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 spin_lock(&xs_state_lock);
110 spin_unlock(&xs_state_lock);
111 wake_up_all(&xs_state_enter_wq);
114 static uint32_t xs_request_enter(struct xb_req_data *req)
118 req->type = req->msg.type;
120 spin_lock(&xs_state_lock);
122 while (!xs_state_users && xs_suspend_active) {
123 spin_unlock(&xs_state_lock);
124 wait_event(xs_state_enter_wq, xs_suspend_active == 0);
125 spin_lock(&xs_state_lock);
128 if (req->type == XS_TRANSACTION_START)
131 rq_id = xs_request_id++;
133 spin_unlock(&xs_state_lock);
138 void xs_request_exit(struct xb_req_data *req)
140 spin_lock(&xs_state_lock);
142 if ((req->type == XS_TRANSACTION_START && req->msg.type == XS_ERROR) ||
143 req->type == XS_TRANSACTION_END)
145 spin_unlock(&xs_state_lock);
147 if (xs_suspend_active && !xs_state_users)
148 wake_up(&xs_state_exit_wq);
151 static int get_error(const char *errorstring)
155 for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) {
156 if (i == ARRAY_SIZE(xsd_errors) - 1) {
157 pr_warn("xen store gave: unknown error %s\n",
162 return xsd_errors[i].errnum;
165 static bool xenbus_ok(void)
167 switch (xen_store_domain_type) {
169 switch (system_state) {
170 case SYSTEM_POWER_OFF:
180 /* FIXME: Could check that the remote domain is alive,
181 * but it is normally initial domain. */
189 static bool test_reply(struct xb_req_data *req)
191 if (req->state == xb_req_state_got_reply || !xenbus_ok()) {
192 /* read req->state before all other fields */
197 /* Make sure to reread req->state each time. */
203 static void *read_reply(struct xb_req_data *req)
206 wait_event(req->wq, test_reply(req));
210 * If we are in the process of being shut-down there is
211 * no point of trying to contact XenBus - it is either
212 * killed (xenstored application) or the other domain
213 * has been killed or is unreachable.
215 return ERR_PTR(-EIO);
217 return ERR_PTR(req->err);
219 } while (req->state != xb_req_state_got_reply);
224 static void xs_send(struct xb_req_data *req, struct xsd_sockmsg *msg)
230 req->state = xb_req_state_queued;
231 init_waitqueue_head(&req->wq);
233 /* Save the caller req_id and restore it later in the reply */
234 req->caller_req_id = req->msg.req_id;
235 req->msg.req_id = xs_request_enter(req);
237 mutex_lock(&xb_write_mutex);
238 list_add_tail(&req->list, &xb_write_list);
239 notify = list_is_singular(&xb_write_list);
240 mutex_unlock(&xb_write_mutex);
246 static void *xs_wait_for_reply(struct xb_req_data *req, struct xsd_sockmsg *msg)
250 ret = read_reply(req);
252 xs_request_exit(req);
254 msg->type = req->msg.type;
255 msg->len = req->msg.len;
257 mutex_lock(&xb_write_mutex);
258 if (req->state == xb_req_state_queued ||
259 req->state == xb_req_state_wait_reply)
260 req->state = xb_req_state_aborted;
263 mutex_unlock(&xb_write_mutex);
268 static void xs_wake_up(struct xb_req_data *req)
273 int xenbus_dev_request_and_reply(struct xsd_sockmsg *msg, void *par)
275 struct xb_req_data *req;
278 req = kmalloc(sizeof(*req) + sizeof(*vec), GFP_KERNEL);
282 vec = (struct kvec *)(req + 1);
283 vec->iov_len = msg->len;
284 vec->iov_base = msg + 1;
288 req->cb = xenbus_dev_queue_reply;
295 EXPORT_SYMBOL(xenbus_dev_request_and_reply);
297 /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */
298 static void *xs_talkv(struct xenbus_transaction t,
299 enum xsd_sockmsg_type type,
300 const struct kvec *iovec,
301 unsigned int num_vecs,
304 struct xb_req_data *req;
305 struct xsd_sockmsg msg;
310 req = kmalloc(sizeof(*req), GFP_NOIO | __GFP_HIGH);
312 return ERR_PTR(-ENOMEM);
315 req->num_vecs = num_vecs;
316 req->cb = xs_wake_up;
322 for (i = 0; i < num_vecs; i++)
323 msg.len += iovec[i].iov_len;
327 ret = xs_wait_for_reply(req, &msg);
334 if (msg.type == XS_ERROR) {
335 err = get_error(ret);
337 return ERR_PTR(-err);
340 if (msg.type != type) {
