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/reboot.h>
47 #include <linux/rwsem.h>
48 #include <linux/mutex.h>
49 #include <asm/xen/hypervisor.h>
50 #include <xen/xenbus.h>
55 * Framework to protect suspend/resume handling against normal Xenstore
57 * During suspend/resume there must be no open transaction and no pending
59 * New watch events happening in this time can be ignored by firing all watches
63 /* Lock protecting enter/exit critical region. */
64 static DEFINE_SPINLOCK(xs_state_lock);
65 /* Number of users in critical region (protected by xs_state_lock). */
66 static unsigned int xs_state_users;
67 /* Suspend handler waiting or already active (protected by xs_state_lock)? */
68 static int xs_suspend_active;
69 /* Unique Xenstore request id (protected by xs_state_lock). */
70 static uint32_t xs_request_id;
72 /* Wait queue for all callers waiting for critical region to become usable. */
73 static DECLARE_WAIT_QUEUE_HEAD(xs_state_enter_wq);
74 /* Wait queue for suspend handling waiting for critical region being empty. */
75 static DECLARE_WAIT_QUEUE_HEAD(xs_state_exit_wq);
77 /* List of registered watches, and a lock to protect it. */
78 static LIST_HEAD(watches);
79 static DEFINE_SPINLOCK(watches_lock);
81 /* List of pending watch callback events, and a lock to protect it. */
82 static LIST_HEAD(watch_events);
83 static DEFINE_SPINLOCK(watch_events_lock);
85 /* Protect watch (de)register against save/restore. */
86 static DECLARE_RWSEM(xs_watch_rwsem);
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
94 static pid_t xenwatch_pid;
95 static DEFINE_MUTEX(xenwatch_mutex);
96 static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq);
98 static void xs_suspend_enter(void)
100 spin_lock(&xs_state_lock);
102 spin_unlock(&xs_state_lock);
103 wait_event(xs_state_exit_wq, xs_state_users == 0);
106 static void xs_suspend_exit(void)
108 xb_dev_generation_id++;
109 spin_lock(&xs_state_lock);
111 spin_unlock(&xs_state_lock);
112 wake_up_all(&xs_state_enter_wq);
115 static uint32_t xs_request_enter(struct xb_req_data *req)
119 req->type = req->msg.type;
121 spin_lock(&xs_state_lock);
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);
129 if (req->type == XS_TRANSACTION_START && !req->user_req)
132 rq_id = xs_request_id++;
134 spin_unlock(&xs_state_lock);
139 void xs_request_exit(struct xb_req_data *req)
141 spin_lock(&xs_state_lock);
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"))))
148 spin_unlock(&xs_state_lock);
150 if (xs_suspend_active && !xs_state_users)
151 wake_up(&xs_state_exit_wq);
154 static int get_error(const char *errorstring)
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",
165 return xsd_errors[i].errnum;
168 static bool xenbus_ok(void)
170 switch (xen_store_domain_type) {
172 switch (system_state) {
173 case SYSTEM_POWER_OFF:
183 /* FIXME: Could check that the remote domain is alive,
184 * but it is normally initial domain. */
192 static bool test_reply(struct xb_req_data *req)
194 if (req->state == xb_req_state_got_reply || !xenbus_ok()) {
195 /* read req->state before all other fields */
200 /* Make sure to reread req->state each time. */
206 static void *read_reply(struct xb_req_data *req)
209 wait_event(req->wq, test_reply(req));
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.
