xref: /openbmc/linux/drivers/xen/xenbus/xenbus_xs.c (revision 4800cd83)
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 "xenbus_comms.h"
49 
50 struct xs_stored_msg {
51 	struct list_head list;
52 
53 	struct xsd_sockmsg hdr;
54 
55 	union {
56 		/* Queued replies. */
57 		struct {
58 			char *body;
59 		} reply;
60 
61 		/* Queued watch events. */
62 		struct {
63 			struct xenbus_watch *handle;
64 			char **vec;
65 			unsigned int vec_size;
66 		} watch;
67 	} u;
68 };
69 
70 struct xs_handle {
71 	/* A list of replies. Currently only one will ever be outstanding. */
72 	struct list_head reply_list;
73 	spinlock_t reply_lock;
74 	wait_queue_head_t reply_waitq;
75 
76 	/*
77 	 * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex.
78 	 * response_mutex is never taken simultaneously with the other three.
79 	 *
80 	 * transaction_mutex must be held before incrementing
81 	 * transaction_count. The mutex is held when a suspend is in
82 	 * progress to prevent new transactions starting.
83 	 *
84 	 * When decrementing transaction_count to zero the wait queue
85 	 * should be woken up, the suspend code waits for count to
86 	 * reach zero.
87 	 */
88 
89 	/* One request at a time. */
90 	struct mutex request_mutex;
91 
92 	/* Protect xenbus reader thread against save/restore. */
93 	struct mutex response_mutex;
94 
95 	/* Protect transactions against save/restore. */
96 	struct mutex transaction_mutex;
97 	atomic_t transaction_count;
98 	wait_queue_head_t transaction_wq;
99 
100 	/* Protect watch (de)register against save/restore. */
101 	struct rw_semaphore watch_mutex;
102 };
103 
104 static struct xs_handle xs_state;
105 
106 /* List of registered watches, and a lock to protect it. */
107 static LIST_HEAD(watches);
108 static DEFINE_SPINLOCK(watches_lock);
109 
110 /* List of pending watch callback events, and a lock to protect it. */
111 static LIST_HEAD(watch_events);
112 static DEFINE_SPINLOCK(watch_events_lock);
113 
114 /*
115  * Details of the xenwatch callback kernel thread. The thread waits on the
116  * watch_events_waitq for work to do (queued on watch_events list). When it
117  * wakes up it acquires the xenwatch_mutex before reading the list and
118  * carrying out work.
119  */
120 static pid_t xenwatch_pid;
121 static DEFINE_MUTEX(xenwatch_mutex);
122 static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq);
123 
124 static int get_error(const char *errorstring)
125 {
126 	unsigned int i;
127 
128 	for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) {
129 		if (i == ARRAY_SIZE(xsd_errors) - 1) {
130 			printk(KERN_WARNING
131 			       "XENBUS xen store gave: unknown error %s",
132 			       errorstring);
133 			return EINVAL;
134 		}
135 	}
136 	return xsd_errors[i].errnum;
137 }
138 
139 static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len)
140 {
141 	struct xs_stored_msg *msg;
142 	char *body;
143 
144 	spin_lock(&xs_state.reply_lock);
145 
146 	while (list_empty(&xs_state.reply_list)) {
147 		spin_unlock(&xs_state.reply_lock);
148 		/* XXX FIXME: Avoid synchronous wait for response here. */
149 		wait_event(xs_state.reply_waitq,
150 			   !list_empty(&xs_state.reply_list));
151 		spin_lock(&xs_state.reply_lock);
