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>
59 #include <xen/xenbus.h>
61 #include <asm/xen/hypervisor.h>
66 * An element of a list of outstanding transactions, for which we're
67 * still waiting a reply.
69 struct xenbus_transaction_holder {
70 struct list_head list;
71 struct xenbus_transaction handle;
75 * A buffer of data on the queue.
78 struct list_head list;
84 struct xenbus_file_priv {
86 * msgbuffer_mutex is held while partial requests are built up
87 * and complete requests are acted on. It therefore protects
88 * the "transactions" and "watches" lists, and the partial
89 * request length and buffer.
91 * reply_mutex protects the reply being built up to return to
92 * usermode. It nests inside msgbuffer_mutex but may be held
93 * alone during a watch callback.
95 struct mutex msgbuffer_mutex;
97 /* In-progress transactions */
98 struct list_head transactions;
100 /* Active watches. */
101 struct list_head watches;
103 /* Partial request. */
106 struct xsd_sockmsg msg;
107 char buffer[XENSTORE_PAYLOAD_MAX];
110 /* Response queue. */
111 struct mutex reply_mutex;
112 struct list_head read_buffers;
113 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 *token_caller;
267 int path_len, tok_len, body_len;
269 LIST_HEAD(staging_q);
271 adap = container_of(watch, struct watch_adapter, watch);
273 token_caller = adap->token;
275 path_len = strlen(path) + 1;
276 tok_len = strlen(token_caller) + 1;
277 body_len = path_len + tok_len;
279 hdr.type = XS_WATCH_EVENT;
282 mutex_lock(&adap->dev_data->reply_mutex);
284 ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
286 ret = queue_reply(&staging_q, path, path_len);
288 ret = queue_reply(&staging_q, token_caller, tok_len);
291 /* success: pass reply list onto watcher */
292 list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
293 wake_up(&adap->dev_data->read_waitq);
295 queue_cleanup(&staging_q);
297 mutex_unlock(&adap->dev_data->reply_mutex);
300 static void xenbus_file_free(struct kref *kref)
302 struct xenbus_file_priv *u;
303 struct xenbus_transaction_holder *trans, *tmp;
304 struct watch_adapter *watch, *tmp_watch;
305 struct read_buffer *rb, *tmp_rb;
307 u = container_of(kref, struct xenbus_file_priv, kref);
310 * No need for locking here because there are no other users,
314 list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
315 xenbus_transaction_end(trans->handle, 1);
316 list_del(&trans->list);
320 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
321 unregister_xenbus_watch(&watch->watch);
322 list_del(&watch->list);
323 free_watch_adapter(watch);
326 list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
333 static struct xenbus_transaction_holder *xenbus_get_transaction(
334 struct xenbus_file_priv *u, uint32_t tx_id)
336 struct xenbus_transaction_holder *trans;
338 list_for_each_entry(trans, &u->transactions, list)
339 if (trans->handle.id == tx_id)
345 void xenbus_dev_queue_reply(struct xb_req_data *req)
347 struct xenbus_file_priv *u = req->par;
348 struct xenbus_transaction_holder *trans = NULL;
350 LIST_HEAD(staging_q);
352 xs_request_exit(req);
354 mutex_lock(&u->msgbuffer_mutex);
356 if (req->type == XS_TRANSACTION_START) {
357 trans = xenbus_get_transaction(u, 0);
360 if (req->msg.type == XS_ERROR) {
361 list_del(&trans->list);
364 rc = kstrtou32(req->body, 10, &trans->handle.id);
368 } else if (req->type == XS_TRANSACTION_END) {
369 trans = xenbus_get_transaction(u, req->msg.tx_id);
372 list_del(&trans->list);
376 mutex_unlock(&u->msgbuffer_mutex);
378 mutex_lock(&u->reply_mutex);
379 rc = queue_reply(&staging_q, &req->msg, sizeof(req->msg));
381 rc = queue_reply(&staging_q, req->body, req->msg.len);
383 list_splice_tail(&staging_q, &u->read_buffers);
384 wake_up(&u->read_waitq);
386 queue_cleanup(&staging_q);
388 mutex_unlock(&u->reply_mutex);
393 kref_put(&u->kref, xenbus_file_free);
398 mutex_unlock(&u->msgbuffer_mutex);
401 static int xenbus_command_reply(struct xenbus_file_priv *u,
402 unsigned int msg_type, const char *reply)
405 struct xsd_sockmsg hdr;
411 msg.hdr.type = msg_type;
412 msg.hdr.len = strlen(reply) + 1;
413 if (msg.hdr.len > sizeof(msg.body))
415 memcpy(&msg.body, reply, msg.hdr.len);
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 struct xsd_sockmsg hdr;
436 } *msg = (void *)u->u.buffer;
438 if (msg_type == XS_TRANSACTION_START) {
439 trans = kzalloc(sizeof(*trans), GFP_KERNEL);
444 list_add(&trans->list, &u->transactions);
445 } else if (msg->hdr.tx_id != 0 &&
446 !xenbus_get_transaction(u, msg->hdr.tx_id))
447 return xenbus_command_reply(u, XS_ERROR, "ENOENT");
