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);
124 while (list_empty(&u->read_buffers)) {
125 mutex_unlock(&u->reply_mutex);
126 if (filp->f_flags & O_NONBLOCK)
129 ret = wait_event_interruptible(u->read_waitq,
130 !list_empty(&u->read_buffers));
133 mutex_lock(&u->reply_mutex);
136 rb = list_entry(u->read_buffers.next, struct read_buffer, list);
139 unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
141 ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
144 rb->cons += sz - ret;
152 /* Clear out buffer if it has been consumed */
153 if (rb->cons == rb->len) {
156 if (list_empty(&u->read_buffers))
158 rb = list_entry(u->read_buffers.next,
159 struct read_buffer, list);
166 mutex_unlock(&u->reply_mutex);
171 * Add a buffer to the queue. Caller must hold the appropriate lock
172 * if the queue is not local. (Commonly the caller will build up
173 * multiple queued buffers on a temporary local list, and then add it
174 * to the appropriate list under lock once all the buffers have een
175 * successfully allocated.)
177 static int queue_reply(struct list_head *queue, const void *data, size_t len)
179 struct read_buffer *rb;
184 rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
191 memcpy(rb->msg, data, len);
193 list_add_tail(&rb->list, queue);
198 * Free all the read_buffer s on a list.
199 * Caller must have sole reference to list.
201 static void queue_cleanup(struct list_head *list)
203 struct read_buffer *rb;
205 while (!list_empty(list)) {
206 rb = list_entry(list->next, struct read_buffer, list);
207 list_del(list->next);
212 struct watch_adapter {
213 struct list_head list;
214 struct xenbus_watch watch;
215 struct xenbus_file_priv *dev_data;
219 static void free_watch_adapter(struct watch_adapter *watch)
221 kfree(watch->watch.node);
226 static struct watch_adapter *alloc_watch_adapter(const char *path,
229 struct watch_adapter *watch;
231 watch = kzalloc(sizeof(*watch), GFP_KERNEL);
235 watch->watch.node = kstrdup(path, GFP_KERNEL);
236 if (watch->watch.node == NULL)
239 watch->token = kstrdup(token, GFP_KERNEL);
240 if (watch->token == NULL)
246 free_watch_adapter(watch);
252 static void watch_fired(struct xenbus_watch *watch,
256 struct watch_adapter *adap;
257 struct xsd_sockmsg hdr;
258 const char *path, *token;
259 int path_len, tok_len, body_len, data_len = 0;
261 LIST_HEAD(staging_q);
263 adap = container_of(watch, struct watch_adapter, watch);
265 path = vec[XS_WATCH_PATH];
268 path_len = strlen(path) + 1;
269 tok_len = strlen(token) + 1;
271 data_len = vec[len] - vec[2] + 1;
272 body_len = path_len + tok_len + data_len;
274 hdr.type = XS_WATCH_EVENT;
277 mutex_lock(&adap->dev_data->reply_mutex);
279 ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
281 ret = queue_reply(&staging_q, path, path_len);
283 ret = queue_reply(&staging_q, token, tok_len);
285 ret = queue_reply(&staging_q, vec[2], data_len);
288 /* success: pass reply list onto watcher */
289 list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
290 wake_up(&adap->dev_data->read_waitq);
292 queue_cleanup(&staging_q);
294 mutex_unlock(&adap->dev_data->reply_mutex);
297 static int xenbus_write_transaction(unsigned msg_type,
298 struct xenbus_file_priv *u)
302 struct xenbus_transaction_holder *trans = NULL;
303 LIST_HEAD(staging_q);
305 if (msg_type == XS_TRANSACTION_START) {
306 trans = kmalloc(sizeof(*trans), GFP_KERNEL);
313 reply = xenbus_dev_request_and_reply(&u->u.msg);
320 if (msg_type == XS_TRANSACTION_START) {
321 trans->handle.id = simple_strtoul(reply, NULL, 0);
323 list_add(&trans->list, &u->transactions);
324 } else if (msg_type == XS_TRANSACTION_END) {
325 list_for_each_entry(trans, &u->transactions, list)
326 if (trans->handle.id == u->u.msg.tx_id)
328 BUG_ON(&trans->list == &u->transactions);
329 list_del(&trans->list);
334 mutex_lock(&u->reply_mutex);
335 rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
337 rc = queue_reply(&staging_q, reply, u->u.msg.len);
339 list_splice_tail(&staging_q, &u->read_buffers);
340 wake_up(&u->read_waitq);
342 queue_cleanup(&staging_q);
344 mutex_unlock(&u->reply_mutex);
352 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
354 struct watch_adapter *watch, *tmp_watch;
357 LIST_HEAD(staging_q);
