Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
xenbus_xs.c
Go to the documentation of this file.
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 <asm/xen/hypervisor.h>
48 #include <xen/xenbus.h>
49 #include <xen/xen.h>
50 #include "xenbus_comms.h"
51 #include <asm/xen/hypervisor.h>
52 
53 struct xs_stored_msg {
54  struct list_head list;
55 
56  struct xsd_sockmsg hdr;
57 
58  union {
59  /* Queued replies. */
60  struct {
61  char *body;
62  } reply;
63 
64  /* Queued watch events. */
65  struct {
67  char **vec;
68  unsigned int vec_size;
69  } watch;
70  } u;
71 };
72 
73 struct xs_handle {
74  /* A list of replies. Currently only one will ever be outstanding. */
78 
79  /*
80  * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex.
81  * response_mutex is never taken simultaneously with the other three.
82  *
83  * transaction_mutex must be held before incrementing
84  * transaction_count. The mutex is held when a suspend is in
85  * progress to prevent new transactions starting.
86  *
87  * When decrementing transaction_count to zero the wait queue
88  * should be woken up, the suspend code waits for count to
89  * reach zero.
90  */
91 
92  /* One request at a time. */
94 
95  /* Protect xenbus reader thread against save/restore. */
97 
98  /* Protect transactions against save/restore. */
102 
103  /* Protect watch (de)register against save/restore. */
105 };
106 
107 static struct xs_handle xs_state;
108 
109 /* List of registered watches, and a lock to protect it. */
110 static LIST_HEAD(watches);
111 static DEFINE_SPINLOCK(watches_lock);
112 
113 /* List of pending watch callback events, and a lock to protect it. */
114 static LIST_HEAD(watch_events);
115 static DEFINE_SPINLOCK(watch_events_lock);
116 
117 /*
118  * Details of the xenwatch callback kernel thread. The thread waits on the
119  * watch_events_waitq for work to do (queued on watch_events list). When it
120  * wakes up it acquires the xenwatch_mutex before reading the list and
121  * carrying out work.
122  */
123 static pid_t xenwatch_pid;
124 static DEFINE_MUTEX(xenwatch_mutex);
125 static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq);
126 
127 static int get_error(const char *errorstring)
128 {
129  unsigned int i;
130 
131  for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) {
132  if (i == ARRAY_SIZE(xsd_errors) - 1) {
134  "XENBUS xen store gave: unknown error %s",
135  errorstring);
136  return EINVAL;
137  }
138  }
139  return xsd_errors[i].errnum;
140 }
141 
142 static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len)
143 {
144  struct xs_stored_msg *msg;
145  char *body;
146 
147  spin_lock(&xs_state.reply_lock);
148 
149  while (list_empty(&xs_state.reply_list)) {
150  spin_unlock(&xs_state.reply_lock);
151  /* XXX FIXME: Avoid synchronous wait for response here. */
152  wait_event(xs_state.reply_waitq,
153  !list_empty(&xs_state.reply_list));
154  spin_lock(&xs_state.reply_lock);
155  }
156 
157  msg = list_entry(xs_state.reply_list.next,
158  struct xs_stored_msg, list);
159  list_del(&msg->list);
160 
161  spin_unlock(&xs_state.reply_lock);
162 
163  *type = msg->hdr.type;
164  if (len)
165  *len = msg->hdr.len;
166  body = msg->u.reply.body;
167 
168  kfree(msg);
169 
170  return body;
171 }
172 
173 static void transaction_start(void)
174 {
175  mutex_lock(&xs_state.transaction_mutex);
176  atomic_inc(&xs_state.transaction_count);
177  mutex_unlock(&xs_state.transaction_mutex);
178 }
179 
180 static void transaction_end(void)
181 {
182  if (atomic_dec_and_test(&xs_state.transaction_count))
183  wake_up(&xs_state.transaction_wq);
184 }
185 
186 static void transaction_suspend(void)
187 {
188  mutex_lock(&xs_state.transaction_mutex);
189  wait_event(xs_state.transaction_wq,
190  atomic_read(&xs_state.transaction_count) == 0);
191 }
192 
193 static void transaction_resume(void)
194 {
195  mutex_unlock(&xs_state.transaction_mutex);
196 }
197 
199 {
200  void *ret;
201  struct xsd_sockmsg req_msg = *msg;
202  int err;
203 
204  if (req_msg.type == XS_TRANSACTION_START)
205  transaction_start();
206 
207  mutex_lock(&xs_state.request_mutex);
208 
209  err = xb_write(msg, sizeof(*msg) + msg->len);
