2 * Driver giving user-space access to the kernel's xenbus connection
5 * Copyright (c) 2005, Christian Limpach
6 * Copyright (c) 2005, Rusty Russell, IBM Corporation
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License version 2
10 * as published by the Free Software Foundation; or, when distributed
11 * separately from the Linux kernel or incorporated into other
12 * software packages, subject to the following license:
14 * Permission is hereby granted, free of charge, to any person obtaining a copy
15 * of this source file (the "Software"), to deal in the Software without
16 * restriction, including without limitation the rights to use, copy, modify,
17 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
18 * and to permit persons to whom the Software is furnished to do so, subject to
19 * the following conditions:
21 * The above copyright notice and this permission notice shall be included in
22 * all copies or substantial portions of the Software.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
27 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
29 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
33 * 2008-10-07 Alex Zeffertt Replaced /proc/xen/xenbus with xenfs filesystem
34 * and /proc/xen compatibility mount point.
35 * Turned xenfs into a loadable module.
38 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40 #include <linux/kernel.h>
41 #include <linux/errno.h>
42 #include <linux/uio.h>
43 #include <linux/notifier.h>
44 #include <linux/wait.h>
46 #include <linux/poll.h>
47 #include <linux/mutex.h>
48 #include <linux/sched.h>
49 #include <linux/spinlock.h>
50 #include <linux/mount.h>
51 #include <linux/pagemap.h>
52 #include <linux/uaccess.h>
53 #include <linux/init.h>
54 #include <linux/namei.h>
55 #include <linux/string.h>
56 #include <linux/slab.h>
57 #include <linux/miscdevice.h>
58 #include <linux/init.h>
60 #include <xen/xenbus.h>
62 #include <asm/xen/hypervisor.h>
67 * An element of a list of outstanding transactions, for which we're
68 * still waiting a reply.
70 struct xenbus_transaction_holder
{
71 struct list_head list
;
72 struct xenbus_transaction handle
;
76 * A buffer of data on the queue.
79 struct list_head list
;
85 struct xenbus_file_priv
{
87 * msgbuffer_mutex is held while partial requests are built up
88 * and complete requests are acted on. It therefore protects
89 * the "transactions" and "watches" lists, and the partial
90 * request length and buffer.
92 * reply_mutex protects the reply being built up to return to
93 * usermode. It nests inside msgbuffer_mutex but may be held
94 * alone during a watch callback.
96 struct mutex msgbuffer_mutex
;
98 /* In-progress transactions */
99 struct list_head transactions
;
101 /* Active watches. */
102 struct list_head watches
;
104 /* Partial request. */
107 struct xsd_sockmsg msg
;
108 char buffer
[XENSTORE_PAYLOAD_MAX
];
111 /* Response queue. */
112 struct mutex reply_mutex
;
113 struct list_head read_buffers
;
114 wait_queue_head_t read_waitq
;
119 /* Read out any raw xenbus messages queued up. */
120 static ssize_t
xenbus_file_read(struct file
*filp
,
122 size_t len
, loff_t
*ppos
)
124 struct xenbus_file_priv
*u
= filp
->private_data
;
125 struct read_buffer
*rb
;
129 mutex_lock(&u
->reply_mutex
);
131 while (list_empty(&u
->read_buffers
)) {
132 mutex_unlock(&u
->reply_mutex
);
133 if (filp
->f_flags
& O_NONBLOCK
)
136 ret
= wait_event_interruptible(u
->read_waitq
,
137 !list_empty(&u
->read_buffers
));
140 mutex_lock(&u
->reply_mutex
);
143 rb
= list_entry(u
->read_buffers
.next
, struct read_buffer
, list
);
146 unsigned sz
= min((unsigned)len
- i
, rb
->len
- rb
->cons
);
148 ret
= copy_to_user(ubuf
+ i
, &rb
->msg
[rb
->cons
], sz
);
151 rb
->cons
+= sz
- ret
;
159 /* Clear out buffer if it has been consumed */
160 if (rb
->cons
== rb
->len
) {
163 if (list_empty(&u
->read_buffers
))
165 rb
= list_entry(u
->read_buffers
.next
,
166 struct read_buffer
, list
);
173 mutex_unlock(&u
->reply_mutex
);
178 * Add a buffer to the queue. Caller must hold the appropriate lock
179 * if the queue is not local. (Commonly the caller will build up
180 * multiple queued buffers on a temporary local list, and then add it
181 * to the appropriate list under lock once all the buffers have een
182 * successfully allocated.)