341 pr_warn_ratelimited("unexpected type [%d], expected [%d]\n",
344 return ERR_PTR(-EINVAL);
349 /* Simplified version of xs_talkv: single message. */
350 static void *xs_single(struct xenbus_transaction t,
351 enum xsd_sockmsg_type type,
357 iovec.iov_base = (void *)string;
358 iovec.iov_len = strlen(string) + 1;
359 return xs_talkv(t, type, &iovec, 1, len);
362 /* Many commands only need an ack, don't care what it says. */
363 static int xs_error(char *reply)
366 return PTR_ERR(reply);
371 static unsigned int count_strings(const char *strings, unsigned int len)
376 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
382 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
383 static char *join(const char *dir, const char *name)
387 if (strlen(name) == 0)
388 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
390 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
391 return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
394 static char **split(char *strings, unsigned int len, unsigned int *num)
398 /* Count the strings. */
399 *num = count_strings(strings, len);
401 /* Transfer to one big alloc for easy freeing. */
402 ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
405 return ERR_PTR(-ENOMEM);
407 memcpy(&ret[*num], strings, len);
410 strings = (char *)&ret[*num];
411 for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
417 char **xenbus_directory(struct xenbus_transaction t,
418 const char *dir, const char *node, unsigned int *num)
420 char *strings, *path;
423 path = join(dir, node);
425 return (char **)path;
427 strings = xs_single(t, XS_DIRECTORY, path, &len);
430 return (char **)strings;
432 return split(strings, len, num);
434 EXPORT_SYMBOL_GPL(xenbus_directory);
436 /* Check if a path exists. Return 1 if it does. */
437 int xenbus_exists(struct xenbus_transaction t,
438 const char *dir, const char *node)
443 d = xenbus_directory(t, dir, node, &dir_n);
449 EXPORT_SYMBOL_GPL(xenbus_exists);
451 /* Get the value of a single file.
452 * Returns a kmalloced value: call free() on it after use.
453 * len indicates length in bytes.
455 void *xenbus_read(struct xenbus_transaction t,
456 const char *dir, const char *node, unsigned int *len)
461 path = join(dir, node);
465 ret = xs_single(t, XS_READ, path, len);
469 EXPORT_SYMBOL_GPL(xenbus_read);
471 /* Write the value of a single file.
472 * Returns -err on failure.
474 int xenbus_write(struct xenbus_transaction t,
475 const char *dir, const char *node, const char *string)
478 struct kvec iovec[2];
481 path = join(dir, node);
483 return PTR_ERR(path);
485 iovec[0].iov_base = (void *)path;
486 iovec[0].iov_len = strlen(path) + 1;
487 iovec[1].iov_base = (void *)string;
488 iovec[1].iov_len = strlen(string);
490 ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
494 EXPORT_SYMBOL_GPL(xenbus_write);
496 /* Create a new directory. */
497 int xenbus_mkdir(struct xenbus_transaction t,
498 const char *dir, const char *node)
503 path = join(dir, node);
505 return PTR_ERR(path);
507 ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
511 EXPORT_SYMBOL_GPL(xenbus_mkdir);
513 /* Destroy a file or directory (directories must be empty). */
514 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
519 path = join(dir, node);
521 return PTR_ERR(path);
523 ret = xs_error(xs_single(t, XS_RM, path, NULL));
527 EXPORT_SYMBOL_GPL(xenbus_rm);
529 /* Start a transaction: changes by others will not be seen during this
530 * transaction, and changes will not be visible to others until end.
532 int xenbus_transaction_start(struct xenbus_transaction *t)
536 id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
538 return PTR_ERR(id_str);
540 t->id = simple_strtoul(id_str, NULL, 0);
544 EXPORT_SYMBOL_GPL(xenbus_transaction_start);
546 /* End a transaction.
547 * If abandon is true, transaction is discarded instead of committed.
