1 /******************************************************************************
4 * This is the kernel equivalent of the "xs" library. We don't need everything
5 * and we use xenbus_comms for communication.
7 * Copyright (C) 2005 Rusty Russell, IBM Corporation
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:
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:
22 * The above copyright notice and this permission notice shall be included in
23 * all copies or substantial portions of the Software.
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
34 #include <linux/unistd.h>
35 #include <linux/errno.h>
36 #include <linux/types.h>
37 #include <linux/uio.h>
38 #include <linux/kernel.h>
39 #include <linux/string.h>
40 #include <linux/err.h>
41 #include <linux/slab.h>
42 #include <linux/fcntl.h>
43 #include <linux/kthread.h>
44 #include <linux/rwsem.h>
45 #include <linux/module.h>
46 #include <linux/mutex.h>
47 #include <xen/xenbus.h>
48 #include "xenbus_comms.h"
50 struct xs_stored_msg {
51 struct list_head list;
53 struct xsd_sockmsg hdr;
61 /* Queued watch events. */
63 struct xenbus_watch *handle;
65 unsigned int vec_size;
71 /* A list of replies. Currently only one will ever be outstanding. */
72 struct list_head reply_list;
73 spinlock_t reply_lock;
74 wait_queue_head_t reply_waitq;
77 * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex.
78 * response_mutex is never taken simultaneously with the other three.
80 * transaction_mutex must be held before incrementing
81 * transaction_count. The mutex is held when a suspend is in
82 * progress to prevent new transactions starting.
84 * When decrementing transaction_count to zero the wait queue
85 * should be woken up, the suspend code waits for count to
89 /* One request at a time. */
90 struct mutex request_mutex;
92 /* Protect xenbus reader thread against save/restore. */
93 struct mutex response_mutex;
95 /* Protect transactions against save/restore. */
96 struct mutex transaction_mutex;
97 atomic_t transaction_count;
98 wait_queue_head_t transaction_wq;
100 /* Protect watch (de)register against save/restore. */
101 struct rw_semaphore watch_mutex;
104 static struct xs_handle xs_state;
106 /* List of registered watches, and a lock to protect it. */
107 static LIST_HEAD(watches);
108 static DEFINE_SPINLOCK(watches_lock);
110 /* List of pending watch callback events, and a lock to protect it. */
111 static LIST_HEAD(watch_events);
112 static DEFINE_SPINLOCK(watch_events_lock);
115 * Details of the xenwatch callback kernel thread. The thread waits on the
116 * watch_events_waitq for work to do (queued on watch_events list). When it
117 * wakes up it acquires the xenwatch_mutex before reading the list and
120 static pid_t xenwatch_pid;
121 static DEFINE_MUTEX(xenwatch_mutex);
122 static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq);
124 static int get_error(const char *errorstring)
128 for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) {
129 if (i == ARRAY_SIZE(xsd_errors) - 1) {
131 "XENBUS xen store gave: unknown error %s",
136 return xsd_errors[i].errnum;
139 static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len)
141 struct xs_stored_msg *msg;
144 spin_lock(&xs_state.reply_lock);
146 while (list_empty(&xs_state.reply_list)) {
147 spin_unlock(&xs_state.reply_lock);
148 /* XXX FIXME: Avoid synchronous wait for response here. */
149 wait_event(xs_state.reply_waitq,
150 !list_empty(&xs_state.reply_list));
151 spin_lock(&xs_state.reply_lock);
154 msg = list_entry(xs_state.reply_list.next,
155 struct xs_stored_msg, list);
156 list_del(&msg->list);
158 spin_unlock(&xs_state.reply_lock);
160 *type = msg->hdr.type;
163 body = msg->u.reply.body;
170 static void transaction_start(void)
172 mutex_lock(&xs_state.transaction_mutex);
173 atomic_inc(&xs_state.transaction_count);
174 mutex_unlock(&xs_state.transaction_mutex);
177 static void transaction_end(void)
179 if (atomic_dec_and_test(&xs_state.transaction_count))
180 wake_up(&xs_state.transaction_wq);
183 static void transaction_suspend(void)
185 mutex_lock(&xs_state.transaction_mutex);
186 wait_event(xs_state.transaction_wq,
187 atomic_read(&xs_state.transaction_count) == 0);
190 static void transaction_resume(void)
192 mutex_unlock(&xs_state.transaction_mutex);
195 void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg)
198 struct xsd_sockmsg req_msg = *msg;
201 if (req_msg.type == XS_TRANSACTION_START)
204 mutex_lock(&xs_state.request_mutex);
206 err = xb_write(msg, sizeof(*msg) + msg->len);
208 msg->type = XS_ERROR;
