[go: nahoru, domu]

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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
36#include <linux/unistd.h>
37#include <linux/errno.h>
38#include <linux/types.h>
39#include <linux/uio.h>
40#include <linux/kernel.h>
41#include <linux/string.h>
42#include <linux/err.h>
43#include <linux/slab.h>
44#include <linux/fcntl.h>
45#include <linux/kthread.h>
46#include <linux/rwsem.h>
47#include <linux/module.h>
48#include <linux/mutex.h>
49#include <asm/xen/hypervisor.h>
50#include <xen/xenbus.h>
51#include <xen/xen.h>
52#include "xenbus_comms.h"
53#include "xenbus_probe.h"
54
55struct xs_stored_msg {
56	struct list_head list;
57
58	struct xsd_sockmsg hdr;
59
60	union {
61		/* Queued replies. */
62		struct {
63			char *body;
64		} reply;
65
66		/* Queued watch events. */
67		struct {
68			struct xenbus_watch *handle;
69			char **vec;
70			unsigned int vec_size;
71		} watch;
72	} u;
73};
74
75struct xs_handle {
76	/* A list of replies. Currently only one will ever be outstanding. */
77	struct list_head reply_list;
78	spinlock_t reply_lock;
79	wait_queue_head_t reply_waitq;
80
81	/*
82	 * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex.
83	 * response_mutex is never taken simultaneously with the other three.
84	 *
85	 * transaction_mutex must be held before incrementing
86	 * transaction_count. The mutex is held when a suspend is in
87	 * progress to prevent new transactions starting.
88	 *
89	 * When decrementing transaction_count to zero the wait queue
90	 * should be woken up, the suspend code waits for count to
91	 * reach zero.
92	 */
93
94	/* One request at a time. */
95	struct mutex request_mutex;
96
97	/* Protect xenbus reader thread against save/restore. */
98	struct mutex response_mutex;
99
100	/* Protect transactions against save/restore. */
101	struct mutex transaction_mutex;
102	atomic_t transaction_count;
103	wait_queue_head_t transaction_wq;
104
105	/* Protect watch (de)register against save/restore. */
106	struct rw_semaphore watch_mutex;
107};
108
109static struct xs_handle xs_state;
110
111/* List of registered watches, and a lock to protect it. */
112static LIST_HEAD(watches);
113static DEFINE_SPINLOCK(watches_lock);
114
115/* List of pending watch callback events, and a lock to protect it. */
116static LIST_HEAD(watch_events);
117static DEFINE_SPINLOCK(watch_events_lock);
118
119/*
120 * Details of the xenwatch callback kernel thread. The thread waits on the
121 * watch_events_waitq for work to do (queued on watch_events list). When it
122 * wakes up it acquires the xenwatch_mutex before reading the list and
123 * carrying out work.
124 */
125static pid_t xenwatch_pid;
126static DEFINE_MUTEX(xenwatch_mutex);
127static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq);
128
129static int get_error(const char *errorstring)
130{
131	unsigned int i;
132
133	for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) {
134		if (i == ARRAY_SIZE(xsd_errors) - 1) {
135			pr_warn("xen store gave: unknown error %s\n",
136				errorstring);
137			return EINVAL;
138		}
139	}
140	return xsd_errors[i].errnum;
141}
142
143static bool xenbus_ok(void)
144{
145	switch (xen_store_domain_type) {
146	case XS_LOCAL:
147		switch (system_state) {
148		case SYSTEM_POWER_OFF:
149		case SYSTEM_RESTART:
150		case SYSTEM_HALT:
151			return false;
152		default:
153			break;
154		}
155		return true;
156	case XS_PV:
157	case XS_HVM:
158		/* FIXME: Could check that the remote domain is alive,
159		 * but it is normally initial domain. */
160		return true;
161	default:
162		break;
163	}
164	return false;
165}
166static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len)
167{
168	struct xs_stored_msg *msg;
169	char *body;
170
171	spin_lock(&xs_state.reply_lock);
172
173	while (list_empty(&xs_state.reply_list)) {
174		spin_unlock(&xs_state.reply_lock);
175		if (xenbus_ok())
176			/* XXX FIXME: Avoid synchronous wait for response here. */
177			wait_event_timeout(xs_state.reply_waitq,
178					   !list_empty(&xs_state.reply_list),
179					   msecs_to_jiffies(500));
180		else {
181			/*
182			 * If we are in the process of being shut-down there is
183			 * no point of trying to contact XenBus - it is either
184			 * killed (xenstored application) or the other domain
185			 * has been killed or is unreachable.