184 static int queue_reply(struct list_head
*queue
, const void *data
, size_t len
)
186 struct read_buffer
*rb
;
190 if (len
> XENSTORE_PAYLOAD_MAX
)
193 rb
= kmalloc(sizeof(*rb
) + len
, GFP_KERNEL
);
200 memcpy(rb
->msg
, data
, len
);
202 list_add_tail(&rb
->list
, queue
);
207 * Free all the read_buffer s on a list.
208 * Caller must have sole reference to list.
210 static void queue_cleanup(struct list_head
*list
)
212 struct read_buffer
*rb
;
214 while (!list_empty(list
)) {
215 rb
= list_entry(list
->next
, struct read_buffer
, list
);
216 list_del(list
->next
);
221 struct watch_adapter
{
222 struct list_head list
;
223 struct xenbus_watch watch
;
224 struct xenbus_file_priv
*dev_data
;
228 static void free_watch_adapter(struct watch_adapter
*watch
)
230 kfree(watch
->watch
.node
);
235 static struct watch_adapter
*alloc_watch_adapter(const char *path
,
238 struct watch_adapter
*watch
;
240 watch
= kzalloc(sizeof(*watch
), GFP_KERNEL
);
244 watch
->watch
.node
= kstrdup(path
, GFP_KERNEL
);
245 if (watch
->watch
.node
== NULL
)
248 watch
->token
= kstrdup(token
, GFP_KERNEL
);
249 if (watch
->token
== NULL
)
255 free_watch_adapter(watch
);
261 static void watch_fired(struct xenbus_watch
*watch
,
265 struct watch_adapter
*adap
;
266 struct xsd_sockmsg hdr
;
267 const char *token_caller
;
268 int path_len
, tok_len
, body_len
;
270 LIST_HEAD(staging_q
);
272 adap
= container_of(watch
, struct watch_adapter
, watch
);
274 token_caller
= adap
->token
;
276 path_len
= strlen(path
) + 1;
277 tok_len
= strlen(token_caller
) + 1;
278 body_len
= path_len
+ tok_len
;
280 hdr
.type
= XS_WATCH_EVENT
;
283 mutex_lock(&adap
->dev_data
->reply_mutex
);
285 ret
= queue_reply(&staging_q
, &hdr
, sizeof(hdr
));
287 ret
= queue_reply(&staging_q
, path
, path_len
);
289 ret
= queue_reply(&staging_q
, token_caller
, tok_len
);
292 /* success: pass reply list onto watcher */
293 list_splice_tail(&staging_q
, &adap
->dev_data
->read_buffers
);
294 wake_up(&adap
->dev_data
->read_waitq
);
296 queue_cleanup(&staging_q
);
298 mutex_unlock(&adap
->dev_data
->reply_mutex
);
301 static void xenbus_file_free(struct kref
*kref
)
303 struct xenbus_file_priv
*u
;
304 struct xenbus_transaction_holder
*trans
, *tmp
;
305 struct watch_adapter
*watch
, *tmp_watch
;
306 struct read_buffer
*rb
, *tmp_rb
;
308 u
= container_of(kref
, struct xenbus_file_priv
, kref
);
311 * No need for locking here because there are no other users,
315 list_for_each_entry_safe(trans
, tmp
, &u
->transactions
, list
) {
316 xenbus_transaction_end(trans
->handle
, 1);
317 list_del(&trans
->list
);
321 list_for_each_entry_safe(watch
, tmp_watch
, &u
->watches
, list
) {
322 unregister_xenbus_watch(&watch
->watch
);
323 list_del(&watch
->list
);
324 free_watch_adapter(watch
);
327 list_for_each_entry_safe(rb
, tmp_rb
, &u
->read_buffers
, list
) {
334 static struct xenbus_transaction_holder
*xenbus_get_transaction(
335 struct xenbus_file_priv
*u
, uint32_t tx_id
)
337 struct xenbus_transaction_holder
*trans
;
339 list_for_each_entry(trans
, &u
->transactions
, list
)
340 if (trans
->handle
.id
== tx_id
)
346 void xenbus_dev_queue_reply(struct xb_req_data
*req
)
348 struct xenbus_file_priv
*u
= req
->par
;
349 struct xenbus_transaction_holder
*trans
= NULL
;
351 LIST_HEAD(staging_q
