
484 
485 /* End a transaction.
486  * If abandon is true, transaction is discarded instead of committed.
487  */
488 int xenbus_transaction_end(struct xenbus_transaction t, int abort)
489 {
490 	char abortstr[2];
491 	int err;
492 
493 	if (abort)
494 		strcpy(abortstr, "F");
495 	else
496 		strcpy(abortstr, "T");
497 
498 	err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
499 
500 	transaction_end();
501 
502 	return err;
503 }
504 EXPORT_SYMBOL_GPL(xenbus_transaction_end);
505 
506 /* Single read and scanf: returns -errno or num scanned. */
507 int xenbus_scanf(struct xenbus_transaction t,
508 		 const char *dir, const char *node, const char *fmt, ...)
509 {
510 	va_list ap;
511 	int ret;
512 	char *val;
513 
514 	val = xenbus_read(t, dir, node, NULL);
515 	if (IS_ERR(val))
516 		return PTR_ERR(val);
517 
518 	va_start(ap, fmt);
519 	ret = vsscanf(val, fmt, ap);
520 	va_end(ap);
521 	kfree(val);
522 	/* Distinctive errno. */
523 	if (ret == 0)
524 		return -ERANGE;
525 	return ret;
526 }
527 EXPORT_SYMBOL_GPL(xenbus_scanf);
528 
529 /* Single printf and write: returns -errno or 0. */
530 int xenbus_printf(struct xenbus_transaction t,
531 		  const char *dir, const char *node, const char *fmt, ...)
532 {
533 	va_list ap;
534 	int ret;
535 	char *buf;
536 
537 	va_start(ap, fmt);
538 	buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
539 	va_end(ap);
540 
541 	if (!buf)
542 		return -ENOMEM;
543 
544 	ret = xenbus_write(t, dir, node, buf);
545 
546 	kfree(buf);
547 
548 	return ret;
549 }
550 EXPORT_SYMBOL_GPL(xenbus_printf);
551 
552 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
553 int xenbus_gather(struct xenbus_transaction t, const char *dir, ...)
554 {
555 	va_list ap;
556 	const char *name;
557 	int ret = 0;
558 
559 	va_start(ap, dir);
560 	while (ret == 0 && (name = va_arg(ap, char *)) != NULL) {
561 		const char *fmt = va_arg(ap, char *);
562 		void *result = va_arg(ap, void *);
563 		char *p;
564 
565 		p = xenbus_read(t, dir, name, NULL);
566 		if (IS_ERR(p)) {
567 			ret = PTR_ERR(p);
568 			break;
569 		}
570 		if (fmt) {
571 			if (sscanf(p, fmt, result) == 0)
572 				ret = -EINVAL;
573 			kfree(p);
574 		} else
575 			*(char **)result = p;
576 	}
577 	va_end(ap);
578 	return ret;
579 }
580 EXPORT_SYMBOL_GPL(xenbus_gather);
581 
582 static int xs_watch(const char *path, const char *token)
583 {
584 	struct kvec iov[2];
585 
586 	iov[0].iov_base = (void *)path;
587 	iov[0].iov_len = strlen(path) + 1;
588 	iov[1].iov_base = (void *)token;
589 	iov[1].iov_len = strlen(token) + 1;
590 
591 	return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
592 				 ARRAY_SIZE(iov), NULL));
593 }
594 
595 static int xs_unwatch(const char *path, const char *token)
596 {
597 	struct kvec iov[2];
598 
599 	iov[0].iov_base = (char *)path;
600 	iov[0].iov_len = strlen(path) + 1;
601 	iov[1].iov_base = (char *)token;
602 	iov[1].iov_len = strlen(token) + 1;
603 
604 	return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
605 				 ARRAY_SIZE(iov), NULL));
606 }
607 
608 static struct xenbus_watch *find_watch(const char *token)
609 {
610 	struct xenbus_watch *i, *cmp;
611 
612 	cmp = (void *)simple_strtoul(token, NULL, 16);
613 
614 	list_for_each_entry(i, &watches, list)
615 		if (i == cmp)
616 			return i;
617 
618 	return NULL;
619 }
620 
621 /* Register callback to watch this node. */
622 int register_xenbus_watch(struct xenbus_watch *watch)
623 {
624 	/* Pointer in ascii is the token. */
625 	char token[sizeof(watch) * 2 + 1];
626 	int err;
627 
628 	sprintf(token, "%lX", (long)watch);
629 
630 	down_read(&xs_state.watch_mutex);
631 
