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
56 #include "xenfs.h"
57 #include "../xenbus/xenbus_comms.h"
58
59 #include <xen/xenbus.h>
60 #include <asm/xen/hypervisor.h>
61
62 /*
63 * An element of a list of outstanding transactions, for which we're
64 * still waiting a reply.
65 */
66 struct xenbus_transaction_holder {
67 struct list_head list;
68 struct xenbus_transaction handle;
69 };
70
71 /*
72 * A buffer of data on the queue.
73 */
74 struct read_buffer {
75 struct list_head list;
76 unsigned int cons;
77 unsigned int len;
78 char msg[];
79 };
80
81 struct xenbus_file_priv {
82 /*
83 * msgbuffer_mutex is held while partial requests are built up
84 * and complete requests are acted on. It therefore protects
85 * the "transactions" and "watches" lists, and the partial
86 * request length and buffer.
87 *
88 * reply_mutex protects the reply being built up to return to
89 * usermode. It nests inside msgbuffer_mutex but may be held
90 * alone during a watch callback.
91 */
92 struct mutex msgbuffer_mutex;
93
94 /* In-progress transactions */
95 struct list_head transactions;
96
97 /* Active watches. */
98 struct list_head watches;
99
100 /* Partial request. */
101 unsigned int len;
102 union {
103 struct xsd_sockmsg msg;
104 char buffer[PAGE_SIZE];
105 } u;
106
107 /* Response queue. */
108 struct mutex reply_mutex;
109 struct list_head read_buffers;
110 wait_queue_head_t read_waitq;
111
112 };
113
114 /* Read out any raw xenbus messages queued up. */
xenbus_file_read(struct file * filp,char __user * ubuf,size_t len,loff_t * ppos)115 static ssize_t xenbus_file_read(struct file *filp,
116 char __user *ubuf,
117 size_t len, loff_t *ppos)
118 {
119 struct xenbus_file_priv *u = filp->private_data;
120 struct read_buffer *rb;
121 unsigned i;
122 int ret;
123
124 mutex_lock(&u->reply_mutex);
125 again:
126 while (list_empty(&u->read_buffers)) {
127 mutex_unlock(&u->reply_mutex);
128 if (filp->f_flags & O_NONBLOCK)
129 return -EAGAIN;
130
131 ret = wait_event_interruptible(u->read_waitq,
132 !list_empty(&u->read_buffers));
133 if (ret)
134 return ret;
135 mutex_lock(&u->reply_mutex);
136 }
137
138 rb = list_entry(u->read_buffers.next, struct read_buffer, list);
139 i = 0;
140 while (i < len) {
141 unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
142
143 ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
144
145 i += sz - ret;
146 rb->cons += sz - ret;
147
148 if (ret != 0) {
149 if (i == 0)
150 i = -EFAULT;
151 goto out;
152 }
153
154 /* Clear out buffer if it has been consumed */
155 if (rb->cons == rb->len) {
156 list_del(&rb->list);
157 kfree(rb);
158 if (list_empty(&u->read_buffers))
159 break;
160 rb = list_entry(u->read_buffers.next,
161 struct read_buffer, list);
162 }
163 }
164 if (i == 0)
165 goto again;
166
167 out:
168 mutex_unlock(&u->reply_mutex);
169 return i;
170 }
171
172 /*
173 * Add a buffer to the queue. Caller must hold the appropriate lock
174 * if the queue is not local. (Commonly the caller will build up
175 * multiple queued buffers on a temporary local list, and then add it
176 * to the appropriate list under lock once all the buffers have een
177 * successfully allocated.)
178 */
queue_reply(struct list_head * queue,const void * data,size_t len)179 static int queue_reply(struct list_head *queue, const void *data, size_t len)
180 {
181 struct read_buffer *rb;
182
183 if (len == 0)
184 return 0;
185
186 rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
187 if (rb == NULL)
188 return -ENOMEM;
189
190 rb->cons = 0;
191 rb->len = len;
192
193 memcpy(rb->msg, data, len);
194
195 list_add_tail(&rb->list, queue);
196 return 0;
197 }
198
199 /*
200 * Free all the read_buffer s on a list.
