]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - drivers/xen/xenfs/xenbus.c
xen: add xenfs to allow usermode <-> Xen interaction
[mirror_ubuntu-zesty-kernel.git] / drivers / xen / xenfs / xenbus.c
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/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>
53
54 #include "xenfs.h"
55 #include "../xenbus/xenbus_comms.h"
56
57 #include <xen/xenbus.h>
58 #include <asm/xen/hypervisor.h>
59
60 /*
61 * An element of a list of outstanding transactions, for which we're
62 * still waiting a reply.
63 */
64 struct xenbus_transaction_holder {
65 struct list_head list;
66 struct xenbus_transaction handle;
67 };
68
69 /*
70 * A buffer of data on the queue.
71 */
72 struct read_buffer {
73 struct list_head list;
74 unsigned int cons;
75 unsigned int len;
76 char msg[];
77 };
78
79 struct xenbus_file_priv {
80 /*
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.
85 *
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.
89 */
90 struct mutex msgbuffer_mutex;
91
92 /* In-progress transactions */
93 struct list_head transactions;
94
95 /* Active watches. */
96 struct list_head watches;
97
98 /* Partial request. */
99 unsigned int len;
100 union {
101 struct xsd_sockmsg msg;
102 char buffer[PAGE_SIZE];
103 } u;
104
105 /* Response queue. */
106 struct mutex reply_mutex;
107 struct list_head read_buffers;
108 wait_queue_head_t read_waitq;
109
110 };
111
112 /* Read out any raw xenbus messages queued up. */
113 static ssize_t xenbus_file_read(struct file *filp,
114 char __user *ubuf,
115 size_t len, loff_t *ppos)
116 {
117 struct xenbus_file_priv *u = filp->private_data;
118 struct read_buffer *rb;
119 unsigned i;
120 int ret;
121
122 mutex_lock(&u->reply_mutex);
123 while (list_empty(&u->read_buffers)) {
124 mutex_unlock(&u->reply_mutex);
125 ret = wait_event_interruptible(u->read_waitq,
126 !list_empty(&u->read_buffers));
127 if (ret)
128 return ret;
129 mutex_lock(&u->reply_mutex);
130 }
131
132 rb = list_entry(u->read_buffers.next, struct read_buffer, list);
133 i = 0;
134 while (i < len) {
135 unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
136
137 ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
138
139 i += sz - ret;
140 rb->cons += sz - ret;
141
142 if (ret != sz) {
143 if (i == 0)
144 i = -EFAULT;
145 goto out;
146 }
147
148 /* Clear out buffer if it has been consumed */
149 if (rb->cons == rb->len) {
150 list_del(&rb->list);
151 kfree(rb);
152 if (list_empty(&u->read_buffers))
153 break;
154 rb = list_entry(u->read_buffers.next,
155 struct read_buffer, list);
156 }
157 }
158
159 out:
160 mutex_unlock(&u->reply_mutex);
161 return i;
162 }
163
164 /*
165 * Add a buffer to the queue. Caller must hold the appropriate lock
166 * if the queue is not local. (Commonly the caller will build up
167 * multiple queued buffers on a temporary local list, and then add it
168 * to the appropriate list under lock once all the buffers have een
169 * successfully allocated.)
170 */
171 static int queue_reply(struct list_head *queue, const void *data, size_t len)
172 {
173 struct read_buffer *rb;
174
175 if (len == 0)
176 return 0;
177
178 rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
179 if (rb == NULL)
180 return -ENOMEM;
181
182 rb->cons = 0;
183 rb->len = len;
184
185 memcpy(rb->msg, data, len);
186
187 list_add_tail(&rb->list, queue);
188 return 0;
189 }
190
191 /*
192 * Free all the read_buffer s on a list.
193 * Caller must have sole reference to list.