632 	spin_lock(&watches_lock);
633 	BUG_ON(find_watch(token));
634 	list_add(&watch->list, &watches);
635 	spin_unlock(&watches_lock);
636 
637 	err = xs_watch(watch->node, token);
638 
639 	if (err) {
640 		spin_lock(&watches_lock);
641 		list_del(&watch->list);
642 		spin_unlock(&watches_lock);
643 	}
644 
645 	up_read(&xs_state.watch_mutex);
646 
647 	return err;
648 }
649 EXPORT_SYMBOL_GPL(register_xenbus_watch);
650 
651 void unregister_xenbus_watch(struct xenbus_watch *watch)
652 {
653 	struct xs_stored_msg *msg, *tmp;
654 	char token[sizeof(watch) * 2 + 1];
655 	int err;
656 
657 	sprintf(token, "%lX", (long)watch);
658 
659 	down_read(&xs_state.watch_mutex);
660 
661 	spin_lock(&watches_lock);
662 	BUG_ON(!find_watch(token));
663 	list_del(&watch->list);
664 	spin_unlock(&watches_lock);
665 
666 	err = xs_unwatch(watch->node, token);
667 	if (err)
668 		printk(KERN_WARNING
669 		       "XENBUS Failed to release watch %s: %i\n",
670 		       watch->node, err);
671 
672 	up_read(&xs_state.watch_mutex);
673 
674 	/* Make sure there are no callbacks running currently (unless
675 	   its us) */
676 	if (current->pid != xenwatch_pid)
677 		mutex_lock(&xenwatch_mutex);
678 
679 	/* Cancel pending watch events. */
680 	spin_lock(&watch_events_lock);
681 	list_for_each_entry_safe(msg, tmp, &watch_events, list) {
682 		if (msg->u.watch.handle != watch)
683 			continue;
684 		list_del(&msg->list);
685 		kfree(msg->u.watch.vec);
686 		kfree(msg);
687 	}
688 	spin_unlock(&watch_events_lock);
689 
690 	if (current->pid != xenwatch_pid)
691 		mutex_unlock(&xenwatch_mutex);
692 }
693 EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
694 
695 void xs_suspend(void)
696 {
697 	transaction_suspend();
698 	down_write(&xs_state.watch_mutex);
699 	mutex_lock(&xs_state.request_mutex);
700 	mutex_lock(&xs_state.response_mutex);
701 }
702 
703 void xs_resume(void)
704 {
705 	struct xenbus_watch *watch;
706 	char token[sizeof(watch) * 2 + 1];
707 
708 	xb_init_comms();
709 
710 	mutex_unlock(&xs_state.response_mutex);
711 	mutex_unlock(&xs_state.request_mutex);
712 	transaction_resume();
713 
714 	/* No need for watches_lock: the watch_mutex is sufficient. */
715 	list_for_each_entry(watch, &watches, list) {
716 		sprintf(token, "%lX", (long)watch);
717 		xs_watch(watch->node, token);
718 	}
719 
720 	up_write(&xs_state.watch_mutex);
721 }
722 
723 void xs_suspend_cancel(void)
724 {
725 	mutex_unlock(&xs_state.response_mutex);
726 	mutex_unlock(&xs_state.request_mutex);
727 	up_write(&xs_state.watch_mutex);
728 	mutex_unlock(&xs_state.transaction_mutex);
729 }
730 
731 static int xenwatch_thread(void *unused)
732 {
733 	struct list_head *ent;
734 	struct xs_stored_msg *msg;
735 
736 	for (;;) {
737 		wait_event_interruptible(watch_events_waitq,
738 					 !list_empty(&watch_events));
739 
740 		if (kthread_should_stop())
741 			break;
742 
743 		mutex_lock(&xenwatch_mutex);
744 
745 		spin_lock(&watch_events_lock);
746 		ent = watch_events.next;
747 		if (ent != &watch_events)
748 			list_del(ent);
749 		spin_unlock(&watch_events_lock);
750 
751 		if (ent != &watch_events) {
752 			msg = list_entry(ent, struct xs_stored_msg, list);
753 			msg->u.watch.handle->callback(
754 				msg->u.watch.handle,
755 				(const char **)msg->u.watch.vec,
756 				msg->u.watch.vec_size);
757 			kfree(msg->u.watch.vec);
758 			kfree(msg);
759 		}
760 
761 		mutex_unlock(&xenwatch_mutex);
762 	}
763 
764 	return 0;
765 }
766 
767 static int process_msg(void)
768 {
769 	struct xs_stored_msg *msg;
770 	char *body;
771 	int err;
772 
773 	/*
774 	 * We must disallow save/restore while reading a xenstore message.
