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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
52 #include "xenbus_comms.h"
53 #include "xenbus_probe.h"
55 struct xs_stored_msg {
56 struct list_head list;
58 struct xsd_sockmsg hdr;
66 /* Queued watch events. */
68 struct xenbus_watch *handle;
70 unsigned int vec_size;
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;
82 * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex.
83 * response_mutex is never taken simultaneously with the other three.
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.
89 * When decrementing transaction_count to zero the wait queue
90 * should be woken up, the suspend code waits for count to
94 /* One request at a time. */
95 struct mutex request_mutex;
97 /* Protect xenbus reader thread against save/restore. */
98 struct mutex response_mutex;
100 /* Protect transactions against save/restore. */
101 struct mutex transaction_mutex;
102 atomic_t transaction_count;
103 wait_queue_head_t transaction_wq;
105 /* Protect watch (de)register against save/restore. */
106 struct rw_semaphore watch_mutex;
109 static struct xs_handle xs_state;
111 /* List of registered watches, and a lock to protect it. */
112 static LIST_HEAD(watches);
113 static DEFINE_SPINLOCK(watches_lock);
115 /* List of pending watch callback events, and a lock to protect it. */
116 static LIST_HEAD(watch_events);
117 static DEFINE_SPINLOCK(watch_events_lock);
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
125 static pid_t xenwatch_pid;
126 static DEFINE_MUTEX(xenwatch_mutex);
127 static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq);
129 static int get_error(const char *errorstring)
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",
140 return xsd_errors[i].errnum;
143 static bool xenbus_ok(void)
145 switch (xen_store_domain_type) {
147 switch (system_state) {
148 case SYSTEM_POWER_OFF:
158 /* FIXME: Could check that the remote domain is alive,
159 * but it is normally initial domain. */
166 static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len)
168 struct xs_stored_msg *msg;
171 spin_lock(&xs_state.reply_lock);
173 while (list_empty(&xs_state.reply_list)) {
174 spin_unlock(&xs_state.reply_lock);
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));
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.
187 return ERR_PTR(-EIO);
189 spin_lock(&xs_state.reply_lock);
192 msg = list_entry(xs_state.reply_list.next,
193 struct xs_stored_msg, list);
194 list_del(&msg->list);
196 spin_unlock(&xs_state.reply_lock);
198 *type = msg->hdr.type;
201 body = msg->u.reply.body;
208 static void transaction_start(void)
210 mutex_lock(&xs_state.transaction_mutex);
211 atomic_inc(&xs_state.transaction_count);
212 mutex_unlock(&xs_state.transaction_mutex);
215 static void transaction_end(void)
217 if (atomic_dec_and_test(&xs_state.transaction_count))
218 wake_up(&xs_state.transaction_wq);
221 static void transaction_suspend(void)
223 mutex_lock(&xs_state.transaction_mutex);
224 wait_event(xs_state.transaction_wq,
225 atomic_read(&xs_state.transaction_count) == 0);
228 static void transaction_resume(void)
230 mutex_unlock(&xs_state.transaction_mutex);
233 void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg)
236 struct xsd_sockmsg req_msg = *msg;
239 if (req_msg.type == XS_TRANSACTION_START)
242 mutex_lock(&xs_state.request_mutex);
244 err = xb_write(msg, sizeof(*msg) + msg->len);
246 msg->type = XS_ERROR;
249 ret = read_reply(&msg->type, &msg->len);
251 mutex_unlock(&xs_state.request_mutex);
256 if ((msg->type == XS_TRANSACTION_END) ||
257 ((req_msg.type == XS_TRANSACTION_START) &&
258 (msg->type == XS_ERROR)))
263 EXPORT_SYMBOL(xenbus_dev_request_and_reply);
265 /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */
266 static void *xs_talkv(struct xenbus_transaction t,
267 enum xsd_sockmsg_type type,
268 const struct kvec *iovec,
269 unsigned int num_vecs,
272 struct xsd_sockmsg msg;
281 for (i = 0; i < num_vecs; i++)
282 msg.len += iovec[i].iov_len;
284 mutex_lock(&xs_state.request_mutex);
286 err = xb_write(&msg, sizeof(msg));
288 mutex_unlock(&xs_state.request_mutex);
292 for (i = 0; i < num_vecs; i++) {
293 err = xb_write(iovec[i].iov_base, iovec[i].iov_len);
295 mutex_unlock(&xs_state.request_mutex);
300 ret = read_reply(&msg.type, len);
302 mutex_unlock(&xs_state.request_mutex);
307 if (msg.type == XS_ERROR) {
308 err = get_error(ret);
310 return ERR_PTR(-err);
313 if (msg.type != type) {
314 pr_warn_ratelimited("unexpected type [%d], expected [%d]\n",
317 return ERR_PTR(-EINVAL);
322 /* Simplified version of xs_talkv: single message. */
323 static void *xs_single(struct xenbus_transaction t,
324 enum xsd_sockmsg_type type,
330 iovec.iov_base = (void *)string;
331 iovec.iov_len = strlen(string) + 1;
332 return xs_talkv(t, type, &iovec, 1, len);
335 /* Many commands only need an ack, don't care what it says. */
336 static int xs_error(char *reply)
339 return PTR_ERR(reply);
344 static unsigned int count_strings(const char *strings, unsigned int len)
349 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
355 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
356 static char *join(const char *dir, const char *name)
360 if (strlen(name) == 0)
361 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
363 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
364 return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
367 static char **split(char *strings, unsigned int len, unsigned int *num)
371 /* Count the strings. */
372 *num = count_strings(strings, len);
374 /* Transfer to one big alloc for easy freeing. */
375 ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
378 return ERR_PTR(-ENOMEM);
380 memcpy(&ret[*num], strings, len);
383 strings = (char *)&ret[*num];
384 for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
390 char **xenbus_directory(struct xenbus_transaction t,
391 const char *dir, const char *node, unsigned int *num)
393 char *strings, *path;
396 path = join(dir, node);
398 return (char **)path;
400 strings = xs_single(t, XS_DIRECTORY, path, &len);
403 return (char **)strings;
405 return split(strings, len, num);
407 EXPORT_SYMBOL_GPL(xenbus_directory);
409 /* Check if a path exists. Return 1 if it does. */
410 int xenbus_exists(struct xenbus_transaction t,
411 const char *dir, const char *node)
416 d = xenbus_directory(t, dir, node, &dir_n);
422 EXPORT_SYMBOL_GPL(xenbus_exists);
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.
