]>
Commit | Line | Data |
---|---|---|
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 | ||
34 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
35 | ||
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/reboot.h> | |
47 | #include <linux/rwsem.h> | |
48 | #include <linux/mutex.h> | |
49 | #include <asm/xen/hypervisor.h> | |
50 | #include <xen/xenbus.h> | |
51 | #include <xen/xen.h> | |
52 | #include "xenbus.h" | |
53 | ||
54 | /* | |
55 | * Framework to protect suspend/resume handling against normal Xenstore | |
56 | * message handling: | |
57 | * During suspend/resume there must be no open transaction and no pending | |
58 | * Xenstore request. | |
59 | * New watch events happening in this time can be ignored by firing all watches | |
60 | * after resume. | |
61 | */ | |
62 | ||
63 | /* Lock protecting enter/exit critical region. */ | |
64 | static DEFINE_SPINLOCK(xs_state_lock); | |
65 | /* Number of users in critical region (protected by xs_state_lock). */ | |
66 | static unsigned int xs_state_users; | |
67 | /* Suspend handler waiting or already active (protected by xs_state_lock)? */ | |
68 | static int xs_suspend_active; | |
69 | /* Unique Xenstore request id (protected by xs_state_lock). */ | |
70 | static uint32_t xs_request_id; | |
71 | ||
72 | /* Wait queue for all callers waiting for critical region to become usable. */ | |
73 | static DECLARE_WAIT_QUEUE_HEAD(xs_state_enter_wq); | |
74 | /* Wait queue for suspend handling waiting for critical region being empty. */ | |
75 | static DECLARE_WAIT_QUEUE_HEAD(xs_state_exit_wq); | |
76 | ||
77 | /* List of registered watches, and a lock to protect it. */ | |
78 | static LIST_HEAD(watches); | |
79 | static DEFINE_SPINLOCK(watches_lock); | |
80 | ||
81 | /* List of pending watch callback events, and a lock to protect it. */ | |
82 | static LIST_HEAD(watch_events); | |
83 | static DEFINE_SPINLOCK(watch_events_lock); | |
84 | ||
85 | /* Protect watch (de)register against save/restore. */ | |
86 | static DECLARE_RWSEM(xs_watch_rwsem); | |
87 | ||
88 | /* | |
89 | * Details of the xenwatch callback kernel thread. The thread waits on the | |
90 | * watch_events_waitq for work to do (queued on watch_events list). When it | |
91 | * wakes up it acquires the xenwatch_mutex before reading the list and | |
92 | * carrying out work. | |
93 | */ | |
94 | static pid_t xenwatch_pid; | |
95 | static DEFINE_MUTEX(xenwatch_mutex); | |
96 | static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq); | |
97 | ||
98 | static void xs_suspend_enter(void) | |
99 | { | |
100 | spin_lock(&xs_state_lock); | |
101 | xs_suspend_active++; | |
102 | spin_unlock(&xs_state_lock); | |
103 | wait_event(xs_state_exit_wq, xs_state_users == 0); | |
104 | } | |
105 | ||
106 | static void xs_suspend_exit(void) | |
107 | { | |
108 | spin_lock(&xs_state_lock); | |
109 | xs_suspend_active--; | |
110 | spin_unlock(&xs_state_lock); | |
111 | wake_up_all(&xs_state_enter_wq); | |
112 | } | |
113 | ||
114 | static uint32_t xs_request_enter(struct xb_req_data *req) | |
115 | { | |
116 | uint32_t rq_id; | |
117 | ||
118 | req->type = req->msg.type; | |
119 | ||
120 | spin_lock(&xs_state_lock); | |
121 | ||
122 | while (!xs_state_users && xs_suspend_active) { | |
123 | spin_unlock(&xs_state_lock); | |
124 | wait_event(xs_state_enter_wq, xs_suspend_active == 0); | |
125 | spin_lock(&xs_state_lock); | |
126 | } | |
127 | ||
128 | if (req->type == XS_TRANSACTION_START) | |
129 | xs_state_users++; | |