194 */
195 static void queue_cleanup(struct list_head *list)
196 {
197 struct read_buffer *rb;
198
199 while (!list_empty(list)) {
200 rb = list_entry(list->next, struct read_buffer, list);
201 list_del(list->next);
202 kfree(rb);
203 }
204 }
205
206 struct watch_adapter {
207 struct list_head list;
208 struct xenbus_watch watch;
209 struct xenbus_file_priv *dev_data;
210 char *token;
211 };
212
213 static void free_watch_adapter(struct watch_adapter *watch)
214 {
215 kfree(watch->watch.node);
216 kfree(watch->token);
217 kfree(watch);
218 }
219
220 static struct watch_adapter *alloc_watch_adapter(const char *path,
221 const char *token)
222 {
223 struct watch_adapter *watch;
224
225 watch = kzalloc(sizeof(*watch), GFP_KERNEL);
226 if (watch == NULL)
227 goto out_fail;
228
229 watch->watch.node = kstrdup(path, GFP_KERNEL);
230 if (watch->watch.node == NULL)
231 goto out_free;
232
233 watch->token = kstrdup(token, GFP_KERNEL);
234 if (watch->token == NULL)
235 goto out_free;
236
237 return watch;
238
239 out_free:
240 free_watch_adapter(watch);
241
242 out_fail:
243 return NULL;
244 }
245
246 static void watch_fired(struct xenbus_watch *watch,
247 const char **vec,
248 unsigned int len)
249 {
250 struct watch_adapter *adap;
251 struct xsd_sockmsg hdr;
252 const char *path, *token;
253 int path_len, tok_len, body_len, data_len = 0;
254 int ret;
255 LIST_HEAD(staging_q);
256
257 adap = container_of(watch, struct watch_adapter, watch);
258
259 path = vec[XS_WATCH_PATH];
260 token = adap->token;
261
262 path_len = strlen(path) + 1;
263 tok_len = strlen(token) + 1;
264 if (len > 2)
265 data_len = vec[len] - vec[2] + 1;
266 body_len = path_len + tok_len + data_len;
267
268 hdr.type = XS_WATCH_EVENT;
269 hdr.len = body_len;
270
271 mutex_lock(&adap->dev_data->reply_mutex);
272
273 ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
274 if (!ret)
275 ret = queue_reply(&staging_q, path, path_len);
276 if (!ret)
277 ret = queue_reply(&staging_q, token, tok_len);
278 if (!ret && len > 2)
279 ret = queue_reply(&staging_q, vec[2], data_len);
280
281 if (!ret) {
282 /* success: pass reply list onto watcher */
283 list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
284 wake_up(&adap->dev_data->read_waitq);
285 } else
286 queue_cleanup(&staging_q);
287
288 mutex_unlock(&adap->dev_data->reply_mutex);
289 }
290
291 static int xenbus_write_transaction(unsigned msg_type,
292 struct xenbus_file_priv *u)
293 {
294 int rc, ret;
295 void *reply;
296 struct xenbus_transaction_holder *trans = NULL;
297 LIST_HEAD(staging_q);
298
299 if (msg_type == XS_TRANSACTION_START) {
300 trans = kmalloc(sizeof(*trans), GFP_KERNEL);
301 if (!trans) {
302 rc = -ENOMEM;
303 goto out;
304 }
305 }
306
307 reply = xenbus_dev_request_and_reply(&u->u.msg);
308 if (IS_ERR(reply)) {
309 kfree(trans);
310 rc = PTR_ERR(reply);
311 goto out;
312 }
313
314 if (msg_type == XS_TRANSACTION_START) {
315 trans->handle.id = simple_strtoul(reply, NULL, 0);
316
317 list_add(&trans->list, &u->transactions);
318 } else if (msg_type == XS_TRANSACTION_END) {
319 list_for_each_entry(trans, &u->transactions, list)
320 if (trans->handle.id == u->u.msg.tx_id)
321 break;
322 BUG_ON(&trans->list == &u->transactions);
323 list_del(&trans->list);
324
325 kfree(trans);
326 }
327
328 mutex_lock(&u->reply_mutex);
329 ret = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
330 if (!ret)
331 ret = queue_reply(&staging_q, reply, u->u.msg.len);
332 if (!ret) {
333 list_splice_tail(&staging_q, &u->read_buffers);
334 wake_up(&u->read_waitq);
335 } else {
336 queue_cleanup(&staging_q);
337 rc = ret;
338 }
339 mutex_unlock(&u->reply_mutex);
340
341 kfree(reply);
342
343 out:
344 return rc;
345 }
346
347 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
348 {
349 struct watch_adapter *watch, *tmp_watch;
350 char *path, *token;
351 int err, rc;
352 LIST_HEAD(staging_q);
353
354 path = u->u.buffer + sizeof(u->u.msg);
355 token = memchr(path, 0, u->u.msg.len);
356 if (token == NULL) {
357 rc = -EILSEQ;
358 goto out;
359 }
360 token++;
361
362 if (msg_type == XS_WATCH) {
363 watch = alloc_watch_adapter(path, token);
364 if (watch == NULL) {
365 rc = -ENOMEM;
366 goto out;
367 }
368
369 watch->watch.callback = watch_fired;
370 watch->dev_data = u;
371
372 err = register_xenbus_watch(&watch->watch);
373 if (err) {
374 free_watch_adapter(watch);
375 rc = err;
376 goto out;
377 }
378 list_add(&watch->list, &u->watches);
379 } else {
380 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
381 if (!strcmp(watch->token, token) &&
382 !strcmp(watch->watch.node, path)) {
383 unregister_xenbus_watch(&watch->watch);
384 list_del(&watch->list);
385 free_watch_adapter(watch);
386 break;
387 }
388 }
389 }
390
391 /* Success. Synthesize a reply to say all is OK. */
392 {
393 struct {
394 struct xsd_sockmsg hdr;
395 char body[3];
396 } __packed reply = {
397 {
398 .type = msg_type,
399 .len = sizeof(reply.body)
400 },
401 "OK"
402 };
403
404 mutex_lock(&u->reply_mutex);
405 rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
406 mutex_unlock(&u->reply_mutex);
407 }
408
409 out:
410 return rc;
411 }
412
413 static ssize_t xenbus_file_write(struct file *filp,
414 const char __user *ubuf,
415 size_t len, loff_t *ppos)
416 {
417 struct xenbus_file_priv *u = filp->private_data;
418 uint32_t msg_type;
419 int rc = len;
420 int ret;
421 LIST_HEAD(staging_q);
422
423 /*
424 * We're expecting usermode to be writing properly formed
425 * xenbus messages. If they write an incomplete message we
426 * buffer it up. Once it is complete, we act on it.
