Loading...
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#include <linux/kernel.h>
39#include <linux/errno.h>
40#include <linux/uio.h>
41#include <linux/notifier.h>
42#include <linux/wait.h>
43#include <linux/fs.h>
44#include <linux/poll.h>
45#include <linux/mutex.h>
46#include <linux/sched.h>
47#include <linux/spinlock.h>
48#include <linux/mount.h>
49#include <linux/pagemap.h>
50#include <linux/uaccess.h>
51#include <linux/init.h>
52#include <linux/namei.h>
53#include <linux/string.h>
54#include <linux/slab.h>
55#include <linux/miscdevice.h>
56#include <linux/module.h>
57
58#include "xenbus_comms.h"
59
60#include <xen/xenbus.h>
61#include <xen/xen.h>
62#include <asm/xen/hypervisor.h>
63
64MODULE_LICENSE("GPL");
65
66/*
67 * An element of a list of outstanding transactions, for which we're
68 * still waiting a reply.
69 */
70struct xenbus_transaction_holder {
71 struct list_head list;
72 struct xenbus_transaction handle;
73};
74
75/*
76 * A buffer of data on the queue.
77 */
78struct read_buffer {
79 struct list_head list;
80 unsigned int cons;
81 unsigned int len;
82 char msg[];
83};
84
85struct xenbus_file_priv {
86 /*
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.
91 *
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.
95 */
96 struct mutex msgbuffer_mutex;
97
98 /* In-progress transactions */
99 struct list_head transactions;
100
101 /* Active watches. */
102 struct list_head watches;
103
104 /* Partial request. */
105 unsigned int len;
106 union {
107 struct xsd_sockmsg msg;
108 char buffer[XENSTORE_PAYLOAD_MAX];
109 } u;
110
111 /* Response queue. */
112 struct mutex reply_mutex;
113 struct list_head read_buffers;
114 wait_queue_head_t read_waitq;
115
116};
117
118/* Read out any raw xenbus messages queued up. */
119static ssize_t xenbus_file_read(struct file *filp,
120 char __user *ubuf,
121 size_t len, loff_t *ppos)
122{
123 struct xenbus_file_priv *u = filp->private_data;
124 struct read_buffer *rb;
125 unsigned i;
126 int ret;
127
128 mutex_lock(&u->reply_mutex);
129again:
130 while (list_empty(&u->read_buffers)) {
131 mutex_unlock(&u->reply_mutex);
132 if (filp->f_flags & O_NONBLOCK)
133 return -EAGAIN;
134
135 ret = wait_event_interruptible(u->read_waitq,
136 !list_empty(&u->read_buffers));
137 if (ret)
138 return ret;
139 mutex_lock(&u->reply_mutex);
140 }
141
142 rb = list_entry(u->read_buffers.next, struct read_buffer, list);
143 i = 0;
144 while (i < len) {
145 unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
146
147 ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
148
149 i += sz - ret;
150 rb->cons += sz - ret;
151
152 if (ret != 0) {
153 if (i == 0)
154 i = -EFAULT;
155 goto out;
156 }
157
158 /* Clear out buffer if it has been consumed */
159 if (rb->cons == rb->len) {
160 list_del(&rb->list);
161 kfree(rb);
162 if (list_empty(&u->read_buffers))
163 break;
164 rb = list_entry(u->read_buffers.next,
165 struct read_buffer, list);
166 }
167 }
168 if (i == 0)
169 goto again;
170
171out:
172 mutex_unlock(&u->reply_mutex);
173 return i;
174}
175
176/*
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.)
182 */
183static int queue_reply(struct list_head *queue, const void *data, size_t len)
184{
185 struct read_buffer *rb;
186
187 if (len == 0)
188 return 0;
189
190 rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
191 if (rb == NULL)
192 return -ENOMEM;
193
194 rb->cons = 0;
195 rb->len = len;
196
197 memcpy(rb->msg, data, len);
198
199 list_add_tail(&rb->list, queue);
200 return 0;
201}
202
203/*
204 * Free all the read_buffer s on a list.