428 void *xenbus_read(struct xenbus_transaction t,
429 const char *dir, const char *node, unsigned int *len)
434 path = join(dir, node);
438 ret = xs_single(t, XS_READ, path, len);
442 EXPORT_SYMBOL_GPL(xenbus_read);
444 /* Write the value of a single file.
445 * Returns -err on failure.
447 int xenbus_write(struct xenbus_transaction t,
448 const char *dir, const char *node, const char *string)
451 struct kvec iovec[2];
454 path = join(dir, node);
456 return PTR_ERR(path);
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);
463 ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
467 EXPORT_SYMBOL_GPL(xenbus_write);
469 /* Create a new directory. */
470 int xenbus_mkdir(struct xenbus_transaction t,
471 const char *dir, const char *node)
476 path = join(dir, node);
478 return PTR_ERR(path);
480 ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
484 EXPORT_SYMBOL_GPL(xenbus_mkdir);
486 /* Destroy a file or directory (directories must be empty). */
487 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
492 path = join(dir, node);
494 return PTR_ERR(path);
496 ret = xs_error(xs_single(t, XS_RM, path, NULL));
500 EXPORT_SYMBOL_GPL(xenbus_rm);
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.
505 int xenbus_transaction_start(struct xenbus_transaction *t)
511 id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
512 if (IS_ERR(id_str)) {
514 return PTR_ERR(id_str);
517 t->id = simple_strtoul(id_str, NULL, 0);
521 EXPORT_SYMBOL_GPL(xenbus_transaction_start);
523 /* End a transaction.
524 * If abandon is true, transaction is discarded instead of committed.
526 int xenbus_transaction_end(struct xenbus_transaction t, int abort)
532 strcpy(abortstr, "F");
534 strcpy(abortstr, "T");
536 err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
542 EXPORT_SYMBOL_GPL(xenbus_transaction_end);
544 /* Single read and scanf: returns -errno or num scanned. */
545 int xenbus_scanf(struct xenbus_transaction t,
546 const char *dir, const char *node, const char *fmt, ...)
552 val = xenbus_read(t, dir, node, NULL);
557 ret = vsscanf(val, fmt, ap);
560 /* Distinctive errno. */
565 EXPORT_SYMBOL_GPL(xenbus_scanf);
567 /* Single printf and write: returns -errno or 0. */
568 int xenbus_printf(struct xenbus_transaction t,
569 const char *dir, const char *node, const char *fmt, ...)
576 buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
582 ret = xenbus_write(t, dir, node, buf);
588 EXPORT_SYMBOL_GPL(xenbus_printf);
590 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
591 int xenbus_gather(struct xenbus_transaction t, const char *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 *);
603 p = xenbus_read(t, dir, name, NULL);
609 if (sscanf(p, fmt, result) == 0)
613 *(char **)result = p;
618 EXPORT_SYMBOL_GPL(xenbus_gather);
620 static int xs_watch(const char *path, const char *token)
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;
629 return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
630 ARRAY_SIZE(iov), NULL));
633 static int xs_unwatch(const char *path, const char *token)
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;
642 return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
643 ARRAY_SIZE(iov), NULL));
646 static struct xenbus_watch *find_watch(const char *token)
648 struct xenbus_watch *i, *cmp;
650 cmp = (void *)simple_strtoul(token, NULL, 16);
652 list_for_each_entry(i, &watches, list)
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.
