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())
194 /* Make sure to reread req->state each time. */
200 static void *read_reply(struct xb_req_data
*req
)
202 while (req
->state
!= xb_req_state_got_reply
) {
203 wait_event(req
->wq
, test_reply(req
));
207 * If we are in the process of being shut-down there is
208 * no point of trying to contact XenBus - it is either
209 * killed (xenstored application) or the other domain
210 * has been killed or is unreachable.
212 return ERR_PTR(-EIO
);
214 return ERR_PTR(req
->err
);
221 static void xs_send(struct xb_req_data
*req
, struct xsd_sockmsg
*msg
)
227 req
->state
= xb_req_state_queued
;
228 init_waitqueue_head(&req
->wq
);
230 req
->msg
.req_id
= xs_request_enter(req
);
232 mutex_lock(&xb_write_mutex
);
233 list_add_tail(&req
->list
, &xb_write_list
);
234 notify
= list_is_singular(&xb_write_list
);
235 mutex_unlock(&xb_write_mutex
);
241 static void *xs_wait_for_reply(struct xb_req_data
*req
, struct xsd_sockmsg
*msg
)
245 ret
= read_reply(req
);
247 xs_request_exit(req
);
249 msg
->type
= req
->msg
.type
;
250 msg
->len
= req
->msg
.len
;
252 mutex_lock(&xb_write_mutex
);
253 if (req
->state
== xb_req_state_queued
||
254 req
->state
== xb_req_state_wait_reply
)
255 req
->state
= xb_req_state_aborted
;
258 mutex_unlock(&xb_write_mutex
);
263 static void xs_wake_up(struct xb_req_data
*req
)
268 int xenbus_dev_request_and_reply(struct xsd_sockmsg
*msg
, void *par
)
270 struct xb_req_data
*req
;
273 req
= kmalloc(sizeof(*req
) + sizeof(*vec
), GFP_KERNEL
);
277 vec
= (struct kvec
*)(req
+ 1);
278 vec
->iov_len
= msg
->len
;
279 vec
->iov_base
= msg
+ 1;
283 req
->cb
= xenbus_dev_queue_reply
;
290 EXPORT_SYMBOL(xenbus_dev_request_and_reply
);
292 /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */
293 static void *xs_talkv(struct xenbus_transaction t
,
294 enum xsd_sockmsg_type type
,
295 const struct kvec
*iovec
,
296 unsigned int num_vecs
,
299 struct xb_req_data
*req
;
300 struct xsd_sockmsg msg
;
305 req
= kmalloc(sizeof(*req
), GFP_NOIO
| __GFP_HIGH
);
307 return ERR_PTR(-ENOMEM
);
310 req
->num_vecs
= num_vecs
;
311 req
->cb
= xs_wake_up
;
316 for (i
= 0; i
< num_vecs
; i
++)
317 msg
.len
+= iovec
[i
].iov_len
;
321 ret
= xs_wait_for_reply(req
, &msg
);
328 if (msg
.type
== XS_ERROR
) {
329 err
= get_error(ret
);
331 return ERR_PTR(-err
);
334 if (msg
.type
!= type
) {
335 pr_warn_ratelimited("unexpected type [%d], expected [%d]\n",
338 return ERR_PTR(-EINVAL
);
343 /* Simplified version of xs_talkv: single message. */
344 static void *xs_single(struct xenbus_transaction t
,
345 enum xsd_sockmsg_type type
,
351 iovec
.iov_base
= (void *)string
;
352 iovec
.iov_len
= strlen(string
) + 1;
353 return xs_talkv(t
, type
, &iovec
, 1, len
);
356 /* Many commands only need an ack, don't care what it says. */
357 static int xs_error(char *reply
)
360 return PTR_ERR(reply
);
365 static unsigned int count_strings(const char *strings
, unsigned int len
)
370 for (p
= strings
, num
= 0; p
< strings
+ len
; p
+= strlen(p
) + 1)
376 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
377 static char *join(const char *dir
, const char *name
)
381 if (strlen(name
) == 0)
382 buffer
= kasprintf(GFP_NOIO
| __GFP_HIGH
, "%s", dir
);
384 buffer
= kasprintf(GFP_NOIO
| __GFP_HIGH
, "%s/%s", dir
, name
);
385 return (!buffer
) ? ERR_PTR(-ENOMEM
) : buffer
;
388 static char **split(char *strings
, unsigned int len
, unsigned int *num
)
392 /* Count the strings. */
393 *num
= count_strings(strings
, len
);
395 /* Transfer to one big alloc for easy freeing. */
396 ret
= kmalloc(*num
* sizeof(char *) + len
, GFP_NOIO
| __GFP_HIGH
);
399 return ERR_PTR(-ENOMEM
);
401 memcpy(&ret
[*num
], strings
, len
);
404 strings
= (char *)&ret
[*num
];
405 for (p
= strings
, *num
= 0; p
< strings
+ len
; p
+= strlen(p
) + 1)
411 char **xenbus_directory(struct xenbus_transaction t
,
412 const char *dir
, const char *node
, unsigned int *num
)
414 char *strings
, *path
;
417 path
= join(dir
, node
);
419 return (char **)path
;
421 strings
= xs_single(t
, XS_DIRECTORY
, path
, &len
);
424 return (char **)strings
;
426 return split(strings
, len
, num
);
428 EXPORT_SYMBOL_GPL(xenbus_directory
);
430 /* Check if a path exists. Return 1 if it does. */
431 int xenbus_exists(struct xenbus_transaction t
,
432 const char *dir
, const char *node
)
437 d
= xenbus_directory(t
, dir
, node
, &dir_n
);
443 EXPORT_SYMBOL_GPL(xenbus_exists
);
445 /* Get the value of a single file.
