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"
54 struct xs_stored_msg {
55 struct list_head list;
57 struct xsd_sockmsg hdr;
65 /* Queued watch events. */
67 struct xenbus_watch *handle;
69 unsigned int vec_size;
75 /* A list of replies. Currently only one will ever be outstanding. */
76 struct list_head reply_list;
77 spinlock_t reply_lock;
78 wait_queue_head_t reply_waitq;
81 * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex.
82 * response_mutex is never taken simultaneously with the other three.
84 * transaction_mutex must be held before incrementing
85 * transaction_count. The mutex is held when a suspend is in
86 * progress to prevent new transactions starting.
88 * When decrementing transaction_count to zero the wait queue
89 * should be woken up, the suspend code waits for count to
93 /* One request at a time. */
94 struct mutex request_mutex;
96 /* Protect xenbus reader thread against save/restore. */
97 struct mutex response_mutex;
99 /* Protect transactions against save/restore. */
100 struct mutex transaction_mutex;
101 atomic_t transaction_count;
102 wait_queue_head_t transaction_wq;
104 /* Protect watch (de)register against save/restore. */
105 struct rw_semaphore watch_mutex;
108 static struct xs_handle xs_state;
110 /* List of registered watches, and a lock to protect it. */
111 static LIST_HEAD(watches);
112 static DEFINE_SPINLOCK(watches_lock);
114 /* List of pending watch callback events, and a lock to protect it. */
115 static LIST_HEAD(watch_events);
116 static DEFINE_SPINLOCK(watch_events_lock);
119 * Details of the xenwatch callback kernel thread. The thread waits on the
120 * watch_events_waitq for work to do (queued on watch_events list). When it
121 * wakes up it acquires the xenwatch_mutex before reading the list and
124 static pid_t xenwatch_pid;
125 static DEFINE_MUTEX(xenwatch_mutex);
126 static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq);
128 static int get_error(const char *errorstring)
132 for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) {
133 if (i == ARRAY_SIZE(xsd_errors) - 1) {
134 pr_warn("xen store gave: unknown error %s\n",
139 return xsd_errors[i].errnum;
142 static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len)
144 struct xs_stored_msg *msg;
147 spin_lock(&xs_state.reply_lock);
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);
157 msg = list_entry(xs_state.reply_list.next,
158 struct xs_stored_msg, list);
159 list_del(&msg->list);
161 spin_unlock(&xs_state.reply_lock);
163 *type = msg->hdr.type;
166 body = msg->u.reply.body;
173 static void transaction_start(void)
175 mutex_lock(&xs_state.transaction_mutex);
176 atomic_inc(&xs_state.transaction_count);
177 mutex_unlock(&xs_state.transaction_mutex);
180 static void transaction_end(void)
182 if (atomic_dec_and_test(&xs_state.transaction_count))
183 wake_up(&xs_state.transaction_wq);
186 static void transaction_suspend(void)
188 mutex_lock(&xs_state.transaction_mutex);
189 wait_event(xs_state.transaction_wq,
190 atomic_read(&xs_state.transaction_count) == 0);
193 static void transaction_resume(void)
195 mutex_unlock(&xs_state.transaction_mutex);
198 void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg)
201 struct xsd_sockmsg req_msg = *msg;
204 if (req_msg.type == XS_TRANSACTION_START)
207 mutex_lock(&xs_state.request_mutex);
209 err = xb_write(msg, sizeof(*msg) + msg->len);
211 msg->type = XS_ERROR;
214 ret = read_reply(&msg->type, &msg->len);
216 mutex_unlock(&xs_state.request_mutex);
218 if ((msg->type == XS_TRANSACTION_END) ||
219 ((req_msg.type == XS_TRANSACTION_START) &&
220 (msg->type == XS_ERROR)))
225 EXPORT_SYMBOL(xenbus_dev_request_and_reply);
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,
234 struct xsd_sockmsg msg;
243 for (i = 0; i < num_vecs; i++)
244 msg.len += iovec[i].iov_len;
246 mutex_lock(&xs_state.request_mutex);
248 err = xb_write(&msg, sizeof(msg));
250 mutex_unlock(&xs_state.request_mutex);
254 for (i = 0; i < num_vecs; i++) {
255 err = xb_write(iovec[i].iov_base, iovec[i].iov_len);
257 mutex_unlock(&xs_state.request_mutex);
262 ret = read_reply(&msg.type, len);
264 mutex_unlock(&xs_state.request_mutex);
269 if (msg.type == XS_ERROR) {
270 err = get_error(ret);
272 return ERR_PTR(-err);
275 if (msg.type != type) {