218 return ERR_PTR(-EIO);
220 return ERR_PTR(req->err);
222 } while (req->state != xb_req_state_got_reply);
227 static void xs_send(struct xb_req_data *req, struct xsd_sockmsg *msg)
233 req->state = xb_req_state_queued;
234 init_waitqueue_head(&req->wq);
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);
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);
249 static void *xs_wait_for_reply(struct xb_req_data *req, struct xsd_sockmsg *msg)
253 ret = read_reply(req);
255 xs_request_exit(req);
257 msg->type = req->msg.type;
258 msg->len = req->msg.len;
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;
266 mutex_unlock(&xb_write_mutex);
271 static void xs_wake_up(struct xb_req_data *req)
276 int xenbus_dev_request_and_reply(struct xsd_sockmsg *msg, void *par)
278 struct xb_req_data *req;
281 req = kmalloc(sizeof(*req) + sizeof(*vec), GFP_KERNEL);
285 vec = (struct kvec *)(req + 1);
286 vec->iov_len = msg->len;
287 vec->iov_base = msg + 1;
291 req->cb = xenbus_dev_queue_reply;
293 req->user_req = true;
299 EXPORT_SYMBOL(xenbus_dev_request_and_reply);
301 /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */
302 static void *xs_talkv(struct xenbus_transaction t,
303 enum xsd_sockmsg_type type,
304 const struct kvec *iovec,
305 unsigned int num_vecs,
308 struct xb_req_data *req;
309 struct xsd_sockmsg msg;
314 req = kmalloc(sizeof(*req), GFP_NOIO | __GFP_HIGH);
316 return ERR_PTR(-ENOMEM);
319 req->num_vecs = num_vecs;
320 req->cb = xs_wake_up;
321 req->user_req = false;
327 for (i = 0; i < num_vecs; i++)
328 msg.len += iovec[i].iov_len;
332 ret = xs_wait_for_reply(req, &msg);
339 if (msg.type == XS_ERROR) {
340 err = get_error(ret);
342 return ERR_PTR(-err);
345 if (msg.type != type) {
346 pr_warn_ratelimited("unexpected type [%d], expected [%d]\n",
349 return ERR_PTR(-EINVAL);
354 /* Simplified version of xs_talkv: single message. */
355 static void *xs_single(struct xenbus_transaction t,
356 enum xsd_sockmsg_type type,
362 iovec.iov_base = (void *)string;
363 iovec.iov_len = strlen(string) + 1;
364 return xs_talkv(t, type, &iovec, 1, len);
367 /* Many commands only need an ack, don't care what it says. */
368 static int xs_error(char *reply)
371 return PTR_ERR(reply);
376 static unsigned int count_strings(const char *strings, unsigned int len)
381 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
387 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
388 static char *join(const char *dir, const char *name)
392 if (strlen(name) == 0)
393 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
395 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
396 return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
399 static char **split(char *strings, unsigned int len, unsigned int *num)
403 /* Count the strings. */
404 *num = count_strings(strings, len);
406 /* Transfer to one big alloc for easy freeing. */
407 ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
410 return ERR_PTR(-ENOMEM);
412 memcpy(&ret[*num], strings, len);
415 strings = (char *)&ret[*num];
416 for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
422 char **xenbus_directory(struct xenbus_transaction t,
423 const char *dir, const char *node, unsigned int *num)
425 char *strings, *path;
428 path = join(dir, node);
430 return (char **)path;
432 strings = xs_single(t, XS_DIRECTORY, path, &len);
435 return (char **)strings;
437 return split(strings, len, num);
439 EXPORT_SYMBOL_GPL(xenbus_directory);
441 /* Check if a path exists. Return 1 if it does. */
442 int xenbus_exists(struct xenbus_transaction t,
443 const char *dir, const char *node)
448 d = xenbus_directory(t, dir, node, &dir_n);
454 EXPORT_SYMBOL_GPL(xenbus_exists);
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.
460 void *xenbus_read(struct xenbus_transaction t,
461 const char *dir, const char *node, unsigned int *len)
466 path = join(dir, node);
470 ret = xs_single(t, XS_READ, path, len);
474 EXPORT_SYMBOL_GPL(xenbus_read);
476 /* Write the value of a single file.
477 * Returns -err on failure.
