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 #include <linux/kernel.h>
39 #include <linux/errno.h>
40 #include <linux/uio.h>
41 #include <linux/notifier.h>
42 #include <linux/wait.h>
44 #include <linux/poll.h>
45 #include <linux/mutex.h>
46 #include <linux/spinlock.h>
47 #include <linux/mount.h>
48 #include <linux/pagemap.h>
49 #include <linux/uaccess.h>
50 #include <linux/init.h>
51 #include <linux/namei.h>
52 #include <linux/string.h>
55 #include "../xenbus/xenbus_comms.h"
57 #include <xen/xenbus.h>
58 #include <asm/xen/hypervisor.h>
61 * An element of a list of outstanding transactions, for which we're
62 * still waiting a reply.
64 struct xenbus_transaction_holder {
65 struct list_head list;
66 struct xenbus_transaction handle;
70 * A buffer of data on the queue.
73 struct list_head list;
79 struct xenbus_file_priv {
81 * msgbuffer_mutex is held while partial requests are built up
82 * and complete requests are acted on. It therefore protects
83 * the "transactions" and "watches" lists, and the partial
84 * request length and buffer.
86 * reply_mutex protects the reply being built up to return to
87 * usermode. It nests inside msgbuffer_mutex but may be held
88 * alone during a watch callback.
90 struct mutex msgbuffer_mutex;
92 /* In-progress transactions */
93 struct list_head transactions;
96 struct list_head watches;
98 /* Partial request. */
101 struct xsd_sockmsg msg;
102 char buffer[PAGE_SIZE];
105 /* Response queue. */
106 struct mutex reply_mutex;
107 struct list_head read_buffers;
108 wait_queue_head_t read_waitq;
112 /* Read out any raw xenbus messages queued up. */
113 static ssize_t xenbus_file_read(struct file *filp,
115 size_t len, loff_t *ppos)
117 struct xenbus_file_priv *u = filp->private_data;
118 struct read_buffer *rb;
122 mutex_lock(&u->reply_mutex);
123 while (list_empty(&u->read_buffers)) {
124 mutex_unlock(&u->reply_mutex);
125 if (filp->f_flags & O_NONBLOCK)
128 ret = wait_event_interruptible(u->read_waitq,
129 !list_empty(&u->read_buffers));
132 mutex_lock(&u->reply_mutex);
135 rb = list_entry(u->read_buffers.next, struct read_buffer, list);
138 unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
140 ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
143 rb->cons += sz - ret;
151 /* Clear out buffer if it has been consumed */
152 if (rb->cons == rb->len) {
155 if (list_empty(&u->read_buffers))
157 rb = list_entry(u->read_buffers.next,
158 struct read_buffer, list);
163 mutex_unlock(&u->reply_mutex);
168 * Add a buffer to the queue. Caller must hold the appropriate lock
169 * if the queue is not local. (Commonly the caller will build up
170 * multiple queued buffers on a temporary local list, and then add it
171 * to the appropriate list under lock once all the buffers have een
172 * successfully allocated.)
174 static int queue_reply(struct list_head *queue, const void *data, size_t len)
176 struct read_buffer *rb;
181 rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
188 memcpy(rb->msg, data, len);
190 list_add_tail(&rb->list, queue);
195 * Free all the read_buffer s on a list.
196 * Caller must have sole reference to list.
198 static void queue_cleanup(struct list_head *list)
200 struct read_buffer *rb;
202 while (!list_empty(list)) {
203 rb = list_entry(list->next, struct read_buffer, list);
204 list_del(list->next);
209 struct watch_adapter {
210 struct list_head list;
211 struct xenbus_watch watch;
212 struct xenbus_file_priv *dev_data;
216 static void free_watch_adapter(struct watch_adapter *watch)
218 kfree(watch->watch.node);
223 static struct watch_adapter *alloc_watch_adapter(const char *path,
226 struct watch_adapter *watch;
228 watch = kzalloc(sizeof(*watch), GFP_KERNEL);
232 watch->watch.node = kstrdup(path, GFP_KERNEL);
233 if (watch->watch.node == NULL)
236 watch->token = kstrdup(token, GFP_KERNEL);
237 if (watch->token == NULL)
243 free_watch_adapter(watch);
249 static void watch_fired(struct xenbus_watch *watch,
253 struct watch_adapter *adap;
254 struct xsd_sockmsg hdr;
255 const char *path, *token;
256 int path_len, tok_len, body_len, data_len = 0;
258 LIST_HEAD(staging_q);
260 adap = container_of(watch, struct watch_adapter, watch);
262 path = vec[XS_WATCH_PATH];
265 path_len = strlen(path) + 1;
266 tok_len = strlen(token) + 1;
268 data_len = vec[len] - vec[2] + 1;
269 body_len = path_len + tok_len + data_len;
271 hdr.type = XS_WATCH_EVENT;
274 mutex_lock(&adap->dev_data->reply_mutex);
276 ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
278 ret = queue_reply(&staging_q, path, path_len);
280 ret = queue_reply(&staging_q, token, tok_len);
282 ret = queue_reply(&staging_q, vec[2], data_len);
285 /* success: pass reply list onto watcher */
286 list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
287 wake_up(&adap->dev_data->read_waitq);
289 queue_cleanup(&staging_q);
291 mutex_unlock(&adap->dev_data->reply_mutex);
294 static int xenbus_write_transaction(unsigned msg_type,
295 struct xenbus_file_priv *u)
299 struct xenbus_transaction_holder *trans = NULL;
300 LIST_HEAD(staging_q);
302 if (msg_type == XS_TRANSACTION_START) {