152 	}
153 
154 	msg = list_entry(xs_state.reply_list.next,
155 			 struct xs_stored_msg, list);
156 	list_del(&msg->list);
157 
158 	spin_unlock(&xs_state.reply_lock);
159 
160 	*type = msg->hdr.type;
161 	if (len)
162 		*len = msg->hdr.len;
163 	body = msg->u.reply.body;
164 
165 	kfree(msg);
166 
167 	return body;
168 }
169 
170 static void transaction_start(void)
171 {
172 	mutex_lock(&xs_state.transaction_mutex);
173 	atomic_inc(&xs_state.transaction_count);
174 	mutex_unlock(&xs_state.transaction_mutex);
175 }
176 
177 static void transaction_end(void)
178 {
179 	if (atomic_dec_and_test(&xs_state.transaction_count))
180 		wake_up(&xs_state.transaction_wq);
181 }
182 
183 static void transaction_suspend(void)
184 {
185 	mutex_lock(&xs_state.transaction_mutex);
186 	wait_event(xs_state.transaction_wq,
187 		   atomic_read(&xs_state.transaction_count) == 0);
188 }
189 
190 static void transaction_resume(void)
191 {
192 	mutex_unlock(&xs_state.transaction_mutex);
193 }
194 
195 void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg)
196 {
197 	void *ret;
198 	struct xsd_sockmsg req_msg = *msg;
199 	int err;
200 
201 	if (req_msg.type == XS_TRANSACTION_START)
202 		transaction_start();
203 
204 	mutex_lock(&xs_state.request_mutex);
205 
206 	err = xb_write(msg, sizeof(*msg) + msg->len);
207 	if (err) {
208 		msg->type = XS_ERROR;
209 		ret = ERR_PTR(err);
210 	} else
211 		ret = read_reply(&msg->type, &msg->len);
212 
213 	mutex_unlock(&xs_state.request_mutex);
214 
215 	if ((msg->type == XS_TRANSACTION_END) ||
216 	    ((req_msg.type == XS_TRANSACTION_START) &&
217 	     (msg->type == XS_ERROR)))
218 		transaction_end();
219 
220 	return ret;
221 }
222 EXPORT_SYMBOL(xenbus_dev_request_and_reply);
223 
224 /* Send message to xs, get kmalloc'ed reply.  ERR_PTR() on error. */
225 static void *xs_talkv(struct xenbus_transaction t,
226 		      enum xsd_sockmsg_type type,
227 		      const struct kvec *iovec,
228 		      unsigned int num_vecs,
229 		      unsigned int *len)
230 {
231 	struct xsd_sockmsg msg;
232 	void *ret = NULL;
233 	unsigned int i;
234 	int err;
235 
236 	msg.tx_id = t.id;
237 	msg.req_id = 0;
238 	msg.type = type;
239 	msg.len = 0;
240 	for (i = 0; i < num_vecs; i++)
241 		msg.len += iovec[i].iov_len;
242 
243 	mutex_lock(&xs_state.request_mutex);
244 
245 	err = xb_write(&msg, sizeof(msg));
246 	if (err) {
247 		mutex_unlock(&xs_state.request_mutex);
248 		return ERR_PTR(err);
249 	}
250 
251 	for (i = 0; i < num_vecs; i++) {
252 		err = xb_write(iovec[i].iov_base, iovec[i].iov_len);
253 		if (err) {
254 			mutex_unlock(&xs_state.request_mutex);
255 			return ERR_PTR(err);
256 		}
257 	}
258 
259 	ret = read_reply(&msg.type, len);
260 
261 	mutex_unlock(&xs_state.request_mutex);
262 
263 	if (IS_ERR(ret))
264 		return ret;
265 
266 	if (msg.type == XS_ERROR) {
267 		err = get_error(ret);
268 		kfree(ret);
269 		return ERR_PTR(-err);
270 	}
271 
272 	if (msg.type != type) {
273 		if (printk_ratelimit())
274 			printk(KERN_WARNING
275 			       "XENBUS unexpected type [%d], expected [%d]\n",
276 			       msg.type, type);
277 		kfree(ret);
278 		return ERR_PTR(-EINVAL);
279 	}
280 	return ret;
281 }
282 
283 /* Simplified version of xs_talkv: single message. */