205 * Caller must have sole reference to list.
206 */
207static void queue_cleanup(struct list_head *list)
208{
209 struct read_buffer *rb;
210
211 while (!list_empty(list)) {
212 rb = list_entry(list->next, struct read_buffer, list);
213 list_del(list->next);
214 kfree(rb);
215 }
216}
217
218struct watch_adapter {
219 struct list_head list;
220 struct xenbus_watch watch;
221 struct xenbus_file_priv *dev_data;
222 char *token;
223};
224
225static void free_watch_adapter(struct watch_adapter *watch)
226{
227 kfree(watch->watch.node);
228 kfree(watch->token);
229 kfree(watch);
230}
231
232static struct watch_adapter *alloc_watch_adapter(const char *path,
233 const char *token)
234{
235 struct watch_adapter *watch;
236
237 watch = kzalloc(sizeof(*watch), GFP_KERNEL);
238 if (watch == NULL)
239 goto out_fail;
240
241 watch->watch.node = kstrdup(path, GFP_KERNEL);
242 if (watch->watch.node == NULL)
243 goto out_free;
244
245 watch->token = kstrdup(token, GFP_KERNEL);
246 if (watch->token == NULL)
247 goto out_free;
248
249 return watch;
250
251out_free:
252 free_watch_adapter(watch);
253
254out_fail:
255 return NULL;
256}
257
258static void watch_fired(struct xenbus_watch *watch,
259 const char **vec,
260 unsigned int len)
261{
262 struct watch_adapter *adap;
263 struct xsd_sockmsg hdr;
264 const char *path, *token;
265 int path_len, tok_len, body_len, data_len = 0;
266 int ret;
267 LIST_HEAD(staging_q);
268
269 adap = container_of(watch, struct watch_adapter, watch);
270
271 path = vec[XS_WATCH_PATH];
272 token = adap->token;
273
274 path_len = strlen(path) + 1;
275 tok_len = strlen(token) + 1;
276 if (len > 2)
277 data_len = vec[len] - vec[2] + 1;
278 body_len = path_len + tok_len + data_len;
279
280 hdr.type = XS_WATCH_EVENT;
281 hdr.len = body_len;
282
283 mutex_lock(&adap->dev_data->reply_mutex);
284
285 ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
286 if (!ret)
287 ret = queue_reply(&staging_q, path, path_len);
288 if (!ret)
289 ret = queue_reply(&staging_q, token, tok_len);
290 if (!ret && len > 2)
291 ret = queue_reply(&staging_q, vec[2], data_len);
292
293 if (!ret) {
294 /* success: pass reply list onto watcher */
295 list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
296 wake_up(&adap->dev_data->read_waitq);
297 } else
298 queue_cleanup(&staging_q);
299
300 mutex_unlock(&adap->dev_data->reply_mutex);
301}
302
303static int xenbus_write_transaction(unsigned msg_type,
304 struct xenbus_file_priv *u)
305{
306 int rc;
307 void *reply;
308 struct xenbus_transaction_holder *trans = NULL;
309 LIST_HEAD(staging_q);
310
311 if (msg_type == XS_TRANSACTION_START) {
312 trans = kmalloc(sizeof(*trans), GFP_KERNEL);
313 if (!trans) {
314 rc = -ENOMEM;
315 goto out;
316 }
317 }
318
319 reply = xenbus_dev_request_and_reply(&u->u.msg);
320 if (IS_ERR(reply)) {
321 kfree(trans);
322 rc = PTR_ERR(reply);
323 goto out;
324 }
325
326 if (msg_type == XS_TRANSACTION_START) {
327 trans->handle.id = simple_strtoul(reply, NULL, 0);
328
329 list_add(&trans->list, &u->transactions);
330 } else if (msg_type == XS_TRANSACTION_END) {
331 list_for_each_entry(trans, &u->transactions, list)
332 if (trans->handle.id == u->u.msg.tx_id)
333 break;
334 BUG_ON(&trans->list == &u->transactions);
335 list_del(&trans->list);
336
337 kfree(trans);
338 }
339
340 mutex_lock(&u->reply_mutex);
341 rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
342 if (!rc)
343 rc = queue_reply(&staging_q, reply, u->u.msg.len);
344 if (!rc) {
345 list_splice_tail(&staging_q, &u->read_buffers);
346 wake_up(&u->read_waitq);
347 } else {
348 queue_cleanup(&staging_q);
349 }