186			 */
187			return ERR_PTR(-EIO);
188		}
189		spin_lock(&xs_state.reply_lock);
190	}
191
192	msg = list_entry(xs_state.reply_list.next,
193			 struct xs_stored_msg, list);
194	list_del(&msg->list);
195
196	spin_unlock(&xs_state.reply_lock);
197
198	*type = msg->hdr.type;
199	if (len)
200		*len = msg->hdr.len;
201	body = msg->u.reply.body;
202
203	kfree(msg);
204
205	return body;
206}
207
208static void transaction_start(void)
209{
210	mutex_lock(&xs_state.transaction_mutex);
211	atomic_inc(&xs_state.transaction_count);
212	mutex_unlock(&xs_state.transaction_mutex);
213}
214
215static void transaction_end(void)
216{
217	if (atomic_dec_and_test(&xs_state.transaction_count))
218		wake_up(&xs_state.transaction_wq);
219}
220
221static void transaction_suspend(void)
222{
223	mutex_lock(&xs_state.transaction_mutex);
224	wait_event(xs_state.transaction_wq,
225		   atomic_read(&xs_state.transaction_count) == 0);
226}
227
228static void transaction_resume(void)
229{
230	mutex_unlock(&xs_state.transaction_mutex);
231}
232
233void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg)
234{
235	void *ret;
236	struct xsd_sockmsg req_msg = *msg;
237	int err;
238
239	if (req_msg.type == XS_TRANSACTION_START)
240		transaction_start();
241
242	mutex_lock(&xs_state.request_mutex);
243
244	err = xb_write(msg, sizeof(*msg) + msg->len);
245	if (err) {
246		msg->type = XS_ERROR;
247		ret = ERR_PTR(err);
248	} else
249		ret = read_reply(&msg->type, &msg->len);
250
251	mutex_unlock(&xs_state.request_mutex);
252
253	if (IS_ERR(ret))
254		return ret;
255
256	if ((msg->type == XS_TRANSACTION_END) ||
257	    ((req_msg.type == XS_TRANSACTION_START) &&
258	     (msg->type == XS_ERROR)))
259		transaction_end();
260
261	return ret;
262}
263EXPORT_SYMBOL(xenbus_dev_request_and_reply);
264
265/* Send message to xs, get kmalloc'ed reply.  ERR_PTR() on error. */
266static void *xs_talkv(struct xenbus_transaction t,
267		      enum xsd_sockmsg_type type,
268		      const struct kvec *iovec,
269		      unsigned int num_vecs,
270		      unsigned int *len)
271{
272	struct xsd_sockmsg msg;
273	void *ret = NULL;
274	unsigned int i;
275	int err;
276
277	msg.tx_id = t.id;
278	msg.req_id = 0;
279	msg.type = type;
280	msg.len = 0;
281	for (i = 0; i < num_vecs; i++)
282		msg.len += iovec[i].iov_len;
283
284	mutex_lock(&xs_state.request_mutex);
285
286	err = xb_write(&msg, sizeof(msg));
287	if (err) {
288		mutex_unlock(&xs_state.request_mutex);
289		return ERR_PTR(err);
290	}
291
292	for (i = 0; i < num_vecs; i++) {
293		err = xb_write(iovec[i].iov_base, iovec[i].iov_len);
294		if (err) {
295			mutex_unlock(&xs_state.request_mutex);
296			return ERR_PTR(err);
297		}
298	}
299
300	ret = read_reply(&msg.type, len);
301
302	mutex_unlock(&xs_state.request_mutex);
303
304	if (IS_ERR(ret))
305		return ret;
306
307	if (msg.type == XS_ERROR) {
308		err = get_error(ret);
309		kfree(ret);
310		return ERR_PTR(-err);
311	}
312
313	if (msg.type != type) {
314		pr_warn_ratelimited("unexpected type [%d], expected [%d]\n",
315				    msg.type, type);
316		kfree(ret);
317		return ERR_PTR(-EINVAL);
318	}
319	return ret;
320}
321
