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 &&
144 !WARN_ON_ONCE(req->msg.type == XS_ERROR &&
145 !strcmp(req->body, "ENOENT"))))
147 spin_unlock(&xs_state_lock);
149 if (xs_suspend_active && !xs_state_users)
150 wake_up(&xs_state_exit_wq);
153 static int get_error(const char *errorstring)
157 for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) {
158 if (i == ARRAY_SIZE(xsd_errors) - 1) {
159 pr_warn("xen store gave: unknown error %s\n",
164 return xsd_errors[i].errnum;
167 static bool xenbus_ok(void)
169 switch (xen_store_domain_type) {
171 switch (system_state) {
172 case SYSTEM_POWER_OFF:
182 /* FIXME: Could check that the remote domain is alive,
183 * but it is normally initial domain. */
191 static bool test_reply(struct xb_req_data *req)
193 if (req->state == xb_req_state_got_reply || !xenbus_ok())
196 /* Make sure to reread req->state each time. */
202 static void *read_reply(struct xb_req_data *req)
204 while (req->state != xb_req_state_got_reply) {
205 wait_event(req->wq, test_reply(req));
209 * If we are in the process of being shut-down there is
210 * no point of trying to contact XenBus - it is either
211 * killed (xenstored application) or the other domain
212 * has been killed or is unreachable.
214 return ERR_PTR(-EIO);
216 return ERR_PTR(req->err);
223 static void xs_send(struct xb_req_data *req, struct xsd_sockmsg *msg)
229 req->state = xb_req_state_queued;
230 init_waitqueue_head(&req->wq);
232 /* Save the caller req_id and restore it later in the reply */
233 req->caller_req_id = req->msg.req_id;
234 req->msg.req_id = xs_request_enter(req);
236 mutex_lock(&xb_write_mutex);
237 list_add_tail(&req->list, &xb_write_list);
238 notify = list_is_singular(&xb_write_list);
239 mutex_unlock(&xb_write_mutex);
245 static void *xs_wait_for_reply(struct xb_req_data *req, struct xsd_sockmsg *msg)
249 ret = read_reply(req);
251 xs_request_exit(req);
253 msg->type = req->msg.type;
254 msg->len = req->msg.len;
256 mutex_lock(&xb_write_mutex);
257 if (req->state == xb_req_state_queued ||
258 req->state == xb_req_state_wait_reply)
259 req->state = xb_req_state_aborted;
262 mutex_unlock(&xb_write_mutex);
267 static void xs_wake_up(struct xb_req_data *req)
272 int xenbus_dev_request_and_reply(struct xsd_sockmsg *msg, void *par)
274 struct xb_req_data *req;
277 req = kmalloc(sizeof(*req) + sizeof(*vec), GFP_KERNEL);
281 vec = (struct kvec *)(req + 1);
282 vec->iov_len = msg->len;
283 vec->iov_base = msg + 1;
287 req->cb = xenbus_dev_queue_reply;
294 EXPORT_SYMBOL(xenbus_dev_request_and_reply);
296 /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */
297 static void *xs_talkv(struct xenbus_transaction t,
298 enum xsd_sockmsg_type type,
299 const struct kvec *iovec,
300 unsigned int num_vecs,
303 struct xb_req_data *req;
304 struct xsd_sockmsg msg;
309 req = kmalloc(sizeof(*req), GFP_NOIO | __GFP_HIGH);
311 return ERR_PTR(-ENOMEM);
314 req->num_vecs = num_vecs;
315 req->cb = xs_wake_up;
321 for (i = 0; i < num_vecs; i++)
322 msg.len += iovec[i].iov_len;
326 ret = xs_wait_for_reply(req, &msg);
333 if (msg.type == XS_ERROR) {
334 err = get_error(ret);
336 return ERR_PTR(-err);
339 if (msg.type != type) {
340 pr_warn_ratelimited("unexpected type [%d], expected [%d]\n",
343 return ERR_PTR(-EINVAL);
348 /* Simplified version of xs_talkv: single message. */
349 static void *xs_single(struct xenbus_transaction t,
350 enum xsd_sockmsg_type type,
356 iovec.iov_base = (void *)string;
357 iovec.iov_len = strlen(string) + 1;
358 return xs_talkv(t, type, &iovec, 1, len);
361 /* Many commands only need an ack, don't care what it says. */
362 static int xs_error(char *reply)
365 return PTR_ERR(reply);
370 static unsigned int count_strings(const char *strings, unsigned int len)