446 * Returns a kmalloced value: call free() on it after use.
447 * len indicates length in bytes.
449 void *xenbus_read(struct xenbus_transaction t
,
450 const char *dir
, const char *node
, unsigned int *len
)
455 path
= join(dir
, node
);
459 ret
= xs_single(t
, XS_READ
, path
, len
);
463 EXPORT_SYMBOL_GPL(xenbus_read
);
465 /* Write the value of a single file.
466 * Returns -err on failure.
468 int xenbus_write(struct xenbus_transaction t
,
469 const char *dir
, const char *node
, const char *string
)
472 struct kvec iovec
[2];
475 path
= join(dir
, node
);
477 return PTR_ERR(path
);
479 iovec
[0].iov_base
= (void *)path
;
480 iovec
[0].iov_len
= strlen(path
) + 1;
481 iovec
[1].iov_base
= (void *)string
;
482 iovec
[1].iov_len
= strlen(string
);
484 ret
= xs_error(xs_talkv(t
, XS_WRITE
, iovec
, ARRAY_SIZE(iovec
), NULL
));
488 EXPORT_SYMBOL_GPL(xenbus_write
);
490 /* Create a new directory. */
491 int xenbus_mkdir(struct xenbus_transaction t
,
492 const char *dir
, const char *node
)
497 path
= join(dir
, node
);
499 return PTR_ERR(path
);
501 ret
= xs_error(xs_single(t
, XS_MKDIR
, path
, NULL
));
505 EXPORT_SYMBOL_GPL(xenbus_mkdir
);
507 /* Destroy a file or directory (directories must be empty). */
508 int xenbus_rm(struct xenbus_transaction t
, const char *dir
, const char *node
)
513 path
= join(dir
, node
);
515 return PTR_ERR(path
);
517 ret
= xs_error(xs_single(t
, XS_RM
, path
, NULL
));
521 EXPORT_SYMBOL_GPL(xenbus_rm
);
523 /* Start a transaction: changes by others will not be seen during this
524 * transaction, and changes will not be visible to others until end.
526 int xenbus_transaction_start(struct xenbus_transaction
*t
)
530 id_str
= xs_single(XBT_NIL
, XS_TRANSACTION_START
, "", NULL
);
532 return PTR_ERR(id_str
);
534 t
->id
= simple_strtoul(id_str
, NULL
, 0);
538 EXPORT_SYMBOL_GPL(xenbus_transaction_start
);
540 /* End a transaction.
541 * If abandon is true, transaction is discarded instead of committed.
543 int xenbus_transaction_end(struct xenbus_transaction t
, int abort
)
548 strcpy(abortstr
, "F");
550 strcpy(abortstr
, "T");
552 return xs_error(xs_single(t
, XS_TRANSACTION_END
, abortstr
, NULL
));
554 EXPORT_SYMBOL_GPL(xenbus_transaction_end
);
556 /* Single read and scanf: returns -errno or num scanned. */
557 int xenbus_scanf(struct xenbus_transaction t
,
558 const char *dir
, const char *node
, const char *fmt
, ...)
