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