211 ret = read_reply(&msg->type, &msg->len);
213 mutex_unlock(&xs_state.request_mutex);
215 if ((msg->type == XS_TRANSACTION_END) ||
216 ((req_msg.type == XS_TRANSACTION_START) &&
217 (msg->type == XS_ERROR)))
222 EXPORT_SYMBOL(xenbus_dev_request_and_reply);
224 /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */
225 static void *xs_talkv(struct xenbus_transaction t,
226 enum xsd_sockmsg_type type,
227 const struct kvec *iovec,
228 unsigned int num_vecs,
231 struct xsd_sockmsg msg;
240 for (i = 0; i < num_vecs; i++)
241 msg.len += iovec[i].iov_len;
243 mutex_lock(&xs_state.request_mutex);
245 err = xb_write(&msg, sizeof(msg));
247 mutex_unlock(&xs_state.request_mutex);
251 for (i = 0; i < num_vecs; i++) {
252 err = xb_write(iovec[i].iov_base, iovec[i].iov_len);
254 mutex_unlock(&xs_state.request_mutex);
259 ret = read_reply(&msg.type, len);
261 mutex_unlock(&xs_state.request_mutex);
266 if (msg.type == XS_ERROR) {
267 err = get_error(ret);
269 return ERR_PTR(-err);
272 if (msg.type != type) {
273 if (printk_ratelimit())
275 "XENBUS unexpected type [%d], expected [%d]\n",
278 return ERR_PTR(-EINVAL);
283 /* Simplified version of xs_talkv: single message. */
284 static void *xs_single(struct xenbus_transaction t,
285 enum xsd_sockmsg_type type,
291 iovec.iov_base = (void *)string;
292 iovec.iov_len = strlen(string) + 1;
293 return xs_talkv(t, type, &iovec, 1, len);
296 /* Many commands only need an ack, don't care what it says. */
297 static int xs_error(char *reply)
300 return PTR_ERR(reply);
305 static unsigned int count_strings(const char *strings, unsigned int len)
310 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
316 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
317 static char *join(const char *dir, const char *name)
321 if (strlen(name) == 0)
322 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
324 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
325 return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
328 static char **split(char *strings, unsigned int len, unsigned int *num)
332 /* Count the strings. */
333 *num = count_strings(strings, len);
335 /* Transfer to one big alloc for easy freeing. */
336 ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
339 return ERR_PTR(-ENOMEM);
341 memcpy(&ret[*num], strings, len);
344 strings = (char *)&ret[*num];
345 for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
351 char **xenbus_directory(struct xenbus_transaction t,
352 const char *dir, const char *node, unsigned int *num)
354 char *strings, *path;
357 path = join(dir, node);
359 return (char **)path;
361 strings = xs_single(t, XS_DIRECTORY, path, &len);
364 return (char **)strings;
366 return split(strings, len, num);
368 EXPORT_SYMBOL_GPL(xenbus_directory);
370 /* Check if a path exists. Return 1 if it does. */
371 int xenbus_exists(struct xenbus_transaction t,
372 const char *dir, const char *node)
377 d = xenbus_directory(t, dir, node, &dir_n);
383 EXPORT_SYMBOL_GPL(xenbus_exists);
385 /* Get the value of a single file.
386 * Returns a kmalloced value: call free() on it after use.
387 * len indicates length in bytes.
389 void *xenbus_read(struct xenbus_transaction t,
390 const char *dir, const char *node, unsigned int *len)
395 path = join(dir, node);
399 ret = xs_single(t, XS_READ, path, len);
403 EXPORT_SYMBOL_GPL(xenbus_read);
405 /* Write the value of a single file.
406 * Returns -err on failure.
408 int xenbus_write(struct xenbus_transaction t,
409 const char *dir, const char *node, const char *string)
412 struct kvec iovec[2];
415 path = join(dir, node);
417 return PTR_ERR(path);
419 iovec[0].iov_base = (void *)path;
420 iovec[0].iov_len = strlen(path) + 1;
421 iovec[1].iov_base = (void *)string;
422 iovec[1].iov_len = strlen(string);
424 ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
428 EXPORT_SYMBOL_GPL(xenbus_write);
430 /* Create a new directory. */
431 int xenbus_mkdir(struct xenbus_transaction t,
432 const char *dir, const char *node)
437 path = join(dir, node);
439 return PTR_ERR(path);
441 ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
445 EXPORT_SYMBOL_GPL(xenbus_mkdir);
447 /* Destroy a file or directory (directories must be empty). */
448 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
453 path = join(dir, node);
455 return PTR_ERR(path);
457 ret = xs_error(xs_single(t, XS_RM, path, NULL));
461 EXPORT_SYMBOL_GPL(xenbus_rm);
463 /* Start a transaction: changes by others will not be seen during this
464 * transaction, and changes will not be visible to others until end.