303 trans = kmalloc(sizeof(*trans), GFP_KERNEL);
310 reply = xenbus_dev_request_and_reply(&u->u.msg);
317 if (msg_type == XS_TRANSACTION_START) {
318 trans->handle.id = simple_strtoul(reply, NULL, 0);
320 list_add(&trans->list, &u->transactions);
321 } else if (msg_type == XS_TRANSACTION_END) {
322 list_for_each_entry(trans, &u->transactions, list)
323 if (trans->handle.id == u->u.msg.tx_id)
325 BUG_ON(&trans->list == &u->transactions);
326 list_del(&trans->list);
331 mutex_lock(&u->reply_mutex);
332 rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
334 rc = queue_reply(&staging_q, reply, u->u.msg.len);
336 list_splice_tail(&staging_q, &u->read_buffers);
337 wake_up(&u->read_waitq);
339 queue_cleanup(&staging_q);
341 mutex_unlock(&u->reply_mutex);
349 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
351 struct watch_adapter *watch, *tmp_watch;
354 LIST_HEAD(staging_q);
356 path = u->u.buffer + sizeof(u->u.msg);
357 token = memchr(path, 0, u->u.msg.len);
364 if (msg_type == XS_WATCH) {
365 watch = alloc_watch_adapter(path, token);
371 watch->watch.callback = watch_fired;
374 err = register_xenbus_watch(&watch->watch);
376 free_watch_adapter(watch);
380 list_add(&watch->list, &u->watches);
382 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
383 if (!strcmp(watch->token, token) &&
384 !strcmp(watch->watch.node, path)) {
385 unregister_xenbus_watch(&watch->watch);
386 list_del(&watch->list);
387 free_watch_adapter(watch);
393 /* Success. Synthesize a reply to say all is OK. */
396 struct xsd_sockmsg hdr;
401 .len = sizeof(reply.body)
406 mutex_lock(&u->reply_mutex);
407 rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
408 wake_up(&u->read_waitq);
409 mutex_unlock(&u->reply_mutex);
416 static ssize_t xenbus_file_write(struct file *filp,
417 const char __user *ubuf,
418 size_t len, loff_t *ppos)
420 struct xenbus_file_priv *u = filp->private_data;
424 LIST_HEAD(staging_q);
427 * We're expecting usermode to be writing properly formed
428 * xenbus messages. If they write an incomplete message we
429 * buffer it up. Once it is complete, we act on it.
433 * Make sure concurrent writers can't stomp all over each
434 * other's messages and make a mess of our partial message
435 * buffer. We don't make any attemppt to stop multiple
436 * writers from making a mess of each other's incomplete
437 * messages; we're just trying to guarantee our own internal
438 * consistency and make sure that single writes are handled
441 mutex_lock(&u->msgbuffer_mutex);
443 /* Get this out of the way early to avoid confusion */
447 /* Can't write a xenbus message larger we can buffer */
448 if ((len + u->len) > sizeof(u->u.buffer)) {
449 /* On error, dump existing buffer */
455 ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
462 /* Deal with a partial copy. */
468 /* Return if we haven't got a full message yet */
469 if (u->len < sizeof(u->u.msg))
470 goto out; /* not even the header yet */
472 /* If we're expecting a message that's larger than we can
473 possibly send, dump what we have and return an error. */
474 if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
480 if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
481 goto out; /* incomplete data portion */
484 * OK, now we have a complete message. Do something with it.
487 msg_type = u->u.msg.type;
492 /* (Un)Ask for some path to be watched for changes */
493 ret = xenbus_write_watch(msg_type, u);
497 /* Send out a transaction */
498 ret = xenbus_write_transaction(msg_type, u);
504 /* Buffered message consumed */
508 mutex_unlock(&u->msgbuffer_mutex);
512 static int xenbus_file_open(struct inode *inode, struct file *filp)
514 struct xenbus_file_priv *u;
516 if (xen_store_evtchn == 0)
519 nonseekable_open(inode, filp);
521 u = kzalloc(sizeof(*u), GFP_KERNEL);
525 INIT_LIST_HEAD(&u->transactions);
526 INIT_LIST_HEAD(&u->watches);
527 INIT_LIST_HEAD(&u->read_buffers);
528 init_waitqueue_head(&u->read_waitq);
530 mutex_init(&u->reply_mutex);
531 mutex_init(&u->msgbuffer_mutex);
533 filp->private_data = u;
538 static int xenbus_file_release(struct inode *inode, struct file *filp)
540 struct xenbus_file_priv *u = filp->private_data;
541 struct xenbus_transaction_holder *trans, *tmp;
542 struct watch_adapter *watch, *tmp_watch;
545 * No need for locking here because there are no other users,
549 list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
550 xenbus_transaction_end(trans->handle, 1);
551 list_del(&trans->list);
555 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
556 unregister_xenbus_watch(&watch->watch);
557 list_del(&watch->list);
558 free_watch_adapter(watch);
566 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
568 struct xenbus_file_priv *u = file->private_data;
570 poll_wait(file, &u->read_waitq, wait);
571 if (!list_empty(&u->read_buffers))
572 return POLLIN | POLLRDNORM;
576 const struct file_operations xenbus_file_ops = {
577 .read = xenbus_file_read,
578 .write = xenbus_file_write,
579 .open = xenbus_file_open,
580 .release = xenbus_file_release,
581 .poll = xenbus_file_poll,