322/* Simplified version of xs_talkv: single message. */
323static void *xs_single(struct xenbus_transaction t,
324		       enum xsd_sockmsg_type type,
325		       const char *string,
326		       unsigned int *len)
327{
328	struct kvec iovec;
329
330	iovec.iov_base = (void *)string;
331	iovec.iov_len = strlen(string) + 1;
332	return xs_talkv(t, type, &iovec, 1, len);
333}
334
335/* Many commands only need an ack, don't care what it says. */
336static int xs_error(char *reply)
337{
338	if (IS_ERR(reply))
339		return PTR_ERR(reply);
340	kfree(reply);
341	return 0;
342}
343
344static unsigned int count_strings(const char *strings, unsigned int len)
345{
346	unsigned int num;
347	const char *p;
348
349	for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
350		num++;
351
352	return num;
353}
354
355/* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
356static char *join(const char *dir, const char *name)
357{
358	char *buffer;
359
360	if (strlen(name) == 0)
361		buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
362	else
363		buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
364	return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
365}
366
367static char **split(char *strings, unsigned int len, unsigned int *num)
368{
369	char *p, **ret;
370
371	/* Count the strings. */
372	*num = count_strings(strings, len);
373
374	/* Transfer to one big alloc for easy freeing. */
375	ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
376	if (!ret) {
377		kfree(strings);
378		return ERR_PTR(-ENOMEM);
379	}
380	memcpy(&ret[*num], strings, len);
381	kfree(strings);
382
383	strings = (char *)&ret[*num];
384	for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
385		ret[(*num)++] = p;
386
387	return ret;
388}
389
390char **xenbus_directory(struct xenbus_transaction t,
391			const char *dir, const char *node, unsigned int *num)
392{
393	char *strings, *path;
394	unsigned int len;
395
396	path = join(dir, node);
397	if (IS_ERR(path))
398		return (char **)path;
399
400	strings = xs_single(t, XS_DIRECTORY, path, &len);
401	kfree(path);
402	if (IS_ERR(strings))
403		return (char **)strings;
404
405	return split(strings, len, num);
406}
407EXPORT_SYMBOL_GPL(xenbus_directory);
408
409/* Check if a path exists. Return 1 if it does. */
410int xenbus_exists(struct xenbus_transaction t,
411		  const char *dir, const char *node)
412{
413	char **d;
414	int dir_n;
415
416	d = xenbus_directory(t, dir, node, &dir_n);
417	if (IS_ERR(d))
418		return 0;
419	kfree(d);
420	return 1;
421}
422EXPORT_SYMBOL_GPL(xenbus_exists);
423
424/* Get the value of a single file.
425 * Returns a kmalloced value: call free() on it after use.
426 * len indicates length in bytes.
427 */
428void *xenbus_read(struct xenbus_transaction t,
429		  const char *dir, const char *node, unsigned int *len)
430{
431	char *path;
432	void *ret;
433
434	path = join(dir, node);
435	if (IS_ERR(path))
436		return (void *)path;
437
438	ret = xs_single(t, XS_READ, path, len);
439	kfree(path);
440	return ret;
441}
442EXPORT_SYMBOL_GPL(xenbus_read);
443
444/* Write the value of a single file.
445 * Returns -err on failure.