664 static bool xen_strict_xenbus_quirk(void)
667 uint32_t eax, ebx, ecx, edx, base;
669 base = xen_cpuid_base();
670 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
678 static void xs_reset_watches(void)
680 int err, supported = 0;
682 if (!xen_hvm_domain() || xen_initial_domain())
685 if (xen_strict_xenbus_quirk())
688 err = xenbus_scanf(XBT_NIL, "control",
689 "platform-feature-xs_reset_watches", "%d", &supported);
690 if (err != 1 || !supported)
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);
698 /* Register callback to watch this node. */
699 int register_xenbus_watch(struct xenbus_watch *watch)
701 /* Pointer in ascii is the token. */
702 char token[sizeof(watch) * 2 + 1];
705 sprintf(token, "%lX", (long)watch);
707 down_read(&xs_state.watch_mutex);
709 spin_lock(&watches_lock);
710 BUG_ON(find_watch(token));
711 list_add(&watch->list, &watches);
712 spin_unlock(&watches_lock);
714 err = xs_watch(watch->node, token);
717 spin_lock(&watches_lock);
718 list_del(&watch->list);
719 spin_unlock(&watches_lock);
722 up_read(&xs_state.watch_mutex);
726 EXPORT_SYMBOL_GPL(register_xenbus_watch);
728 void unregister_xenbus_watch(struct xenbus_watch *watch)
730 struct xs_stored_msg *msg, *tmp;
731 char token[sizeof(watch) * 2 + 1];
734 sprintf(token, "%lX", (long)watch);
736 down_read(&xs_state.watch_mutex);
738 spin_lock(&watches_lock);
739 BUG_ON(!find_watch(token));
740 list_del(&watch->list);
741 spin_unlock(&watches_lock);
743 err = xs_unwatch(watch->node, token);
745 pr_warn("Failed to release watch %s: %i\n", watch->node, err);
747 up_read(&xs_state.watch_mutex);
749 /* Make sure there are no callbacks running currently (unless
751 if (current->pid != xenwatch_pid)
752 mutex_lock(&xenwatch_mutex);
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)
759 list_del(&msg->list);
760 kfree(msg->u.watch.vec);
763 spin_unlock(&watch_events_lock);
765 if (current->pid != xenwatch_pid)
766 mutex_unlock(&xenwatch_mutex);
768 EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
770 void xs_suspend(void)
772 transaction_suspend();
773 down_write(&xs_state.watch_mutex);
774 mutex_lock(&xs_state.request_mutex);
775 mutex_lock(&xs_state.response_mutex);
780 struct xenbus_watch *watch;
781 char token[sizeof(watch) * 2 + 1];
785 mutex_unlock(&xs_state.response_mutex);
786 mutex_unlock(&xs_state.request_mutex);
787 transaction_resume();
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);
795 up_write(&xs_state.watch_mutex);
798 void xs_suspend_cancel(void)
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);
806 static int xenwatch_thread(void *unused)
808 struct list_head *ent;
809 struct xs_stored_msg *msg;
812 wait_event_interruptible(watch_events_waitq,
813 !list_empty(&watch_events));
815 if (kthread_should_stop())
818 mutex_lock(&xenwatch_mutex);
820 spin_lock(&watch_events_lock);
821 ent = watch_events.next;
822 if (ent != &watch_events)
824 spin_unlock(&watch_events_lock);
826 if (ent != &watch_events) {
827 msg = list_entry(ent, struct xs_stored_msg, list);
828 msg->u.watch.handle->callback(
830 (const char **)msg->u.watch.vec,
831 msg->u.watch.vec_size);
832 kfree(msg->u.watch.vec);
836 mutex_unlock(&xenwatch_mutex);
842 static int process_msg(void)
844 struct xs_stored_msg *msg;
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.
853 err = xb_wait_for_data_to_read();
856 mutex_lock(&xs_state.response_mutex);
857 if (xb_data_to_read())
859 /* We raced with save/restore: pending data 'disappeared'. */
860 mutex_unlock(&xs_state.response_mutex);
864 msg = kmalloc(sizeof(*msg), GFP_NOIO | __GFP_HIGH);
870 err = xb_read(&msg->hdr, sizeof(msg->hdr));
876 if (msg->hdr.len > XENSTORE_PAYLOAD_MAX) {
882 body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH);
889 err = xb_read(body, msg->hdr.len);
895 body[msg->hdr.len] = '\0';
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);
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);
915 kfree(msg->u.watch.vec);
918 spin_unlock(&watches_lock);
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);
928 mutex_unlock(&xs_state.response_mutex);
932 static int xenbus_thread(void *unused)
939 pr_warn("error %d while reading message\n", err);
940 if (kthread_should_stop())
950 struct task_struct *task;
952 INIT_LIST_HEAD(&xs_state.reply_list);
953 spin_lock_init(&xs_state.reply_lock);
954 init_waitqueue_head(&xs_state.reply_waitq);
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);
963 /* Initialize the shared memory rings to talk to xenstored */
964 err = xb_init_comms();
968 task = kthread_run(xenwatch_thread, NULL, "xenwatch");
970 return PTR_ERR(task);
971 xenwatch_pid = task->pid;
973 task = kthread_run(xenbus_thread, NULL, "xenbus");
975 return PTR_ERR(task);
977 /* shutdown watches for kexec boot */