]>
Commit | Line | Data |
---|---|---|
4bac07c9 JF |
1 | /****************************************************************************** |
2 | * xenbus_xs.c | |
3 | * | |
4 | * This is the kernel equivalent of the "xs" library. We don't need everything | |
5 | * and we use xenbus_comms for communication. | |
6 | * | |
7 | * Copyright (C) 2005 Rusty Russell, IBM Corporation | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU General Public License version 2 | |
11 | * as published by the Free Software Foundation; or, when distributed | |
12 | * separately from the Linux kernel or incorporated into other | |
13 | * software packages, subject to the following license: | |
14 | * | |
15 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
16 | * of this source file (the "Software"), to deal in the Software without | |
17 | * restriction, including without limitation the rights to use, copy, modify, | |
18 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
19 | * and to permit persons to whom the Software is furnished to do so, subject to | |
20 | * the following conditions: | |
21 | * | |
22 | * The above copyright notice and this permission notice shall be included in | |
23 | * all copies or substantial portions of the Software. | |
24 | * | |
25 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
26 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
27 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
28 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
29 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
30 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
31 | * IN THE SOFTWARE. | |
32 | */ | |
33 | ||
283c0972 JP |
34 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
35 | ||
4bac07c9 JF |
36 | #include <linux/unistd.h> |
37 | #include <linux/errno.h> | |
38 | #include <linux/types.h> | |
39 | #include <linux/uio.h> | |
40 | #include <linux/kernel.h> | |
41 | #include <linux/string.h> | |
42 | #include <linux/err.h> | |
43 | #include <linux/slab.h> | |
44 | #include <linux/fcntl.h> | |
45 | #include <linux/kthread.h> | |
46 | #include <linux/rwsem.h> | |
4bac07c9 | 47 | #include <linux/mutex.h> |
ecc635f9 | 48 | #include <asm/xen/hypervisor.h> |
4bac07c9 | 49 | #include <xen/xenbus.h> |
5b25d89e | 50 | #include <xen/xen.h> |
4bac07c9 | 51 | #include "xenbus_comms.h" |
027bd7e8 | 52 | #include "xenbus_probe.h" |
4bac07c9 JF |
53 | |
54 | struct xs_stored_msg { | |
55 | struct list_head list; | |
56 | ||
57 | struct xsd_sockmsg hdr; | |
58 | ||
59 | union { | |
60 | /* Queued replies. */ | |
61 | struct { | |
62 | char *body; | |
63 | } reply; | |
64 | ||
65 | /* Queued watch events. */ | |
66 | struct { | |
67 | struct xenbus_watch *handle; | |
68 | char **vec; | |
69 | unsigned int vec_size; | |
70 | } watch; | |
71 | } u; | |
72 | }; | |
73 | ||
74 | struct xs_handle { | |
75 | /* A list of replies. Currently only one will ever be outstanding. */ | |
76 | struct list_head reply_list; | |
77 | spinlock_t reply_lock; | |
78 | wait_queue_head_t reply_waitq; | |
79 | ||
80 | /* | |
81 | * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex. | |
82 | * response_mutex is never taken simultaneously with the other three. | |
4c31a781 IC |
83 | * |
84 | * transaction_mutex must be held before incrementing | |
85 | * transaction_count. The mutex is held when a suspend is in | |
86 | * progress to prevent new transactions starting. | |
87 | * | |
88 | * When decrementing transaction_count to zero the wait queue | |
89 | * should be woken up, the suspend code waits for count to | |
90 | * reach zero. | |
4bac07c9 JF |
91 | */ |
92 | ||
93 | /* One request at a time. */ | |
94 | struct mutex request_mutex; | |
95 | ||
96 | /* Protect xenbus reader thread against save/restore. */ | |
97 | struct mutex response_mutex; | |
98 | ||
99 | /* Protect transactions against save/restore. */ | |
4c31a781 IC |
100 | struct mutex transaction_mutex; |
101 | atomic_t transaction_count; | |