375 for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1)
381 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
382 static char *join(const char *dir, const char *name)
386 if (strlen(name) == 0)
387 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir);
389 buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name);
390 return (!buffer) ? ERR_PTR(-ENOMEM) : buffer;
393 static char **split(char *strings, unsigned int len, unsigned int *num)
397 /* Count the strings. */
398 *num = count_strings(strings, len);
400 /* Transfer to one big alloc for easy freeing. */
401 ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH);
404 return ERR_PTR(-ENOMEM);
406 memcpy(&ret[*num], strings, len);
409 strings = (char *)&ret[*num];
410 for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1)
416 char **xenbus_directory(struct xenbus_transaction t,
417 const char *dir, const char *node, unsigned int *num)
419 char *strings, *path;
422 path = join(dir, node);
424 return (char **)path;
426 strings = xs_single(t, XS_DIRECTORY, path, &len);
429 return (char **)strings;
431 return split(strings, len, num);
433 EXPORT_SYMBOL_GPL(xenbus_directory);
435 /* Check if a path exists. Return 1 if it does. */
436 int xenbus_exists(struct xenbus_transaction t,
437 const char *dir, const char *node)
442 d = xenbus_directory(t, dir, node, &dir_n);
448 EXPORT_SYMBOL_GPL(xenbus_exists);
450 /* Get the value of a single file.
451 * Returns a kmalloced value: call free() on it after use.
452 * len indicates length in bytes.
454 void *xenbus_read(struct xenbus_transaction t,
455 const char *dir, const char *node, unsigned int *len)
460 path = join(dir, node);
464 ret = xs_single(t, XS_READ, path, len);
468 EXPORT_SYMBOL_GPL(xenbus_read);
470 /* Write the value of a single file.
471 * Returns -err on failure.
473 int xenbus_write(struct xenbus_transaction t,
474 const char *dir, const char *node, const char *string)
477 struct kvec iovec[2];
480 path = join(dir, node);
482 return PTR_ERR(path);
484 iovec[0].iov_base = (void *)path;
485 iovec[0].iov_len = strlen(path) + 1;
486 iovec[1].iov_base = (void *)string;
487 iovec[1].iov_len = strlen(string);
489 ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL));
493 EXPORT_SYMBOL_GPL(xenbus_write);
495 /* Create a new directory. */
496 int xenbus_mkdir(struct xenbus_transaction t,
497 const char *dir, const char *node)
502 path = join(dir, node);
504 return PTR_ERR(path);
506 ret = xs_error(xs_single(t, XS_MKDIR, path, NULL));
510 EXPORT_SYMBOL_GPL(xenbus_mkdir);
512 /* Destroy a file or directory (directories must be empty). */
513 int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node)
518 path = join(dir, node);
520 return PTR_ERR(path);
522 ret = xs_error(xs_single(t, XS_RM, path, NULL));
526 EXPORT_SYMBOL_GPL(xenbus_rm);
528 /* Start a transaction: changes by others will not be seen during this
529 * transaction, and changes will not be visible to others until end.
531 int xenbus_transaction_start(struct xenbus_transaction *t)
535 id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL);
537 return PTR_ERR(id_str);
539 t->id = simple_strtoul(id_str, NULL, 0);
543 EXPORT_SYMBOL_GPL(xenbus_transaction_start);
545 /* End a transaction.
546 * If abandon is true, transaction is discarded instead of committed.
548 int xenbus_transaction_end(struct xenbus_transaction t, int abort)
553 strcpy(abortstr, "F");
555 strcpy(abortstr, "T");
557 return xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL));
559 EXPORT_SYMBOL_GPL(xenbus_transaction_end);
561 /* Single read and scanf: returns -errno or num scanned. */
562 int xenbus_scanf(struct xenbus_transaction t,
563 const char *dir, const char *node, const char *fmt, ...)