284 static void *xs_single(struct xenbus_transaction t,
285 		       enum xsd_sockmsg_type type,
286 		       const char *string,
287 		       unsigned int *len)
288 {
289 	struct kvec iovec;
290 
291 	iovec.iov_base = (void *)string;
292 	iovec.iov_len = strlen(string) + 1;
293 	return xs_talkv(t, type, &iovec, 1, len);
294 }
295 
296 /* Many commands only need an ack, don't care what it says. */
297 static int xs_error(char *reply)
298 {
299 	if (IS_ERR(reply))
300 		return PTR_ERR(reply);
301 	kfree(reply);
302 	return 0;
303 }
304 
305 static unsigned int count_strings(const char *strings, unsigned int len)
306 {
307 	unsigned int num;
308 	const char *p;
309 
310 	for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
311 		num++;
312 
313 	return num;
314 }
315 
316 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
317 static char *join(const char *dir, const char *name)
318 {
319 	char *buffer;
320 
321 	if (strlen(name) == 0)
322 		buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
323 	else
324 		buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
325 	return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
326 }
327 
328 static char **split(char *strings, unsigned int len, unsigned int *num)
329 {
330 	char *p, **ret;
331 
332 	/* Count the strings. */
333 	*num = count_strings(strings, len);
334 
335 	/* Transfer to one big alloc for easy freeing. */
336 	ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
337 	if (!ret) {
338 		kfree(strings);
339 		return ERR_PTR(-ENOMEM);
340 	}
341 	memcpy(&ret[*num], strings, len);
342 	kfree(strings);
343 
344 	strings = (char *)&ret[*num];
345 	for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
346 		ret[(*num)++] = p;
347 
348 	return ret;
349 }
350 
351 char **xenbus_directory(struct xenbus_transaction t,
352 			const char *dir, const char *node, unsigned int *num)
353 {
354 	char *strings, *path;
355 	unsigned int len;
356 
357 	path = join(dir, node);
358 	if (IS_ERR(path))
359 		return (char **)path;
360 
361 	strings = xs_single(t, XS_DIRECTORY, path, &len);
362 	kfree(path);
363 	if (IS_ERR(strings))
364 		return (char **)strings;
365 
366 	return split(strings, len, num);
367 }
368 EXPORT_SYMBOL_GPL(xenbus_directory);
369 
370 /* Check if a path exists. Return 1 if it does. */
371 int xenbus_exists(struct xenbus_transaction t,
372 		  const char *dir, const char *node)
373 {
374 	char **d;
375 	int dir_n;
376 
377 	d = xenbus_directory(t, dir, node, &dir_n);
378 	if (IS_ERR(d))
379 		return 0;
380 	kfree(d);
381 	return 1;
382 }
383 EXPORT_SYMBOL_GPL(xenbus_exists);
384 
385 /* Get the value of a single file.
386  * Returns a kmalloced value: call free() on it after use.
387  * len indicates length in bytes.
388  */
389 void *xenbus_read(struct xenbus_transaction t,
390 		  const char *dir, const char *node, unsigned int *len)
391 {
392 	char *path;
393 	void *ret;
394 
395 	path = join(dir, node);
396 	if (IS_ERR(path))
397 		return (void *)path;
398 
399 	ret = xs_single(t, XS_READ, path, len);
400 	kfree(path);
401 	return ret;
402 }
403 EXPORT_SYMBOL_GPL(xenbus_read);
404 
405 /* Write the value of a single file.
406  * Returns -err on failure.