102 | wait_queue_head_t transaction_wq; | |
4bac07c9 JF |
103 | |
104 | /* Protect watch (de)register against save/restore. */ | |
105 | struct rw_semaphore watch_mutex; | |
106 | }; | |
107 | ||
108 | static struct xs_handle xs_state; | |
109 | ||
110 | /* List of registered watches, and a lock to protect it. */ | |
111 | static LIST_HEAD(watches); | |
112 | static DEFINE_SPINLOCK(watches_lock); | |
113 | ||
114 | /* List of pending watch callback events, and a lock to protect it. */ | |
115 | static LIST_HEAD(watch_events); | |
116 | static DEFINE_SPINLOCK(watch_events_lock); | |
117 | ||
118 | /* | |
119 | * Details of the xenwatch callback kernel thread. The thread waits on the | |
120 | * watch_events_waitq for work to do (queued on watch_events list). When it | |
121 | * wakes up it acquires the xenwatch_mutex before reading the list and | |
122 | * carrying out work. | |
123 | */ | |
124 | static pid_t xenwatch_pid; | |
125 | static DEFINE_MUTEX(xenwatch_mutex); | |
126 | static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq); | |
127 | ||
128 | static int get_error(const char *errorstring) | |
129 | { | |
130 | unsigned int i; | |
131 | ||
132 | for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) { | |
133 | if (i == ARRAY_SIZE(xsd_errors) - 1) { | |
283c0972 JP |
134 | pr_warn("xen store gave: unknown error %s\n", |
135 | errorstring); | |
4bac07c9 JF |
136 | return EINVAL; |
137 | } | |
138 | } | |
139 | return xsd_errors[i].errnum; | |
140 | } | |
141 | ||
027bd7e8 KRW |
142 | static bool xenbus_ok(void) |
143 | { | |
144 | switch (xen_store_domain_type) { | |
145 | case XS_LOCAL: | |
146 | switch (system_state) { | |
147 | case SYSTEM_POWER_OFF: | |
148 | case SYSTEM_RESTART: | |
149 | case SYSTEM_HALT: | |
150 | return false; | |
151 | default: | |
152 | break; | |
153 | } | |
154 | return true; | |
155 | case XS_PV: | |
156 | case XS_HVM: | |
157 | /* FIXME: Could check that the remote domain is alive, | |
158 | * but it is normally initial domain. */ | |
159 | return true; | |
160 | default: | |
161 | break; | |
162 | } | |
163 | return false; | |
164 | } | |
4bac07c9 JF |
165 | static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len) |
166 | { | |
167 | struct xs_stored_msg *msg; | |
168 | char *body; | |
169 | ||
170 | spin_lock(&xs_state.reply_lock); | |
171 | ||
172 | while (list_empty(&xs_state.reply_list)) { | |
173 | spin_unlock(&xs_state.reply_lock); | |
027bd7e8 KRW |
174 | if (xenbus_ok()) |
175 | /* XXX FIXME: Avoid synchronous wait for response here. */ | |
176 | wait_event_timeout(xs_state.reply_waitq, | |
177 | !list_empty(&xs_state.reply_list), | |
178 | msecs_to_jiffies(500)); | |
179 | else { | |
180 | /* | |
181 | * If we are in the process of being shut-down there is | |
182 | * no point of trying to contact XenBus - it is either | |
183 | * killed (xenstored application) or the other domain | |
184 | * has been killed or is unreachable. | |
185 | */ | |
186 | return ERR_PTR(-EIO); | |
187 | } | |
4bac07c9 JF |
188 | spin_lock(&xs_state.reply_lock); |
189 | } | |
190 | ||
191 | msg = list_entry(xs_state.reply_list.next, | |
192 | struct xs_stored_msg, list); | |
193 | list_del(&msg->list); | |
194 | ||
195 | spin_unlock(&xs_state.reply_lock); | |
196 | ||
197 | *type = msg->hdr.type; | |
198 | if (len) | |
199 | *len = msg->hdr.len; | |
200 | body = msg->u.reply.body; | |
201 | ||
202 | kfree(msg); | |
203 | ||
204 | return body; | |
205 | } | |
206 | ||
4c31a781 IC |
207 | static void transaction_start(void) |
208 | { | |
209 | mutex_lock(&xs_state.transaction_mutex); | |
210 | atomic_inc(&xs_state.transaction_count); | |
211 | mutex_unlock(&xs_state.transaction_mutex); | |
212 | } | |
213 | ||
214 | static void transaction_end(void) | |
215 | { | |
216 | if (atomic_dec_and_test(&xs_state.transaction_count)) | |
217 | wake_up(&xs_state.transaction_wq); | |
218 | } | |
219 | ||
220 | static void transaction_suspend(void) | |
221 | { | |
222 | mutex_lock(&xs_state.transaction_mutex); | |