569 val = xenbus_read(t, dir, node, NULL);
574 ret = vsscanf(val, fmt, ap);
577 /* Distinctive errno. */
582 EXPORT_SYMBOL_GPL(xenbus_scanf);
584 /* Read an (optional) unsigned value. */
585 unsigned int xenbus_read_unsigned(const char *dir, const char *node,
586 unsigned int default_val)
591 ret = xenbus_scanf(XBT_NIL, dir, node, "%u", &val);
597 EXPORT_SYMBOL_GPL(xenbus_read_unsigned);
599 /* Single printf and write: returns -errno or 0. */
600 int xenbus_printf(struct xenbus_transaction t,
601 const char *dir, const char *node, const char *fmt, ...)
608 buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
614 ret = xenbus_write(t, dir, node, buf);
620 EXPORT_SYMBOL_GPL(xenbus_printf);
622 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
623 int xenbus_gather(struct xenbus_transaction t, const char *dir, ...)
630 while (ret == 0 && (name = va_arg(ap, char *)) != NULL) {
631 const char *fmt = va_arg(ap, char *);
632 void *result = va_arg(ap, void *);
635 p = xenbus_read(t, dir, name, NULL);
641 if (sscanf(p, fmt, result) == 0)
645 *(char **)result = p;
650 EXPORT_SYMBOL_GPL(xenbus_gather);
652 static int xs_watch(const char *path, const char *token)
656 iov[0].iov_base = (void *)path;
657 iov[0].iov_len = strlen(path) + 1;
658 iov[1].iov_base = (void *)token;
659 iov[1].iov_len = strlen(token) + 1;
661 return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov,
662 ARRAY_SIZE(iov), NULL));
665 static int xs_unwatch(const char *path, const char *token)
669 iov[0].iov_base = (char *)path;
670 iov[0].iov_len = strlen(path) + 1;
671 iov[1].iov_base = (char *)token;
672 iov[1].iov_len = strlen(token) + 1;
674 return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov,
675 ARRAY_SIZE(iov), NULL));
678 static struct xenbus_watch *find_watch(const char *token)
680 struct xenbus_watch *i, *cmp;
682 cmp = (void *)simple_strtoul(token, NULL, 16);
684 list_for_each_entry(i, &watches, list)
691 int xs_watch_msg(struct xs_watch_event *event)
693 if (count_strings(event->body, event->len) != 2) {
697 event->path = (const char *)event->body;
698 event->token = (const char *)strchr(event->body, '\0') + 1;
700 spin_lock(&watches_lock);
701 event->handle = find_watch(event->token);
702 if (event->handle != NULL) {
703 spin_lock(&watch_events_lock);
704 list_add_tail(&event->list, &watch_events);
705 wake_up(&watch_events_waitq);
706 spin_unlock(&watch_events_lock);
709 spin_unlock(&watches_lock);
715 * Certain older XenBus toolstack cannot handle reading values that are
716 * not populated. Some Xen 3.4 installation are incapable of doing this
717 * so if we are running on anything older than 4 do not attempt to read
718 * control/platform-feature-xs_reset_watches.