466 int xenbus_transaction_start(struct xenbus_transaction *t)
472 id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
473 if (IS_ERR(id_str)) {
475 return PTR_ERR(id_str);
478 t->id = simple_strtoul(id_str, NULL, 0);
482 EXPORT_SYMBOL_GPL(xenbus_transaction_start);
484 /* End a transaction.
485 * If abandon is true, transaction is discarded instead of committed.
487 int xenbus_transaction_end(struct xenbus_transaction t, int abort)
493 strcpy(abortstr, "F");
495 strcpy(abortstr, "T");
497 err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
503 EXPORT_SYMBOL_GPL(xenbus_transaction_end);
505 /* Single read and scanf: returns -errno or num scanned. */
506 int xenbus_scanf(struct xenbus_transaction t,
507 const char *dir, const char *node, const char *fmt, ...)
513 val = xenbus_read(t, dir, node, NULL);
518 ret = vsscanf(val, fmt, ap);
521 /* Distinctive errno. */
526 EXPORT_SYMBOL_GPL(xenbus_scanf);
528 /* Single printf and write: returns -errno or 0. */
529 int xenbus_printf(struct xenbus_transaction t,
530 const char *dir, const char *node, const char *fmt, ...)
534 #define PRINTF_BUFFER_SIZE 4096
537 printf_buffer = kmalloc(PRINTF_BUFFER_SIZE, GFP_NOIO | __GFP_HIGH);
538 if (printf_buffer == NULL)
542 ret = vsnprintf(printf_buffer, PRINTF_BUFFER_SIZE, fmt, ap);
545 BUG_ON(ret > PRINTF_BUFFER_SIZE-1);
546 ret = xenbus_write(t, dir, node, printf_buffer);
548 kfree(printf_buffer);
552 EXPORT_SYMBOL_GPL(xenbus_printf);
554 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
555 int xenbus_gather(struct xenbus_transaction t, const char *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 *);
567 p = xenbus_read(t, dir, name, NULL);
573 if (sscanf(p, fmt, result) == 0)
577 *(char **)result = p;
582 EXPORT_SYMBOL_GPL(xenbus_gather);
584 static int xs_watch(const char *path, const char *token)
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;
593 return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
594 ARRAY_SIZE(iov), NULL));
597 static int xs_unwatch(const char *path, const char *token)
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;
606 return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
607 ARRAY_SIZE(iov), NULL));
610 static struct xenbus_watch *find_watch(const char *token)
612 struct xenbus_watch *i, *cmp;
614 cmp = (void *)simple_strtoul(token, NULL, 16);
616 list_for_each_entry(i, &watches, list)
623 /* Register callback to watch this node. */
624 int register_xenbus_watch(struct xenbus_watch *watch)
626 /* Pointer in ascii is the token. */
627 char token[sizeof(watch) * 2 + 1];
630 sprintf(token, "%lX", (long)watch);
632 down_read(&xs_state.watch_mutex);
634 spin_lock(&watches_lock);
635 BUG_ON(find_watch(token));
636 list_add(&watch->list, &watches);
637 spin_unlock(&watches_lock);
639 err = xs_watch(watch->node, token);
641 /* Ignore errors due to multiple registration. */
642 if ((err != 0) && (err != -EEXIST)) {
643 spin_lock(&watches_lock);
644 list_del(&watch->list);
645 spin_unlock(&watches_lock);
648 up_read(&xs_state.watch_mutex);
652 EXPORT_SYMBOL_GPL(register_xenbus_watch);
654 void unregister_xenbus_watch(struct xenbus_watch *watch)
656 struct xs_stored_msg *msg, *tmp;
657 char token[sizeof(watch) * 2 + 1];
660 sprintf(token, "%lX", (long)watch);
662 down_read(&xs_state.watch_mutex);
664 spin_lock(&watches_lock);
665 BUG_ON(!find_watch(token));
666 list_del(&watch->list);
667 spin_unlock(&watches_lock);
669 err = xs_unwatch(watch->node, token);
672 "XENBUS Failed to release watch %s: %i\n",
675 up_read(&xs_state.watch_mutex);
677 /* Make sure there are no callbacks running currently (unless
679 if (current->pid != xenwatch_pid)
680 mutex_lock(&xenwatch_mutex);
682 /* Cancel pending watch events. */
683 spin_lock(&watch_events_lock);
684 list_for_each_entry_safe(msg, tmp, &watch_events, list) {
685 if (msg->u.watch.handle != watch)
687 list_del(&msg->list);
688 kfree(msg->u.watch.vec);