407  */
408 int xenbus_write(struct xenbus_transaction t,
409 		 const char *dir, const char *node, const char *string)
410 {
411 	const char *path;
412 	struct kvec iovec[2];
413 	int ret;
414 
415 	path = join(dir, node);
416 	if (IS_ERR(path))
417 		return PTR_ERR(path);
418 
419 	iovec[0].iov_base = (void *)path;
420 	iovec[0].iov_len = strlen(path) + 1;
421 	iovec[1].iov_base = (void *)string;
422 	iovec[1].iov_len = strlen(string);
423 
424 	ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
425 	kfree(path);
426 	return ret;
427 }
428 EXPORT_SYMBOL_GPL(xenbus_write);
429 
430 /* Create a new directory. */
431 int xenbus_mkdir(struct xenbus_transaction t,
432 		 const char *dir, const char *node)
433 {
434 	char *path;
435 	int ret;
436 
437 	path = join(dir, node);
438 	if (IS_ERR(path))
439 		return PTR_ERR(path);
440 
441 	ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
442 	kfree(path);
443 	return ret;
444 }
445 EXPORT_SYMBOL_GPL(xenbus_mkdir);
446 
447 /* Destroy a file or directory (directories must be empty). */
448 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
449 {
450 	char *path;
451 	int ret;
452 
453 	path = join(dir, node);
454 	if (IS_ERR(path))
455 		return PTR_ERR(path);
456 
457 	ret = xs_error(xs_single(t, XS_RM, path, NULL));
458 	kfree(path);
459 	return ret;
460 }
461 EXPORT_SYMBOL_GPL(xenbus_rm);
462 
463 /* Start a transaction: changes by others will not be seen during this
464  * transaction, and changes will not be visible to others until end.
465  */
466 int xenbus_transaction_start(struct xenbus_transaction *t)
467 {
468 	char *id_str;
469 
470 	transaction_start();
471 
472 	id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
473 	if (IS_ERR(id_str)) {
474 		transaction_end();
475 		return PTR_ERR(id_str);
476 	}
477 
478 	t->id = simple_strtoul(id_str, NULL, 0);
479 	kfree(id_str);
480 	return 0;
481 }
482 EXPORT_SYMBOL_GPL(xenbus_transaction_start);
483 
484 /* End a transaction.
485  * If abandon is true, transaction is discarded instead of committed.
486  */
487 int xenbus_transaction_end(struct xenbus_transaction t, int abort)
488 {
489 	char abortstr[2];
490 	int err;
491 
492 	if (abort)
493 		strcpy(abortstr, "F");
494 	else
495 		strcpy(abortstr, "T");
496 
497 	err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
498 
499 	transaction_end();
500 
501 	return err;
502 }
503 EXPORT_SYMBOL_GPL(xenbus_transaction_end);
504 
505 /* Single read and scanf: returns -errno or num scanned. */
506 int xenbus_scanf(struct xenbus_transaction t,
507 		 const char *dir, const char *node, const char *fmt, ...)
508 {
509 	va_list ap;
510 	int ret;
511 	char *val;
512 
513 	val = xenbus_read(t, dir, node, NULL);
514 	if (IS_ERR(val))
515 		return PTR_ERR(val);
516 
517 	va_start(ap, fmt);
518 	ret = vsscanf(val, fmt, ap);
519 	va_end(ap);
520 	kfree(val);
521 	/* Distinctive errno. */
522 	if (ret == 0)
523 		return -ERANGE;
524 	return ret;
525 }
526 EXPORT_SYMBOL_GPL(xenbus_scanf);
527 
528 /* Single printf and write: returns -errno or 0. */
529 int xenbus_printf(struct xenbus_transaction t,
530 		  const char *dir, const char *node, const char *fmt, ...)
531 {
532 	va_list ap;
533 	int ret;
534 #define PRINTF_BUFFER_SIZE 4096
535 	char *printf_buffer;
536 
537 	printf_buffer = kmalloc(PRINTF_BUFFER_SIZE, GFP_NOIO | __GFP_HIGH);
538 	if (printf_buffer == NULL)
539 		return -ENOMEM;
540 
541 	va_start(ap, fmt);
542 	ret = vsnprintf(printf_buffer, PRINTF_BUFFER_SIZE, fmt, ap);
543 	va_end(ap);
544 
545 	BUG_ON(ret > PRINTF_BUFFER_SIZE-1);
546 	ret = xenbus_write(t, dir, node, printf_buffer);
547 
548 	kfree(printf_buffer);
549 
550 	return ret;
551 }
552 EXPORT_SYMBOL_GPL(xenbus_printf);
553 
554 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
555 int xenbus_gather(struct xenbus_transaction t, const char *dir, ...)