446 */
447int xenbus_write(struct xenbus_transaction t,
448		 const char *dir, const char *node, const char *string)
449{
450	const char *path;
451	struct kvec iovec[2];
452	int ret;
453
454	path = join(dir, node);
455	if (IS_ERR(path))
456		return PTR_ERR(path);
457
458	iovec[0].iov_base = (void *)path;
459	iovec[0].iov_len = strlen(path) + 1;
460	iovec[1].iov_base = (void *)string;
461	iovec[1].iov_len = strlen(string);
462
463	ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
464	kfree(path);
465	return ret;
466}
467EXPORT_SYMBOL_GPL(xenbus_write);
468
469/* Create a new directory. */
470int xenbus_mkdir(struct xenbus_transaction t,
471		 const char *dir, const char *node)
472{
473	char *path;
474	int ret;
475
476	path = join(dir, node);
477	if (IS_ERR(path))
478		return PTR_ERR(path);
479
480	ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
481	kfree(path);
482	return ret;
483}
484EXPORT_SYMBOL_GPL(xenbus_mkdir);
485
486/* Destroy a file or directory (directories must be empty). */
487int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
488{
489	char *path;
490	int ret;
491
492	path = join(dir, node);
493	if (IS_ERR(path))
494		return PTR_ERR(path);
495
496	ret = xs_error(xs_single(t, XS_RM, path, NULL));
497	kfree(path);
498	return ret;
499}
500EXPORT_SYMBOL_GPL(xenbus_rm);
501
502/* Start a transaction: changes by others will not be seen during this
503 * transaction, and changes will not be visible to others until end.
504 */
505int xenbus_transaction_start(struct xenbus_transaction *t)
506{
507	char *id_str;
508
509	transaction_start();
510
511	id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
512	if (IS_ERR(id_str)) {
513		transaction_end();
514		return PTR_ERR(id_str);
515	}
516
517	t->id = simple_strtoul(id_str, NULL, 0);
518	kfree(id_str);
519	return 0;
520}
521EXPORT_SYMBOL_GPL(xenbus_transaction_start);
522
523/* End a transaction.
524 * If abandon is true, transaction is discarded instead of committed.
525 */
526int xenbus_transaction_end(struct xenbus_transaction t, int abort)
527{
528	char abortstr[2];
529	int err;
530
531	if (abort)
532		strcpy(abortstr, "F");
533	else
534		strcpy(abortstr, "T");
535
536	err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
537
538	transaction_end();
539
540	return err;
541}
542EXPORT_SYMBOL_GPL(xenbus_transaction_end);
543
544/* Single read and scanf: returns -errno or num scanned. */
545int xenbus_scanf(struct xenbus_transaction t,
546		 const char *dir, const char *node, const char *fmt, ...)
547{
548	va_list ap;
549	int ret;
550	char *val;
551
552	val = xenbus_read(t, dir, node, NULL);
553	if (IS_ERR(val))
554		return PTR_ERR(val);
555
556	va_start(ap, fmt);
557	ret = vsscanf(val, fmt, ap);
558	va_end(ap);
559	kfree(val);
560	/* Distinctive errno. */
561	if (ret == 0)
562		return -ERANGE;
563	return ret;
564}
565EXPORT_SYMBOL_GPL(xenbus_scanf);
566
567/* Single printf and write: returns -errno or 0. */
568int xenbus_printf(struct xenbus_transaction t,
569		  const char *dir, const char *node, const char *fmt, ...)
570{
571	va_list ap;
572	int ret;
573	char *buf;
574
575	va_start(ap, fmt);
576	buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
577	va_end(ap);
578
579	if (!buf)
580		return -ENOMEM;
581
582	ret = xenbus_write(t, dir, node, buf);
583
584	kfree(buf);
585
586	return ret;
587}
588EXPORT_SYMBOL_GPL(xenbus_printf);
589
590/* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
591int xenbus_gather(struct xenbus_transaction t, const char *dir, ...)
