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 #include <linux/unistd.h>
35 #include <linux/errno.h>
36 #include <linux/types.h>
37 #include <linux/uio.h>
38 #include <linux/kernel.h>
39 #include <linux/string.h>
40 #include <linux/err.h>
41 #include <linux/slab.h>
42 #include <linux/fcntl.h>
43 #include <linux/kthread.h>
44 #include <linux/rwsem.h>
45 #include <linux/module.h>
46 #include <linux/mutex.h>
47 #include <asm/xen/hypervisor.h>
48 #include <xen/xenbus.h>
50 #include "xenbus_comms.h"
52 struct xs_stored_msg
{
53 struct list_head list
;
55 struct xsd_sockmsg hdr
;
63 /* Queued watch events. */
65 struct xenbus_watch
*handle
;
67 unsigned int vec_size
;
73 /* A list of replies. Currently only one will ever be outstanding. */
74 struct list_head reply_list
;
75 spinlock_t reply_lock
;
76 wait_queue_head_t reply_waitq
;
79 * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex.
80 * response_mutex is never taken simultaneously with the other three.
82 * transaction_mutex must be held before incrementing
83 * transaction_count. The mutex is held when a suspend is in
84 * progress to prevent new transactions starting.
86 * When decrementing transaction_count to zero the wait queue
87 * should be woken up, the suspend code waits for count to
91 /* One request at a time. */
92 struct mutex request_mutex
;
94 /* Protect xenbus reader thread against save/restore. */
95 struct mutex response_mutex
;
97 /* Protect transactions against save/restore. */
98 struct mutex transaction_mutex
;
99 atomic_t transaction_count
;
100 wait_queue_head_t transaction_wq
;
102 /* Protect watch (de)register against save/restore. */
103 struct rw_semaphore watch_mutex
;
106 static struct xs_handle xs_state
;
108 /* List of registered watches, and a lock to protect it. */
109 static LIST_HEAD(watches
);
110 static DEFINE_SPINLOCK(watches_lock
);
112 /* List of pending watch callback events, and a lock to protect it. */
113 static LIST_HEAD(watch_events
);
114 static DEFINE_SPINLOCK(watch_events_lock
);
117 * Details of the xenwatch callback kernel thread. The thread waits on the
118 * watch_events_waitq for work to do (queued on watch_events list). When it
119 * wakes up it acquires the xenwatch_mutex before reading the list and
122 static pid_t xenwatch_pid
;
123 static DEFINE_MUTEX(xenwatch_mutex
);
124 static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq
);
126 static int get_error(const char *errorstring
)
130 for (i
= 0; strcmp(errorstring
, xsd_errors
[i
].errstring
) != 0; i
++) {
131 if (i
== ARRAY_SIZE(xsd_errors
) - 1) {
133 "XENBUS xen store gave: unknown error %s",
138 return xsd_errors
[i
].errnum
;
141 static void *read_reply(enum xsd_sockmsg_type
*type
, unsigned int *len
)
143 struct xs_stored_msg
*msg
;
146 spin_lock(&xs_state
.reply_lock
);
148 while (list_empty(&xs_state
.reply_list
)) {
149 spin_unlock(&xs_state
.reply_lock
);
150 /* XXX FIXME: Avoid synchronous wait for response here. */
151 wait_event(xs_state
.reply_waitq
,
152 !list_empty(&xs_state
.reply_list
));
153 spin_lock(&xs_state
.reply_lock
);
156 msg
= list_entry(xs_state
.reply_list