549 int xenbus_transaction_end(struct xenbus_transaction t, int abort)
554 strcpy(abortstr, "F");
556 strcpy(abortstr, "T");
558 return xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
560 EXPORT_SYMBOL_GPL(xenbus_transaction_end);
562 /* Single read and scanf: returns -errno or num scanned. */
563 int xenbus_scanf(struct xenbus_transaction t,
564 const char *dir, const char *node, const char *fmt, ...)
570 val = xenbus_read(t, dir, node, NULL);
575 ret = vsscanf(val, fmt, ap);
578 /* Distinctive errno. */
583 EXPORT_SYMBOL_GPL(xenbus_scanf);
585 /* Read an (optional) unsigned value. */
586 unsigned int xenbus_read_unsigned(const char *dir, const char *node,
587 unsigned int default_val)
592 ret = xenbus_scanf(XBT_NIL, dir, node, "%u", &val);
598 EXPORT_SYMBOL_GPL(xenbus_read_unsigned);
600 /* Single printf and write: returns -errno or 0. */
601 int xenbus_printf(struct xenbus_transaction t,
602 const char *dir, const char *node, const char *fmt, ...)
609 buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
615 ret = xenbus_write(t, dir, node, buf);
621 EXPORT_SYMBOL_GPL(xenbus_printf);
623 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
624 int xenbus_gather(struct xenbus_transaction t, const char *dir, ...)
631 while (ret == 0 && (name = va_arg(ap, char *)) != NULL) {
632 const char *fmt = va_arg(ap, char *);
633 void *result = va_arg(ap, void *);
636 p = xenbus_read(t, dir, name, NULL);
642 if (sscanf(p, fmt, result) == 0)
646 *(char **)result = p;
651 EXPORT_SYMBOL_GPL(xenbus_gather);
653 static int xs_watch(const char *path, const char *token)
657 iov[0].iov_base = (void *)path;
658 iov[0].iov_len = strlen(path) + 1;
659 iov[1].iov_base = (void *)token;
660 iov[1].iov_len = strlen(token) + 1;
662 return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
663 ARRAY_SIZE(iov), NULL));
666 static int xs_unwatch(const char *path, const char *token)
670 iov[0].iov_base = (char *)path;
671 iov[0].iov_len = strlen(path) + 1;
672 iov[1].iov_base = (char *)token;
673 iov[1].iov_len = strlen(token) + 1;
675 return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
676 ARRAY_SIZE(iov), NULL));
679 static struct xenbus_watch *find_watch(const char *token)
681 struct xenbus_watch *i, *cmp;
683 cmp = (void *)simple_strtoul(token, NULL, 16);
685 list_for_each_entry(i, &watches, list)
692 int xs_watch_msg(struct xs_watch_event *event)
694 if (count_strings(event->body, event->len) != 2) {
698 event->path = (const char *)event->body;
699 event->token = (const char *)strchr(event->body, '\0') + 1;
701 spin_lock(&watches_lock);
702 event->handle = find_watch(event->token);
703 if (event->handle != NULL &&
704 (!event->handle->will_handle ||
705 event->handle->will_handle(event->handle,
706 event->path, event->token))) {
707 spin_lock(&watch_events_lock);
708 list_add_tail(&event->list, &watch_events);
709 event->handle->nr_pending++;
710 wake_up(&watch_events_waitq);
711 spin_unlock(&watch_events_lock);
714 spin_unlock(&watches_lock);
720 * Certain older XenBus toolstack cannot handle reading values that are
721 * not populated. Some Xen 3.4 installation are incapable of doing this
722 * so if we are running on anything older than 4 do not attempt to read
723 * control/platform-feature-xs_reset_watches.