223 | wait_event(xs_state.transaction_wq, | |
224 | atomic_read(&xs_state.transaction_count) == 0); | |
225 | } | |
226 | ||
227 | static void transaction_resume(void) | |
228 | { | |
229 | mutex_unlock(&xs_state.transaction_mutex); | |
230 | } | |
231 | ||
4bac07c9 JF |
232 | void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg) |
233 | { | |
234 | void *ret; | |
e5a79475 | 235 | enum xsd_sockmsg_type type = msg->type; |
4bac07c9 JF |
236 | int err; |
237 | ||
e5a79475 | 238 | if (type == XS_TRANSACTION_START) |
4c31a781 | 239 | transaction_start(); |
4bac07c9 JF |
240 | |
241 | mutex_lock(&xs_state.request_mutex); | |
242 | ||
243 | err = xb_write(msg, sizeof(*msg) + msg->len); | |
244 | if (err) { | |
245 | msg->type = XS_ERROR; | |
246 | ret = ERR_PTR(err); | |
247 | } else | |
248 | ret = read_reply(&msg->type, &msg->len); | |
249 | ||
250 | mutex_unlock(&xs_state.request_mutex); | |
251 | ||
252 | if ((msg->type == XS_TRANSACTION_END) || | |
e5a79475 | 253 | ((type == XS_TRANSACTION_START) && (msg->type == XS_ERROR))) |
4c31a781 | 254 | transaction_end(); |
4bac07c9 JF |
255 | |
256 | return ret; | |
257 | } | |
1107ba88 | 258 | EXPORT_SYMBOL(xenbus_dev_request_and_reply); |
4bac07c9 JF |
259 | |
260 | /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */ | |
261 | static void *xs_talkv(struct xenbus_transaction t, | |
262 | enum xsd_sockmsg_type type, | |
263 | const struct kvec *iovec, | |
264 | unsigned int num_vecs, | |
265 | unsigned int *len) | |
266 | { | |
267 | struct xsd_sockmsg msg; | |
268 | void *ret = NULL; | |
269 | unsigned int i; | |
270 | int err; | |
271 | ||
272 | msg.tx_id = t.id; | |
273 | msg.req_id = 0; | |
274 | msg.type = type; | |
275 | msg.len = 0; | |
276 | for (i = 0; i < num_vecs; i++) | |
277 | msg.len += iovec[i].iov_len; | |
278 | ||
279 | mutex_lock(&xs_state.request_mutex); | |
280 | ||
281 | err = xb_write(&msg, sizeof(msg)); | |
282 | if (err) { | |
283 | mutex_unlock(&xs_state.request_mutex); | |
284 | return ERR_PTR(err); | |
285 | } | |
286 | ||
287 | for (i = 0; i < num_vecs; i++) { | |
288 | err = xb_write(iovec[i].iov_base, iovec[i].iov_len); | |
289 | if (err) { | |
290 | mutex_unlock(&xs_state.request_mutex); | |
291 | return ERR_PTR(err); | |
292 | } | |
293 | } | |
294 | ||
295 | ret = read_reply(&msg.type, len); | |
296 | ||
297 | mutex_unlock(&xs_state.request_mutex); | |
298 | ||
299 | if (IS_ERR(ret)) | |
300 | return ret; | |
301 | ||
302 | if (msg.type == XS_ERROR) { | |
303 | err = get_error(ret); | |
304 | kfree(ret); | |
305 | return ERR_PTR(-err); | |
306 | } | |
307 | ||
308 | if (msg.type != type) { | |
283c0972 JP |
309 | pr_warn_ratelimited("unexpected type [%d], expected [%d]\n", |
310 | msg.type, type); | |
4bac07c9 JF |
311 | kfree(ret); |
312 | return ERR_PTR(-EINVAL); | |
313 | } | |
314 | return ret; | |
315 | } | |
316 | ||
317 | /* Simplified version of xs_talkv: single message. */ | |
318 | static void *xs_single(struct xenbus_transaction t, | |
319 | enum xsd_sockmsg_type type, | |
320 | const char *string, | |
321 | unsigned int *len) | |
322 | { | |
323 | struct kvec iovec; | |
324 | ||
325 | iovec.iov_base = (void *)string; | |
326 | iovec.iov_len = strlen(string) + 1; | |
327 | return xs_talkv(t, type, &iovec, 1, len); | |
328 | } | |
329 | ||
330 | /* Many commands only need an ack, don't care what it says. */ | |
331 | static int xs_error(char *reply) | |
332 | { | |
333 | if (IS_ERR(reply)) | |
334 | return PTR_ERR(reply); | |
335 | kfree(reply); | |
336 | return 0; | |
337 | } | |
338 | ||
339 | static unsigned int count_strings(const char *strings, unsigned int len) | |
340 | { | |
341 | unsigned int num; | |
342 | const char *p; | |
343 | ||
344 | for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1) | |
345 | num++; | |
346 | ||
347 | return num; | |
348 | } | |
349 | ||
350 | /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */ | |
351 | static char *join(const char *dir, const char *name) | |