691 spin_unlock(&watch_events_lock);
693 if (current->pid != xenwatch_pid)
694 mutex_unlock(&xenwatch_mutex);
696 EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
698 void xs_suspend(void)
700 transaction_suspend();
701 down_write(&xs_state.watch_mutex);
702 mutex_lock(&xs_state.request_mutex);
703 mutex_lock(&xs_state.response_mutex);
708 struct xenbus_watch *watch;
709 char token[sizeof(watch) * 2 + 1];
713 mutex_unlock(&xs_state.response_mutex);
714 mutex_unlock(&xs_state.request_mutex);
715 transaction_resume();
717 /* No need for watches_lock: the watch_mutex is sufficient. */
718 list_for_each_entry(watch, &watches, list) {
719 sprintf(token, "%lX", (long)watch);
720 xs_watch(watch->node, token);
723 up_write(&xs_state.watch_mutex);
726 void xs_suspend_cancel(void)
728 mutex_unlock(&xs_state.response_mutex);
729 mutex_unlock(&xs_state.request_mutex);
730 up_write(&xs_state.watch_mutex);
731 mutex_unlock(&xs_state.transaction_mutex);
734 static int xenwatch_thread(void *unused)
736 struct list_head *ent;
737 struct xs_stored_msg *msg;
740 wait_event_interruptible(watch_events_waitq,
741 !list_empty(&watch_events));
743 if (kthread_should_stop())
746 mutex_lock(&xenwatch_mutex);
748 spin_lock(&watch_events_lock);
749 ent = watch_events.next;
750 if (ent != &watch_events)
752 spin_unlock(&watch_events_lock);
754 if (ent != &watch_events) {
755 msg = list_entry(ent, struct xs_stored_msg, list);
756 msg->u.watch.handle->callback(
758 (const char **)msg->u.watch.vec,
759 msg->u.watch.vec_size);
760 kfree(msg->u.watch.vec);
764 mutex_unlock(&xenwatch_mutex);
770 static int process_msg(void)
772 struct xs_stored_msg *msg;
777 * We must disallow save/restore while reading a xenstore message.
778 * A partial read across s/r leaves us out of sync with xenstored.
781 err = xb_wait_for_data_to_read();
784 mutex_lock(&xs_state.response_mutex);
785 if (xb_data_to_read())
787 /* We raced with save/restore: pending data 'disappeared'. */
788 mutex_unlock(&xs_state.response_mutex);
792 msg = kmalloc(sizeof(*msg), GFP_NOIO | __GFP_HIGH);
798 err = xb_read(&msg->hdr, sizeof(msg->hdr));
804 body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH);
811 err = xb_read(body, msg->hdr.len);
817 body[msg->hdr.len] = '\0';
819 if (msg->hdr.type == XS_WATCH_EVENT) {
820 msg->u.watch.vec = split(body, msg->hdr.len,
821 &msg->u.watch.vec_size);
822 if (IS_ERR(msg->u.watch.vec)) {
823 err = PTR_ERR(msg->u.watch.vec);
828 spin_lock(&watches_lock);
829 msg->u.watch.handle = find_watch(
830 msg->u.watch.vec[XS_WATCH_TOKEN]);
831 if (msg->u.watch.handle != NULL) {
832 spin_lock(&watch_events_lock);
833 list_add_tail(&msg->list, &watch_events);
834 wake_up(&watch_events_waitq);
835 spin_unlock(&watch_events_lock);
837 kfree(msg->u.watch.vec);
840 spin_unlock(&watches_lock);
842 msg->u.reply.body = body;
843 spin_lock(&xs_state.reply_lock);
844 list_add_tail(&msg->list, &xs_state.reply_list);
845 spin_unlock(&xs_state.reply_lock);
846 wake_up(&xs_state.reply_waitq);
850 mutex_unlock(&xs_state.response_mutex);
854 static int xenbus_thread(void *unused)
861 printk(KERN_WARNING "XENBUS error %d while reading "
863 if (kthread_should_stop())
873 struct task_struct *task;
875 INIT_LIST_HEAD(&xs_state.reply_list);
876 spin_lock_init(&xs_state.reply_lock);
877 init_waitqueue_head(&xs_state.reply_waitq);
879 mutex_init(&xs_state.request_mutex);
880 mutex_init(&xs_state.response_mutex);
881 mutex_init(&xs_state.transaction_mutex);
882 init_rwsem(&xs_state.watch_mutex);
883 atomic_set(&xs_state.transaction_count, 0);
884 init_waitqueue_head(&xs_state.transaction_wq);
886 /* Initialize the shared memory rings to talk to xenstored */
887 err = xb_init_comms();
891 task = kthread_run(xenwatch_thread, NULL, "xenwatch");
893 return PTR_ERR(task);
894 xenwatch_pid = task->pid;
896 task = kthread_run(xenbus_thread, NULL, "xenbus");
898 return PTR_ERR(task);