592{
593	va_list ap;
594	const char *name;
595	int ret = 0;
596
597	va_start(ap, dir);
598	while (ret == 0 && (name = va_arg(ap, char *)) != NULL) {
599		const char *fmt = va_arg(ap, char *);
600		void *result = va_arg(ap, void *);
601		char *p;
602
603		p = xenbus_read(t, dir, name, NULL);
604		if (IS_ERR(p)) {
605			ret = PTR_ERR(p);
606			break;
607		}
608		if (fmt) {
609			if (sscanf(p, fmt, result) == 0)
610				ret = -EINVAL;
611			kfree(p);
612		} else
613			*(char **)result = p;
614	}
615	va_end(ap);
616	return ret;
617}
618EXPORT_SYMBOL_GPL(xenbus_gather);
619
620static int xs_watch(const char *path, const char *token)
621{
622	struct kvec iov[2];
623
624	iov[0].iov_base = (void *)path;
625	iov[0].iov_len = strlen(path) + 1;
626	iov[1].iov_base = (void *)token;
627	iov[1].iov_len = strlen(token) + 1;
628
629	return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
630				 ARRAY_SIZE(iov), NULL));
631}
632
633static int xs_unwatch(const char *path, const char *token)
634{
635	struct kvec iov[2];
636
637	iov[0].iov_base = (char *)path;
638	iov[0].iov_len = strlen(path) + 1;
639	iov[1].iov_base = (char *)token;
640	iov[1].iov_len = strlen(token) + 1;
641
642	return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
643				 ARRAY_SIZE(iov), NULL));
644}
645
646static struct xenbus_watch *find_watch(const char *token)
647{
648	struct xenbus_watch *i, *cmp;
649
650	cmp = (void *)simple_strtoul(token, NULL, 16);
651
652	list_for_each_entry(i, &watches, list)
653		if (i == cmp)
654			return i;
655
656	return NULL;
657}
658/*
659 * Certain older XenBus toolstack cannot handle reading values that are
660 * not populated. Some Xen 3.4 installation are incapable of doing this
661 * so if we are running on anything older than 4 do not attempt to read
662 * control/platform-feature-xs_reset_watches.
663 */
664static bool xen_strict_xenbus_quirk(void)
665{
666#ifdef CONFIG_X86
667	uint32_t eax, ebx, ecx, edx, base;
668
669	base = xen_cpuid_base();
670	cpuid(base + 1, &eax, &ebx, &ecx, &edx);
671
672	if ((eax >> 16) < 4)
673		return true;
674#endif
675	return false;
676
677}
678static void xs_reset_watches(void)
679{
680	int err, supported = 0;
681
682	if (!xen_hvm_domain() || xen_initial_domain())
683		return;
684
685	if (xen_strict_xenbus_quirk())
686		return;
687
688	err = xenbus_scanf(XBT_NIL, "control",
689			"platform-feature-xs_reset_watches", "%d", &supported);
690	if (err != 1 || !supported)
691		return;
692
693	err = xs_error(xs_single(XBT_NIL, XS_RESET_WATCHES, "", NULL));
694	if (err && err != -EEXIST)
695		pr_warn("xs_reset_watches failed: %d\n", err);
696}
697
698/* Register callback to watch this node. */
699int register_xenbus_watch(struct xenbus_watch *watch)
700{
701	/* Pointer in ascii is the token. */
702	char token[sizeof(watch) * 2 + 1];
703	int err;
704
705	sprintf(token, "%lX", (long)watch);
706
707	down_read(&xs_state.watch_mutex);
708
709	spin_lock(&watches_lock);
710	BUG_ON(find_watch(token));
711	list_add(&watch->list, &watches);
712	spin_unlock(&watches_lock);
713
714	err = xs_watch(watch->node, token);
715
716	if (err) {
717		spin_lock(&watches_lock);
718		list_del(&watch->list);
719		spin_unlock(&watches_lock);
720	}
721