.next
,
157 struct xs_stored_msg
, list
);
158 list_del(&msg
->list
);
160 spin_unlock(&xs_state
.reply_lock
);
162 *type
= msg
->hdr
.type
;
165 body
= msg
->u
.reply
.body
;
172 static void transaction_start(void)
174 mutex_lock(&xs_state
.transaction_mutex
);
175 atomic_inc(&xs_state
.transaction_count
);
176 mutex_unlock(&xs_state
.transaction_mutex
);
179 static void transaction_end(void)
181 if (atomic_dec_and_test(&xs_state
.transaction_count
))
182 wake_up(&xs_state
.transaction_wq
);
185 static void transaction_suspend(void)
187 mutex_lock(&xs_state
.transaction_mutex
);
188 wait_event(xs_state
.transaction_wq
,
189 atomic_read(&xs_state
.transaction_count
) == 0);
192 static void transaction_resume(void)
194 mutex_unlock(&xs_state
.transaction_mutex
);
197 void *xenbus_dev_request_and_reply(struct xsd_sockmsg
*msg
)
200 struct xsd_sockmsg req_msg
= *msg
;
203 if (req_msg
.type
== XS_TRANSACTION_START
)
206 mutex_lock(&xs_state
.request_mutex
);
208 err
= xb_write(msg
, sizeof(*msg
) + msg
->len
);
210 msg
->type
= XS_ERROR
;
213 ret
= read_reply(&msg
->type
, &msg
->len
);
215 mutex_unlock(&xs_state
.request_mutex
);
217 if ((msg
->type
== XS_TRANSACTION_END
) ||
218 ((req_msg
.type
== XS_TRANSACTION_START
) &&
219 (msg
->type
== XS_ERROR
)))
224 EXPORT_SYMBOL(xenbus_dev_request_and_reply
);
226 /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */
227 static void *xs_talkv(struct xenbus_transaction t
,
228 enum xsd_sockmsg_type type
,
229 const struct kvec
*iovec
,
230 unsigned int num_vecs
,
233 struct xsd_sockmsg msg
;
242 for (i
= 0; i
< num_vecs
; i
++)
243 msg
.len
+= iovec
[i
].iov_len
;
245 mutex_lock(&xs_state
.request_mutex
);
247 err
= xb_write(&msg
, sizeof(msg
));
249 mutex_unlock(&xs_state
.request_mutex
);
253 for (i
= 0; i
< num_vecs
; i
++) {
254 err
= xb_write(iovec
[i
].iov_base
, iovec
[i
].iov_len
);
256 mutex_unlock(&xs_state
.request_mutex
);
261 ret
= read_reply(&msg
.type
, len
);
263 mutex_unlock(&xs_state
.request_mutex
);
268 if (msg
.type
== XS_ERROR
) {
269 err
= get_error(ret
);
271 return ERR_PTR(-err
);
274 if (msg
.type
!= type
) {
275 if (printk_ratelimit())
277 "XENBUS unexpected type [%d], expected [%d]\n",
280 return ERR_PTR(-EINVAL
);
285 /* Simplified version of xs_talkv: single message. */
286 static void *xs_single(struct xenbus_transaction t
,
287 enum xsd_sockmsg_type type
,
293 iovec
.iov_base
= (void *)string
;
294 iovec
.iov_len
= strlen(string
) + 1;
295 return xs_talkv(t
, type
, &iovec
, 1, len
);
298 /* Many commands only need an ack, don't care what it says. */
299 static int xs_error(char *reply
)
302 return PTR_ERR(reply
);
307 static unsigned int count_strings(const char *strings
, unsigned int len
)
312 for (p
= strings
, num
= 0; p
< strings
+ len
; p
+= strlen(p
) + 1)
318 /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */
319 static char *join(const char *dir
, const char *name
)
323 if (strlen(name
) == 0)
324 buffer
= kasprintf(GFP_NOIO
| __GFP_HIGH
, "%s", dir
);
326 buffer
= kasprintf(GFP_NOIO
| __GFP_HIGH
, "%s/%s", dir
, name
);
327 return (!buffer
) ? ERR_PTR(-ENOMEM