130 | xs_state_users++; | |
131 | rq_id = xs_request_id++; | |
132 | ||
133 | spin_unlock(&xs_state_lock); | |
134 | ||
135 | return rq_id; | |
136 | } | |
137 | ||
138 | void xs_request_exit(struct xb_req_data *req) | |
139 | { | |
140 | spin_lock(&xs_state_lock); | |
141 | xs_state_users--; | |
142 | if ((req->type == XS_TRANSACTION_START && req->msg.type == XS_ERROR) || | |
143 | req->type == XS_TRANSACTION_END) | |
144 | xs_state_users--; | |
145 | spin_unlock(&xs_state_lock); | |
146 | ||
147 | if (xs_suspend_active && !xs_state_users) | |
148 | wake_up(&xs_state_exit_wq); | |
149 | } | |
150 | ||
151 | static int get_error(const char *errorstring) | |
152 | { | |
153 | unsigned int i; | |
154 | ||
155 | for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) { | |
156 | if (i == ARRAY_SIZE(xsd_errors) - 1) { | |
157 | pr_warn("xen store gave: unknown error %s\n", | |
158 | errorstring); | |
159 | return EINVAL; | |
160 | } | |
161 | } | |
162 | return xsd_errors[i].errnum; | |
163 | } | |
164 | ||
165 | static bool xenbus_ok(void) | |
166 | { | |
167 | switch (xen_store_domain_type) { | |
168 | case XS_LOCAL: | |
169 | switch (system_state) { | |
170 | case SYSTEM_POWER_OFF: | |
171 | case SYSTEM_RESTART: | |
172 | case SYSTEM_HALT: | |
173 | return false; | |
174 | default: | |
175 | break; | |
176 | } | |
177 | return true; | |
178 | case XS_PV: | |
179 | case XS_HVM: | |
180 | /* FIXME: Could check that the remote domain is alive, | |
181 | * but it is normally initial domain. */ | |
182 | return true; | |
183 | default: | |
184 | break; | |
185 | } | |
186 | return false; | |
187 | } | |
188 | ||
189 | static bool test_reply(struct xb_req_data *req) | |
190 | { | |
191 | if (req->state == xb_req_state_got_reply || !xenbus_ok()) | |
192 | return true; | |
193 | ||
194 | /* Make sure to reread req->state each time. */ | |
195 | barrier(); | |
196 | ||
197 | return false; | |
198 | } | |
199 | ||
200 | static void *read_reply(struct xb_req_data *req) | |
201 | { | |
202 | while (req->state != xb_req_state_got_reply) { | |
203 | wait_event(req->wq, test_reply(req)); | |
204 | ||
205 | if (!xenbus_ok()) | |
206 | /* | |
207 | * If we are in the process of being shut-down there is | |
208 | * no point of trying to contact XenBus - it is either | |
209 | * killed (xenstored application) or the other domain | |
210 | * has been killed or is unreachable. | |
211 | */ | |
212 | return ERR_PTR(-EIO); | |
213 | if (req->err) | |
214 | return ERR_PTR(req->err); | |
215 | ||
216 | } | |
217 | ||
218 | return req->body; | |
219 | } | |
220 | ||
221 | static void xs_send(struct xb_req_data *req, struct xsd_sockmsg *msg) | |
222 | { | |
223 | bool notify; | |
224 | ||
225 | req->msg = *msg; | |
226 | req->err = 0; | |
227 | req->state = xb_req_state_queued; | |
228 | init_waitqueue_head(&req->wq); | |
229 | ||
230 | req->msg.req_id = xs_request_enter(req); | |
231 | ||
232 | mutex_lock(&xb_write_mutex); | |
233 | list_add_tail(&req->list, &xb_write_list); | |
234 | notify = list_is_singular(&xb_write_list); | |
235 | mutex_unlock(&xb_write_mutex); | |
236 | ||
237 | if (notify) | |
238 | wake_up(&xb_waitq); | |
239 | } | |
240 | ||
241 | static void *xs_wait_for_reply(struct xb_req_data *req, struct xsd_sockmsg *msg) | |
242 | { | |
243 | void *ret; | |
244 | ||
245 | ret = read_reply(req); | |
246 | ||
247 | xs_request_exit(req); | |
248 | ||
249 | msg->type = req->msg.type; | |
250 | msg->len = req->msg.len; | |
251 | ||