350 mutex_unlock(&u->reply_mutex);
351
352 kfree(reply);
353
354out:
355 return rc;
356}
357
358static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
359{
360 struct watch_adapter *watch, *tmp_watch;
361 char *path, *token;
362 int err, rc;
363 LIST_HEAD(staging_q);
364
365 path = u->u.buffer + sizeof(u->u.msg);
366 token = memchr(path, 0, u->u.msg.len);
367 if (token == NULL) {
368 rc = -EILSEQ;
369 goto out;
370 }
371 token++;
372 if (memchr(token, 0, u->u.msg.len - (token - path)) == NULL) {
373 rc = -EILSEQ;
374 goto out;
375 }
376
377 if (msg_type == XS_WATCH) {
378 watch = alloc_watch_adapter(path, token);
379 if (watch == NULL) {
380 rc = -ENOMEM;
381 goto out;
382 }
383
384 watch->watch.callback = watch_fired;
385 watch->dev_data = u;
386
387 err = register_xenbus_watch(&watch->watch);
388 if (err) {
389 free_watch_adapter(watch);
390 rc = err;
391 goto out;
392 }
393 list_add(&watch->list, &u->watches);
394 } else {
395 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
396 if (!strcmp(watch->token, token) &&
397 !strcmp(watch->watch.node, path)) {
398 unregister_xenbus_watch(&watch->watch);
399 list_del(&watch->list);
400 free_watch_adapter(watch);
401 break;
402 }
403 }
404 }
405
406 /* Success. Synthesize a reply to say all is OK. */
407 {
408 struct {
409 struct xsd_sockmsg hdr;
410 char body[3];
411 } __packed reply = {
412 {
413 .type = msg_type,
414 .len = sizeof(reply.body)
415 },
416 "OK"
417 };
418
419 mutex_lock(&u->reply_mutex);
420 rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
421 wake_up(&u->read_waitq);
422 mutex_unlock(&u->reply_mutex);
423 }
424
425out:
426 return rc;
427}
428
429static ssize_t xenbus_file_write(struct file *filp,
430 const char __user *ubuf,
431 size_t len, loff_t *ppos)
432{
433 struct xenbus_file_priv *u = filp->private_data;
434 uint32_t msg_type;
435 int rc = len;
436 int ret;
437 LIST_HEAD(staging_q);
438
439 /*
440 * We're expecting usermode to be writing properly formed
441 * xenbus messages. If they write an incomplete message we
442 * buffer it up. Once it is complete, we act on it.
443 */
444
445 /*
446 * Make sure concurrent writers can't stomp all over each
447 * other's messages and make a mess of our partial message
448 * buffer. We don't make any attemppt to stop multiple
449 * writers from making a mess of each other's incomplete
450 * messages; we're just trying to guarantee our own internal
451 * consistency and make sure that single writes are handled
452 * atomically.
453 */
454 mutex_lock(&u->msgbuffer_mutex);
455
456 /* Get this out of the way early to avoid confusion */
457 if (len == 0)
458 goto out;
459
460 /* Can't write a xenbus message larger we can buffer */
461 if ((len + u->len) > sizeof(u->u.buffer)) {
462 /* On error, dump existing buffer */
463 u->len = 0;
464 rc = -EINVAL;
465 goto out;
466 }
467
468 ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
469
470 if (ret != 0) {
471 rc = -EFAULT;
472 goto out;
473 }
474
475 /* Deal with a partial copy. */
476 len -= ret;
477 rc = len;
478
479 u->len += len;
480
481 /* Return if we haven't got a full message yet */
482 if (u->len < sizeof(u->u.msg))
483 goto out; /* not even the header yet */
484
485 /* If we're expecting a message that's larger than we can
486 possibly send, dump what we have and return an error. */
487 if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
488 rc = -E2BIG;
489 u->len = 0;
490 goto out;
491 }
492
493 if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
494 goto out; /* incomplete data portion */
495
496 /*
497 * OK, now we have a complete message. Do something with it.