479 int xenbus_write(struct xenbus_transaction t,
480 const char *dir, const char *node, const char *string)
483 struct kvec iovec[2];
486 path = join(dir, node);
488 return PTR_ERR(path);
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);
495 ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
499 EXPORT_SYMBOL_GPL(xenbus_write);
501 /* Create a new directory. */
502 int xenbus_mkdir(struct xenbus_transaction t,
503 const char *dir, const char *node)
508 path = join(dir, node);
510 return PTR_ERR(path);
512 ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
516 EXPORT_SYMBOL_GPL(xenbus_mkdir);
518 /* Destroy a file or directory (directories must be empty). */
519 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
524 path = join(dir, node);
526 return PTR_ERR(path);
528 ret = xs_error(xs_single(t, XS_RM, path, NULL));
532 EXPORT_SYMBOL_GPL(xenbus_rm);
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.
537 int xenbus_transaction_start(struct xenbus_transaction *t)
541 id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
543 return PTR_ERR(id_str);
545 t->id = simple_strtoul(id_str, NULL, 0);
549 EXPORT_SYMBOL_GPL(xenbus_transaction_start);
551 /* End a transaction.
552 * If abandon is true, transaction is discarded instead of committed.
554 int xenbus_transaction_end(struct xenbus_transaction t, int abort)
559 strcpy(abortstr, "F");
561 strcpy(abortstr, "T");
563 return xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
565 EXPORT_SYMBOL_GPL(xenbus_transaction_end);
567 /* Single read and scanf: returns -errno or num scanned. */
568 int xenbus_scanf(struct xenbus_transaction t,
569 const char *dir, const char *node, const char *fmt, ...)
575 val = xenbus_read(t, dir, node, NULL);
580 ret = vsscanf(val, fmt, ap);
583 /* Distinctive errno. */
588 EXPORT_SYMBOL_GPL(xenbus_scanf);
590 /* Read an (optional) unsigned value. */
591 unsigned int xenbus_read_unsigned(const char *dir, const char *node,
592 unsigned int default_val)
597 ret = xenbus_scanf(XBT_NIL, dir, node, "%u", &val);
603 EXPORT_SYMBOL_GPL(xenbus_read_unsigned);
605 /* Single printf and write: returns -errno or 0. */
606 int xenbus_printf(struct xenbus_transaction t,
607 const char *dir, const char *node, const char *fmt, ...)
614 buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
620 ret = xenbus_write(t, dir, node, buf);
626 EXPORT_SYMBOL_GPL(xenbus_printf);
628 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
629 int xenbus_gather(struct xenbus_transaction t, const char *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 *);
641 p = xenbus_read(t, dir, name, NULL);
647 if (sscanf(p, fmt, result) == 0)
651 *(char **)result = p;
656 EXPORT_SYMBOL_GPL(xenbus_gather);
658 static int xs_watch(const char *path, const char *token)
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;
667 return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
668 ARRAY_SIZE(iov), NULL));
671 static int xs_unwatch(const char *path, const char *token)
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;
680 return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
681 ARRAY_SIZE(iov), NULL));
684 static struct xenbus_watch *find_watch(const char *token)
686 struct xenbus_watch *i, *cmp;
688 cmp = (void *)simple_strtoul(token, NULL, 16);
690 list_for_each_entry(i, &watches, list)
697 int xs_watch_msg(struct xs_watch_event *event)
699 if (count_strings(event->body, event->len) != 2) {
703 event->path = (const char *)event->body;
704 event->token = (const char *)strchr(event->body, '\0') + 1;
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);
719 spin_unlock(&watches_lock);
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.
