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 "xenbus_comms.h"
62 #include <xen/xenbus.h>
64 #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;
118 /* Read out any raw xenbus messages queued up. */
119 static ssize_t xenbus_file_read(struct file *filp,
121 size_t len, loff_t *ppos)
123 struct xenbus_file_priv *u = filp->private_data;
124 struct read_buffer *rb;
128 mutex_lock(&u->reply_mutex);
130 while (list_empty(&u->read_buffers)) {
131 mutex_unlock(&u->reply_mutex);
132 if (filp->f_flags & O_NONBLOCK)
135 ret = wait_event_interruptible(u->read_waitq,
136 !list_empty(&u->read_buffers));
139 mutex_lock(&u->reply_mutex);
142 rb = list_entry(u->read_buffers.next, struct read_buffer, list);
145 unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
147 ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
150 rb->cons += sz - ret;
158 /* Clear out buffer if it has been consumed */
159 if (rb->cons == rb->len) {
162 if (list_empty(&u->read_buffers))
164 rb = list_entry(u->read_buffers.next,
165 struct read_buffer, list);
172 mutex_unlock(&u->reply_mutex);
177 * Add a buffer to the queue. Caller must hold the appropriate lock
178 * if the queue is not local. (Commonly the caller will build up
179 * multiple queued buffers on a temporary local list, and then add it
180 * to the appropriate list under lock once all the buffers have een
181 * successfully allocated.)
183 static int queue_reply(struct list_head *queue, const void *data, size_t len)
185 struct read_buffer *rb;
189 if (len > XENSTORE_PAYLOAD_MAX)
192 rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
199 memcpy(rb->msg, data, len);
201 list_add_tail(&rb->list, queue);
206 * Free all the read_buffer s on a list.
207 * Caller must have sole reference to list.
209 static void queue_cleanup(struct list_head *list)
211 struct read_buffer *rb;
213 while (!list_empty(list)) {
214 rb = list_entry(list->next, struct read_buffer, list);
215 list_del(list->next);
220 struct watch_adapter {
221 struct list_head list;
222 struct xenbus_watch watch;
223 struct xenbus_file_priv *dev_data;
227 static void free_watch_adapter(struct watch_adapter *watch)
229 kfree(watch->watch.node);
234 static struct watch_adapter *alloc_watch_adapter(const char *path,
237 struct watch_adapter *watch;
239 watch = kzalloc(sizeof(*watch), GFP_KERNEL);
243 watch->watch.node = kstrdup(path, GFP_KERNEL);
244 if (watch->watch.node == NULL)
247 watch->token = kstrdup(token, GFP_KERNEL);
248 if (watch->token == NULL)
254 free_watch_adapter(watch);
260 static void watch_fired(struct xenbus_watch *watch,
264 struct watch_adapter *adap;
265 struct xsd_sockmsg hdr;
266 const char *path, *token;
267 int path_len, tok_len, body_len, data_len = 0;
269 LIST_HEAD(staging_q);
271 adap = container_of(watch, struct watch_adapter, watch);
273 path = vec[XS_WATCH_PATH];
276 path_len = strlen(path) + 1;
277 tok_len = strlen(token) + 1;
279 data_len = vec[len] - vec[2] + 1;
280 body_len = path_len + tok_len + data_len;
282 hdr.type = XS_WATCH_EVENT;
285 mutex_lock(&adap->dev_data->reply_mutex);
287 ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
289 ret = queue_reply(&staging_q, path, path_len);
291 ret = queue_reply(&staging_q, token, tok_len);
293 ret = queue_reply(&staging_q, vec[2], data_len);
296 /* success: pass reply list onto watcher */
297 list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
298 wake_up(&adap->dev_data->read_waitq);
300 queue_cleanup(&staging_q);
302 mutex_unlock(&adap->dev_data->reply_mutex);
305 static int xenbus_write_transaction(unsigned msg_type,
306 struct xenbus_file_priv *u)
310 struct xenbus_transaction_holder *trans = NULL;
311 LIST_HEAD(staging_q);
313 if (msg_type == XS_TRANSACTION_START) {
314 trans = kmalloc(sizeof(*trans), GFP_KERNEL);
320 list_for_each_entry(trans, &u->transactions, list)
321 if (trans->handle.id == u->u.msg.tx_id)
323 if (&trans->list == &u->transactions)
327 reply = xenbus_dev_request_and_reply(&u->u.msg);
329 if (msg_type == XS_TRANSACTION_START)
335 if (msg_type == XS_TRANSACTION_START) {
336 if (u->u.msg.type == XS_ERROR)
339 trans->handle.id = simple_strtoul(reply, NULL, 0);
340 list_add(&trans->list, &u->transactions);
342 } else if (u->u.msg.type == XS_TRANSACTION_END) {
343 list_del(&trans->list);
347 mutex_lock(&u->reply_mutex);
348 rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
350 rc = queue_reply(&staging_q, reply, u->u.msg.len);
352 list_splice_tail(&staging_q, &u->read_buffers);
353 wake_up(&u->read_waitq);
355 queue_cleanup(&staging_q);
357 mutex_unlock(&u->reply_mutex);
365 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
367 struct watch_adapter *watch, *tmp_watch;
370 LIST_HEAD(staging_q);