427 */
428
429 /*
430 * Make sure concurrent writers can't stomp all over each
431 * other's messages and make a mess of our partial message
432 * buffer. We don't make any attemppt to stop multiple
433 * writers from making a mess of each other's incomplete
434 * messages; we're just trying to guarantee our own internal
435 * consistency and make sure that single writes are handled
436 * atomically.
437 */
438 mutex_lock(&u->msgbuffer_mutex);
439
440 /* Get this out of the way early to avoid confusion */
441 if (len == 0)
442 goto out;
443
444 /* Can't write a xenbus message larger we can buffer */
445 if ((len + u->len) > sizeof(u->u.buffer)) {
446 /* On error, dump existing buffer */
447 u->len = 0;
448 rc = -EINVAL;
449 goto out;
450 }
451
452 ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
453
454 if (ret == len) {
455 rc = -EFAULT;
456 goto out;
457 }
458
459 /* Deal with a partial copy. */
460 len -= ret;
461 rc = len;
462
463 u->len += len;
464
465 /* Return if we haven't got a full message yet */
466 if (u->len < sizeof(u->u.msg))
467 goto out; /* not even the header yet */
468
469 /* If we're expecting a message that's larger than we can
470 possibly send, dump what we have and return an error. */
471 if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
472 rc = -E2BIG;
473 u->len = 0;
474 goto out;
475 }
476
477 if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
478 goto out; /* incomplete data portion */
479
480 /*
481 * OK, now we have a complete message. Do something with it.
482 */
483
484 msg_type = u->u.msg.type;
485
486 switch (msg_type) {
487 case XS_TRANSACTION_START:
488 case XS_TRANSACTION_END:
489 case XS_DIRECTORY:
490 case XS_READ:
491 case XS_GET_PERMS:
492 case XS_RELEASE:
493 case XS_GET_DOMAIN_PATH:
494 case XS_WRITE:
495 case XS_MKDIR:
496 case XS_RM:
497 case XS_SET_PERMS:
498 /* Send out a transaction */
499 ret = xenbus_write_transaction(msg_type, u);
500 break;
501
502 case XS_WATCH:
503 case XS_UNWATCH:
504 /* (Un)Ask for some path to be watched for changes */
505 ret = xenbus_write_watch(msg_type, u);
506 break;
507
508 default:
509 ret = -EINVAL;
510 break;
511 }
512 if (ret != 0)
513 rc = ret;
514
515 /* Buffered message consumed */
516 u->len = 0;
517
518 out:
519 mutex_unlock(&u->msgbuffer_mutex);
520 return rc;
521 }
522
523 static int xenbus_file_open(struct inode *inode, struct file *filp)
524 {
525 struct xenbus_file_priv *u;
526
527 if (xen_store_evtchn == 0)
528 return -ENOENT;
529
530 nonseekable_open(inode, filp);
531
532 u = kzalloc(sizeof(*u), GFP_KERNEL);
533 if (u == NULL)
534 return -ENOMEM;
535
536 INIT_LIST_HEAD(&u->transactions);
537 INIT_LIST_HEAD(&u->watches);
538 INIT_LIST_HEAD(&u->read_buffers);
539 init_waitqueue_head(&u->read_waitq);
540
541 mutex_init(&u->reply_mutex);
542 mutex_init(&u->msgbuffer_mutex);
543
544 filp->private_data = u;
545
546 return 0;
547 }
548
549 static int xenbus_file_release(struct inode *inode, struct file *filp)
550 {
551 struct xenbus_file_priv *u = filp->private_data;
552 struct xenbus_transaction_holder *trans, *tmp;
553 struct watch_adapter *watch, *tmp_watch;
554
555 /*
556 * No need for locking here because there are no other users,
557 * by definition.
558 */
559
560 list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
561 xenbus_transaction_end(trans->handle, 1);
562 list_del(&trans->list);
563 kfree(trans);
564 }
565
566 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
567 unregister_xenbus_watch(&watch->watch);
568 list_del(&watch->list);
569 free_watch_adapter(watch);
570 }
571
572 kfree(u);
573
574 return 0;
575 }
576
577 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
578 {
579 struct xenbus_file_priv *u = file->private_data;
580
581 poll_wait(file, &u->read_waitq, wait);
582 if (!list_empty(&u->read_buffers))
583 return POLLIN | POLLRDNORM;
584 return 0;
585 }
586
587 const struct file_operations xenbus_file_ops = {
588 .read = xenbus_file_read,
589 .write = xenbus_file_write,
590 .open = xenbus_file_open,
591 .release = xenbus_file_release,
592 .poll = xenbus_file_poll,
593 };