) : buffer
;
330 static char **split(char *strings
, unsigned int len
, unsigned int *num
)
334 /* Count the strings. */
335 *num
= count_strings(strings
, len
);
337 /* Transfer to one big alloc for easy freeing. */
338 ret
= kmalloc(*num
* sizeof(char *) + len
, GFP_NOIO
| __GFP_HIGH
);
341 return ERR_PTR(-ENOMEM
);
343 memcpy(&ret
[*num
], strings
, len
);
346 strings
= (char *)&ret
[*num
];
347 for (p
= strings
, *num
= 0; p
< strings
+ len
; p
+= strlen(p
) + 1)
353 char **xenbus_directory(struct xenbus_transaction t
,
354 const char *dir
, const char *node
, unsigned int *num
)
356 char *strings
, *path
;
359 path
= join(dir
, node
);
361 return (char **)path
;
363 strings
= xs_single(t
, XS_DIRECTORY
, path
, &len
);
366 return (char **)strings
;
368 return split(strings
, len
, num
);
370 EXPORT_SYMBOL_GPL(xenbus_directory
);
372 /* Check if a path exists. Return 1 if it does. */
373 int xenbus_exists(struct xenbus_transaction t
,
374 const char *dir
, const char *node
)
379 d
= xenbus_directory(t
, dir
, node
, &dir_n
);
385 EXPORT_SYMBOL_GPL(xenbus_exists
);
387 /* Get the value of a single file.
388 * Returns a kmalloced value: call free() on it after use.
389 * len indicates length in bytes.
391 void *xenbus_read(struct xenbus_transaction t
,
392 const char *dir
, const char *node
, unsigned int *len
)
397 path
= join(dir
, node
);
401 ret
= xs_single(t
, XS_READ
, path
, len
);
405 EXPORT_SYMBOL_GPL(xenbus_read
);
407 /* Write the value of a single file.
408 * Returns -err on failure.
410 int xenbus_write(struct xenbus_transaction t
,
411 const char *dir
, const char *node
, const char *string
)
414 struct kvec iovec
[2];
417 path
= join(dir
, node
);
419 return PTR_ERR(path
);
421 iovec
[0].iov_base
= (void *)path
;
422 iovec
[0].iov_len
= strlen(path
) + 1;
423 iovec
[1].iov_base
= (void *)string
;
424 iovec
[1].iov_len
= strlen(string
);
426 ret
= xs_error(xs_talkv(t
, XS_WRITE
, iovec
, ARRAY_SIZE(iovec
), NULL
));
430 EXPORT_SYMBOL_GPL(xenbus_write
);
432 /* Create a new directory. */
433 int xenbus_mkdir(struct xenbus_transaction t
,
434 const char *dir
, const char *node
)
439 path
= join(dir
, node
);
441 return PTR_ERR(path
);
443 ret
= xs_error(xs_single(t
, XS_MKDIR
, path
, NULL
));
447 EXPORT_SYMBOL_GPL(xenbus_mkdir
);
449 /* Destroy a file or directory (directories must be empty). */
450 int xenbus_rm(struct xenbus_transaction t
, const char *dir
, const char *node
)
455 path
= join(dir
, node
);
457 return PTR_ERR(path
);
459 ret
= xs_error(xs_single(t
, XS_RM
, path
, NULL
));
463 EXPORT_SYMBOL_GPL(xenbus_rm
);
465 /* Start a transaction: changes by others will not be seen during this
466 * transaction, and changes will not be visible to others until end.
468 int xenbus_transaction_start(struct xenbus_transaction
*t
)
474 id_str
= xs_single(XBT_NIL
, XS_TRANSACTION_START
, "", NULL
);
475 if (IS_ERR(id_str
)) {
477 return PTR_ERR(id_str
);
480 t
->id
= simple_strtoul(id_str
, NULL
, 0);
484 EXPORT_SYMBOL_GPL(xenbus_transaction_start
);
486 /* End a transaction.
487 * If abandon is true, transaction is discarded instead of committed.