352 | { | |
353 | char *buffer; | |
354 | ||
355 | if (strlen(name) == 0) | |
a144ff09 | 356 | buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir); |
4bac07c9 | 357 | else |
a144ff09 | 358 | buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name); |
4bac07c9 JF |
359 | return (!buffer) ? ERR_PTR(-ENOMEM) : buffer; |
360 | } | |
361 | ||
362 | static char **split(char *strings, unsigned int len, unsigned int *num) | |
363 | { | |
364 | char *p, **ret; | |
365 | ||
366 | /* Count the strings. */ | |
367 | *num = count_strings(strings, len); | |
368 | ||
369 | /* Transfer to one big alloc for easy freeing. */ | |
a144ff09 | 370 | ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH); |
4bac07c9 JF |
371 | if (!ret) { |
372 | kfree(strings); | |
373 | return ERR_PTR(-ENOMEM); | |
374 | } | |
375 | memcpy(&ret[*num], strings, len); | |
376 | kfree(strings); | |
377 | ||
378 | strings = (char *)&ret[*num]; | |
379 | for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1) | |
380 | ret[(*num)++] = p; | |
381 | ||
382 | return ret; | |
383 | } | |
384 | ||
385 | char **xenbus_directory(struct xenbus_transaction t, | |
386 | const char *dir, const char *node, unsigned int *num) | |
387 | { | |
388 | char *strings, *path; | |
389 | unsigned int len; | |
390 | ||
391 | path = join(dir, node); | |
392 | if (IS_ERR(path)) | |
393 | return (char **)path; | |
394 | ||
395 | strings = xs_single(t, XS_DIRECTORY, path, &len); | |
396 | kfree(path); | |
397 | if (IS_ERR(strings)) | |
398 | return (char **)strings; | |
399 | ||
400 | return split(strings, len, num); | |
401 | } | |
402 | EXPORT_SYMBOL_GPL(xenbus_directory); | |
403 | ||
404 | /* Check if a path exists. Return 1 if it does. */ | |
405 | int xenbus_exists(struct xenbus_transaction t, | |
406 | const char *dir, const char *node) | |
407 | { | |
408 | char **d; | |
409 | int dir_n; | |
410 | ||
411 | d = xenbus_directory(t, dir, node, &dir_n); | |
412 | if (IS_ERR(d)) | |
413 | return 0; | |
414 | kfree(d); | |
415 | return 1; | |
416 | } | |
417 | EXPORT_SYMBOL_GPL(xenbus_exists); | |
418 | ||
419 | /* Get the value of a single file. | |
420 | * Returns a kmalloced value: call free() on it after use. | |
421 | * len indicates length in bytes. | |
422 | */ | |
423 | void *xenbus_read(struct xenbus_transaction t, | |
424 | const char *dir, const char *node, unsigned int *len) | |
425 | { | |
426 | char *path; | |
427 | void *ret; | |
428 | ||
429 | path = join(dir, node); | |
430 | if (IS_ERR(path)) | |
431 | return (void *)path; | |
432 | ||
433 | ret = xs_single(t, XS_READ, path, len); | |
434 | kfree(path); | |
435 | return ret; | |
436 | } | |
437 | EXPORT_SYMBOL_GPL(xenbus_read); | |
438 | ||
439 | /* Write the value of a single file. | |
440 | * Returns -err on failure. | |
441 | */ | |
442 | int xenbus_write(struct xenbus_transaction t, | |
443 | const char *dir, const char *node, const char *string) | |
444 | { | |
445 | const char *path; | |
446 | struct kvec iovec[2]; | |
447 | int ret; | |
448 | ||
449 | path = join(dir, node); | |
450 | if (IS_ERR(path)) | |
451 | return PTR_ERR(path); | |
452 | ||
453 | iovec[0].iov_base = (void *)path; | |
454 | iovec[0].iov_len = strlen(path) + 1; | |
455 | iovec[1].iov_base = (void *)string; | |
456 | iovec[1].iov_len = strlen(string); | |
457 | ||
458 | ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL)); | |
459 | kfree(path); | |
460 | return ret; | |
461 | } | |
462 | EXPORT_SYMBOL_GPL(xenbus_write); | |
463 | ||
464 | /* Create a new directory. */ | |
465 | int xenbus_mkdir(struct xenbus_transaction t, | |
466 | const char *dir, const char *node) | |
467 | { | |
468 | char *path; | |
469 | int ret; | |
470 | ||
471 | path = join(dir, node); | |
472 | if (IS_ERR(path)) | |
473 | return PTR_ERR(path); | |
474 | ||
475 | ret = xs_error(xs_single(t, XS_MKDIR, path, NULL)); | |
476 | kfree(path); | |
477 | return ret; | |
478 | } | |
479 | EXPORT_SYMBOL_GPL(xenbus_mkdir); | |
480 | ||