556 {
557 	va_list ap;
558 	const char *name;
559 	int ret = 0;
560 
561 	va_start(ap, dir);
562 	while (ret == 0 && (name = va_arg(ap, char *)) != NULL) {
563 		const char *fmt = va_arg(ap, char *);
564 		void *result = va_arg(ap, void *);
565 		char *p;
566 
567 		p = xenbus_read(t, dir, name, NULL);
568 		if (IS_ERR(p)) {
569 			ret = PTR_ERR(p);
570 			break;
571 		}
572 		if (fmt) {
573 			if (sscanf(p, fmt, result) == 0)
574 				ret = -EINVAL;
575 			kfree(p);
576 		} else
577 			*(char **)result = p;
578 	}
579 	va_end(ap);
580 	return ret;
581 }
582 EXPORT_SYMBOL_GPL(xenbus_gather);
583 
584 static int xs_watch(const char *path, const char *token)
585 {
586 	struct kvec iov[2];
587 
588 	iov[0].iov_base = (void *)path;
589 	iov[0].iov_len = strlen(path) + 1;
590 	iov[1].iov_base = (void *)token;
591 	iov[1].iov_len = strlen(token) + 1;
592 
593 	return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
594 				 ARRAY_SIZE(iov), NULL));
595 }
596 
597 static int xs_unwatch(const char *path, const char *token)
598 {
599 	struct kvec iov[2];
600 
601 	iov[0].iov_base = (char *)path;
602 	iov[0].iov_len = strlen(path) + 1;
603 	iov[1].iov_base = (char *)token;
604 	iov[1].iov_len = strlen(token) + 1;
605 
606 	return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
607 				 ARRAY_SIZE(iov), NULL));
608 }
609 
610 static struct xenbus_watch *find_watch(const char *token)
611 {
612 	struct xenbus_watch *i, *cmp;
613 
614 	cmp = (void *)simple_strtoul(token, NULL, 16);
615 
616 	list_for_each_entry(i, &watches, list)
617 		if (i == cmp)
618 			return i;
619 
620 	return NULL;
621 }
622 
623 /* Register callback to watch this node. */
624 int register_xenbus_watch(struct xenbus_watch *watch)
625 {
626 	/* Pointer in ascii is the token. */
627 	char token[sizeof(watch) * 2 + 1];
628 	int err;
629 
630 	sprintf(token, "%lX", (long)watch);
631 
632 	down_read(&xs_state.watch_mutex);
633 
634 	spin_lock(&watches_lock);
635 	BUG_ON(find_watch(token));
636 	list_add(&watch->list, &watches);
637 	spin_unlock(&watches_lock);
638 
639 	err = xs_watch(watch->node, token);
640 
641 	/* Ignore errors due to multiple registration. */
642 	if ((err != 0) && (err != -EEXIST)) {
643 		spin_lock(&watches_lock);
644 		list_del(&watch->list);
645 		spin_unlock(&watches_lock);
646 	}
647 
648 	up_read(&xs_state.watch_mutex);
649 
650 	return err;
651 }
652 EXPORT_SYMBOL_GPL(register_xenbus_watch);
653 
654 void unregister_xenbus_watch(struct xenbus_watch *watch)
655 {
656 	struct xs_stored_msg *msg, *tmp;
657 	char token[sizeof(watch) * 2 + 1];
658 	int err;
659 
660 	sprintf(token, "%lX", (long)watch);
661 
662 	down_read(&xs_state.watch_mutex);
663 
664 	spin_lock(&watches_lock);
665 	BUG_ON(!find_watch(token));
666 	list_del(&watch->list);
667 	spin_unlock(&watches_lock);
668 
669 	err = xs_unwatch(watch->node, token);
