1/*
2 * Driver giving user-space access to the kernel's xenbus connection
3 * to xenstore.
4 *
5 * Copyright (c) 2005, Christian Limpach
6 * Copyright (c) 2005, Rusty Russell, IBM Corporation
7 *
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:
13 *
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:
20 *
21 * The above copyright notice and this permission notice shall be included in
22 * all copies or substantial portions of the Software.
23 *
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
30 * IN THE SOFTWARE.
31 *
32 * Changes:
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.
36 */
37
38#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39
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>
45#include <linux/fs.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/module.h>
59
60#include "xenbus_comms.h"
61
62#include <xen/xenbus.h>
63#include <xen/xen.h>
64#include <asm/xen/hypervisor.h>
65
66MODULE_LICENSE("GPL");
67
68/*
69 * An element of a list of outstanding transactions, for which we're
70 * still waiting a reply.
71 */
72struct xenbus_transaction_holder {
73	struct list_head list;
74	struct xenbus_transaction handle;
75};
76
77/*
78 * A buffer of data on the queue.
79 */
80struct read_buffer {
81	struct list_head list;
82	unsigned int cons;
83	unsigned int len;
84	char msg[];
85};
86
87struct xenbus_file_priv {
88	/*
89	 * msgbuffer_mutex is held while partial requests are built up
90	 * and complete requests are acted on.  It therefore protects
91	 * the "transactions" and "watches" lists, and the partial
92	 * request length and buffer.
93	 *
94	 * reply_mutex protects the reply being built up to return to
95	 * usermode.  It nests inside msgbuffer_mutex but may be held
96	 * alone during a watch callback.
97	 */
98	struct mutex msgbuffer_mutex;
99
100	/* In-progress transactions */
101	struct list_head transactions;
102
103	/* Active watches. */
104	struct list_head watches;
105
106	/* Partial request. */
107	unsigned int len;
108	union {
109		struct xsd_sockmsg msg;
110		char buffer[XENSTORE_PAYLOAD_MAX];
111	} u;
112
113	/* Response queue. */
114	struct mutex reply_mutex;
115	struct list_head read_buffers;
116	wait_queue_head_t read_waitq;
117
118};
119
120/* Read out any raw xenbus messages queued up. */
121static ssize_t xenbus_file_read(struct file *filp,
122			       char __user *ubuf,
123			       size_t len, loff_t *ppos)
124{
125	struct xenbus_file_priv *u = filp->private_data;
126	struct read_buffer *rb;
127	unsigned i;
128	int ret;
129
130	mutex_lock(&u->reply_mutex);
131again:
132	while (list_empty(&u->read_buffers)) {
133		mutex_unlock(&u->reply_mutex);
134		if (filp->f_flags & O_NONBLOCK)
135			return -EAGAIN;
136
137		ret = wait_event_interruptible(u->read_waitq,
138					       !list_empty(&u->read_buffers));
139		if (ret)
140			return ret;
141		mutex_lock(&u->reply_mutex);
142	}
143
144	rb = list_entry(u->read_buffers.next, struct read_buffer, list);
145	i = 0;
146	while (i < len) {
147		unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
148
149		ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
150
151		i += sz - ret;
152		rb->cons += sz - ret;
153
154		if (ret != 0) {
155			if (i == 0)
156				i = -EFAULT;
157			goto out;
158		}
159
160		/* Clear out buffer if it has been consumed */
161		if (rb->cons == rb->len) {
162			list_del(&rb->list);
163			kfree(rb);
164			if (list_empty(&u->read_buffers))
165				break;
166			rb = list_entry(u->read_buffers.next,
167					struct read_buffer, list);
168		}
169	}
170	if (i == 0)
171		goto again;
172
173out:
174	mutex_unlock(&u->reply_mutex);
175	return i;
176}
177
178/*
179 * Add a buffer to the queue.  Caller must hold the appropriate lock
180 * if the queue is not local.  (Commonly the caller will build up
181 * multiple queued buffers on a temporary local list, and then add it
182 * to the appropriate list under lock once all the buffers have een
183 * successfully allocated.)