201 * Caller must have sole reference to list.
202 */
queue_cleanup(struct list_head * list)203 static void queue_cleanup(struct list_head *list)
204 {
205 struct read_buffer *rb;
206
207 while (!list_empty(list)) {
208 rb = list_entry(list->next, struct read_buffer, list);
209 list_del(list->next);
210 kfree(rb);
211 }
212 }
213
214 struct watch_adapter {
215 struct list_head list;
216 struct xenbus_watch watch;
217 struct xenbus_file_priv *dev_data;
218 char *token;
219 };
220
free_watch_adapter(struct watch_adapter * watch)221 static void free_watch_adapter(struct watch_adapter *watch)
222 {
223 kfree(watch->watch.node);
224 kfree(watch->token);
225 kfree(watch);
226 }
227
alloc_watch_adapter(const char * path,const char * token)228 static struct watch_adapter *alloc_watch_adapter(const char *path,
229 const char *token)
230 {
231 struct watch_adapter *watch;
232
233 watch = kzalloc(sizeof(*watch), GFP_KERNEL);
234 if (watch == NULL)
235 goto out_fail;
236
237 watch->watch.node = kstrdup(path, GFP_KERNEL);
238 if (watch->watch.node == NULL)
239 goto out_free;
240
241 watch->token = kstrdup(token, GFP_KERNEL);
242 if (watch->token == NULL)
243 goto out_free;
244
245 return watch;
246
247 out_free:
248 free_watch_adapter(watch);
249
250 out_fail:
251 return NULL;
252 }
253
watch_fired(struct xenbus_watch * watch,const char ** vec,unsigned int len)254 static void watch_fired(struct xenbus_watch *watch,
255 const char **vec,
256 unsigned int len)
257 {
258 struct watch_adapter *adap;
259 struct xsd_sockmsg hdr;
260 const char *path, *token;
261 int path_len, tok_len, body_len, data_len = 0;
262 int ret;
263 LIST_HEAD(staging_q);
264
265 adap = container_of(watch, struct watch_adapter, watch);
266
267 path = vec[XS_WATCH_PATH];
268 token = adap->token;
269
270 path_len = strlen(path) + 1;
271 tok_len = strlen(token) + 1;
272 if (len > 2)
273 data_len = vec[len] - vec[2] + 1;
274 body_len = path_len + tok_len + data_len;
275
276 hdr.type = XS_WATCH_EVENT;
277 hdr.len = body_len;
278
279 mutex_lock(&adap->dev_data->reply_mutex);
280
281 ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
282 if (!ret)
283 ret = queue_reply(&staging_q, path, path_len);
284 if (!ret)
285 ret = queue_reply(&staging_q, token, tok_len);
286 if (!ret && len > 2)
287 ret = queue_reply(&staging_q, vec[2], data_len);
288
289 if (!ret) {
290 /* success: pass reply list onto watcher */
291 list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
292 wake_up(&adap->dev_data->read_waitq);
293 } else
294 queue_cleanup(&staging_q);
295
296 mutex_unlock(&adap->dev_data->reply_mutex);
297 }
298
xenbus_write_transaction(unsigned msg_type,struct xenbus_file_priv * u)299 static int xenbus_write_transaction(unsigned msg_type,
300 struct xenbus_file_priv *u)
301 {
302 int rc;
303 void *reply;
304 struct xenbus_transaction_holder *trans = NULL;
305 LIST_HEAD(staging_q);
306
307 if (msg_type == XS_TRANSACTION_START) {
308 trans = kmalloc(sizeof(*trans), GFP_KERNEL);
309 if (!trans) {
310 rc = -ENOMEM;
311 goto out;
312 }
313 }
314
315 reply = xenbus_dev_request_and_reply(&u->u.msg);
316 if (IS_ERR(reply)) {
317 kfree(trans);
318 rc = PTR_ERR(reply);
319 goto out;
320 }
321
322 if (msg_type == XS_TRANSACTION_START) {