670 	if (err)
671 		printk(KERN_WARNING
672 		       "XENBUS Failed to release watch %s: %i\n",
673 		       watch->node, err);
674 
675 	up_read(&xs_state.watch_mutex);
676 
677 	/* Make sure there are no callbacks running currently (unless
678 	   its us) */
679 	if (current->pid != xenwatch_pid)
680 		mutex_lock(&xenwatch_mutex);
681 
682 	/* Cancel pending watch events. */
683 	spin_lock(&watch_events_lock);
684 	list_for_each_entry_safe(msg, tmp, &watch_events, list) {
685 		if (msg->u.watch.handle != watch)
686 			continue;
687 		list_del(&msg->list);
688 		kfree(msg->u.watch.vec);
689 		kfree(msg);
690 	}
691 	spin_unlock(&watch_events_lock);
692 
693 	if (current->pid != xenwatch_pid)
694 		mutex_unlock(&xenwatch_mutex);
695 }
696 EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
697 
698 void xs_suspend(void)
699 {
700 	transaction_suspend();
701 	down_write(&xs_state.watch_mutex);
702 	mutex_lock(&xs_state.request_mutex);
703 	mutex_lock(&xs_state.response_mutex);
704 }
705 
706 void xs_resume(void)
707 {
708 	struct xenbus_watch *watch;
709 	char token[sizeof(watch) * 2 + 1];
710 
711 	xb_init_comms();
712 
713 	mutex_unlock(&xs_state.response_mutex);
714 	mutex_unlock(&xs_state.request_mutex);
715 	transaction_resume();
716 
717 	/* No need for watches_lock: the watch_mutex is sufficient. */
718 	list_for_each_entry(watch, &watches, list) {
719 		sprintf(token, "%lX", (long)watch);
720 		xs_watch(watch->node, token);
721 	}
722 
723 	up_write(&xs_state.watch_mutex);
724 }
725 
726 void xs_suspend_cancel(void)
727 {
728 	mutex_unlock(&xs_state.response_mutex);
729 	mutex_unlock(&xs_state.request_mutex);
730 	up_write(&xs_state.watch_mutex);
731 	mutex_unlock(&xs_state.transaction_mutex);
732 }
733 
734 static int xenwatch_thread(void *unused)
735 {
736 	struct list_head *ent;
737 	struct xs_stored_msg *msg;
738 
739 	for (;;) {
740 		wait_event_interruptible(watch_events_waitq,
741 					 !list_empty(&watch_events));
742 
743 		if (kthread_should_stop())
744 			break;
745 
746 		mutex_lock(&xenwatch_mutex);
747 
748 		spin_lock(&watch_events_lock);
749 		ent = watch_events.next;
750 		if (ent != &watch_events)
751 			list_del(ent);
752 		spin_unlock(&watch_events_lock);
753 
754 		if (ent != &watch_events) {
755 			msg = list_entry(ent, struct xs_stored_msg, list);
756 			msg->u.watch.handle->callback(
757 				msg->u.watch.handle,
758 				(const char **)msg->u.watch.vec,
759 				msg->u.watch.vec_size);
760 			kfree(msg->u.watch.vec);
761 			kfree(msg);
762 		}
763 
764 		mutex_unlock(&xenwatch_mutex);
765 	}
766 
767 	return 0;
768 }
769 
770 static int process_msg(void)
771 {
772 	struct xs_stored_msg *msg;
773 	char *body;
774 	int err;
775 
776 	/*
777 	 * We must disallow save/restore while reading a xenstore message.
778 	 * A partial read across s/r leaves us out of sync with xenstored.