276 pr_warn_ratelimited("unexpected type [%d], expected [%d]\n",
279 return ERR_PTR(-EINVAL);
284 /* Simplified version of xs_talkv: single message. */
285 static void *xs_single(struct xenbus_transaction t,
286 enum xsd_sockmsg_type type,
292 iovec.iov_base = (void *)string;
293 iovec.iov_len = strlen(string) + 1;
294 return xs_talkv(t, type, &iovec, 1, len);
297 /* Many commands only need an ack, don't care what it says. */
298 static int xs_error(char *reply)
301 return PTR_ERR(reply);
306 static unsigned int count_strings(const char *strings, unsigned int len)
311 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
317 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
318 static char *join(const char *dir, const char *name)
322 if (strlen(name) == 0)
323 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
325 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
326 return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
329 static char **split(char *strings, unsigned int len, unsigned int *num)
333 /* Count the strings. */
334 *num = count_strings(strings, len);
336 /* Transfer to one big alloc for easy freeing. */
337 ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
340 return ERR_PTR(-ENOMEM);
342 memcpy(&ret[*num], strings, len);
345 strings = (char *)&ret[*num];
346 for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
352 char **xenbus_directory(struct xenbus_transaction t,
353 const char *dir, const char *node, unsigned int *num)
355 char *strings, *path;
358 path = join(dir, node);
360 return (char **)path;
362 strings = xs_single(t, XS_DIRECTORY, path, &len);
365 return (char **)strings;
367 return split(strings, len, num);
369 EXPORT_SYMBOL_GPL(xenbus_directory);
371 /* Check if a path exists. Return 1 if it does. */
372 int xenbus_exists(struct xenbus_transaction t,
373 const char *dir, const char *node)
378 d = xenbus_directory(t, dir, node, &dir_n);
384 EXPORT_SYMBOL_GPL(xenbus_exists);
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.
390 void *xenbus_read(struct xenbus_transaction t,
391 const char *dir, const char *node, unsigned int *len)
396 path = join(dir, node);
400 ret = xs_single(t, XS_READ, path, len);
404 EXPORT_SYMBOL_GPL(xenbus_read);
406 /* Write the value of a single file.
407 * Returns -err on failure.
409 int xenbus_write(struct xenbus_transaction t,
410 const char *dir, const char *node, const char *string)
413 struct kvec iovec[2];
416 path = join(dir, node);
418 return PTR_ERR(path);
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);
425 ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
429 EXPORT_SYMBOL_GPL(xenbus_write);
431 /* Create a new directory. */
432 int xenbus_mkdir(struct xenbus_transaction t,
433 const char *dir, const char *node)
438 path = join(dir, node);
440 return PTR_ERR(path);
442 ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
446 EXPORT_SYMBOL_GPL(xenbus_mkdir);
448 /* Destroy a file or directory (directories must be empty). */
449 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
454 path = join(dir, node);
456 return PTR_ERR(path);
458 ret = xs_error(xs_single(t, XS_RM, path, NULL));
462 EXPORT_SYMBOL_GPL(xenbus_rm);
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.
467 int xenbus_transaction_start(struct xenbus_transaction *t)
473 id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
474 if (IS_ERR(id_str)) {
476 return PTR_ERR(id_str);
479 t->id = simple_strtoul(id_str, NULL, 0);
483 EXPORT_SYMBOL_GPL(xenbus_transaction_start);
485 /* End a transaction.
486 * If abandon is true, transaction is discarded instead of committed.
488 int xenbus_transaction_end(struct xenbus_transaction t, int abort)
494 strcpy(abortstr, "F");
496 strcpy(abortstr, "T");
498 err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
504 EXPORT_SYMBOL_GPL(xenbus_transaction_end);
506 /* Single read and scanf: returns -errno or num scanned. */
507 int xenbus_scanf(struct xenbus_transaction t,
508 const char *dir, const char *node, const char *fmt, ...)
514 val = xenbus_read(t, dir, node, NULL);
519 ret = vsscanf(val, fmt, ap);
522 /* Distinctive errno. */
527 EXPORT_SYMBOL_GPL(xenbus_scanf);
529 /* Single printf and write: returns -errno or 0. */
530 int xenbus_printf(struct xenbus_transaction t,
531 const char *dir, const char *node, const char *fmt, ...)