564 val
= xenbus_read(t
, dir
, node
, NULL
);
569 ret
= vsscanf(val
, fmt
, ap
);
572 /* Distinctive errno. */
577 EXPORT_SYMBOL_GPL(xenbus_scanf
);
579 /* Read an (optional) unsigned value. */
580 unsigned int xenbus_read_unsigned(const char *dir
, const char *node
,
581 unsigned int default_val
)
586 ret
= xenbus_scanf(XBT_NIL
, dir
, node
, "%u", &val
);
592 EXPORT_SYMBOL_GPL(xenbus_read_unsigned
);
594 /* Single printf and write: returns -errno or 0. */
595 int xenbus_printf(struct xenbus_transaction t
,
596 const char *dir
, const char *node
, const char *fmt
, ...)
603 buf
= kvasprintf(GFP_NOIO
| __GFP_HIGH
, fmt
, ap
);
609 ret
= xenbus_write(t
, dir
, node
, buf
);
615 EXPORT_SYMBOL_GPL(xenbus_printf
);
617 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
618 int xenbus_gather(struct xenbus_transaction t
, const char *dir
, ...)
625 while (ret
== 0 && (name
= va_arg(ap
, char *)) != NULL
) {
626 const char *fmt
= va_arg(ap
, char *);
627 void *result
= va_arg(ap
, void *);
630 p
= xenbus_read(t
, dir
, name
, NULL
);
636 if (sscanf(p
, fmt
, result
) == 0)
640 *(char **)result
= p
;
645 EXPORT_SYMBOL_GPL(xenbus_gather
);
647 static int xs_watch(const char *path
, const char *token
)
651 iov
[0].iov_base
= (void *)path
;
652 iov
[0].iov_len
= strlen(path
) + 1;
653 iov
[1].iov_base
= (void *)token
;
654 iov
[1].iov_len
= strlen(token
) + 1;
656 return xs_error(xs_talkv(XBT_NIL
, XS_WATCH
, iov
,
657 ARRAY_SIZE(iov
), NULL
));
660 static int xs_unwatch(const char *path
, const char *token
)
664 iov
[0].iov_base
= (char *)path
;
665 iov
[0].iov_len
= strlen(path
) + 1;
666 iov
[1].iov_base
= (char *)token
;
667 iov
[1].iov_len
= strlen(token
) + 1;
669 return xs_error(xs_talkv(XBT_NIL
, XS_UNWATCH
, iov
,
670 ARRAY_SIZE(iov
), NULL
));
673 static struct xenbus_watch
*find_watch(const char *token
)
675 struct xenbus_watch
*i
, *cmp
;
677 cmp
= (void *)simple_strtoul(token
, NULL
, 16);
679 list_for_each_entry(i
, &watches
, list
)
686 int xs_watch_msg(struct xs_watch_event
*event
)
688 if (count_strings(event
->body
, event
->len
) != 2) {
692 event
->path
= (const char *)event
->body
;
693 event
->token
= (const char *)strchr(event
->body
, '\0') + 1;
695 spin_lock(&watches_lock
);
696 event
->handle
= find_watch(event
->token
);
697 if (event
->handle
!= NULL
) {
698 spin_lock(&watch_events_lock
);
699 list_add_tail(&event
->list
, &watch_events
);
700 wake_up(&watch_events_waitq
);
701 spin_unlock(&watch_events_lock
);
704 spin_unlock(&watches_lock
);
710 * Certain older XenBus toolstack cannot handle reading values that are
711 * not populated. Some Xen 3.4 installation are incapable of doing this
712 * so if we are running on anything older than 4 do not attempt to read
713 * control/platform-feature-xs_reset_watches.