498 */
499
500 msg_type = u->u.msg.type;
501
502 switch (msg_type) {
503 case XS_WATCH:
504 case XS_UNWATCH:
505 /* (Un)Ask for some path to be watched for changes */
506 ret = xenbus_write_watch(msg_type, u);
507 break;
508
509 default:
510 /* Send out a transaction */
511 ret = xenbus_write_transaction(msg_type, u);
512 break;
513 }
514 if (ret != 0)
515 rc = ret;
516
517 /* Buffered message consumed */
518 u->len = 0;
519
520 out:
521 mutex_unlock(&u->msgbuffer_mutex);
522 return rc;
523}
524
525static int xenbus_file_open(struct inode *inode, struct file *filp)
526{
527 struct xenbus_file_priv *u;
528
529 if (xen_store_evtchn == 0)
530 return -ENOENT;
531
532 nonseekable_open(inode, filp);
533
534 u = kzalloc(sizeof(*u), GFP_KERNEL);
535 if (u == NULL)
536 return -ENOMEM;
537
538 INIT_LIST_HEAD(&u->transactions);
539 INIT_LIST_HEAD(&u->watches);
540 INIT_LIST_HEAD(&u->read_buffers);
541 init_waitqueue_head(&u->read_waitq);
542
543 mutex_init(&u->reply_mutex);
544 mutex_init(&u->msgbuffer_mutex);
545
546 filp->private_data = u;
547
548 return 0;
549}
550
551static int xenbus_file_release(struct inode *inode, struct file *filp)
552{
553 struct xenbus_file_priv *u = filp->private_data;
554 struct xenbus_transaction_holder *trans, *tmp;
555 struct watch_adapter *watch, *tmp_watch;
556 struct read_buffer *rb, *tmp_rb;
557
558 /*
559 * No need for locking here because there are no other users,
560 * by definition.
561 */
562
563 list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
564 xenbus_transaction_end(trans->handle, 1);
565 list_del(&trans->list);
566 kfree(trans);
567 }
568
569 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
570 unregister_xenbus_watch(&watch->watch);
571 list_del(&watch->list);
572 free_watch_adapter(watch);
573 }
574
575 list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
576 list_del(&rb->list);
577 kfree(rb);
578 }
579 kfree(u);
580
581 return 0;
582}
583
584static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
585{
586 struct xenbus_file_priv *u = file->private_data;
587
588 poll_wait(file, &u->read_waitq, wait);
589 if (!list_empty(&u->read_buffers))
590 return POLLIN | POLLRDNORM;
591 return 0;
592}
593
594const struct file_operations xen_xenbus_fops = {
595 .read = xenbus_file_read,
596 .write = xenbus_file_write,
597 .open = xenbus_file_open,
598 .release = xenbus_file_release,
599 .poll = xenbus_file_poll,
600 .llseek = no_llseek,
601};
602EXPORT_SYMBOL_GPL(xen_xenbus_fops);
603
604static struct miscdevice xenbus_dev = {
605 .minor = MISC_DYNAMIC_MINOR,
606 .name = "xen/xenbus",
607 .fops = &xen_xenbus_fops,
608};
609
610static int __init xenbus_init(void)
611{
612 int err;
613
614 if (!xen_domain())
615 return -ENODEV;
616
617 err = misc_register(&xenbus_dev);
618 if (err)
619 printk(KERN_ERR "Could not register xenbus frontend device\n");
620 return err;
621}
622
623static void __exit xenbus_exit(void)
624{
625 misc_deregister(&xenbus_dev);
626}
627
628module_init(xenbus_init);
629module_exit(xenbus_exit);