210  if (err) {
211  msg->type = XS_ERROR;
212  ret = ERR_PTR(err);
213  } else
214  ret = read_reply(&msg->type, &msg->len);
215 
216  mutex_unlock(&xs_state.request_mutex);
217 
218  if ((msg->type == XS_TRANSACTION_END) ||
219  ((req_msg.type == XS_TRANSACTION_START) &&
220  (msg->type == XS_ERROR)))
221  transaction_end();
222 
223  return ret;
224 }
226 
227 /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */
228 static void *xs_talkv(struct xenbus_transaction t,
229  enum xsd_sockmsg_type type,
230  const struct kvec *iovec,
231  unsigned int num_vecs,
232  unsigned int *len)
233 {
234  struct xsd_sockmsg msg;
235  void *ret = NULL;
236  unsigned int i;
237  int err;
238 
239  msg.tx_id = t.id;
240  msg.req_id = 0;
241  msg.type = type;
242  msg.len = 0;
243  for (i = 0; i < num_vecs; i++)
244  msg.len += iovec[i].iov_len;
245 
246  mutex_lock(&xs_state.request_mutex);
247 
248  err = xb_write(&msg, sizeof(msg));
249  if (err) {
250  mutex_unlock(&xs_state.request_mutex);
251  return ERR_PTR(err);
252  }
253 
254  for (i = 0; i < num_vecs; i++) {
255  err = xb_write(iovec[i].iov_base, iovec[i].iov_len);
256  if (err) {
257  mutex_unlock(&xs_state.request_mutex);
258  return ERR_PTR(err);
259  }
260  }
261 
262  ret = read_reply(&msg.type, len);
263 
264  mutex_unlock(&xs_state.request_mutex);
265 
266  if (IS_ERR(ret))
267  return ret;
268 
269  if (msg.type == XS_ERROR) {
270  err = get_error(ret);
271  kfree(ret);
272  return ERR_PTR(-err);
273  }
274 
275  if (msg.type != type) {
276  if (printk_ratelimit())
278  "XENBUS unexpected type [%d], expected [%d]\n",
279  msg.type, type);
280  kfree(ret);
281  return ERR_PTR(-EINVAL);
282  }
283  return ret;
284 }
285 
286 /* Simplified version of xs_talkv: single message. */
287 static void *xs_single(struct xenbus_transaction t,
288  enum xsd_sockmsg_type type,
289  const char *string,
290  unsigned int *len)
291 {
292  struct kvec iovec;
293 
294  iovec.iov_base = (void *)string;
295  iovec.iov_len = strlen(string) + 1;
296  return xs_talkv(t, type, &iovec, 1, len);
297 }
298 
299 /* Many commands only need an ack, don't care what it says. */
300 static int xs_error(char *reply)
301 {
302  if (IS_ERR(reply))
303  return PTR_ERR(reply);
304  kfree(reply);
305  return 0;
306 }
307 
308 static unsigned int count_strings(const char *strings, unsigned int len)
309 {
310  unsigned int num;
311  const char *p;
312 
313  for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
314  num++;
315 
316  return num;
317 }
318 
319 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
320 static char *join(const char *dir, const char *name)
321 {
322  char *buffer;
323 
324  if (strlen(name) == 0)
325  buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
326  else
327  buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
328  return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
329 }
330 
331 static char **split(char *strings, unsigned int len, unsigned int *num)
332 {
333  char *p, **ret;
334 
335  /* Count the strings. */
336  *num = count_strings(strings, len);
337 
338  /* Transfer to one big alloc for easy freeing. */
339  ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
340  if (!ret) {
341  kfree(strings);
342  return ERR_PTR(-ENOMEM);
343  }
344  memcpy(&ret[*num], strings, len);
345  kfree(strings);
346 
347  strings = (char *)&ret[*num];
348  for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
349  ret[(*num)++] = p;
350 
351  return ret;
352 }
353 
355  const char *dir, const char *node, unsigned int *num)
356 {
357  char *strings, *path;
358  unsigned int len;
359 
360  path = join(dir, node);
361  if (IS_ERR(path))
362  return (char **)path;
363 
364  strings = xs_single(t, XS_DIRECTORY, path, &len);
365  kfree(path);
366  if (IS_ERR(strings))
367  return (char **)strings;
368 
369  return split(strings, len, num);
370 }
372 
373 /* Check if a path exists. Return 1 if it does. */
375  const char *dir, const char *node)
376 {
377  char **d;
378  int dir_n;
379 
380  d = xenbus_directory(t, dir, node, &dir_n);
381  if (IS_ERR(d))
382  return 0;
383  kfree(d);
384  return 1;
385 }
387 
388 /* Get the value of a single file.