481 | /* Destroy a file or directory (directories must be empty). */ | |
482 | int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node) | |
483 | { | |
484 | char *path; | |
485 | int ret; | |
486 | ||
487 | path = join(dir, node); | |
488 | if (IS_ERR(path)) | |
489 | return PTR_ERR(path); | |
490 | ||
491 | ret = xs_error(xs_single(t, XS_RM, path, NULL)); | |
492 | kfree(path); | |
493 | return ret; | |
494 | } | |
495 | EXPORT_SYMBOL_GPL(xenbus_rm); | |
496 | ||
497 | /* Start a transaction: changes by others will not be seen during this | |
498 | * transaction, and changes will not be visible to others until end. | |
499 | */ | |
500 | int xenbus_transaction_start(struct xenbus_transaction *t) | |
501 | { | |
502 | char *id_str; | |
503 | ||
4c31a781 | 504 | transaction_start(); |
4bac07c9 JF |
505 | |
506 | id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL); | |
507 | if (IS_ERR(id_str)) { | |
4c31a781 | 508 | transaction_end(); |
4bac07c9 JF |
509 | return PTR_ERR(id_str); |
510 | } | |
511 | ||
512 | t->id = simple_strtoul(id_str, NULL, 0); | |
513 | kfree(id_str); | |
514 | return 0; | |
515 | } | |
516 | EXPORT_SYMBOL_GPL(xenbus_transaction_start); | |
517 | ||
518 | /* End a transaction. | |
519 | * If abandon is true, transaction is discarded instead of committed. | |
520 | */ | |
521 | int xenbus_transaction_end(struct xenbus_transaction t, int abort) | |
522 | { | |
523 | char abortstr[2]; | |
524 | int err; | |
525 | ||
526 | if (abort) | |
527 | strcpy(abortstr, "F"); | |
528 | else | |
529 | strcpy(abortstr, "T"); | |
530 | ||
531 | err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL)); | |
532 | ||
4c31a781 | 533 | transaction_end(); |
4bac07c9 JF |
534 | |
535 | return err; | |
536 | } | |
537 | EXPORT_SYMBOL_GPL(xenbus_transaction_end); | |
538 | ||
539 | /* Single read and scanf: returns -errno or num scanned. */ | |
540 | int xenbus_scanf(struct xenbus_transaction t, | |
541 | const char *dir, const char *node, const char *fmt, ...) | |
542 | { | |
543 | va_list ap; | |
544 | int ret; | |
545 | char *val; | |
546 | ||
547 | val = xenbus_read(t, dir, node, NULL); | |
548 | if (IS_ERR(val)) | |
549 | return PTR_ERR(val); | |
550 | ||
551 | va_start(ap, fmt); | |
552 | ret = vsscanf(val, fmt, ap); | |
553 | va_end(ap); | |
554 | kfree(val); | |
555 | /* Distinctive errno. */ | |
556 | if (ret == 0) | |
557 | return -ERANGE; | |
558 | return ret; | |
559 | } | |
560 | EXPORT_SYMBOL_GPL(xenbus_scanf); | |
561 | ||
9c53a179 JG |
562 | /* Read an (optional) unsigned value. */ |
563 | unsigned int xenbus_read_unsigned(const char *dir, const char *node, | |
564 | unsigned int default_val) | |
565 | { | |
566 | unsigned int val; | |
567 | int ret; | |
568 | ||
569 | ret = xenbus_scanf(XBT_NIL, dir, node, "%u", &val); | |
570 | if (ret <= 0) | |
571 | val = default_val; | |
572 | ||
573 | return val; | |
574 | } | |
575 | EXPORT_SYMBOL_GPL(xenbus_read_unsigned); | |
576 | ||
4bac07c9 JF |
577 | /* Single printf and write: returns -errno or 0. */ |
578 | int xenbus_printf(struct xenbus_transaction t, | |
579 | const char *dir, const char *node, const char *fmt, ...) | |
580 | { | |
581 | va_list ap; | |
582 | int ret; | |
a800651e | 583 | char *buf; |
4bac07c9 JF |
584 | |
585 | va_start(ap, fmt); | |
a800651e | 586 | buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap); |
4bac07c9 JF |
587 | va_end(ap); |
588 | ||
a800651e IC |
589 | if (!buf) |
590 | return -ENOMEM; | |
591 | ||
592 | ret = xenbus_write(t, dir, node, buf); | |
4bac07c9 | 593 | |
a800651e | 594 | kfree(buf); |
4bac07c9 JF |
595 | |
596 | return ret; | |
597 | } | |
598 | EXPORT_SYMBOL_GPL(xenbus_printf); | |
599 | ||
600 | /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */ | |
601 | int xenbus_gather(struct xenbus_transaction t, const char *dir, ...) | |
602 | { | |
603 | va_list ap; | |
604 | const char *name; | |
605 | int ret = 0; | |
606 | ||
607 | va_start(ap, dir); | |
608 | while (ret == 0 && (name = va_arg(ap, char *)) != NULL) { | |