489 int xenbus_transaction_end(struct xenbus_transaction t
, int abort
)
495 strcpy(abortstr
, "F");
497 strcpy(abortstr
, "T");
499 err
= xs_error(xs_single(t
, XS_TRANSACTION_END
, abortstr
, NULL
));
505 EXPORT_SYMBOL_GPL(xenbus_transaction_end
);
507 /* Single read and scanf: returns -errno or num scanned. */
508 int xenbus_scanf(struct xenbus_transaction t
,
509 const char *dir
, const char *node
, const char *fmt
, ...)
515 val
= xenbus_read(t
, dir
, node
, NULL
);
520 ret
= vsscanf(val
, fmt
, ap
);
523 /* Distinctive errno. */
528 EXPORT_SYMBOL_GPL(xenbus_scanf
);
530 /* Single printf and write: returns -errno or 0. */
531 int xenbus_printf(struct xenbus_transaction t
,
532 const char *dir
, const char *node
, const char *fmt
, ...)
539 buf
= kvasprintf(GFP_NOIO
| __GFP_HIGH
, fmt
, ap
);
545 ret
= xenbus_write(t
, dir
, node
, buf
);
551 EXPORT_SYMBOL_GPL(xenbus_printf
);
553 /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */
554 int xenbus_gather(struct xenbus_transaction t
, const char *dir
, ...)
561 while (ret
== 0 && (name
= va_arg(ap
, char *)) != NULL
) {
562 const char *fmt
= va_arg(ap
, char *);
563 void *result
= va_arg(ap
, void *);
566 p
= xenbus_read(t
, dir
, name
, NULL
);
572 if (sscanf(p
, fmt
, result
) == 0)
576 *(char **)result
= p
;
581 EXPORT_SYMBOL_GPL(xenbus_gather
);
583 static int xs_watch(const char *path
, const char *token
)
587 iov
[0].iov_base
= (void *)path
;
588 iov
[0].iov_len
= strlen(path
) + 1;
589 iov
[1].iov_base
= (void *)token
;
590 iov
[1].iov_len
= strlen(token
) + 1;
592 return xs_error(xs_talkv(XBT_NIL
, XS_WATCH
, iov
,
593 ARRAY_SIZE(iov
), NULL
));
596 static int xs_unwatch(const char *path
, const char *token
)
600 iov
[0].iov_base
= (char *)path
;
601 iov
[0].iov_len
= strlen(path
) + 1;
602 iov
[1].iov_base
= (char *)token
;
603 iov
[1].iov_len
= strlen(token
) + 1;
605 return xs_error(xs_talkv(XBT_NIL
, XS_UNWATCH
, iov
,
606 ARRAY_SIZE(iov
), NULL
));
609 static struct xenbus_watch
*find_watch(const char *token
)
611 struct xenbus_watch
*i
, *cmp
;
613 cmp
= (void *)simple_strtoul(token
, NULL
, 16);
615 list_for_each_entry(i
, &watches
, list
)
622 * Certain older XenBus toolstack cannot handle reading values that are
623 * not populated. Some Xen 3.4 installation are incapable of doing this
624 * so if we are running on anything older than 4 do not attempt to read
625 * control/platform-feature-xs_reset_watches.