);
353 xs_request_exit(req
);
355 mutex_lock(&u
->msgbuffer_mutex
);
357 if (req
->type
== XS_TRANSACTION_START
) {
358 trans
= xenbus_get_transaction(u
, 0);
361 if (req
->msg
.type
== XS_ERROR
) {
362 list_del(&trans
->list
);
365 rc
= kstrtou32(req
->body
, 10, &trans
->handle
.id
);
369 } else if (req
->msg
.type
== XS_TRANSACTION_END
) {
370 trans
= xenbus_get_transaction(u
, req
->msg
.tx_id
);
373 list_del(&trans
->list
);
377 mutex_unlock(&u
->msgbuffer_mutex
);
379 mutex_lock(&u
->reply_mutex
);
380 rc
= queue_reply(&staging_q
, &req
->msg
, sizeof(req
->msg
));
382 rc
= queue_reply(&staging_q
, req
->body
, req
->msg
.len
);
384 list_splice_tail(&staging_q
, &u
->read_buffers
);
385 wake_up(&u
->read_waitq
);
387 queue_cleanup(&staging_q
);
389 mutex_unlock(&u
->reply_mutex
);
394 kref_put(&u
->kref
, xenbus_file_free
);
399 mutex_unlock(&u
->msgbuffer_mutex
);
402 static int xenbus_command_reply(struct xenbus_file_priv
*u
,
403 unsigned int msg_type
, const char *reply
)
406 struct xsd_sockmsg hdr
;
412 msg
.hdr
.type
= msg_type
;
413 msg
.hdr
.len
= strlen(reply
) + 1;
414 if (msg
.hdr
.len
> sizeof(msg
.body
))
417 mutex_lock(&u
->reply_mutex
);
418 rc
= queue_reply(&u
->read_buffers
, &msg
, sizeof(msg
.hdr
) + msg
.hdr
.len
);
419 wake_up(&u
->read_waitq
);
420 mutex_unlock(&u
->reply_mutex
);
423 kref_put(&u
->kref
, xenbus_file_free
);
428 static int xenbus_write_transaction(unsigned msg_type
,
429 struct xenbus_file_priv
*u
)
432 struct xenbus_transaction_holder
*trans
= NULL
;
434 if (msg_type
== XS_TRANSACTION_START
) {
435 trans
= kzalloc(sizeof(*trans
), GFP_KERNEL
);
440 list_add(&trans
->list
, &u
->transactions
);
441 } else if (u
->u
.msg
.tx_id
!= 0 &&
442 !xenbus_get_transaction(u
, u
->u
.msg
.tx_id
))
443 return xenbus_command_reply(u
, XS_ERROR
, "ENOENT");
445 rc
= xenbus_dev_request_and_reply(&u
->u
.msg
, u
);
453 static int xenbus_write_watch(unsigned msg_type
, struct xenbus_file_priv
*u
)
455 struct watch_adapter
*watch
;
458 LIST_HEAD(staging_q
);
460 path
= u
->u
.buffer
+ sizeof(u
->u
.msg
);
461 token
= memchr(path
, 0, u
->u
.msg
.len
);
463 rc
= xenbus_command_reply(u
, XS_ERROR
, "EINVAL");
467 if (memchr(token
, 0, u
->u
.msg
.len
- (token
- path
)) == NULL
) {
468 rc
= xenbus_command_reply(u
, XS_ERROR
, "EINVAL");
472 if (msg_type
== XS_WATCH
) {
473 watch
= alloc_watch_adapter(path
, token
);
479 watch
->watch
.callback
= watch_fired
;
482 err
= register_xenbus_watch(&watch
->watch
);
484 free_watch_adapter(watch
);
488 list_add(&watch
->list
, &u
->watches
);
490 list_for_each_entry(watch
, &u
->watches
, list
) {
491 if (!strcmp(watch
->token
, token
) &&
492 !strcmp(watch
->watch
.node
, path
)) {
493 unregister_xenbus_watch(&watch
->watch
);
494 list_del(&watch
->list
);
495 free_watch_adapter(watch
);
501 /* Success. Synthesize a reply to say all is OK. */
502 rc
= xenbus_command_reply(u
, msg_type
, "OK");
508 static ssize_t
xenbus_file_write(struct file
*filp
,
509 const char __user
*ubuf
,
510 size_t len
, loff_t
*ppos
)
512 struct xenbus_file_priv
*u
= filp
->private_data
;
516 LIST_HEAD(staging_q
);
519 * We're expecting usermode to be writing properly formed
520 * xenbus messages. If they write an incomplete message we
521 * buffer it up. Once it is complete, we act on it.