720 static bool xen_strict_xenbus_quirk(void)
723 uint32_t eax, ebx, ecx, edx, base;
725 base = xen_cpuid_base();
726 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
734 static void xs_reset_watches(void)
738 if (!xen_hvm_domain() || xen_initial_domain())
741 if (xen_strict_xenbus_quirk())
744 if (!xenbus_read_unsigned("control",
745 "platform-feature-xs_reset_watches", 0))
748 err = xs_error(xs_single(XBT_NIL, XS_RESET_WATCHES, "", NULL));
749 if (err && err != -EEXIST)
750 pr_warn("xs_reset_watches failed: %d\n", err);
753 /* Register callback to watch this node. */
754 int register_xenbus_watch(struct xenbus_watch *watch)
756 /* Pointer in ascii is the token. */
757 char token[sizeof(watch) * 2 + 1];
760 sprintf(token, "%lX", (long)watch);
762 down_read(&xs_watch_rwsem);
764 spin_lock(&watches_lock);
765 BUG_ON(find_watch(token));
766 list_add(&watch->list, &watches);
767 spin_unlock(&watches_lock);
769 err = xs_watch(watch->node, token);
772 spin_lock(&watches_lock);
773 list_del(&watch->list);
774 spin_unlock(&watches_lock);
777 up_read(&xs_watch_rwsem);
781 EXPORT_SYMBOL_GPL(register_xenbus_watch);
783 void unregister_xenbus_watch(struct xenbus_watch *watch)
785 struct xs_watch_event *event, *tmp;
786 char token[sizeof(watch) * 2 + 1];
789 sprintf(token, "%lX", (long)watch);
791 down_read(&xs_watch_rwsem);
793 spin_lock(&watches_lock);
794 BUG_ON(!find_watch(token));
795 list_del(&watch->list);
796 spin_unlock(&watches_lock);
798 err = xs_unwatch(watch->node, token);
800 pr_warn("Failed to release watch %s: %i\n", watch->node, err);
802 up_read(&xs_watch_rwsem);
804 /* Make sure there are no callbacks running currently (unless
806 if (current->pid != xenwatch_pid)
807 mutex_lock(&xenwatch_mutex);
809 /* Cancel pending watch events. */
810 spin_lock(&watch_events_lock);
811 list_for_each_entry_safe(event, tmp, &watch_events, list) {
812 if (event->handle != watch)
814 list_del(&event->list);
817 spin_unlock(&watch_events_lock);
819 if (current->pid != xenwatch_pid)
820 mutex_unlock(&xenwatch_mutex);
822 EXPORT_SYMBOL_GPL(unregister_xenbus_watch);
824 void xs_suspend(void)
828 down_write(&xs_watch_rwsem);
829 mutex_lock(&xs_response_mutex);
834 struct xenbus_watch *watch;
835 char token[sizeof(watch) * 2 + 1];
839 mutex_unlock(&xs_response_mutex);
843 /* No need for watches_lock: the xs_watch_rwsem is sufficient. */
844 list_for_each_entry(watch, &watches, list) {
845 sprintf(token, "%lX", (long)watch);
846 xs_watch(watch->node, token);
849 up_write(&xs_watch_rwsem);
852 void xs_suspend_cancel(void)
854 mutex_unlock(&xs_response_mutex);
855 up_write(&xs_watch_rwsem);
860 static int xenwatch_thread(void *unused)
862 struct list_head *ent;
863 struct xs_watch_event *event;
865 xenwatch_pid = current->pid;
868 wait_event_interruptible(watch_events_waitq,
869 !list_empty(&watch_events));
871 if (kthread_should_stop())
874 mutex_lock(&xenwatch_mutex);
876 spin_lock(&watch_events_lock);
877 ent = watch_events.next;
878 if (ent != &watch_events)
880 spin_unlock(&watch_events_lock);
882 if (ent != &watch_events) {
883 event = list_entry(ent, struct xs_watch_event, list);
884 event->handle->callback(event->handle, event->path,
889 mutex_unlock(&xenwatch_mutex);
896 * Wake up all threads waiting for a xenstore reply. In case of shutdown all
897 * pending replies will be marked as "aborted" in order to let the waiters
898 * return in spite of xenstore possibly no longer being able to reply. This
899 * will avoid blocking shutdown by a thread waiting for xenstore but being
900 * necessary for shutdown processing to proceed.
902 static int xs_reboot_notify(struct notifier_block *nb,
903 unsigned long code, void *unused)
905 struct xb_req_data *req;
907 mutex_lock(&xb_write_mutex);
908 list_for_each_entry(req, &xs_reply_list, list)
910 list_for_each_entry(req, &xb_write_list, list)
912 mutex_unlock(&xb_write_mutex);
916 static struct notifier_block xs_reboot_nb = {
917 .notifier_call = xs_reboot_notify,
923 struct task_struct *task;
925 register_reboot_notifier(&xs_reboot_nb);
927 /* Initialize the shared memory rings to talk to xenstored */
928 err = xb_init_comms();
932 task = kthread_run(xenwatch_thread, NULL, "xenwatch");
934 return PTR_ERR(task);
936 /* shutdown watches for kexec boot */