252 | mutex_lock(&xb_write_mutex); | |
253 | if (req->state == xb_req_state_queued || | |
254 | req->state == xb_req_state_wait_reply) | |
255 | req->state = xb_req_state_aborted; | |
256 | else | |
257 | kfree(req); | |
258 | mutex_unlock(&xb_write_mutex); | |
259 | ||
260 | return ret; | |
261 | } | |
262 | ||
263 | static void xs_wake_up(struct xb_req_data *req) | |
264 | { | |
265 | wake_up(&req->wq); | |
266 | } | |
267 | ||
268 | int xenbus_dev_request_and_reply(struct xsd_sockmsg *msg, void *par) | |
269 | { | |
270 | struct xb_req_data *req; | |
271 | struct kvec *vec; | |
272 | ||
273 | req = kmalloc(sizeof(*req) + sizeof(*vec), GFP_KERNEL); | |
274 | if (!req) | |
275 | return -ENOMEM; | |
276 | ||
277 | vec = (struct kvec *)(req + 1); | |
278 | vec->iov_len = msg->len; | |
279 | vec->iov_base = msg + 1; | |
280 | ||
281 | req->vec = vec; | |
282 | req->num_vecs = 1; | |
283 | req->cb = xenbus_dev_queue_reply; | |
284 | req->par = par; | |
285 | ||
286 | xs_send(req, msg); | |
287 | ||
288 | return 0; | |
289 | } | |
290 | EXPORT_SYMBOL(xenbus_dev_request_and_reply); | |
291 | ||
292 | /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */ | |
293 | static void *xs_talkv(struct xenbus_transaction t, | |
294 | enum xsd_sockmsg_type type, | |
295 | const struct kvec *iovec, | |
296 | unsigned int num_vecs, | |
297 | unsigned int *len) | |
298 | { | |
299 | struct xb_req_data *req; | |
300 | struct xsd_sockmsg msg; | |
301 | void *ret = NULL; | |
302 | unsigned int i; | |
303 | int err; | |
304 | ||
305 | req = kmalloc(sizeof(*req), GFP_NOIO | __GFP_HIGH); | |
306 | if (!req) | |
307 | return ERR_PTR(-ENOMEM); | |
308 | ||
309 | req->vec = iovec; | |
310 | req->num_vecs = num_vecs; | |
311 | req->cb = xs_wake_up; | |
312 | ||
313 | msg.tx_id = t.id; | |
314 | msg.type = type; | |
315 | msg.len = 0; | |
316 | for (i = 0; i < num_vecs; i++) | |
317 | msg.len += iovec[i].iov_len; | |
318 | ||
319 | xs_send(req, &msg); | |
320 | ||
321 | ret = xs_wait_for_reply(req, &msg); | |
322 | if (len) | |
323 | *len = msg.len; | |
324 | ||
325 | if (IS_ERR(ret)) | |
326 | return ret; | |
327 | ||
328 | if (msg.type == XS_ERROR) { | |
329 | err = get_error(ret); | |
330 | kfree(ret); | |
331 | return ERR_PTR(-err); | |
332 | } | |
333 | ||
334 | if (msg.type != type) { | |
335 | pr_warn_ratelimited("unexpected type [%d], expected [%d]\n", | |
336 | msg.type, type); | |
337 | kfree(ret); | |
338 | return ERR_PTR(-EINVAL); | |
339 | } | |
340 | return ret; | |
341 | } | |
342 | ||
343 | /* Simplified version of xs_talkv: single message. */ | |
344 | static void *xs_single(struct xenbus_transaction t, | |
345 | enum xsd_sockmsg_type type, | |
346 | const char *string, | |
347 | unsigned int *len) | |
348 | { | |
349 | struct kvec iovec; | |
350 | ||
351 | iovec.iov_base = (void *)string; | |
352 | iovec.iov_len = strlen(string) + 1; | |
353 | return xs_talkv(t, type, &iovec, 1, len); | |
354 | } | |
355 | ||
356 | /* Many commands only need an ack, don't care what it says. */ | |
357 | static int xs_error(char *reply) | |
358 | { | |
359 | if (IS_ERR(reply)) | |
360 | return PTR_ERR(reply); | |
361 | kfree(reply); | |
362 | return 0; | |
363 | } | |
364 | ||
365 | static unsigned int count_strings(const char *strings, unsigned int len) | |
366 | { | |
367 | unsigned int num; | |
368 | const char *p; | |
369 | ||
370 | for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1) | |
371 | num++; | |
372 | ||
373 | return num; | |
374 | } | |
375 | ||