538 buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
544 ret = xenbus_write(t, dir, node, buf);
550 EXPORT_SYMBOL_GPL(xenbus_printf);
552 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
553 int xenbus_gather(struct xenbus_transaction t, const char *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 *);
565 p = xenbus_read(t, dir, name, NULL);
571 if (sscanf(p, fmt, result) == 0)
575 *(char **)result = p;
580 EXPORT_SYMBOL_GPL(xenbus_gather);
582 static int xs_watch(const char *path, const char *token)
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;
591 return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
592 ARRAY_SIZE(iov), NULL));
595 static int xs_unwatch(const char *path, const char *token)
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;
604 return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
605 ARRAY_SIZE(iov), NULL));
608 static struct xenbus_watch *find_watch(const char *token)
610 struct xenbus_watch *i, *cmp;
612 cmp = (void *)simple_strtoul(token, NULL, 16);
614 list_for_each_entry(i, &watches, list)
621 * Certain older XenBus toolstack cannot handle reading values that are
622 * not populated. Some Xen 3.4 installation are incapable of doing this
623 * so if we are running on anything older than 4 do not attempt to read
624 * control/platform-feature-xs_reset_watches.
626 static bool xen_strict_xenbus_quirk(void)
629 uint32_t eax, ebx, ecx, edx, base;
631 base = xen_cpuid_base();
632 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
640 static void xs_reset_watches(void)
642 int err, supported = 0;
644 if (!xen_hvm_domain() || xen_initial_domain())
647 if (xen_strict_xenbus_quirk())
650 err = xenbus_scanf(XBT_NIL, "control",
651 "platform-feature-xs_reset_watches", "%d", &supported);
652 if (err != 1 || !supported)
655 err = xs_error(xs_single(XBT_NIL, XS_RESET_WATCHES, "", NULL));
656 if (err && err != -EEXIST)
657 pr_warn("xs_reset_watches failed: %d\n", err);
660 /* Register callback to watch this node. */
661 int register_xenbus_watch(struct xenbus_watch *watch)
663 /* Pointer in ascii is the token. */
664 char token[sizeof(watch) * 2 + 1];
667 sprintf(token, "%lX", (long)watch);
669 down_read(&xs_state.watch_mutex);
671 spin_lock(&watches_lock);
672 BUG_ON(find_watch(token));
673 list_add(&watch->list, &watches);
674 spin_unlock(&watches_lock);
676 err = xs_watch(watch->node, token);
679 spin_lock(&watches_lock);
680 list_del(&watch->list);
681 spin_unlock(&watches_lock);
684 up_read(&xs_state.watch_mutex);
688 EXPORT_SYMBOL_GPL(register_xenbus_watch);
690 void unregister_xenbus_watch(struct xenbus_watch *watch)
692 struct xs_stored_msg *msg, *tmp;
693 char token[sizeof(watch) * 2 + 1];
696 sprintf(token, "%lX", (long)watch);
698 down_read(&xs_state.watch_mutex);
700 spin_lock(&watches_lock);
701 BUG_ON(!find_watch(token));
702 list_del(&watch->list);
703 spin_unlock(&watches_lock);
705 err = xs_unwatch(watch->node, token);
707 pr_warn("Failed to release watch %s: %i\n", watch->node, err);
709 up_read(&xs_state.watch_mutex);
711 /* Make sure there are no callbacks running currently (unless
713 if (current->pid != xenwatch_pid)
714 mutex_lock(&xenwatch_mutex);
716 /* Cancel pending watch events. */
717 spin_lock(&watch_events_lock);
718 list_for_each_entry_safe(msg, tmp, &watch_events, list) {
719 if (msg->u.watch.handle != watch)
721 list_del(&msg->list);
722 kfree(msg->u.watch.vec);
725 spin_unlock(&watch_events_lock);
727 if (current->pid != xenwatch_pid)
728 mutex_unlock(&xenwatch_mutex);
730 EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
732 void xs_suspend(void)
734 transaction_suspend();
735 down_write(&xs_state.watch_mutex);
736 mutex_lock(&xs_state.request_mutex);