389  * Returns a kmalloced value: call free() on it after use.
390  * len indicates length in bytes.
391  */
393  const char *dir, const char *node, unsigned int *len)
394 {
395  char *path;
396  void *ret;
397 
398  path = join(dir, node);
399  if (IS_ERR(path))
400  return (void *)path;
401 
402  ret = xs_single(t, XS_READ, path, len);
403  kfree(path);
404  return ret;
405 }
407 
408 /* Write the value of a single file.
409  * Returns -err on failure.
410  */
412  const char *dir, const char *node, const char *string)
413 {
414  const char *path;
415  struct kvec iovec[2];
416  int ret;
417 
418  path = join(dir, node);
419  if (IS_ERR(path))
420  return PTR_ERR(path);
421 
422  iovec[0].iov_base = (void *)path;
423  iovec[0].iov_len = strlen(path) + 1;
424  iovec[1].iov_base = (void *)string;
425  iovec[1].iov_len = strlen(string);
426 
427  ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
428  kfree(path);
429  return ret;
430 }
432 
433 /* Create a new directory. */
435  const char *dir, const char *node)
436 {
437  char *path;
438  int ret;
439 
440  path = join(dir, node);
441  if (IS_ERR(path))
442  return PTR_ERR(path);
443 
444  ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
445  kfree(path);
446  return ret;
447 }
449 
450 /* Destroy a file or directory (directories must be empty). */
451 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
452 {
453  char *path;
454  int ret;
455 
456  path = join(dir, node);
457  if (IS_ERR(path))
458  return PTR_ERR(path);
459 
460  ret = xs_error(xs_single(t, XS_RM, path, NULL));
461  kfree(path);
462  return ret;
463 }
465 
466 /* Start a transaction: changes by others will not be seen during this
467  * transaction, and changes will not be visible to others until end.
468  */
470 {
471  char *id_str;
472 
473  transaction_start();
474 
475  id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
476  if (IS_ERR(id_str)) {
477  transaction_end();
478  return PTR_ERR(id_str);
479  }
480 
481  t->id = simple_strtoul(id_str, NULL, 0);
482  kfree(id_str);
483  return 0;
484 }
486 
487 /* End a transaction.
488  * If abandon is true, transaction is discarded instead of committed.
489  */
491 {
492  char abortstr[2];
493  int err;
494 
495  if (abort)
496  strcpy(abortstr, "F");
497  else
498  strcpy(abortstr, "T");
499 
500  err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
501 
502  transaction_end();
503 
504  return err;
505 }
507 
508 /* Single read and scanf: returns -errno or num scanned. */
510  const char *dir, const char *node, const char *fmt, ...)
511 {
512  va_list ap;
513  int ret;
514  char *val;
515 
516  val = xenbus_read(t, dir, node, NULL);
517  if (IS_ERR(val))
518  return PTR_ERR(val);
519 
520  va_start(ap, fmt);
521  ret = vsscanf(val, fmt, ap);
522  va_end(ap);
523  kfree(val);
524  /* Distinctive errno. */
525  if (ret == 0)
526  return -ERANGE;
527  return ret;
528 }
530 
531 /* Single printf and write: returns -errno or 0. */
533  const char *dir, const char *node, const char *fmt, ...)
534 {
535  va_list ap;
536  int ret;
537  char *buf;
538 
539  va_start(ap, fmt);
540  buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
541  va_end(ap);
542 
543  if (!buf)
544  return -ENOMEM;
545 
546  ret = xenbus_write(t, dir, node, buf);
547 
548  kfree(buf);
549 
550  return ret;
551 }
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 }
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  * Certain older XenBus toolstack cannot handle reading values that are
624  * not populated. Some Xen 3.4 installation are incapable of doing this
625  * so if we are running on anything older than 4 do not attempt to read
626  * control/platform-feature-xs_reset_watches.