376 | /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */ | |
377 | static char *join(const char *dir, const char *name) | |
378 | { | |
379 | char *buffer; | |
380 | ||
381 | if (strlen(name) == 0) | |
382 | buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s", dir); | |
383 | else | |
384 | buffer = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/%s", dir, name); | |
385 | return (!buffer) ? ERR_PTR(-ENOMEM) : buffer; | |
386 | } | |
387 | ||
388 | static char **split(char *strings, unsigned int len, unsigned int *num) | |
389 | { | |
390 | char *p, **ret; | |
391 | ||
392 | /* Count the strings. */ | |
393 | *num = count_strings(strings, len); | |
394 | ||
395 | /* Transfer to one big alloc for easy freeing. */ | |
396 | ret = kmalloc(*num * sizeof(char *) + len, GFP_NOIO | __GFP_HIGH); | |
397 | if (!ret) { | |
398 | kfree(strings); | |
399 | return ERR_PTR(-ENOMEM); | |
400 | } | |
401 | memcpy(&ret[*num], strings, len); | |
402 | kfree(strings); | |
403 | ||
404 | strings = (char *)&ret[*num]; | |
405 | for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1) | |
406 | ret[(*num)++] = p; | |
407 | ||
408 | return ret; | |
409 | } | |
410 | ||
411 | char **xenbus_directory(struct xenbus_transaction t, | |
412 | const char *dir, const char *node, unsigned int *num) | |
413 | { | |
414 | char *strings, *path; | |
415 | unsigned int len; | |
416 | ||
417 | path = join(dir, node); | |
418 | if (IS_ERR(path)) | |
419 | return (char **)path; | |
420 | ||
421 | strings = xs_single(t, XS_DIRECTORY, path, &len); | |
422 | kfree(path); | |
423 | if (IS_ERR(strings)) | |
424 | return (char **)strings; | |
425 | ||
426 | return split(strings, len, num); | |
427 | } | |
428 | EXPORT_SYMBOL_GPL(xenbus_directory); | |
429 | ||
430 | /* Check if a path exists. Return 1 if it does. */ | |
431 | int xenbus_exists(struct xenbus_transaction t, | |
432 | const char *dir, const char *node) | |
433 | { | |
434 | char **d; | |
435 | int dir_n; | |
436 | ||
437 | d = xenbus_directory(t, dir, node, &dir_n); | |
438 | if (IS_ERR(d)) | |
439 | return 0; | |
440 | kfree(d); | |
441 | return 1; | |
442 | } | |
443 | EXPORT_SYMBOL_GPL(xenbus_exists); | |
444 | ||
445 | /* Get the value of a single file. | |
446 | * Returns a kmalloced value: call free() on it after use. | |
447 | * len indicates length in bytes. | |
448 | */ | |
449 | void *xenbus_read(struct xenbus_transaction t, | |
450 | const char *dir, const char *node, unsigned int *len) | |
451 | { | |
452 | char *path; | |
453 | void *ret; | |
454 | ||
455 | path = join(dir, node); | |
456 | if (IS_ERR(path)) | |
457 | return (void *)path; | |
458 | ||
459 | ret = xs_single(t, XS_READ, path, len); | |
460 | kfree(path); | |
461 | return ret; | |
462 | } | |
463 | EXPORT_SYMBOL_GPL(xenbus_read); | |
464 | ||
465 | /* Write the value of a single file. | |
466 | * Returns -err on failure. | |
467 | */ | |
468 | int xenbus_write(struct xenbus_transaction t, | |
469 | const char *dir, const char *node, const char *string) | |
470 | { | |
471 | const char *path; | |
472 | struct kvec iovec[2]; | |
473 | int ret; | |
474 | ||
475 | path = join(dir, node); | |
476 | if (IS_ERR(path)) | |
477 | return PTR_ERR(path); | |
478 | ||
479 | iovec[0].iov_base = (void *)path; | |
480 | iovec[0].iov_len = strlen(path) + 1; | |
481 | iovec[1].iov_base = (void *)string; | |
482 | iovec[1].iov_len = strlen(string); | |
483 | ||
484 | ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL)); | |