609 | const char *fmt = va_arg(ap, char *); | |
610 | void *result = va_arg(ap, void *); | |
611 | char *p; | |
612 | ||
613 | p = xenbus_read(t, dir, name, NULL); | |
614 | if (IS_ERR(p)) { | |
615 | ret = PTR_ERR(p); | |
616 | break; | |
617 | } | |
618 | if (fmt) { | |
619 | if (sscanf(p, fmt, result) == 0) | |
620 | ret = -EINVAL; | |
621 | kfree(p); | |
622 | } else | |
623 | *(char **)result = p; | |
624 | } | |
625 | va_end(ap); | |
626 | return ret; | |
627 | } | |
628 | EXPORT_SYMBOL_GPL(xenbus_gather); | |
629 | ||
630 | static int xs_watch(const char *path, const char *token) | |
631 | { | |
632 | struct kvec iov[2]; | |
633 | ||
634 | iov[0].iov_base = (void *)path; | |
635 | iov[0].iov_len = strlen(path) + 1; | |
636 | iov[1].iov_base = (void *)token; | |
637 | iov[1].iov_len = strlen(token) + 1; | |
638 | ||
639 | return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov, | |
640 | ARRAY_SIZE(iov), NULL)); | |
641 | } | |
642 | ||
643 | static int xs_unwatch(const char *path, const char *token) | |
644 | { | |
645 | struct kvec iov[2]; | |
646 | ||
647 | iov[0].iov_base = (char *)path; | |
648 | iov[0].iov_len = strlen(path) + 1; | |
649 | iov[1].iov_base = (char *)token; | |
650 | iov[1].iov_len = strlen(token) + 1; | |
651 | ||
652 | return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov, | |
653 | ARRAY_SIZE(iov), NULL)); | |
654 | } | |
655 | ||
656 | static struct xenbus_watch *find_watch(const char *token) | |
657 | { | |
658 | struct xenbus_watch *i, *cmp; | |
659 | ||
660 | cmp = (void *)simple_strtoul(token, NULL, 16); | |
661 | ||
662 | list_for_each_entry(i, &watches, list) | |
663 | if (i == cmp) | |
664 | return i; | |
665 | ||
666 | return NULL; | |
667 | } | |
cb6b6df1 KRW |
668 | /* |
669 | * Certain older XenBus toolstack cannot handle reading values that are | |
670 | * not populated. Some Xen 3.4 installation are incapable of doing this | |
671 | * so if we are running on anything older than 4 do not attempt to read | |
672 | * control/platform-feature-xs_reset_watches. | |
673 | */ | |
e9d1aa05 | 674 | static bool xen_strict_xenbus_quirk(void) |
cb6b6df1 | 675 | { |
7644bdac | 676 | #ifdef CONFIG_X86 |
cb6b6df1 KRW |
677 | uint32_t eax, ebx, ecx, edx, base; |
678 | ||
679 | base = xen_cpuid_base(); | |
680 | cpuid(base + 1, &eax, &ebx, &ecx, &edx); | |
4bac07c9 | 681 | |
cb6b6df1 KRW |
682 | if ((eax >> 16) < 4) |
683 | return true; | |
7644bdac | 684 | #endif |
cb6b6df1 KRW |
685 | return false; |
686 | ||
687 | } | |
254d1a3f OH |
688 | static void xs_reset_watches(void) |
689 | { | |
999c9af9 | 690 | int err; |
254d1a3f | 691 | |
ecc635f9 | 692 | if (!xen_hvm_domain() || xen_initial_domain()) |
254d1a3f OH |
693 | return; |
694 | ||
cb6b6df1 KRW |
695 | if (xen_strict_xenbus_quirk()) |
696 | return; | |
697 | ||
999c9af9 JG |
698 | if (!xenbus_read_unsigned("control", |
699 | "platform-feature-xs_reset_watches", 0)) | |
254d1a3f OH |
700 | return; |
701 | ||
702 | err = xs_error(xs_single(XBT_NIL, XS_RESET_WATCHES, "", NULL)); | |
703 | if (err && err != -EEXIST) | |
283c0972 | 704 | pr_warn("xs_reset_watches failed: %d\n", err); |
254d1a3f OH |
705 | } |
706 | ||
4bac07c9 JF |
707 | /* Register callback to watch this node. */ |
708 | int register_xenbus_watch(struct xenbus_watch *watch) | |
709 | { | |
710 | /* Pointer in ascii is the token. */ | |
711 | char token[sizeof(watch) * 2 + 1]; | |
712 | int err; | |
713 | ||
714 | sprintf(token, "%lX", (long)watch); | |
715 | ||
716 | down_read(&xs_state.watch_mutex); | |
717 | ||
718 | spin_lock(&watches_lock); | |
719 | BUG_ON(find_watch(token)); | |
720 | list_add(&watch->list, &watches); | |
721 | spin_unlock(&watches_lock); | |
722 | ||
723 | err = xs_watch(watch->node, token); | |
724 | ||
c4c303c7 | 725 | if (err) { |
4bac07c9 JF |
726 | spin_lock(&watches_lock); |
727 | list_del(&watch->list); | |
728 | spin_unlock(&watches_lock); | |
729 | } | |
730 | ||
731 | up_read(&xs_state.watch_mutex); | |