725 static bool xen_strict_xenbus_quirk(void)
728 uint32_t eax, ebx, ecx, edx, base;
730 base = xen_cpuid_base();
731 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
739 static void xs_reset_watches(void)
743 if (!xen_hvm_domain() || xen_initial_domain())
746 if (xen_strict_xenbus_quirk())
749 if (!xenbus_read_unsigned("control",
750 "platform-feature-xs_reset_watches", 0))
753 err = xs_error(xs_single(XBT_NIL, XS_RESET_WATCHES, "", NULL));
754 if (err && err != -EEXIST)
755 pr_warn("xs_reset_watches failed: %d\n", err);
758 /* Register callback to watch this node. */
759 int register_xenbus_watch(struct xenbus_watch *watch)
761 /* Pointer in ascii is the token. */
762 char token[sizeof(watch) * 2 + 1];
765 sprintf(token, "%lX", (long)watch);
767 watch->nr_pending = 0;
769 down_read(&xs_watch_rwsem);
771 spin_lock(&watches_lock);
772 BUG_ON(find_watch(token));
773 list_add(&watch->list, &watches);
774 spin_unlock(&watches_lock);
776 err = xs_watch(watch->node, token);
779 spin_lock(&watches_lock);
780 list_del(&watch->list);
781 spin_unlock(&watches_lock);
784 up_read(&xs_watch_rwsem);
788 EXPORT_SYMBOL_GPL(register_xenbus_watch);
790 void unregister_xenbus_watch(struct xenbus_watch *watch)
792 struct xs_watch_event *event, *tmp;
793 char token[sizeof(watch) * 2 + 1];
796 sprintf(token, "%lX", (long)watch);
798 down_read(&xs_watch_rwsem);
800 spin_lock(&watches_lock);
801 BUG_ON(!find_watch(token));
802 list_del(&watch->list);
803 spin_unlock(&watches_lock);
805 err = xs_unwatch(watch->node, token);
807 pr_warn("Failed to release watch %s: %i\n", watch->node, err);
809 up_read(&xs_watch_rwsem);
811 /* Make sure there are no callbacks running currently (unless
813 if (current->pid != xenwatch_pid)
814 mutex_lock(&xenwatch_mutex);
816 /* Cancel pending watch events. */
817 spin_lock(&watch_events_lock);
818 if (watch->nr_pending) {
819 list_for_each_entry_safe(event, tmp, &watch_events, list) {
820 if (event->handle != watch)
822 list_del(&event->list);
825 watch->nr_pending = 0;
827 spin_unlock(&watch_events_lock);
829 if (current->pid != xenwatch_pid)
830 mutex_unlock(&xenwatch_mutex);
832 EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
834 void xs_suspend(void)
838 down_write(&xs_watch_rwsem);
839 mutex_lock(&xs_response_mutex);
844 struct xenbus_watch *watch;
845 char token[sizeof(watch) * 2 + 1];
849 mutex_unlock(&xs_response_mutex);
853 /* No need for watches_lock: the xs_watch_rwsem is sufficient. */
854 list_for_each_entry(watch, &watches, list) {
855 sprintf(token, "%lX", (long)watch);
856 xs_watch(watch->node, token);
859 up_write(&xs_watch_rwsem);
862 void xs_suspend_cancel(void)
864 mutex_unlock(&xs_response_mutex);
865 up_write(&xs_watch_rwsem);
870 static int xenwatch_thread(void *unused)
872 struct xs_watch_event *event;
874 xenwatch_pid = current->pid;
877 wait_event_interruptible(watch_events_waitq,
878 !list_empty(&watch_events));
880 if (kthread_should_stop())
883 mutex_lock(&xenwatch_mutex);
885 spin_lock(&watch_events_lock);
886 event = list_first_entry_or_null(&watch_events,
887 struct xs_watch_event, list);
889 list_del(&event->list);
890 event->handle->nr_pending--;
892 spin_unlock(&watch_events_lock);
895 event->handle->callback(event->handle, event->path,
900 mutex_unlock(&xenwatch_mutex);
907 * Wake up all threads waiting for a xenstore reply. In case of shutdown all
908 * pending replies will be marked as "aborted" in order to let the waiters
909 * return in spite of xenstore possibly no longer being able to reply. This
910 * will avoid blocking shutdown by a thread waiting for xenstore but being
911 * necessary for shutdown processing to proceed.
913 static int xs_reboot_notify(struct notifier_block *nb,
914 unsigned long code, void *unused)
916 struct xb_req_data *req;
918 mutex_lock(&xb_write_mutex);
919 list_for_each_entry(req, &xs_reply_list, list)
921 list_for_each_entry(req, &xb_write_list, list)
923 mutex_unlock(&xb_write_mutex);
927 static struct notifier_block xs_reboot_nb = {
928 .notifier_call = xs_reboot_notify,
934 struct task_struct *task;
936 register_reboot_notifier(&xs_reboot_nb);
938 /* Initialize the shared memory rings to talk to xenstored */
939 err = xb_init_comms();
943 task = kthread_run(xenwatch_thread, NULL, "xenwatch");
945 return PTR_ERR(task);
947 /* shutdown watches for kexec boot */