485 | kfree(path); | |
486 | return ret; | |
487 | } | |
488 | EXPORT_SYMBOL_GPL(xenbus_write); | |
489 | ||
490 | /* Create a new directory. */ | |
491 | int xenbus_mkdir(struct xenbus_transaction t, | |
492 | const char *dir, const char *node) | |
493 | { | |
494 | char *path; | |
495 | int ret; | |
496 | ||
497 | path = join(dir, node); | |
498 | if (IS_ERR(path)) | |
499 | return PTR_ERR(path); | |
500 | ||
501 | ret = xs_error(xs_single(t, XS_MKDIR, path, NULL)); | |
502 | kfree(path); | |
503 | return ret; | |
504 | } | |
505 | EXPORT_SYMBOL_GPL(xenbus_mkdir); | |
506 | ||
507 | /* Destroy a file or directory (directories must be empty). */ | |
508 | int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node) | |
509 | { | |
510 | char *path; | |
511 | int ret; | |
512 | ||
513 | path = join(dir, node); | |
514 | if (IS_ERR(path)) | |
515 | return PTR_ERR(path); | |
516 | ||
517 | ret = xs_error(xs_single(t, XS_RM, path, NULL)); | |
518 | kfree(path); | |
519 | return ret; | |
520 | } | |
521 | EXPORT_SYMBOL_GPL(xenbus_rm); | |
522 | ||
523 | /* Start a transaction: changes by others will not be seen during this | |
524 | * transaction, and changes will not be visible to others until end. | |
525 | */ | |
526 | int xenbus_transaction_start(struct xenbus_transaction *t) | |
527 | { | |
528 | char *id_str; | |
529 | ||
530 | id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL); | |
531 | if (IS_ERR(id_str)) | |
532 | return PTR_ERR(id_str); | |
533 | ||
534 | t->id = simple_strtoul(id_str, NULL, 0); | |
535 | kfree(id_str); | |
536 | return 0; | |
537 | } | |
538 | EXPORT_SYMBOL_GPL(xenbus_transaction_start); | |
539 | ||
540 | /* End a transaction. | |
541 | * If abandon is true, transaction is discarded instead of committed. | |
542 | */ | |
543 | int xenbus_transaction_end(struct xenbus_transaction t, int abort) | |
544 | { | |
545 | char abortstr[2]; | |
546 | ||
547 | if (abort) | |
548 | strcpy(abortstr, "F"); | |
549 | else | |
550 | strcpy(abortstr, "T"); | |
551 | ||
552 | return xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL)); | |
553 | } | |
554 | EXPORT_SYMBOL_GPL(xenbus_transaction_end); | |
555 | ||
556 | /* Single read and scanf: returns -errno or num scanned. */ | |
557 | int xenbus_scanf(struct xenbus_transaction t, | |
558 | const char *dir, const char *node, const char *fmt, ...) | |
559 | { | |
560 | va_list ap; | |
561 | int ret; | |
562 | char *val; | |
563 | ||
564 | val = xenbus_read(t, dir, node, NULL); | |
565 | if (IS_ERR(val)) | |
566 | return PTR_ERR(val); | |
567 | ||
568 | va_start(ap, fmt); | |
569 | ret = vsscanf(val, fmt, ap); | |
570 | va_end(ap); | |
571 | kfree(val); | |
572 | /* Distinctive errno. */ | |
573 | if (ret == 0) | |
574 | return -ERANGE; | |
575 | return ret; | |
576 | } | |
577 | EXPORT_SYMBOL_GPL(xenbus_scanf); | |
578 | ||
579 | /* Read an (optional) unsigned value. */ | |
580 | unsigned int xenbus_read_unsigned(const char *dir, const char *node, | |
581 | unsigned int default_val) | |
582 | { | |
583 | unsigned int val; | |
584 | int ret; | |
585 | ||
586 | ret = xenbus_scanf(XBT_NIL, dir, node, "%u", &val); | |
587 | if (ret <= 0) | |
588 | val = default_val; | |
589 | ||
590 | return val; | |
591 | } | |
592 | EXPORT_SYMBOL_GPL(xenbus_read_unsigned); | |
593 | ||
594 | /* Single printf and write: returns -errno or 0. */ | |
595 | int xenbus_printf(struct xenbus_transaction t, | |
596 | const char *dir, const char *node, const char *fmt, ...) | |
597 | { | |