627  */
628 static bool xen_strict_xenbus_quirk(void)
629 {
630 #ifdef CONFIG_X86
631  uint32_t eax, ebx, ecx, edx, base;
632 
633  base = xen_cpuid_base();
634  cpuid(base + 1, &eax, &ebx, &ecx, &edx);
635 
636  if ((eax >> 16) < 4)
637  return true;
638 #endif
639  return false;
640 
641 }
642 static void xs_reset_watches(void)
643 {
644  int err, supported = 0;
645 
647  return;
648 
649  if (xen_strict_xenbus_quirk())
650  return;
651 
652  err = xenbus_scanf(XBT_NIL, "control",
653  "platform-feature-xs_reset_watches", "%d", &supported);
654  if (err != 1 || !supported)
655  return;
656 
657  err = xs_error(xs_single(XBT_NIL, XS_RESET_WATCHES, "", NULL));
658  if (err && err != -EEXIST)
659  printk(KERN_WARNING "xs_reset_watches failed: %d\n", err);
660 }
661 
662 /* Register callback to watch this node. */
664 {
665  /* Pointer in ascii is the token. */
666  char token[sizeof(watch) * 2 + 1];
667  int err;
668 
669  sprintf(token, "%lX", (long)watch);
670 
671  down_read(&xs_state.watch_mutex);
672 
673  spin_lock(&watches_lock);
674  BUG_ON(find_watch(token));
675  list_add(&watch->list, &watches);
676  spin_unlock(&watches_lock);
677 
678  err = xs_watch(watch->node, token);
679 
680  if (err) {
681  spin_lock(&watches_lock);
682  list_del(&watch->list);
683  spin_unlock(&watches_lock);
684  }
685 
686  up_read(&xs_state.watch_mutex);
687 
688  return err;
689 }
691 
693 {
694  struct xs_stored_msg *msg, *tmp;
695  char token[sizeof(watch) * 2 + 1];
696  int err;
697 
698  sprintf(token, "%lX", (long)watch);
699 
700  down_read(&xs_state.watch_mutex);
701 
702  spin_lock(&watches_lock);
703  BUG_ON(!find_watch(token));
704  list_del(&watch->list);
705  spin_unlock(&watches_lock);
706 
707  err = xs_unwatch(watch->node, token);
708  if (err)
710  "XENBUS Failed to release watch %s: %i\n",
711  watch->node, err);
712 
713  up_read(&xs_state.watch_mutex);
714 
715  /* Make sure there are no callbacks running currently (unless
716  its us) */
717  if (current->pid != xenwatch_pid)
718  mutex_lock(&xenwatch_mutex);
719 
720  /* Cancel pending watch events. */
721  spin_lock(&watch_events_lock);
722  list_for_each_entry_safe(msg, tmp, &watch_events, list) {
723  if (msg->u.watch.handle != watch)
724  continue;
725  list_del(&msg->list);
726  kfree(msg->u.watch.vec);
727  kfree(msg);
728  }
729  spin_unlock(&watch_events_lock);
730 
731  if (current->pid != xenwatch_pid)
732  mutex_unlock(&xenwatch_mutex);
733 }
735 
736 void xs_suspend(void)
737 {
738  transaction_suspend();
739  down_write(&xs_state.watch_mutex);
740  mutex_lock(&xs_state.request_mutex);
741  mutex_lock(&xs_state.response_mutex);
742 }
743 
744 void xs_resume(void)
745 {
746  struct xenbus_watch *watch;
747  char token[sizeof(watch) * 2 + 1];
748 
749  xb_init_comms();
750 
751  mutex_unlock(&xs_state.response_mutex);
752  mutex_unlock(&xs_state.request_mutex);
753  transaction_resume();
754 
755  /* No need for watches_lock: the watch_mutex is sufficient. */
756  list_for_each_entry(watch, &watches, list) {
757  sprintf(token, "%lX", (long)watch);
758  xs_watch(watch->node, token);
759  }
760 
761  up_write(&xs_state.watch_mutex);
762 }
763 
765 {
766  mutex_unlock(&xs_state.response_mutex);
767  mutex_unlock(&xs_state.request_mutex);
768  up_write(&xs_state.watch_mutex);
769  mutex_unlock(&xs_state.transaction_mutex);
770 }
771 
772 static int xenwatch_thread(void *unused)
773 {
774  struct list_head *ent;
775  struct xs_stored_msg *msg;
776 
777  for (;;) {
778  wait_event_interruptible(watch_events_waitq,
779  !list_empty(&watch_events));
780 
781  if (kthread_should_stop())
782  break;
783 
784  mutex_lock(&xenwatch_mutex);
785 
786  spin_lock(&watch_events_lock);
787  ent = watch_events.next;
788  if (ent != &watch_events)
789  list_del(ent);
790  spin_unlock(&watch_events_lock);
791 
792  if (ent != &watch_events) {
793  msg = list_entry(ent, struct xs_stored_msg, list);
794  msg->u.watch.handle->callback(
795  msg->u.watch.handle,
796  (const char **)msg->u.watch.vec,
797  msg->u.watch.vec_size);
798  kfree(msg->u.watch.vec);
799  kfree(msg);
800  }
801 
802  mutex_unlock(&xenwatch_mutex);
803  }
804 
805  return 0;
806 }
807 
808 static int process_msg(void)
809 {
810  struct xs_stored_msg *msg;
811  char *body;
812  int err;
813 
814  /*
815  * We must disallow save/restore while reading a xenstore message.