323 trans->handle.id = simple_strtoul(reply, NULL, 0);
324
325 list_add(&trans->list, &u->transactions);
326 } else if (msg_type == XS_TRANSACTION_END) {
327 list_for_each_entry(trans, &u->transactions, list)
328 if (trans->handle.id == u->u.msg.tx_id)
329 break;
330 BUG_ON(&trans->list == &u->transactions);
331 list_del(&trans->list);
332
333 kfree(trans);
334 }
335
336 mutex_lock(&u->reply_mutex);
337 rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
338 if (!rc)
339 rc = queue_reply(&staging_q, reply, u->u.msg.len);
340 if (!rc) {
341 list_splice_tail(&staging_q, &u->read_buffers);
342 wake_up(&u->read_waitq);
343 } else {
344 queue_cleanup(&staging_q);
345 }
346 mutex_unlock(&u->reply_mutex);
347
348 kfree(reply);
349
350 out:
351 return rc;
352 }
353
xenbus_write_watch(unsigned msg_type,struct xenbus_file_priv * u)354 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
355 {
356 struct watch_adapter *watch, *tmp_watch;
357 char *path, *token;
358 int err, rc;
359 LIST_HEAD(staging_q);
360
361 path = u->u.buffer + sizeof(u->u.msg);
362 token = memchr(path, 0, u->u.msg.len);
363 if (token == NULL) {
364 rc = -EILSEQ;
365 goto out;
366 }
367 token++;
368
369 if (msg_type == XS_WATCH) {
370 watch = alloc_watch_adapter(path, token);
371 if (watch == NULL) {
372 rc = -ENOMEM;
373 goto out;
374 }
375
376 watch->watch.callback = watch_fired;
377 watch->dev_data = u;
378
379 err = register_xenbus_watch(&watch->watch);
380 if (err) {
381 free_watch_adapter(watch);
382 rc = err;
383 goto out;
384 }
385 list_add(&watch->list, &u->watches);
386 } else {
387 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
388 if (!strcmp(watch->token, token) &&
389 !strcmp(watch->watch.node, path)) {
390 unregister_xenbus_watch(&watch->watch);
391 list_del(&watch->list);
392 free_watch_adapter(watch);
393 break;
394 }
395 }
396 }
397
398 /* Success. Synthesize a reply to say all is OK. */
399 {
400 struct {
401 struct xsd_sockmsg hdr;
402 char body[3];
403 } __packed reply = {
404 {
405 .type = msg_type,
406 .len = sizeof(reply.body)
407 },
408 "OK"
409 };
410
411 mutex_lock(&u->reply_mutex);
412 rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
413 wake_up(&u->read_waitq);
414 mutex_unlock(&u->reply_mutex);
415 }
416
417 out:
418 return rc;
419 }
420
xenbus_file_write(struct file * filp,const char __user * ubuf,size_t len,loff_t * ppos)421 static ssize_t xenbus_file_write(struct file *filp,
422 const char __user *ubuf,
423 size_t len, loff_t *ppos)
424 {
425 struct xenbus_file_priv *u = filp->private_data;
426 uint32_t msg_type;
427 int rc = len;
428 int ret;
429 LIST_HEAD(staging_q);
430
431 /*
432 * We're expecting usermode to be writing properly formed
433 * xenbus messages. If they write an incomplete message we
434 * buffer it up. Once it is complete, we act on it.
435 */
436
437 /*
438 * Make sure concurrent writers can't stomp all over each
439 * other's messages and make a mess of our partial message
440 * buffer. We don't make any attemppt to stop multiple
441 * writers from making a mess of each other's incomplete
442 * messages; we're just trying to guarantee our own internal
443 * consistency and make sure that single writes are handled
444 * atomically.