737 mutex_lock(&xs_state.response_mutex);
742 struct xenbus_watch *watch;
743 char token[sizeof(watch) * 2 + 1];
747 mutex_unlock(&xs_state.response_mutex);
748 mutex_unlock(&xs_state.request_mutex);
749 transaction_resume();
751 /* No need for watches_lock: the watch_mutex is sufficient. */
752 list_for_each_entry(watch, &watches, list) {
753 sprintf(token, "%lX", (long)watch);
754 xs_watch(watch->node, token);
757 up_write(&xs_state.watch_mutex);
760 void xs_suspend_cancel(void)
762 mutex_unlock(&xs_state.response_mutex);
763 mutex_unlock(&xs_state.request_mutex);
764 up_write(&xs_state.watch_mutex);
765 mutex_unlock(&xs_state.transaction_mutex);
768 static int xenwatch_thread(void *unused)
770 struct list_head *ent;
771 struct xs_stored_msg *msg;
774 wait_event_interruptible(watch_events_waitq,
775 !list_empty(&watch_events));
777 if (kthread_should_stop())
780 mutex_lock(&xenwatch_mutex);
782 spin_lock(&watch_events_lock);
783 ent = watch_events.next;
784 if (ent != &watch_events)
786 spin_unlock(&watch_events_lock);
788 if (ent != &watch_events) {
789 msg = list_entry(ent, struct xs_stored_msg, list);
790 msg->u.watch.handle->callback(
792 (const char **)msg->u.watch.vec,
793 msg->u.watch.vec_size);
794 kfree(msg->u.watch.vec);
798 mutex_unlock(&xenwatch_mutex);
804 static int process_msg(void)
806 struct xs_stored_msg *msg;
811 * We must disallow save/restore while reading a xenstore message.
812 * A partial read across s/r leaves us out of sync with xenstored.
815 err = xb_wait_for_data_to_read();
818 mutex_lock(&xs_state.response_mutex);
819 if (xb_data_to_read())
821 /* We raced with save/restore: pending data 'disappeared'. */
822 mutex_unlock(&xs_state.response_mutex);
826 msg = kmalloc(sizeof(*msg), GFP_NOIO | __GFP_HIGH);
832 err = xb_read(&msg->hdr, sizeof(msg->hdr));
838 if (msg->hdr.len > XENSTORE_PAYLOAD_MAX) {
844 body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH);
851 err = xb_read(body, msg->hdr.len);
857 body[msg->hdr.len] = '\0';
859 if (msg->hdr.type == XS_WATCH_EVENT) {
860 msg->u.watch.vec = split(body, msg->hdr.len,
861 &msg->u.watch.vec_size);
862 if (IS_ERR(msg->u.watch.vec)) {
863 err = PTR_ERR(msg->u.watch.vec);
868 spin_lock(&watches_lock);
869 msg->u.watch.handle = find_watch(
870 msg->u.watch.vec[XS_WATCH_TOKEN]);
871 if (msg->u.watch.handle != NULL) {
872 spin_lock(&watch_events_lock);
873 list_add_tail(&msg->list, &watch_events);
874 wake_up(&watch_events_waitq);
875 spin_unlock(&watch_events_lock);
877 kfree(msg->u.watch.vec);
880 spin_unlock(&watches_lock);
882 msg->u.reply.body = body;
883 spin_lock(&xs_state.reply_lock);
884 list_add_tail(&msg->list, &xs_state.reply_list);
885 spin_unlock(&xs_state.reply_lock);
886 wake_up(&xs_state.reply_waitq);
890 mutex_unlock(&xs_state.response_mutex);
894 static int xenbus_thread(void *unused)
901 pr_warn("error %d while reading message\n", err);
902 if (kthread_should_stop())
912 struct task_struct *task;
914 INIT_LIST_HEAD(&xs_state.reply_list);
915 spin_lock_init(&xs_state.reply_lock);
916 init_waitqueue_head(&xs_state.reply_waitq);
918 mutex_init(&xs_state.request_mutex);
919 mutex_init(&xs_state.response_mutex);
920 mutex_init(&xs_state.transaction_mutex);
921 init_rwsem(&xs_state.watch_mutex);
922 atomic_set(&xs_state.transaction_count, 0);
923 init_waitqueue_head(&xs_state.transaction_wq);
925 /* Initialize the shared memory rings to talk to xenstored */
926 err = xb_init_comms();
930 task = kthread_run(xenwatch_thread, NULL, "xenwatch");
932 return PTR_ERR(task);
933 xenwatch_pid = task->pid;
935 task = kthread_run(xenbus_thread, NULL, "xenbus");
937 return PTR_ERR(task);
939 /* shutdown watches for kexec boot */