184 */
185static int queue_reply(struct list_head *queue, const void *data, size_t len)
186{
187	struct read_buffer *rb;
188
189	if (len == 0)
190		return 0;
191
192	rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
193	if (rb == NULL)
194		return -ENOMEM;
195
196	rb->cons = 0;
197	rb->len = len;
198
199	memcpy(rb->msg, data, len);
200
201	list_add_tail(&rb->list, queue);
202	return 0;
203}
204
205/*
206 * Free all the read_buffer s on a list.
207 * Caller must have sole reference to list.
208 */
209static void queue_cleanup(struct list_head *list)
210{
211	struct read_buffer *rb;
212
213	while (!list_empty(list)) {
214		rb = list_entry(list->next, struct read_buffer, list);
215		list_del(list->next);
216		kfree(rb);
217	}
218}
219
220struct watch_adapter {
221	struct list_head list;
222	struct xenbus_watch watch;
223	struct xenbus_file_priv *dev_data;
224	char *token;
225};
226
227static void free_watch_adapter(struct watch_adapter *watch)
228{
229	kfree(watch->watch.node);
230	kfree(watch->token);
231	kfree(watch);
232}
233
234static struct watch_adapter *alloc_watch_adapter(const char *path,
235						 const char *token)
236{
237	struct watch_adapter *watch;
238
239	watch = kzalloc(sizeof(*watch), GFP_KERNEL);
240	if (watch == NULL)
241		goto out_fail;
242
243	watch->watch.node = kstrdup(path, GFP_KERNEL);
244	if (watch->watch.node == NULL)
245		goto out_free;
246
247	watch->token = kstrdup(token, GFP_KERNEL);
248	if (watch->token == NULL)
249		goto out_free;
250
251	return watch;
252
253out_free:
254	free_watch_adapter(watch);
255
256out_fail:
257	return NULL;
258}
259
260static void watch_fired(struct xenbus_watch *watch,
261			const char **vec,
262			unsigned int len)
263{
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;
268	int ret;
269	LIST_HEAD(staging_q);
270
271	adap = container_of(watch, struct watch_adapter, watch);
272
273	path = vec[XS_WATCH_PATH];
274	token = adap->token;
275
276	path_len = strlen(path) + 1;
277	tok_len = strlen(token) + 1;
278	if (len > 2)
279		data_len = vec[len] - vec[2] + 1;
280	body_len = path_len + tok_len + data_len;
281
282	hdr.type = XS_WATCH_EVENT;
283	hdr.len = body_len;
284
285	mutex_lock(&adap->dev_data->reply_mutex);
286
287	ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
288	if (!ret)
289		ret = queue_reply(&staging_q, path, path_len);
290	if (!ret)
291		ret = queue_reply(&staging_q, token, tok_len);
292	if (!ret && len > 2)
293		ret = queue_reply(&staging_q, vec[2], data_len);
294
295	if (!ret) {
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);
299	} else
300		queue_cleanup(&staging_q);
301
302	mutex_unlock(&adap->dev_data->reply_mutex);
303}
304
305static int xenbus_write_transaction(unsigned msg_type,
306				    struct xenbus_file_priv *u)
307{
308	int rc;
309	void *reply;
310	struct xenbus_transaction_holder *trans = NULL;
311	LIST_HEAD(staging_q);
312
313	if (msg_type == XS_TRANSACTION_START) {
314		trans = kmalloc(sizeof(*trans), GFP_KERNEL);
315		if (!trans) {
316			rc = -ENOMEM;
317			goto out;
318		}
319	}
320
321	reply = xenbus_dev_request_and_reply(&u->u.msg);
322	if (IS_ERR(reply)) {
323		kfree(trans);
324		rc = PTR_ERR(reply);
325		goto out;
326	}
327
328	if (msg_type == XS_TRANSACTION_START) {
329		if (u->u.msg.type == XS_ERROR)
330			kfree(trans);
331		else {
332			trans->handle.id = simple_strtoul(reply, NULL, 0);
333			list_add(&trans->list, &u->transactions);
334		}
335	} else if (u->u.msg.type == XS_TRANSACTION_END) {
336		list_for_each_entry(trans, &u->transactions, list)
337			if (trans->handle.id == u->u.msg.tx_id)
338				break;
339		BUG_ON(&trans->list == &u->transactions);
340		list_del(&trans->list);
341
342		kfree(trans);
343	}
344
345	mutex_lock(&u->reply_mutex);
346	rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
347	if (!rc)
348		rc = queue_reply(&staging_q, reply, u->u.msg.len);
349	if (!rc) {