775 	 * A partial read across s/r leaves us out of sync with xenstored.
776 	 */
777 	for (;;) {
778 		err = xb_wait_for_data_to_read();
779 		if (err)
780 			return err;
781 		mutex_lock(&xs_state.response_mutex);
782 		if (xb_data_to_read())
783 			break;
784 		/* We raced with save/restore: pending data 'disappeared'. */
785 		mutex_unlock(&xs_state.response_mutex);
786 	}
787 
788 
789 	msg = kmalloc(sizeof(*msg), GFP_NOIO | __GFP_HIGH);
790 	if (msg == NULL) {
791 		err = -ENOMEM;
792 		goto out;
793 	}
794 
795 	err = xb_read(&msg->hdr, sizeof(msg->hdr));
796 	if (err) {
797 		kfree(msg);
798 		goto out;
799 	}
800 
801 	if (msg->hdr.len > XENSTORE_PAYLOAD_MAX) {
802 		kfree(msg);
803 		err = -EINVAL;
804 		goto out;
805 	}
806 
807 	body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH);
808 	if (body == NULL) {
809 		kfree(msg);
810 		err = -ENOMEM;
811 		goto out;
812 	}
813 
814 	err = xb_read(body, msg->hdr.len);
815 	if (err) {
816 		kfree(body);
817 		kfree(msg);
818 		goto out;
819 	}
820 	body[msg->hdr.len] = '\0';
821 
822 	if (msg->hdr.type == XS_WATCH_EVENT) {
823 		msg->u.watch.vec = split(body, msg->hdr.len,
824 					 &msg->u.watch.vec_size);
825 		if (IS_ERR(msg->u.watch.vec)) {
826 			err = PTR_ERR(msg->u.watch.vec);
827 			kfree(msg);
828 			goto out;
829 		}
830 
831 		spin_lock(&watches_lock);
832 		msg->u.watch.handle = find_watch(
833 			msg->u.watch.vec[XS_WATCH_TOKEN]);
834 		if (msg->u.watch.handle != NULL) {
835 			spin_lock(&watch_events_lock);
836 			list_add_tail(&msg->list, &watch_events);
837 			wake_up(&watch_events_waitq);
838 			spin_unlock(&watch_events_lock);
839 		} else {
840 			kfree(msg->u.watch.vec);
841 			kfree(msg);
842 		}
843 		spin_unlock(&watches_lock);
844 	} else {
845 		msg->u.reply.body = body;
846 		spin_lock(&xs_state.reply_lock);
847 		list_add_tail(&msg->list, &xs_state.reply_list);
848 		spin_unlock(&xs_state.reply_lock);
849 		wake_up(&xs_state.reply_waitq);
850 	}
851 
852  out:
853 	mutex_unlock(&xs_state.response_mutex);
854 	return err;
855 }
856 
857 static int xenbus_thread(void *unused)
858 {
859 	int err;
860 
861 	for (;;) {
862 		err = process_msg();
863 		if (err)
864 			printk(KERN_WARNING "XENBUS error %d while reading "
865 			       "message\n", err);
866 		if (kthread_should_stop())
867 			break;
868 	}
869 
870 	return 0;
871 }
872 
873 int xs_init(void)
874 {
875 	int err;
876 	struct task_struct *task;
877 
878 	INIT_LIST_HEAD(&xs_state.reply_list);
879 	spin_lock_init(&xs_state.reply_lock);
880 	init_waitqueue_head(&xs_state.reply_waitq);
881 
882 	mutex_init(&xs_state.request_mutex);
883 	mutex_init(&xs_state.response_mutex);
884 	mutex_init(&xs_state.transaction_mutex);
885 	init_rwsem(&xs_state.watch_mutex);
886 	atomic_set(&xs_state.transaction_count, 0);
887 	init_waitqueue_head(&xs_state.transaction_wq);
888 
889 	/* Initialize the shared memory rings to talk to xenstored */
890 	err = xb_init_comms();
891 	if (err)
892 		return err;
893 
894 	task = kthread_run(xenwatch_thread, NULL, "xenwatch");
895 	if (IS_ERR(task))
896 		return PTR_ERR(task);
897 	xenwatch_pid = task->pid;
898 
899 	task = kthread_run(xenbus_thread, NULL, "xenbus");
900 	if (IS_ERR(task))
901 		return PTR_ERR(task);
902 
903 	return 0;
904 }
905