722	up_read(&xs_state.watch_mutex);
723
724	return err;
725}
726EXPORT_SYMBOL_GPL(register_xenbus_watch);
727
728void unregister_xenbus_watch(struct xenbus_watch *watch)
729{
730	struct xs_stored_msg *msg, *tmp;
731	char token[sizeof(watch) * 2 + 1];
732	int err;
733
734	sprintf(token, "%lX", (long)watch);
735
736	down_read(&xs_state.watch_mutex);
737
738	spin_lock(&watches_lock);
739	BUG_ON(!find_watch(token));
740	list_del(&watch->list);
741	spin_unlock(&watches_lock);
742
743	err = xs_unwatch(watch->node, token);
744	if (err)
745		pr_warn("Failed to release watch %s: %i\n", watch->node, err);
746
747	up_read(&xs_state.watch_mutex);
748
749	/* Make sure there are no callbacks running currently (unless
750	   its us) */
751	if (current->pid != xenwatch_pid)
752		mutex_lock(&xenwatch_mutex);
753
754	/* Cancel pending watch events. */
755	spin_lock(&watch_events_lock);
756	list_for_each_entry_safe(msg, tmp, &watch_events, list) {
757		if (msg->u.watch.handle != watch)
758			continue;
759		list_del(&msg->list);
760		kfree(msg->u.watch.vec);
761		kfree(msg);
762	}
763	spin_unlock(&watch_events_lock);
764
765	if (current->pid != xenwatch_pid)
766		mutex_unlock(&xenwatch_mutex);
767}
768EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
769
770void xs_suspend(void)
771{
772	transaction_suspend();
773	down_write(&xs_state.watch_mutex);
774	mutex_lock(&xs_state.request_mutex);
775	mutex_lock(&xs_state.response_mutex);
776}
777
778void xs_resume(void)
779{
780	struct xenbus_watch *watch;
781	char token[sizeof(watch) * 2 + 1];
782
783	xb_init_comms();
784
785	mutex_unlock(&xs_state.response_mutex);
786	mutex_unlock(&xs_state.request_mutex);
787	transaction_resume();
788
789	/* No need for watches_lock: the watch_mutex is sufficient. */
790	list_for_each_entry(watch, &watches, list) {
791		sprintf(token, "%lX", (long)watch);
792		xs_watch(watch->node, token);
793	}
794
795	up_write(&xs_state.watch_mutex);
796}
797
798void xs_suspend_cancel(void)
799{
800	mutex_unlock(&xs_state.response_mutex);
801	mutex_unlock(&xs_state.request_mutex);
802	up_write(&xs_state.watch_mutex);
803	mutex_unlock(&xs_state.transaction_mutex);
804}
805
806static int xenwatch_thread(void *unused)
807{
808	struct list_head *ent;
809	struct xs_stored_msg *msg;
810
811	for (;;) {
812		wait_event_interruptible(watch_events_waitq,
813					 !list_empty(&watch_events));
814
815		if (kthread_should_stop())
816			break;
817
818		mutex_lock(&xenwatch_mutex);
819
820		spin_lock(&watch_events_lock);
821		ent = watch_events.next;
822		if (ent != &watch_events)
823			list_del(ent);
824		spin_unlock(&watch_events_lock);
825
826		if (ent != &watch_events) {
827			msg = list_entry(ent, struct xs_stored_msg, list);
828			msg->u.watch.handle->callback(
829				msg->u.watch.handle,
830				(const char **)msg->u.watch.vec,
831				msg->u.watch.vec_size);
832			kfree(msg->u.watch.vec);
833			kfree(msg);
834		}
835
836		mutex_unlock(&xenwatch_mutex);
837	}
838
839	return 0;
840}
841
842static int process_msg(void)
843{
844	struct xs_stored_msg *msg;
845	char *body;
846	int err;
847
848	/*
849	 * We must disallow save/restore while reading a xenstore message.
850	 * A partial read across s/r leaves us out of sync with xenstored.