715 static bool xen_strict_xenbus_quirk(void)
718 uint32_t eax
, ebx
, ecx
, edx
, base
;
720 base
= xen_cpuid_base();
721 cpuid(base
+ 1, &eax
, &ebx
, &ecx
, &edx
);
729 static void xs_reset_watches(void)
733 if (!xen_hvm_domain() || xen_initial_domain())
736 if (xen_strict_xenbus_quirk())
739 if (!xenbus_read_unsigned("control",
740 "platform-feature-xs_reset_watches", 0))
743 err
= xs_error(xs_single(XBT_NIL
, XS_RESET_WATCHES
, "", NULL
));
744 if (err
&& err
!= -EEXIST
)
745 pr_warn("xs_reset_watches failed: %d\n", err
);
748 /* Register callback to watch this node. */
749 int register_xenbus_watch(struct xenbus_watch
*watch
)
751 /* Pointer in ascii is the token. */
752 char token
[sizeof(watch
) * 2 + 1];
755 sprintf(token
, "%lX", (long)watch
);
757 down_read(&xs_watch_rwsem
);
759 spin_lock(&watches_lock
);
760 BUG_ON(find_watch(token
));
761 list_add(&watch
->list
, &watches
);
762 spin_unlock(&watches_lock
);
764 err
= xs_watch(watch
->node
, token
);
767 spin_lock(&watches_lock
);
768 list_del(&watch
->list
);
769 spin_unlock(&watches_lock
);
772 up_read(&xs_watch_rwsem
);
776 EXPORT_SYMBOL_GPL(register_xenbus_watch
);
778 void unregister_xenbus_watch(struct xenbus_watch
*watch
)
780 struct xs_watch_event
*event
, *tmp
;
781 char token
[sizeof(watch
) * 2 + 1];
784 sprintf(token
, "%lX", (long)watch
);
786 down_read(&xs_watch_rwsem
);
788 spin_lock(&watches_lock
);
789 BUG_ON(!find_watch(token
));
790 list_del(&watch
->list
);
791 spin_unlock(&watches_lock
);
793 err
= xs_unwatch(watch
->node
, token
);
795 pr_warn("Failed to release watch %s: %i\n", watch
->node
, err
);
797 up_read(&xs_watch_rwsem
);
799 /* Make sure there are no callbacks running currently (unless
801 if (current
->pid
!= xenwatch_pid
)
802 mutex_lock(&xenwatch_mutex
);
804 /* Cancel pending watch events. */
805 spin_lock(&watch_events_lock
);
806 list_for_each_entry_safe(event
, tmp
, &watch_events
, list
) {
807 if (event
->handle
!= watch
)
809 list_del(&event
->list
);
812 spin_unlock(&watch_events_lock
);
814 if (current
->pid
!= xenwatch_pid
)
815 mutex_unlock(&xenwatch_mutex
);
817 EXPORT_SYMBOL_GPL(unregister_xenbus_watch
);
819 void xs_suspend(void)
823 down_write(&xs_watch_rwsem
);
824 mutex_lock(&xs_response_mutex
);
829 struct xenbus_watch
*watch
;
830 char token
[sizeof(watch
) * 2 + 1];
834 mutex_unlock(&xs_response_mutex
);
838 /* No need for watches_lock: the xs_watch_rwsem is sufficient. */
839 list_for_each_entry(watch
, &watches
, list
) {
840 sprintf(token
, "%lX", (long)watch
);
841 xs_watch(watch
->node
, token
);
844 up_write(&xs_watch_rwsem
);
847 void xs_suspend_cancel(void)
849 mutex_unlock(&xs_response_mutex
);
850 up_write(&xs_watch_rwsem
);
855 static int xenwatch_thread(void *unused
)
857 struct list_head
*ent
;
858 struct xs_watch_event
*event
;
861 wait_event_interruptible(watch_events_waitq
,
862 !list_empty(&watch_events
));
864 if (kthread_should_stop())
867 mutex_lock(&xenwatch_mutex
);
869 spin_lock(&watch_events_lock
);
870 ent
= watch_events
.next
;
871 if (ent
!= &watch_events
)
873 spin_unlock(&watch_events_lock
);
875 if (ent
!= &watch_events
) {
876 event
= list_entry(ent
, struct xs_watch_event
, list
);
877 event
->handle
->callback(event
->handle
, event
->path
,
882 mutex_unlock(&xenwatch_mutex
);
889 * Wake up all threads waiting for a xenstore reply. In case of shutdown all
890 * pending replies will be marked as "aborted" in order to let the waiters
891 * return in spite of xenstore possibly no longer being able to reply. This
892 * will avoid blocking shutdown by a thread waiting for xenstore but being
893 * necessary for shutdown processing to proceed.
895 static int xs_reboot_notify(struct notifier_block
*nb
,
896 unsigned long code
, void *unused
)
898 struct xb_req_data
*req
;
900 mutex_lock(&xb_write_mutex
);
901 list_for_each_entry(req
, &xs_reply_list
, list
)
903 list_for_each_entry(req
, &xb_write_list
, list
)
905 mutex_unlock(&xb_write_mutex
);
909 static struct notifier_block xs_reboot_nb
= {
910 .notifier_call
= xs_reboot_notify
,
916 struct task_struct
*task
;
918 register_reboot_notifier(&xs_reboot_nb
);
920 /* Initialize the shared memory rings to talk to xenstored */
921 err
= xb_init_comms();
925 task
= kthread_run(xenwatch_thread
, NULL
, "xenwatch");
927 return PTR_ERR(task
);
928 xenwatch_pid
= task
->pid
;
930 /* shutdown watches for kexec boot */