732 | ||
733 | return err; | |
734 | } | |
735 | EXPORT_SYMBOL_GPL(register_xenbus_watch); | |
736 | ||
737 | void unregister_xenbus_watch(struct xenbus_watch *watch) | |
738 | { | |
739 | struct xs_stored_msg *msg, *tmp; | |
740 | char token[sizeof(watch) * 2 + 1]; | |
741 | int err; | |
742 | ||
743 | sprintf(token, "%lX", (long)watch); | |
744 | ||
745 | down_read(&xs_state.watch_mutex); | |
746 | ||
747 | spin_lock(&watches_lock); | |
748 | BUG_ON(!find_watch(token)); | |
749 | list_del(&watch->list); | |
750 | spin_unlock(&watches_lock); | |
751 | ||
752 | err = xs_unwatch(watch->node, token); | |
753 | if (err) | |
283c0972 | 754 | pr_warn("Failed to release watch %s: %i\n", watch->node, err); |
4bac07c9 JF |
755 | |
756 | up_read(&xs_state.watch_mutex); | |
757 | ||
758 | /* Make sure there are no callbacks running currently (unless | |
759 | its us) */ | |
760 | if (current->pid != xenwatch_pid) | |
761 | mutex_lock(&xenwatch_mutex); | |
762 | ||
763 | /* Cancel pending watch events. */ | |
764 | spin_lock(&watch_events_lock); | |
765 | list_for_each_entry_safe(msg, tmp, &watch_events, list) { | |
766 | if (msg->u.watch.handle != watch) | |
767 | continue; | |
768 | list_del(&msg->list); | |
769 | kfree(msg->u.watch.vec); | |
770 | kfree(msg); | |
771 | } | |
772 | spin_unlock(&watch_events_lock); | |
773 | ||
774 | if (current->pid != xenwatch_pid) | |
775 | mutex_unlock(&xenwatch_mutex); | |
776 | } | |
777 | EXPORT_SYMBOL_GPL(unregister_xenbus_watch); | |
778 | ||
779 | void xs_suspend(void) | |
780 | { | |
4c31a781 | 781 | transaction_suspend(); |
4bac07c9 JF |
782 | down_write(&xs_state.watch_mutex); |
783 | mutex_lock(&xs_state.request_mutex); | |
784 | mutex_lock(&xs_state.response_mutex); | |
785 | } | |
786 | ||
787 | void xs_resume(void) | |
788 | { | |
789 | struct xenbus_watch *watch; | |
790 | char token[sizeof(watch) * 2 + 1]; | |
791 | ||
de5b31bd IC |
792 | xb_init_comms(); |
793 | ||
4bac07c9 JF |
794 | mutex_unlock(&xs_state.response_mutex); |
795 | mutex_unlock(&xs_state.request_mutex); | |
4c31a781 | 796 | transaction_resume(); |
4bac07c9 JF |
797 | |
798 | /* No need for watches_lock: the watch_mutex is sufficient. */ | |
799 | list_for_each_entry(watch, &watches, list) { | |
800 | sprintf(token, "%lX", (long)watch); | |
801 | xs_watch(watch->node, token); | |
802 | } | |
803 | ||
804 | up_write(&xs_state.watch_mutex); | |
805 | } | |
806 | ||
807 | void xs_suspend_cancel(void) | |
808 | { | |
809 | mutex_unlock(&xs_state.response_mutex); | |
810 | mutex_unlock(&xs_state.request_mutex); | |
811 | up_write(&xs_state.watch_mutex); | |
4c31a781 | 812 | mutex_unlock(&xs_state.transaction_mutex); |
4bac07c9 JF |
813 | } |
814 | ||
815 | static int xenwatch_thread(void *unused) | |
816 | { | |
817 | struct list_head *ent; | |
818 | struct xs_stored_msg *msg; | |
819 | ||
820 | for (;;) { | |
821 | wait_event_interruptible(watch_events_waitq, | |
822 | !list_empty(&watch_events)); | |
823 | ||
824 | if (kthread_should_stop()) | |
825 | break; | |
826 | ||
827 | mutex_lock(&xenwatch_mutex); | |
828 | ||
829 | spin_lock(&watch_events_lock); | |
830 | ent = watch_events.next; | |
831 | if (ent != &watch_events) | |
832 | list_del(ent); | |
833 | spin_unlock(&watch_events_lock); | |
834 | ||
835 | if (ent != &watch_events) { | |
836 | msg = list_entry(ent, struct xs_stored_msg, list); | |
837 | msg->u.watch.handle->callback( | |
838 | msg->u.watch.handle, | |
839 | (const char **)msg->u.watch.vec, | |
840 | msg->u.watch.vec_size); | |
841 | kfree(msg->u.watch.vec); | |
842 | kfree(msg); | |
843 | } | |
844 | ||
845 | mutex_unlock(&xenwatch_mutex); | |
846 | } | |
847 | ||
848 | return 0; | |
849 | } | |
850 | ||
851 | static int process_msg(void) | |
852 | { | |
853 | struct xs_stored_msg *msg; | |
854 | char *body; | |
855 | int err; | |
856 | ||
857 | /* | |
858 | * We must disallow save/restore while reading a xenstore message. | |
859 | * A partial read across s/r leaves us out of sync with xenstored. | |