730 static bool xen_strict_xenbus_quirk(void)
733 uint32_t eax, ebx, ecx, edx, base;
735 base = xen_cpuid_base();
736 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
744 static void xs_reset_watches(void)
748 if (!xen_hvm_domain() || xen_initial_domain())
751 if (xen_strict_xenbus_quirk())
754 if (!xenbus_read_unsigned("control",
755 "platform-feature-xs_reset_watches", 0))
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);
763 /* Register callback to watch this node. */
764 int register_xenbus_watch(struct xenbus_watch *watch)
766 /* Pointer in ascii is the token. */
767 char token[sizeof(watch) * 2 + 1];
770 sprintf(token, "%lX", (long)watch);
772 watch->nr_pending = 0;
774 down_read(&xs_watch_rwsem);
776 spin_lock(&watches_lock);
777 BUG_ON(find_watch(token));
778 list_add(&watch->list, &watches);
779 spin_unlock(&watches_lock);
781 err = xs_watch(watch->node, token);
784 spin_lock(&watches_lock);
785 list_del(&watch->list);
786 spin_unlock(&watches_lock);
789 up_read(&xs_watch_rwsem);
793 EXPORT_SYMBOL_GPL(register_xenbus_watch);
795 void unregister_xenbus_watch(struct xenbus_watch *watch)
797 struct xs_watch_event *event, *tmp;
798 char token[sizeof(watch) * 2 + 1];
801 sprintf(token, "%lX", (long)watch);
803 down_read(&xs_watch_rwsem);
805 spin_lock(&watches_lock);
806 BUG_ON(!find_watch(token));
807 list_del(&watch->list);
808 spin_unlock(&watches_lock);
810 err = xs_unwatch(watch->node, token);
812 pr_warn("Failed to release watch %s: %i\n", watch->node, err);
814 up_read(&xs_watch_rwsem);
816 /* Make sure there are no callbacks running currently (unless
818 if (current->pid != xenwatch_pid)
819 mutex_lock(&xenwatch_mutex);
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)
827 list_del(&event->list);
830 watch->nr_pending = 0;
832 spin_unlock(&watch_events_lock);
834 if (current->pid != xenwatch_pid)
835 mutex_unlock(&xenwatch_mutex);
837 EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
839 void xs_suspend(void)
843 down_write(&xs_watch_rwsem);
844 mutex_lock(&xs_response_mutex);
849 struct xenbus_watch *watch;
850 char token[sizeof(watch) * 2 + 1];
854 mutex_unlock(&xs_response_mutex);
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);
864 up_write(&xs_watch_rwsem);
867 void xs_suspend_cancel(void)
869 mutex_unlock(&xs_response_mutex);
870 up_write(&xs_watch_rwsem);
875 static int xenwatch_thread(void *unused)
877 struct xs_watch_event *event;
879 xenwatch_pid = current->pid;
882 wait_event_interruptible(watch_events_waitq,
883 !list_empty(&watch_events));
885 if (kthread_should_stop())
888 mutex_lock(&xenwatch_mutex);
890 spin_lock(&watch_events_lock);
891 event = list_first_entry_or_null(&watch_events,
892 struct xs_watch_event, list);
894 list_del(&event->list);
895 event->handle->nr_pending--;
897 spin_unlock(&watch_events_lock);
900 event->handle->callback(event->handle, event->path,
905 mutex_unlock(&xenwatch_mutex);
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.
918 static int xs_reboot_notify(struct notifier_block *nb,
919 unsigned long code, void *unused)
921 struct xb_req_data *req;
923 mutex_lock(&xb_write_mutex);
924 list_for_each_entry(req, &xs_reply_list, list)
926 list_for_each_entry(req, &xb_write_list, list)
928 mutex_unlock(&xb_write_mutex);
932 static struct notifier_block xs_reboot_nb = {
933 .notifier_call = xs_reboot_notify,
939 struct task_struct *task;
941 register_reboot_notifier(&xs_reboot_nb);
943 /* Initialize the shared memory rings to talk to xenstored */
944 err = xb_init_comms();
948 task = kthread_run(xenwatch_thread, NULL, "xenwatch");
950 return PTR_ERR(task);
952 /* shutdown watches for kexec boot */