627 static bool xen_strict_xenbus_quirk(void)
630 uint32_t eax
, ebx
, ecx
, edx
, base
;
632 base
= xen_cpuid_base();
633 cpuid(base
+ 1, &eax
, &ebx
, &ecx
, &edx
);
641 static void xs_reset_watches(void)
643 int err
, supported
= 0;
645 if (!xen_hvm_domain() || xen_initial_domain())
648 if (xen_strict_xenbus_quirk())
651 err
= xenbus_scanf(XBT_NIL
, "control",
652 "platform-feature-xs_reset_watches", "%d", &supported
);
653 if (err
!= 1 || !supported
)
656 err
= xs_error(xs_single(XBT_NIL
, XS_RESET_WATCHES
, "", NULL
));
657 if (err
&& err
!= -EEXIST
)
658 printk(KERN_WARNING
"xs_reset_watches failed: %d\n", err
);
661 /* Register callback to watch this node. */
662 int register_xenbus_watch(struct xenbus_watch
*watch
)
664 /* Pointer in ascii is the token. */
665 char token
[sizeof(watch
) * 2 + 1];
668 sprintf(token
, "%lX", (long)watch
);
670 down_read(&xs_state
.watch_mutex
);
672 spin_lock(&watches_lock
);
673 BUG_ON(find_watch(token
));
674 list_add(&watch
->list
, &watches
);
675 spin_unlock(&watches_lock
);
677 err
= xs_watch(watch
->node
, token
);
680 spin_lock(&watches_lock
);
681 list_del(&watch
->list
);
682 spin_unlock(&watches_lock
);
685 up_read(&xs_state
.watch_mutex
);
689 EXPORT_SYMBOL_GPL(register_xenbus_watch
);
691 void unregister_xenbus_watch(struct xenbus_watch
*watch
)
693 struct xs_stored_msg
*msg
, *tmp
;
694 char token
[sizeof(watch
) * 2 + 1];
697 sprintf(token
, "%lX", (long)watch
);
699 down_read(&xs_state
.watch_mutex
);
701 spin_lock(&watches_lock
);
702 BUG_ON(!find_watch(token
));
703 list_del(&watch
->list
);
704 spin_unlock(&watches_lock
);
706 err
= xs_unwatch(watch
->node
, token
);
709 "XENBUS Failed to release watch %s: %i\n",
712 up_read(&xs_state
.watch_mutex
);
714 /* Make sure there are no callbacks running currently (unless
716 if (current
->pid
!= xenwatch_pid
)
717 mutex_lock(&xenwatch_mutex
);
719 /* Cancel pending watch events. */
720 spin_lock(&watch_events_lock
);
721 list_for_each_entry_safe(msg
, tmp
, &watch_events
, list
) {
722 if (msg
->u
.watch
.handle
!= watch
)
724 list_del(&msg
->list
);
725 kfree(msg
->u
.watch
.vec
);
728 spin_unlock(&watch_events_lock
);
730 if (current
->pid
!= xenwatch_pid
)
731 mutex_unlock(&xenwatch_mutex
);
733 EXPORT_SYMBOL_GPL(unregister_xenbus_watch
);
735 void xs_suspend(void)
737 transaction_suspend();
738 down_write(&xs_state
.watch_mutex
);
739 mutex_lock(&xs_state
.request_mutex
);
740 mutex_lock(&xs_state
.response_mutex
);
745 struct xenbus_watch
*watch
;
746 char token
[sizeof(watch
) * 2 + 1];
750 mutex_unlock(&xs_state
.response_mutex
);
751 mutex_unlock(&xs_state
.request_mutex
);
752 transaction_resume();
754 /* No need for watches_lock: the watch_mutex is sufficient. */
755 list_for_each_entry(watch
, &watches
, list
) {
756 sprintf(token
, "%lX", (long)watch
);
757 xs_watch(watch
->node
, token
);
760 up_write(&xs_state
.watch_mutex
);
763 void xs_suspend_cancel(void)
765 mutex_unlock(&xs_state
.response_mutex
);
766 mutex_unlock(&xs_state
.request_mutex
);
767 up_write(&xs_state
.watch_mutex
);
768 mutex_unlock(&xs_state
.transaction_mutex
);
771 static int xenwatch_thread(void *unused
)
773 struct list_head
*ent
;
774 struct xs_stored_msg
*msg
;
777 wait_event_interruptible(watch_events_waitq
,
778 !list_empty(&watch_events
));
780 if (kthread_should_stop())
783 mutex_lock(&xenwatch_mutex
);
785 spin_lock(&watch_events_lock
);
786 ent
= watch_events
.next
;
787 if (ent
!= &watch_events
)
789 spin_unlock(&watch_events_lock
);
791 if (ent
!= &watch_events
) {
792 msg
= list_entry(ent
, struct xs_stored_msg
, list
);
793 msg
->u
.watch
.handle
->callback(
795 (const char **)msg
->u
.watch
.vec
,
796 msg
->u
.watch
.vec_size
);
797 kfree(msg
->u
.watch
.vec
);
801 mutex_unlock(&xenwatch_mutex
);
807 static int process_msg(void)
809 struct xs_stored_msg
*msg
;
814 * We must disallow save/restore while reading a xenstore message.