816  * A partial read across s/r leaves us out of sync with xenstored.
817  */
818  for (;;) {
819  err = xb_wait_for_data_to_read();
820  if (err)
821  return err;
822  mutex_lock(&xs_state.response_mutex);
823  if (xb_data_to_read())
824  break;
825  /* We raced with save/restore: pending data 'disappeared'. */
826  mutex_unlock(&xs_state.response_mutex);
827  }
828 
829 
830  msg = kmalloc(sizeof(*msg), GFP_NOIO | __GFP_HIGH);
831  if (msg == NULL) {
832  err = -ENOMEM;
833  goto out;
834  }
835 
836  err = xb_read(&msg->hdr, sizeof(msg->hdr));
837  if (err) {
838  kfree(msg);
839  goto out;
840  }
841 
842  if (msg->hdr.len > XENSTORE_PAYLOAD_MAX) {
843  kfree(msg);
844  err = -EINVAL;
845  goto out;
846  }
847 
848  body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH);
849  if (body == NULL) {
850  kfree(msg);
851  err = -ENOMEM;
852  goto out;
853  }
854 
855  err = xb_read(body, msg->hdr.len);
856  if (err) {
857  kfree(body);
858  kfree(msg);
859  goto out;
860  }
861  body[msg->hdr.len] = '\0';
862 
863  if (msg->hdr.type == XS_WATCH_EVENT) {
864  msg->u.watch.vec = split(body, msg->hdr.len,
865  &msg->u.watch.vec_size);
866  if (IS_ERR(msg->u.watch.vec)) {
867  err = PTR_ERR(msg->u.watch.vec);
868  kfree(msg);
869  goto out;
870  }
871 
872  spin_lock(&watches_lock);
873  msg->u.watch.handle = find_watch(
874  msg->u.watch.vec[XS_WATCH_TOKEN]);
875  if (msg->u.watch.handle != NULL) {
876  spin_lock(&watch_events_lock);
877  list_add_tail(&msg->list, &watch_events);
878  wake_up(&watch_events_waitq);
879  spin_unlock(&watch_events_lock);
880  } else {
881  kfree(msg->u.watch.vec);
882  kfree(msg);
883  }
884  spin_unlock(&watches_lock);
885  } else {
886  msg->u.reply.body = body;
887  spin_lock(&xs_state.reply_lock);
888  list_add_tail(&msg->list, &xs_state.reply_list);
889  spin_unlock(&xs_state.reply_lock);
890  wake_up(&xs_state.reply_waitq);
891  }
892 
893  out:
894  mutex_unlock(&xs_state.response_mutex);
895  return err;
896 }
897 
898 static int xenbus_thread(void *unused)
899 {
900  int err;
901 
902  for (;;) {
903  err = process_msg();
904  if (err)
905  printk(KERN_WARNING "XENBUS error %d while reading "
906  "message\n", err);
907  if (kthread_should_stop())
908  break;
909  }
910 
911  return 0;
912 }
913 
914 int xs_init(void)
915 {
916  int err;
917  struct task_struct *task;
918 
919  INIT_LIST_HEAD(&xs_state.reply_list);
920  spin_lock_init(&xs_state.reply_lock);
921  init_waitqueue_head(&xs_state.reply_waitq);
922 
923  mutex_init(&xs_state.request_mutex);
924  mutex_init(&xs_state.response_mutex);
925  mutex_init(&xs_state.transaction_mutex);
926  init_rwsem(&xs_state.watch_mutex);
927  atomic_set(&xs_state.transaction_count, 0);
928  init_waitqueue_head(&xs_state.transaction_wq);
929 
930  /* Initialize the shared memory rings to talk to xenstored */
931  err = xb_init_comms();
932  if (err)
933  return err;
934 
935  task = kthread_run(xenwatch_thread, NULL, "xenwatch");
936  if (IS_ERR(task))
937  return PTR_ERR(task);
938  xenwatch_pid = task->pid;
939 
940  task = kthread_run(xenbus_thread, NULL, "xenbus");
941  if (IS_ERR(task))
942  return PTR_ERR(task);
943 
944  /* shutdown watches for kexec boot */
945  xs_reset_watches();
946 
947  return 0;
948 }