448 else if (msg_type == XS_TRANSACTION_END &&
449 !(msg->hdr.len == 2 &&
450 (!strcmp(msg->body, "T") || !strcmp(msg->body, "F"))))
451 return xenbus_command_reply(u, XS_ERROR, "EINVAL");
453 rc = xenbus_dev_request_and_reply(&msg->hdr, u);
455 list_del(&trans->list);
463 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
465 struct watch_adapter *watch;
468 LIST_HEAD(staging_q);
470 path = u->u.buffer + sizeof(u->u.msg);
471 token = memchr(path, 0, u->u.msg.len);
473 rc = xenbus_command_reply(u, XS_ERROR, "EINVAL");
477 if (memchr(token, 0, u->u.msg.len - (token - path)) == NULL) {
478 rc = xenbus_command_reply(u, XS_ERROR, "EINVAL");
482 if (msg_type == XS_WATCH) {
483 watch = alloc_watch_adapter(path, token);
489 watch->watch.callback = watch_fired;
492 err = register_xenbus_watch(&watch->watch);
494 free_watch_adapter(watch);
498 list_add(&watch->list, &u->watches);
500 list_for_each_entry(watch, &u->watches, list) {
501 if (!strcmp(watch->token, token) &&
502 !strcmp(watch->watch.node, path)) {
503 unregister_xenbus_watch(&watch->watch);
504 list_del(&watch->list);
505 free_watch_adapter(watch);
511 /* Success. Synthesize a reply to say all is OK. */
512 rc = xenbus_command_reply(u, msg_type, "OK");
518 static ssize_t xenbus_file_write(struct file *filp,
519 const char __user *ubuf,
520 size_t len, loff_t *ppos)
522 struct xenbus_file_priv *u = filp->private_data;
526 LIST_HEAD(staging_q);
529 * We're expecting usermode to be writing properly formed
530 * xenbus messages. If they write an incomplete message we
531 * buffer it up. Once it is complete, we act on it.
535 * Make sure concurrent writers can't stomp all over each
536 * other's messages and make a mess of our partial message
537 * buffer. We don't make any attemppt to stop multiple
538 * writers from making a mess of each other's incomplete
539 * messages; we're just trying to guarantee our own internal
540 * consistency and make sure that single writes are handled
543 mutex_lock(&u->msgbuffer_mutex);
545 /* Get this out of the way early to avoid confusion */
549 /* Can't write a xenbus message larger we can buffer */
550 if (len > sizeof(u->u.buffer) - u->len) {
551 /* On error, dump existing buffer */
557 ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
564 /* Deal with a partial copy. */
570 /* Return if we haven't got a full message yet */
571 if (u->len < sizeof(u->u.msg))
572 goto out; /* not even the header yet */
574 /* If we're expecting a message that's larger than we can
575 possibly send, dump what we have and return an error. */
576 if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
582 if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
583 goto out; /* incomplete data portion */
586 * OK, now we have a complete message. Do something with it.
591 msg_type = u->u.msg.type;
596 /* (Un)Ask for some path to be watched for changes */
597 ret = xenbus_write_watch(msg_type, u);
601 /* Send out a transaction */
602 ret = xenbus_write_transaction(msg_type, u);
607 kref_put(&u->kref, xenbus_file_free);
610 /* Buffered message consumed */
614 mutex_unlock(&u->msgbuffer_mutex);
618 static int xenbus_file_open(struct inode *inode, struct file *filp)
620 struct xenbus_file_priv *u;
622 if (xen_store_evtchn == 0)
625 nonseekable_open(inode, filp);
627 filp->f_mode &= ~FMODE_ATOMIC_POS; /* cdev-style semantics */
629 u = kzalloc(sizeof(*u), GFP_KERNEL);
635 INIT_LIST_HEAD(&u->transactions);
636 INIT_LIST_HEAD(&u->watches);
637 INIT_LIST_HEAD(&u->read_buffers);
638 init_waitqueue_head(&u->read_waitq);
640 mutex_init(&u->reply_mutex);
641 mutex_init(&u->msgbuffer_mutex);
643 filp->private_data = u;
648 static int xenbus_file_release(struct inode *inode, struct file *filp)
650 struct xenbus_file_priv *u = filp->private_data;
652 kref_put(&u->kref, xenbus_file_free);
657 static __poll_t xenbus_file_poll(struct file *file, poll_table *wait)
659 struct xenbus_file_priv *u = file->private_data;
661 poll_wait(file, &u->read_waitq, wait);
662 if (!list_empty(&u->read_buffers))
663 return EPOLLIN | EPOLLRDNORM;
667 const struct file_operations xen_xenbus_fops = {
668 .read = xenbus_file_read,
669 .write = xenbus_file_write,
670 .open = xenbus_file_open,
671 .release = xenbus_file_release,
672 .poll = xenbus_file_poll,
675 EXPORT_SYMBOL_GPL(xen_xenbus_fops);
677 static struct miscdevice xenbus_dev = {
678 .minor = MISC_DYNAMIC_MINOR,
679 .name = "xen/xenbus",
680 .fops = &xen_xenbus_fops,
683 static int __init xenbus_init(void)
690 err = misc_register(&xenbus_dev);
692 pr_err("Could not register xenbus frontend device\n");
695 device_initcall(xenbus_init);