525 * Make sure concurrent writers can't stomp all over each
526 * other's messages and make a mess of our partial message
527 * buffer. We don't make any attemppt to stop multiple
528 * writers from making a mess of each other's incomplete
529 * messages; we're just trying to guarantee our own internal
530 * consistency and make sure that single writes are handled
533 mutex_lock(&u
->msgbuffer_mutex
);
535 /* Get this out of the way early to avoid confusion */
539 /* Can't write a xenbus message larger we can buffer */
540 if (len
> sizeof(u
->u
.buffer
) - u
->len
) {
541 /* On error, dump existing buffer */
547 ret
= copy_from_user(u
->u
.buffer
+ u
->len
, ubuf
, len
);
554 /* Deal with a partial copy. */
560 /* Return if we haven't got a full message yet */
561 if (u
->len
< sizeof(u
->u
.msg
))
562 goto out
; /* not even the header yet */
564 /* If we're expecting a message that's larger than we can
565 possibly send, dump what we have and return an error. */
566 if ((sizeof(u
->u
.msg
) + u
->u
.msg
.len
) > sizeof(u
->u
.buffer
)) {
572 if (u
->len
< (sizeof(u
->u
.msg
) + u
->u
.msg
.len
))
573 goto out
; /* incomplete data portion */
576 * OK, now we have a complete message. Do something with it.
581 msg_type
= u
->u
.msg
.type
;
586 /* (Un)Ask for some path to be watched for changes */
587 ret
= xenbus_write_watch(msg_type
, u
);
591 /* Send out a transaction */
592 ret
= xenbus_write_transaction(msg_type
, u
);
597 kref_put(&u
->kref
, xenbus_file_free
);
600 /* Buffered message consumed */
604 mutex_unlock(&u
->msgbuffer_mutex
);
608 static int xenbus_file_open(struct inode
*inode
, struct file
*filp
)
610 struct xenbus_file_priv
*u
;
612 if (xen_store_evtchn
== 0)
615 nonseekable_open(inode
, filp
);
617 filp
->f_mode
&= ~FMODE_ATOMIC_POS
; /* cdev-style semantics */
619 u
= kzalloc(sizeof(*u
), GFP_KERNEL
);
625 INIT_LIST_HEAD(&u
->transactions
);
626 INIT_LIST_HEAD(&u
->watches
);
627 INIT_LIST_HEAD(&u
->read_buffers
);
628 init_waitqueue_head(&u
->read_waitq
);
630 mutex_init(&u
->reply_mutex
);
631 mutex_init(&u
->msgbuffer_mutex
);
633 filp
->private_data
= u
;
638 static int xenbus_file_release(struct inode
*inode
, struct file
*filp
)
640 struct xenbus_file_priv
*u
= filp
->private_data
;
642 kref_put(&u
->kref
, xenbus_file_free
);
647 static unsigned int xenbus_file_poll(struct file
*file
, poll_table
*wait
)
649 struct xenbus_file_priv
*u
= file
->private_data
;
651 poll_wait(file
, &u
->read_waitq
, wait
);
652 if (!list_empty(&u
->read_buffers
))
653 return POLLIN
| POLLRDNORM
;
657 const struct file_operations xen_xenbus_fops
= {
658 .read
= xenbus_file_read
,
659 .write
= xenbus_file_write
,
660 .open
= xenbus_file_open
,
661 .release
= xenbus_file_release
,
662 .poll
= xenbus_file_poll
,
665 EXPORT_SYMBOL_GPL(xen_xenbus_fops
);
667 static struct miscdevice xenbus_dev
= {
668 .minor
= MISC_DYNAMIC_MINOR
,
669 .name
= "xen/xenbus",
670 .fops
= &xen_xenbus_fops
,
673 static int __init
xenbus_init(void)
680 err
= misc_register(&xenbus_dev
);
682 pr_err("Could not register xenbus frontend device\n");
685 device_initcall(xenbus_init
);