598 | va_list ap; | |
599 | int ret; | |
600 | char *buf; | |
601 | ||
602 | va_start(ap, fmt); | |
603 | buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap); | |
604 | va_end(ap); | |
605 | ||
606 | if (!buf) | |
607 | return -ENOMEM; | |
608 | ||
609 | ret = xenbus_write(t, dir, node, buf); | |
610 | ||
611 | kfree(buf); | |
612 | ||
613 | return ret; | |
614 | } | |
615 | EXPORT_SYMBOL_GPL(xenbus_printf); | |
616 | ||
617 | /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */ | |
618 | int xenbus_gather(struct xenbus_transaction t, const char *dir, ...) | |
619 | { | |
620 | va_list ap; | |
621 | const char *name; | |
622 | int ret = 0; | |
623 | ||
624 | va_start(ap, dir); | |
625 | while (ret == 0 && (name = va_arg(ap, char *)) != NULL) { | |
626 | const char *fmt = va_arg(ap, char *); | |
627 | void *result = va_arg(ap, void *); | |
628 | char *p; | |
629 | ||
630 | p = xenbus_read(t, dir, name, NULL); | |
631 | if (IS_ERR(p)) { | |
632 | ret = PTR_ERR(p); | |
633 | break; | |
634 | } | |
635 | if (fmt) { | |
636 | if (sscanf(p, fmt, result) == 0) | |
637 | ret = -EINVAL; | |
638 | kfree(p); | |
639 | } else | |
640 | *(char **)result = p; | |
641 | } | |
642 | va_end(ap); | |
643 | return ret; | |
644 | } | |
645 | EXPORT_SYMBOL_GPL(xenbus_gather); | |
646 | ||
647 | static int xs_watch(const char *path, const char *token) | |
648 | { | |
649 | struct kvec iov[2]; | |
650 | ||
651 | iov[0].iov_base = (void *)path; | |
652 | iov[0].iov_len = strlen(path) + 1; | |
653 | iov[1].iov_base = (void *)token; | |
654 | iov[1].iov_len = strlen(token) + 1; | |
655 | ||
656 | return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov, | |
657 | ARRAY_SIZE(iov), NULL)); | |
658 | } | |
659 | ||
660 | static int xs_unwatch(const char *path, const char *token) | |
661 | { | |
662 | struct kvec iov[2]; | |
663 | ||
664 | iov[0].iov_base = (char *)path; | |
665 | iov[0].iov_len = strlen(path) + 1; | |
666 | iov[1].iov_base = (char *)token; | |
667 | iov[1].iov_len = strlen(token) + 1; | |
668 | ||
669 | return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov, | |
670 | ARRAY_SIZE(iov), NULL)); | |
671 | } | |
672 | ||
673 | static struct xenbus_watch *find_watch(const char *token) | |
674 | { | |
675 | struct xenbus_watch *i, *cmp; | |
676 | ||
677 | cmp = (void *)simple_strtoul(token, NULL, 16); | |
678 | ||
679 | list_for_each_entry(i, &watches, list) | |
680 | if (i == cmp) | |
681 | return i; | |
682 | ||
683 | return NULL; | |
684 | } | |
685 | ||
686 | int xs_watch_msg(struct xs_watch_event *event) | |
687 | { | |
688 | if (count_strings(event->body, event->len) != 2) { | |
689 | kfree(event); | |
690 | return -EINVAL; | |
691 | } | |
692 | event->path = (const char *)event->body; | |
693 | event->token = (const char *)strchr(event->body, '\0') + 1; | |
694 | ||
695 | spin_lock(&watches_lock); | |
696 | event->handle = find_watch(event->token); | |
697 | if (event->handle != NULL) { | |
698 | spin_lock(&watch_events_lock); | |
699 | list_add_tail(&event->list, &watch_events); | |
700 | wake_up(&watch_events_waitq); | |
701 | spin_unlock(&watch_events_lock); | |
702 | } else | |
703 | kfree(event); | |
704 | spin_unlock(&watches_lock); | |
705 | ||
706 | return 0; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Certain older XenBus toolstack cannot handle reading values that are | |
711 | * not populated. Some Xen 3.4 installation are incapable of doing this | |
712 | * so if we are running on anything older than 4 do not attempt to read | |