372 path = u->u.buffer + sizeof(u->u.msg);
373 token = memchr(path, 0, u->u.msg.len);
379 if (memchr(token, 0, u->u.msg.len - (token - path)) == NULL) {
384 if (msg_type == XS_WATCH) {
385 watch = alloc_watch_adapter(path, token);
391 watch->watch.callback = watch_fired;
394 err = register_xenbus_watch(&watch->watch);
396 free_watch_adapter(watch);
400 list_add(&watch->list, &u->watches);
402 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
403 if (!strcmp(watch->token, token) &&
404 !strcmp(watch->watch.node, path)) {
405 unregister_xenbus_watch(&watch->watch);
406 list_del(&watch->list);
407 free_watch_adapter(watch);
413 /* Success. Synthesize a reply to say all is OK. */
416 struct xsd_sockmsg hdr;
421 .len = sizeof(reply.body)
426 mutex_lock(&u->reply_mutex);
427 rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
428 wake_up(&u->read_waitq);
429 mutex_unlock(&u->reply_mutex);
436 static ssize_t xenbus_file_write(struct file *filp,
437 const char __user *ubuf,
438 size_t len, loff_t *ppos)
440 struct xenbus_file_priv *u = filp->private_data;
444 LIST_HEAD(staging_q);
447 * We're expecting usermode to be writing properly formed
448 * xenbus messages. If they write an incomplete message we
449 * buffer it up. Once it is complete, we act on it.
453 * Make sure concurrent writers can't stomp all over each
454 * other's messages and make a mess of our partial message
455 * buffer. We don't make any attemppt to stop multiple
456 * writers from making a mess of each other's incomplete
457 * messages; we're just trying to guarantee our own internal
458 * consistency and make sure that single writes are handled
461 mutex_lock(&u->msgbuffer_mutex);
463 /* Get this out of the way early to avoid confusion */
467 /* Can't write a xenbus message larger we can buffer */
468 if (len > sizeof(u->u.buffer) - u->len) {
469 /* On error, dump existing buffer */
475 ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
482 /* Deal with a partial copy. */
488 /* Return if we haven't got a full message yet */
489 if (u->len < sizeof(u->u.msg))
490 goto out; /* not even the header yet */
492 /* If we're expecting a message that's larger than we can
493 possibly send, dump what we have and return an error. */
494 if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
500 if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
501 goto out; /* incomplete data portion */
504 * OK, now we have a complete message. Do something with it.
507 msg_type = u->u.msg.type;
512 /* (Un)Ask for some path to be watched for changes */
513 ret = xenbus_write_watch(msg_type, u);
517 /* Send out a transaction */
518 ret = xenbus_write_transaction(msg_type, u);
524 /* Buffered message consumed */
528 mutex_unlock(&u->msgbuffer_mutex);
532 static int xenbus_file_open(struct inode *inode, struct file *filp)
534 struct xenbus_file_priv *u;
536 if (xen_store_evtchn == 0)
539 nonseekable_open(inode, filp);
541 u = kzalloc(sizeof(*u), GFP_KERNEL);
545 INIT_LIST_HEAD(&u->transactions);
546 INIT_LIST_HEAD(&u->watches);
547 INIT_LIST_HEAD(&u->read_buffers);
548 init_waitqueue_head(&u->read_waitq);
550 mutex_init(&u->reply_mutex);
551 mutex_init(&u->msgbuffer_mutex);
553 filp->private_data = u;
558 static int xenbus_file_release(struct inode *inode, struct file *filp)
560 struct xenbus_file_priv *u = filp->private_data;
561 struct xenbus_transaction_holder *trans, *tmp;
562 struct watch_adapter *watch, *tmp_watch;
563 struct read_buffer *rb, *tmp_rb;
566 * No need for locking here because there are no other users,
570 list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
571 xenbus_transaction_end(trans->handle, 1);
572 list_del(&trans->list);
576 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
577 unregister_xenbus_watch(&watch->watch);
578 list_del(&watch->list);
579 free_watch_adapter(watch);
582 list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
591 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
593 struct xenbus_file_priv *u = file->private_data;
595 poll_wait(file, &u->read_waitq, wait);
596 if (!list_empty(&u->read_buffers))
597 return POLLIN | POLLRDNORM;
601 const struct file_operations xen_xenbus_fops = {
602 .read = xenbus_file_read,
603 .write = xenbus_file_write,
604 .open = xenbus_file_open,
605 .release = xenbus_file_release,
606 .poll = xenbus_file_poll,
609 EXPORT_SYMBOL_GPL(xen_xenbus_fops);
611 static struct miscdevice xenbus_dev = {
612 .minor = MISC_DYNAMIC_MINOR,
613 .name = "xen/xenbus",
614 .fops = &xen_xenbus_fops,
617 static int __init xenbus_init(void)
624 err = misc_register(&xenbus_dev);
626 pr_err("Could not register xenbus frontend device\n");
629 device_initcall(xenbus_init);