851	 */
852	for (;;) {
853		err = xb_wait_for_data_to_read();
854		if (err)
855			return err;
856		mutex_lock(&xs_state.response_mutex);
857		if (xb_data_to_read())
858			break;
859		/* We raced with save/restore: pending data 'disappeared'. */
860		mutex_unlock(&xs_state.response_mutex);
861	}
862
863
864	msg = kmalloc(sizeof(*msg), GFP_NOIO | __GFP_HIGH);
865	if (msg == NULL) {
866		err = -ENOMEM;
867		goto out;
868	}
869
870	err = xb_read(&msg->hdr, sizeof(msg->hdr));
871	if (err) {
872		kfree(msg);
873		goto out;
874	}
875
876	if (msg->hdr.len > XENSTORE_PAYLOAD_MAX) {
877		kfree(msg);
878		err = -EINVAL;
879		goto out;
880	}
881
882	body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH);
883	if (body == NULL) {
884		kfree(msg);
885		err = -ENOMEM;
886		goto out;
887	}
888
889	err = xb_read(body, msg->hdr.len);
890	if (err) {
891		kfree(body);
892		kfree(msg);
893		goto out;
894	}
895	body[msg->hdr.len] = '\0';
896
897	if (msg->hdr.type == XS_WATCH_EVENT) {
898		msg->u.watch.vec = split(body, msg->hdr.len,
899					 &msg->u.watch.vec_size);
900		if (IS_ERR(msg->u.watch.vec)) {
901			err = PTR_ERR(msg->u.watch.vec);
902			kfree(msg);
903			goto out;
904		}
905
906		spin_lock(&watches_lock);
907		msg->u.watch.handle = find_watch(
908			msg->u.watch.vec[XS_WATCH_TOKEN]);
909		if (msg->u.watch.handle != NULL) {
910			spin_lock(&watch_events_lock);
911			list_add_tail(&msg->list, &watch_events);
912			wake_up(&watch_events_waitq);
913			spin_unlock(&watch_events_lock);
914		} else {
915			kfree(msg->u.watch.vec);
916			kfree(msg);
917		}
918		spin_unlock(&watches_lock);
919	} else {
920		msg->u.reply.body = body;
921		spin_lock(&xs_state.reply_lock);
922		list_add_tail(&msg->list, &xs_state.reply_list);
923		spin_unlock(&xs_state.reply_lock);
924		wake_up(&xs_state.reply_waitq);
925	}
926
927 out:
928	mutex_unlock(&xs_state.response_mutex);
929	return err;
930}
931
932static int xenbus_thread(void *unused)
933{
934	int err;
935
936	for (;;) {
937		err = process_msg();
938		if (err)
939			pr_warn("error %d while reading message\n", err);
940		if (kthread_should_stop())
941			break;
942	}
943
944	return 0;
945}
946
947int xs_init(void)
948{
949	int err;
950	struct task_struct *task;
951
952	INIT_LIST_HEAD(&xs_state.reply_list);
953	spin_lock_init(&xs_state.reply_lock);
954	init_waitqueue_head(&xs_state.reply_waitq);
955
956	mutex_init(&xs_state.request_mutex);
957	mutex_init(&xs_state.response_mutex);
958	mutex_init(&xs_state.transaction_mutex);
959	init_rwsem(&xs_state.watch_mutex);
960	atomic_set(&xs_state.transaction_count, 0);
961	init_waitqueue_head(&xs_state.transaction_wq);
962
963	/* Initialize the shared memory rings to talk to xenstored */
964	err = xb_init_comms();
965	if (err)
966		return err;
967
968	task = kthread_run(xenwatch_thread, NULL, "xenwatch");
969	if (IS_ERR(task))
970		return PTR_ERR(task);
971	xenwatch_pid = task->pid;
972
973	task = kthread_run(xenbus_thread, NULL, "xenbus");
974	if (IS_ERR(task))
975		return PTR_ERR(task);
976
977	/* shutdown watches for kexec boot */
978	xs_reset_watches();
979
980	return 0;
981}
982