359 path = u->u.buffer + sizeof(u->u.msg);
360 token = memchr(path, 0, u->u.msg.len);
367 if (msg_type == XS_WATCH) {
368 watch = alloc_watch_adapter(path, token);
374 watch->watch.callback = watch_fired;
377 err = register_xenbus_watch(&watch->watch);
379 free_watch_adapter(watch);
383 list_add(&watch->list, &u->watches);
385 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
386 if (!strcmp(watch->token, token) &&
387 !strcmp(watch->watch.node, path)) {
388 unregister_xenbus_watch(&watch->watch);
389 list_del(&watch->list);
390 free_watch_adapter(watch);
396 /* Success. Synthesize a reply to say all is OK. */
399 struct xsd_sockmsg hdr;
404 .len = sizeof(reply.body)
409 mutex_lock(&u->reply_mutex);
410 rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
411 wake_up(&u->read_waitq);
412 mutex_unlock(&u->reply_mutex);
419 static ssize_t xenbus_file_write(struct file *filp,
420 const char __user *ubuf,
421 size_t len, loff_t *ppos)
423 struct xenbus_file_priv *u = filp->private_data;
427 LIST_HEAD(staging_q);
430 * We're expecting usermode to be writing properly formed
431 * xenbus messages. If they write an incomplete message we
432 * buffer it up. Once it is complete, we act on it.
436 * Make sure concurrent writers can't stomp all over each
437 * other's messages and make a mess of our partial message
438 * buffer. We don't make any attemppt to stop multiple
439 * writers from making a mess of each other's incomplete
440 * messages; we're just trying to guarantee our own internal
441 * consistency and make sure that single writes are handled
444 mutex_lock(&u->msgbuffer_mutex);
446 /* Get this out of the way early to avoid confusion */
450 /* Can't write a xenbus message larger we can buffer */
451 if ((len + u->len) > sizeof(u->u.buffer)) {
452 /* On error, dump existing buffer */
458 ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
465 /* Deal with a partial copy. */
471 /* Return if we haven't got a full message yet */
472 if (u->len < sizeof(u->u.msg))
473 goto out; /* not even the header yet */
475 /* If we're expecting a message that's larger than we can
476 possibly send, dump what we have and return an error. */
477 if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
483 if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
484 goto out; /* incomplete data portion */
487 * OK, now we have a complete message. Do something with it.
490 msg_type = u->u.msg.type;
495 /* (Un)Ask for some path to be watched for changes */
496 ret = xenbus_write_watch(msg_type, u);
500 /* Send out a transaction */
501 ret = xenbus_write_transaction(msg_type, u);
507 /* Buffered message consumed */
511 mutex_unlock(&u->msgbuffer_mutex);
515 static int xenbus_file_open(struct inode *inode, struct file *filp)
517 struct xenbus_file_priv *u;
519 if (xen_store_evtchn == 0)
522 nonseekable_open(inode, filp);
524 u = kzalloc(sizeof(*u), GFP_KERNEL);
528 INIT_LIST_HEAD(&u->transactions);
529 INIT_LIST_HEAD(&u->watches);
530 INIT_LIST_HEAD(&u->read_buffers);
531 init_waitqueue_head(&u->read_waitq);
533 mutex_init(&u->reply_mutex);
534 mutex_init(&u->msgbuffer_mutex);
536 filp->private_data = u;
541 static int xenbus_file_release(struct inode *inode, struct file *filp)
543 struct xenbus_file_priv *u = filp->private_data;
544 struct xenbus_transaction_holder *trans, *tmp;
545 struct watch_adapter *watch, *tmp_watch;
546 struct read_buffer *rb, *tmp_rb;
549 * No need for locking here because there are no other users,
553 list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
554 xenbus_transaction_end(trans->handle, 1);
555 list_del(&trans->list);
559 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
560 unregister_xenbus_watch(&watch->watch);
561 list_del(&watch->list);
562 free_watch_adapter(watch);
565 list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
574 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
576 struct xenbus_file_priv *u = file->private_data;
578 poll_wait(file, &u->read_waitq, wait);
579 if (!list_empty(&u->read_buffers))
580 return POLLIN | POLLRDNORM;
584 const struct file_operations xenbus_file_ops = {
585 .read = xenbus_file_read,
586 .write = xenbus_file_write,
587 .open = xenbus_file_open,
588 .release = xenbus_file_release,
589 .poll = xenbus_file_poll,