350		list_splice_tail(&staging_q, &u->read_buffers);
351		wake_up(&u->read_waitq);
352	} else {
353		queue_cleanup(&staging_q);
354	}
355	mutex_unlock(&u->reply_mutex);
356
357	kfree(reply);
358
359out:
360	return rc;
361}
362
363static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
364{
365	struct watch_adapter *watch, *tmp_watch;
366	char *path, *token;
367	int err, rc;
368	LIST_HEAD(staging_q);
369
370	path = u->u.buffer + sizeof(u->u.msg);
371	token = memchr(path, 0, u->u.msg.len);
372	if (token == NULL) {
373		rc = -EILSEQ;
374		goto out;
375	}
376	token++;
377	if (memchr(token, 0, u->u.msg.len - (token - path)) == NULL) {
378		rc = -EILSEQ;
379		goto out;
380	}
381
382	if (msg_type == XS_WATCH) {
383		watch = alloc_watch_adapter(path, token);
384		if (watch == NULL) {
385			rc = -ENOMEM;
386			goto out;
387		}
388
389		watch->watch.callback = watch_fired;
390		watch->dev_data = u;
391
392		err = register_xenbus_watch(&watch->watch);
393		if (err) {
394			free_watch_adapter(watch);
395			rc = err;
396			goto out;
397		}
398		list_add(&watch->list, &u->watches);
399	} else {
400		list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
401			if (!strcmp(watch->token, token) &&
402			    !strcmp(watch->watch.node, path)) {
403				unregister_xenbus_watch(&watch->watch);
404				list_del(&watch->list);
405				free_watch_adapter(watch);
406				break;
407			}
408		}
409	}
410
411	/* Success.  Synthesize a reply to say all is OK. */
412	{
413		struct {
414			struct xsd_sockmsg hdr;
415			char body[3];
416		} __packed reply = {
417			{
418				.type = msg_type,
419				.len = sizeof(reply.body)
420			},
421			"OK"
422		};
423
424		mutex_lock(&u->reply_mutex);
425		rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
426		wake_up(&u->read_waitq);
427		mutex_unlock(&u->reply_mutex);
428	}
429
430out:
431	return rc;
432}
433
434static ssize_t xenbus_file_write(struct file *filp,
435				const char __user *ubuf,
436				size_t len, loff_t *ppos)
437{
438	struct xenbus_file_priv *u = filp->private_data;
439	uint32_t msg_type;
440	int rc = len;
441	int ret;
442	LIST_HEAD(staging_q);
443
444	/*
445	 * We're expecting usermode to be writing properly formed
446	 * xenbus messages.  If they write an incomplete message we
447	 * buffer it up.  Once it is complete, we act on it.
448	 */
449
450	/*
451	 * Make sure concurrent writers can't stomp all over each
452	 * other's messages and make a mess of our partial message
453	 * buffer.  We don't make any attemppt to stop multiple
454	 * writers from making a mess of each other's incomplete
455	 * messages; we're just trying to guarantee our own internal
456	 * consistency and make sure that single writes are handled
457	 * atomically.
458	 */
459	mutex_lock(&u->msgbuffer_mutex);
460
461	/* Get this out of the way early to avoid confusion */
462	if (len == 0)
463		goto out;
464
465	/* Can't write a xenbus message larger we can buffer */
466	if (len > sizeof(u->u.buffer) - u->len) {
467		/* On error, dump existing buffer */
468		u->len = 0;
469		rc = -EINVAL;
470		goto out;
471	}
472
473	ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
474
475	if (ret != 0) {
476		rc = -EFAULT;
477		goto out;
478	}
479
480	/* Deal with a partial copy. */
481	len -= ret;
482	rc = len;
483
484	u->len += len;
485
486	/* Return if we haven't got a full message yet */
487	if (u->len < sizeof(u->u.msg))
488		goto out;	/* not even the header yet */
489
490	/* If we're expecting a message that's larger than we can
491	   possibly send, dump what we have and return an error. */
492	if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
493		rc = -E2BIG;
494		u->len = 0;
495		goto out;
496	}
497
498	if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
499		goto out;	/* incomplete data portion */
500
501	/*
502	 * OK, now we have a complete message.  Do something with it.