815 * A partial read across s/r leaves us out of sync with xenstored.
818 err
= xb_wait_for_data_to_read();
821 mutex_lock(&xs_state
.response_mutex
);
822 if (xb_data_to_read())
824 /* We raced with save/restore: pending data 'disappeared'. */
825 mutex_unlock(&xs_state
.response_mutex
);
829 msg
= kmalloc(sizeof(*msg
), GFP_NOIO
| __GFP_HIGH
);
835 err
= xb_read(&msg
->hdr
, sizeof(msg
->hdr
));
841 if (msg
->hdr
.len
> XENSTORE_PAYLOAD_MAX
) {
847 body
= kmalloc(msg
->hdr
.len
+ 1, GFP_NOIO
| __GFP_HIGH
);
854 err
= xb_read(body
, msg
->hdr
.len
);
860 body
[msg
->hdr
.len
] = '\0';
862 if (msg
->hdr
.type
== XS_WATCH_EVENT
) {
863 msg
->u
.watch
.vec
= split(body
, msg
->hdr
.len
,
864 &msg
->u
.watch
.vec_size
);
865 if (IS_ERR(msg
->u
.watch
.vec
)) {
866 err
= PTR_ERR(msg
->u
.watch
.vec
);
871 spin_lock(&watches_lock
);
872 msg
->u
.watch
.handle
= find_watch(
873 msg
->u
.watch
.vec
[XS_WATCH_TOKEN
]);
874 if (msg
->u
.watch
.handle
!= NULL
) {
875 spin_lock(&watch_events_lock
);
876 list_add_tail(&msg
->list
, &watch_events
);
877 wake_up(&watch_events_waitq
);
878 spin_unlock(&watch_events_lock
);
880 kfree(msg
->u
.watch
.vec
);
883 spin_unlock(&watches_lock
);
885 msg
->u
.reply
.body
= body
;
886 spin_lock(&xs_state
.reply_lock
);
887 list_add_tail(&msg
->list
, &xs_state
.reply_list
);
888 spin_unlock(&xs_state
.reply_lock
);
889 wake_up(&xs_state
.reply_waitq
);
893 mutex_unlock(&xs_state
.response_mutex
);
897 static int xenbus_thread(void *unused
)
904 printk(KERN_WARNING
"XENBUS error %d while reading "
906 if (kthread_should_stop())
916 struct task_struct
*task
;
918 INIT_LIST_HEAD(&xs_state
.reply_list
);
919 spin_lock_init(&xs_state
.reply_lock
);
920 init_waitqueue_head(&xs_state
.reply_waitq
);
922 mutex_init(&xs_state
.request_mutex
);
923 mutex_init(&xs_state
.response_mutex
);
924 mutex_init(&xs_state
.transaction_mutex
);
925 init_rwsem(&xs_state
.watch_mutex
);
926 atomic_set(&xs_state
.transaction_count
, 0);
927 init_waitqueue_head(&xs_state
.transaction_wq
);
929 /* Initialize the shared memory rings to talk to xenstored */
930 err
= xb_init_comms();
934 task
= kthread_run(xenwatch_thread
, NULL
, "xenwatch");
936 return PTR_ERR(task
);
937 xenwatch_pid
= task
->pid
;
939 task
= kthread_run(xenbus_thread
, NULL
, "xenbus");
941 return PTR_ERR(task
);
943 /* shutdown watches for kexec boot */