860 | */ | |
861 | for (;;) { | |
862 | err = xb_wait_for_data_to_read(); | |
863 | if (err) | |
864 | return err; | |
865 | mutex_lock(&xs_state.response_mutex); | |
866 | if (xb_data_to_read()) | |
867 | break; | |
868 | /* We raced with save/restore: pending data 'disappeared'. */ | |
869 | mutex_unlock(&xs_state.response_mutex); | |
870 | } | |
871 | ||
872 | ||
a144ff09 | 873 | msg = kmalloc(sizeof(*msg), GFP_NOIO | __GFP_HIGH); |
4bac07c9 JF |
874 | if (msg == NULL) { |
875 | err = -ENOMEM; | |
876 | goto out; | |
877 | } | |
878 | ||
879 | err = xb_read(&msg->hdr, sizeof(msg->hdr)); | |
880 | if (err) { | |
881 | kfree(msg); | |
882 | goto out; | |
883 | } | |
884 | ||
9e7860ce IC |
885 | if (msg->hdr.len > XENSTORE_PAYLOAD_MAX) { |
886 | kfree(msg); | |
887 | err = -EINVAL; | |
888 | goto out; | |
889 | } | |
890 | ||
a144ff09 | 891 | body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH); |
4bac07c9 JF |
892 | if (body == NULL) { |
893 | kfree(msg); | |
894 | err = -ENOMEM; | |
895 | goto out; | |
896 | } | |
897 | ||
898 | err = xb_read(body, msg->hdr.len); | |
899 | if (err) { | |
900 | kfree(body); | |
901 | kfree(msg); | |
902 | goto out; | |
903 | } | |
904 | body[msg->hdr.len] = '\0'; | |
905 | ||
906 | if (msg->hdr.type == XS_WATCH_EVENT) { | |
907 | msg->u.watch.vec = split(body, msg->hdr.len, | |
908 | &msg->u.watch.vec_size); | |
909 | if (IS_ERR(msg->u.watch.vec)) { | |
4bac07c9 | 910 | err = PTR_ERR(msg->u.watch.vec); |
98ac0e53 | 911 | kfree(msg); |
4bac07c9 JF |
912 | goto out; |
913 | } | |
914 | ||
915 | spin_lock(&watches_lock); | |
916 | msg->u.watch.handle = find_watch( | |
917 | msg->u.watch.vec[XS_WATCH_TOKEN]); | |
918 | if (msg->u.watch.handle != NULL) { | |
919 | spin_lock(&watch_events_lock); | |
920 | list_add_tail(&msg->list, &watch_events); | |
921 | wake_up(&watch_events_waitq); | |
922 | spin_unlock(&watch_events_lock); | |
923 | } else { | |
924 | kfree(msg->u.watch.vec); | |
925 | kfree(msg); | |
926 | } | |
927 | spin_unlock(&watches_lock); | |
928 | } else { | |
929 | msg->u.reply.body = body; | |
930 | spin_lock(&xs_state.reply_lock); | |
931 | list_add_tail(&msg->list, &xs_state.reply_list); | |
932 | spin_unlock(&xs_state.reply_lock); | |
933 | wake_up(&xs_state.reply_waitq); | |
934 | } | |
935 | ||
936 | out: | |
937 | mutex_unlock(&xs_state.response_mutex); | |
938 | return err; | |
939 | } | |
940 | ||
941 | static int xenbus_thread(void *unused) | |
942 | { | |
943 | int err; | |
944 | ||
945 | for (;;) { | |
946 | err = process_msg(); | |
947 | if (err) | |
283c0972 | 948 | pr_warn("error %d while reading message\n", err); |
4bac07c9 JF |
949 | if (kthread_should_stop()) |
950 | break; | |
951 | } | |
952 | ||
953 | return 0; | |
954 | } | |
955 | ||
956 | int xs_init(void) | |
957 | { | |
958 | int err; | |
959 | struct task_struct *task; | |
960 | ||
961 | INIT_LIST_HEAD(&xs_state.reply_list); | |
962 | spin_lock_init(&xs_state.reply_lock); | |
963 | init_waitqueue_head(&xs_state.reply_waitq); | |
964 | ||
965 | mutex_init(&xs_state.request_mutex); | |
966 | mutex_init(&xs_state.response_mutex); | |
4c31a781 | 967 | mutex_init(&xs_state.transaction_mutex); |
4bac07c9 | 968 | init_rwsem(&xs_state.watch_mutex); |
4c31a781 IC |
969 | atomic_set(&xs_state.transaction_count, 0); |
970 | init_waitqueue_head(&xs_state.transaction_wq); | |
4bac07c9 JF |
971 | |
972 | /* Initialize the shared memory rings to talk to xenstored */ | |
973 | err = xb_init_comms(); | |
974 | if (err) | |
975 | return err; | |
976 | ||
977 | task = kthread_run(xenwatch_thread, NULL, "xenwatch"); | |
978 | if (IS_ERR(task)) | |
979 | return PTR_ERR(task); | |
980 | xenwatch_pid = task->pid; | |
981 | ||
982 | task = kthread_run(xenbus_thread, NULL, "xenbus"); | |
983 | if (IS_ERR(task)) | |
984 | return PTR_ERR(task); | |
985 | ||
254d1a3f OH |
986 | /* shutdown watches for kexec boot */ |
987 | xs_reset_watches(); | |
988 | ||
4bac07c9 JF |
989 | return 0; |
990 | } |