445 */
446 mutex_lock(&u->msgbuffer_mutex);
447
448 /* Get this out of the way early to avoid confusion */
449 if (len == 0)
450 goto out;
451
452 /* Can't write a xenbus message larger we can buffer */
453 if ((len + u->len) > sizeof(u->u.buffer)) {
454 /* On error, dump existing buffer */
455 u->len = 0;
456 rc = -EINVAL;
457 goto out;
458 }
459
460 ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
461
462 if (ret != 0) {
463 rc = -EFAULT;
464 goto out;
465 }
466
467 /* Deal with a partial copy. */
468 len -= ret;
469 rc = len;
470
471 u->len += len;
472
473 /* Return if we haven't got a full message yet */
474 if (u->len < sizeof(u->u.msg))
475 goto out; /* not even the header yet */
476
477 /* If we're expecting a message that's larger than we can
478 possibly send, dump what we have and return an error. */
479 if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
480 rc = -E2BIG;
481 u->len = 0;
482 goto out;
483 }
484
485 if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
486 goto out; /* incomplete data portion */
487
488 /*
489 * OK, now we have a complete message. Do something with it.
490 */
491
492 msg_type = u->u.msg.type;
493
494 switch (msg_type) {
495 case XS_WATCH:
496 case XS_UNWATCH:
497 /* (Un)Ask for some path to be watched for changes */
498 ret = xenbus_write_watch(msg_type, u);
499 break;
500
501 default:
502 /* Send out a transaction */
503 ret = xenbus_write_transaction(msg_type, u);
504 break;
505 }
506 if (ret != 0)
507 rc = ret;
508
509 /* Buffered message consumed */
510 u->len = 0;
511
512 out:
513 mutex_unlock(&u->msgbuffer_mutex);
514 return rc;
515 }
516
xenbus_file_open(struct inode * inode,struct file * filp)517 static int xenbus_file_open(struct inode *inode, struct file *filp)
518 {
519 struct xenbus_file_priv *u;
520
521 if (xen_store_evtchn == 0)
522 return -ENOENT;
523
524 nonseekable_open(inode, filp);
525
526 u = kzalloc(sizeof(*u), GFP_KERNEL);
527 if (u == NULL)
528 return -ENOMEM;
529
530 INIT_LIST_HEAD(&u->transactions);
531 INIT_LIST_HEAD(&u->watches);
532 INIT_LIST_HEAD(&u->read_buffers);
533 init_waitqueue_head(&u->read_waitq);
534
535 mutex_init(&u->reply_mutex);
536 mutex_init(&u->msgbuffer_mutex);
537
538 filp->private_data = u;
539
540 return 0;
541 }
542
xenbus_file_release(struct inode * inode,struct file * filp)543 static int xenbus_file_release(struct inode *inode, struct file *filp)
544 {
545 struct xenbus_file_priv *u = filp->private_data;
546 struct xenbus_transaction_holder *trans, *tmp;
547 struct watch_adapter *watch, *tmp_watch;
548 struct read_buffer *rb, *tmp_rb;
549
550 /*
551 * No need for locking here because there are no other users,
552 * by definition.
553 */
554
555 list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
556 xenbus_transaction_end(trans->handle, 1);
557 list_del(&trans->list);
558 kfree(trans);
559 }
560
561 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
562 unregister_xenbus_watch(&watch->watch);
563 list_del(&watch->list);
564 free_watch_adapter(watch);
565 }
566
567 list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) {
568 list_del(&rb->list);
569 kfree(rb);
570 }
571 kfree(u);
572
573 return 0;
574 }
575
xenbus_file_poll(struct file * file,poll_table * wait)576 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
577 {
578 struct xenbus_file_priv *u = file->private_data;
579
580 poll_wait(file, &u->read_waitq, wait);
581 if (!list_empty(&u->read_buffers))
582 return POLLIN | POLLRDNORM;
583 return 0;
584 }
585
586 const struct file_operations xenbus_file_ops = {
587 .read = xenbus_file_read,
588 .write = xenbus_file_write,
589 .open = xenbus_file_open,
590 .release = xenbus_file_release,
591 .poll = xenbus_file_poll,
592 .llseek = no_llseek,
593 };
594