503	 */
504
505	msg_type = u->u.msg.type;
506
507	switch (msg_type) {
508	case XS_WATCH:
509	case XS_UNWATCH:
510		/* (Un)Ask for some path to be watched for changes */
511		ret = xenbus_write_watch(msg_type, u);
512		break;
513
514	default:
515		/* Send out a transaction */
516		ret = xenbus_write_transaction(msg_type, u);
517		break;
518	}
519	if (ret != 0)
520		rc = ret;
521
522	/* Buffered message consumed */
523	u->len = 0;
524
525 out:
526	mutex_unlock(&u->msgbuffer_mutex);
527	return rc;
528}
529
530static int xenbus_file_open(struct inode *inode, struct file *filp)
531{
532	struct xenbus_file_priv *u;
533
534	if (xen_store_evtchn == 0)
535		return -ENOENT;
536
537	nonseekable_open(inode, filp);
538
539	u = kzalloc(sizeof(*u), GFP_KERNEL);
540	if (u == NULL)
541		return -ENOMEM;
542
543	INIT_LIST_HEAD(&u->transactions);
544	INIT_LIST_HEAD(&u->watches);
545	INIT_LIST_HEAD(&u->read_buffers);
546	init_waitqueue_head(&u->read_waitq);
547
548	mutex_init(&u->reply_mutex);
549	mutex_init(&u->msgbuffer_mutex);
550
551	filp->private_data = u;
552
553	return 0;
554}
555
556static int xenbus_file_release(struct inode *inode, struct file *filp)
557{
558	struct xenbus_file_priv *u = filp->private_data;
559	struct xenbus_transaction_holder *trans, *tmp;
560	struct watch_adapter *watch, *tmp_watch;
561	struct read_buffer *rb, *tmp_rb;
562
563	/*
564	 * No need for locking here because there are no other users,
565	 * by definition.
566	 */
567
568	list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
569		xenbus_transaction_end(trans->handle, 1);
570		list_del(&trans->list);
571		kfree(trans);
572	}
573
574	list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
575		unregister_xenbus_watch(&watch->watch);
576		list_del(&watch->list);
577		free_watch_adapter(watch);
578	}
579
580	list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
581		list_del(&rb->list);
582		kfree(rb);
583	}
584	kfree(u);
585
586	return 0;
587}
588
589static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
590{
591	struct xenbus_file_priv *u = file->private_data;
592
593	poll_wait(file, &u->read_waitq, wait);
594	if (!list_empty(&u->read_buffers))
595		return POLLIN | POLLRDNORM;
596	return 0;
597}
598
599const struct file_operations xen_xenbus_fops = {
600	.read = xenbus_file_read,
601	.write = xenbus_file_write,
602	.open = xenbus_file_open,
603	.release = xenbus_file_release,
604	.poll = xenbus_file_poll,
605	.llseek = no_llseek,
606};
607EXPORT_SYMBOL_GPL(xen_xenbus_fops);
608
609static struct miscdevice xenbus_dev = {
610	.minor = MISC_DYNAMIC_MINOR,
611	.name = "xen/xenbus",
612	.fops = &xen_xenbus_fops,
613};
614
615static int __init xenbus_init(void)
616{
617	int err;
618
619	if (!xen_domain())
620		return -ENODEV;
621
622	err = misc_register(&xenbus_dev);
623	if (err)
624		pr_err("Could not register xenbus frontend device\n");
625	return err;
626}
627
628static void __exit xenbus_exit(void)
629{
630	misc_deregister(&xenbus_dev);
631}
632
633module_init(xenbus_init);
634module_exit(xenbus_exit);
635