779 	 */
780 	for (;;) {
781 		err = xb_wait_for_data_to_read();
782 		if (err)
783 			return err;
784 		mutex_lock(&xs_state.response_mutex);
785 		if (xb_data_to_read())
786 			break;
787 		/* We raced with save/restore: pending data 'disappeared'. */
788 		mutex_unlock(&xs_state.response_mutex);
789 	}
790 
791 
792 	msg = kmalloc(sizeof(*msg), GFP_NOIO | __GFP_HIGH);
793 	if (msg == NULL) {
794 		err = -ENOMEM;
795 		goto out;
796 	}
797 
798 	err = xb_read(&msg->hdr, sizeof(msg->hdr));
799 	if (err) {
800 		kfree(msg);
801 		goto out;
802 	}
803 
804 	body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH);
805 	if (body == NULL) {
806 		kfree(msg);
807 		err = -ENOMEM;
808 		goto out;
809 	}
810 
811 	err = xb_read(body, msg->hdr.len);
812 	if (err) {
813 		kfree(body);
814 		kfree(msg);
815 		goto out;
816 	}
817 	body[msg->hdr.len] = '\0';
818 
819 	if (msg->hdr.type == XS_WATCH_EVENT) {
820 		msg->u.watch.vec = split(body, msg->hdr.len,
821 					 &msg->u.watch.vec_size);
822 		if (IS_ERR(msg->u.watch.vec)) {
823 			err = PTR_ERR(msg->u.watch.vec);
824 			kfree(msg);
825 			goto out;
826 		}
827 
828 		spin_lock(&watches_lock);
829 		msg->u.watch.handle = find_watch(
830 			msg->u.watch.vec[XS_WATCH_TOKEN]);
831 		if (msg->u.watch.handle != NULL) {
832 			spin_lock(&watch_events_lock);
833 			list_add_tail(&msg->list, &watch_events);
834 			wake_up(&watch_events_waitq);
835 			spin_unlock(&watch_events_lock);
836 		} else {
837 			kfree(msg->u.watch.vec);
838 			kfree(msg);
839 		}
840 		spin_unlock(&watches_lock);
841 	} else {
842 		msg->u.reply.body = body;
843 		spin_lock(&xs_state.reply_lock);
844 		list_add_tail(&msg->list, &xs_state.reply_list);
845 		spin_unlock(&xs_state.reply_lock);
846 		wake_up(&xs_state.reply_waitq);
847 	}
848 
849  out:
850 	mutex_unlock(&xs_state.response_mutex);
851 	return err;
852 }
853 
854 static int xenbus_thread(void *unused)
855 {
856 	int err;
857 
858 	for (;;) {
859 		err = process_msg();
860 		if (err)
861 			printk(KERN_WARNING "XENBUS error %d while reading "
862 			       "message\n", err);
863 		if (kthread_should_stop())
864 			break;
865 	}
866 
867 	return 0;
868 }
869 
870 int xs_init(void)
871 {
872 	int err;
873 	struct task_struct *task;
874 
875 	INIT_LIST_HEAD(&xs_state.reply_list);
876 	spin_lock_init(&xs_state.reply_lock);
877 	init_waitqueue_head(&xs_state.reply_waitq);
878 
879 	mutex_init(&xs_state.request_mutex);
880 	mutex_init(&xs_state.response_mutex);
881 	mutex_init(&xs_state.transaction_mutex);
882 	init_rwsem(&xs_state.watch_mutex);
883 	atomic_set(&xs_state.transaction_count, 0);
884 	init_waitqueue_head(&xs_state.transaction_wq);
885 
886 	/* Initialize the shared memory rings to talk to xenstored */
887 	err = xb_init_comms();
888 	if (err)
889 		return err;
890 
891 	task = kthread_run(xenwatch_thread, NULL, "xenwatch");
892 	if (IS_ERR(task))
893 		return PTR_ERR(task);
894 	xenwatch_pid = task->pid;
895 
896 	task = kthread_run(xenbus_thread, NULL, "xenbus");
897 	if (IS_ERR(task))
898 		return PTR_ERR(task);
899 
900 	return 0;
901 }
902