713 | * control/platform-feature-xs_reset_watches. | |
714 | */ | |
715 | static bool xen_strict_xenbus_quirk(void) | |
716 | { | |
717 | #ifdef CONFIG_X86 | |
718 | uint32_t eax, ebx, ecx, edx, base; | |
719 | ||
720 | base = xen_cpuid_base(); | |
721 | cpuid(base + 1, &eax, &ebx, &ecx, &edx); | |
722 | ||
723 | if ((eax >> 16) < 4) | |
724 | return true; | |
725 | #endif | |
726 | return false; | |
727 | ||
728 | } | |
729 | static void xs_reset_watches(void) | |
730 | { | |
731 | int err; | |
732 | ||
733 | if (!xen_hvm_domain() || xen_initial_domain()) | |
734 | return; | |
735 | ||
736 | if (xen_strict_xenbus_quirk()) | |
737 | return; | |
738 | ||
739 | if (!xenbus_read_unsigned("control", | |
740 | "platform-feature-xs_reset_watches", 0)) | |
741 | return; | |
742 | ||
743 | err = xs_error(xs_single(XBT_NIL, XS_RESET_WATCHES, "", NULL)); | |
744 | if (err && err != -EEXIST) | |
745 | pr_warn("xs_reset_watches failed: %d\n", err); | |
746 | } | |
747 | ||
748 | /* Register callback to watch this node. */ | |
749 | int register_xenbus_watch(struct xenbus_watch *watch) | |
750 | { | |
751 | /* Pointer in ascii is the token. */ | |
752 | char token[sizeof(watch) * 2 + 1]; | |
753 | int err; | |
754 | ||
755 | sprintf(token, "%lX", (long)watch); | |
756 | ||
757 | down_read(&xs_watch_rwsem); | |
758 | ||
759 | spin_lock(&watches_lock); | |
760 | BUG_ON(find_watch(token)); | |
761 | list_add(&watch->list, &watches); | |
762 | spin_unlock(&watches_lock); | |
763 | ||
764 | err = xs_watch(watch->node, token); | |
765 | ||
766 | if (err) { | |
767 | spin_lock(&watches_lock); | |
768 | list_del(&watch->list); | |
769 | spin_unlock(&watches_lock); | |
770 | } | |
771 | ||
772 | up_read(&xs_watch_rwsem); | |
773 | ||
774 | return err; | |
775 | } | |
776 | EXPORT_SYMBOL_GPL(register_xenbus_watch); | |
777 | ||
778 | void unregister_xenbus_watch(struct xenbus_watch *watch) | |
779 | { | |
780 | struct xs_watch_event *event, *tmp; | |
781 | char token[sizeof(watch) * 2 + 1]; | |
782 | int err; | |
783 | ||
784 | sprintf(token, "%lX", (long)watch); | |
785 | ||
786 | down_read(&xs_watch_rwsem); | |
787 | ||
788 | spin_lock(&watches_lock); | |
789 | BUG_ON(!find_watch(token)); | |
790 | list_del(&watch->list); | |
791 | spin_unlock(&watches_lock); | |
792 | ||
793 | err = xs_unwatch(watch->node, token); | |
794 | if (err) | |
795 | pr_warn("Failed to release watch %s: %i\n", watch->node, err); | |
796 | ||
797 | up_read(&xs_watch_rwsem); | |
798 | ||
799 | /* Make sure there are no callbacks running currently (unless | |
800 | its us) */ | |
801 | if (current->pid != xenwatch_pid) | |
802 | mutex_lock(&xenwatch_mutex); | |
803 | ||
804 | /* Cancel pending watch events. */ | |
805 | spin_lock(&watch_events_lock); | |
806 | list_for_each_entry_safe(event, tmp, &watch_events, list) { | |
807 | if (event->handle != watch) | |
808 | continue; | |
809 | list_del(&event->list); | |
810 | kfree(event); | |
811 | } | |
812 | spin_unlock(&watch_events_lock); | |
813 | ||
814 | if (current->pid != xenwatch_pid) | |
815 | mutex_unlock(&xenwatch_mutex); | |
816 | } | |
817 | EXPORT_SYMBOL_GPL(unregister_xenbus_watch); | |
818 | ||
819 | void xs_suspend(void) | |
820 | { | |
821 | xs_suspend_enter(); | |
822 | ||
823 | down_write(&xs_watch_rwsem); | |
824 | mutex_lock(&xs_response_mutex); | |
825 | } | |
826 | ||
827 | void xs_resume(void) | |
828 | { | |
829 | struct xenbus_watch *watch; | |
830 | char token[sizeof(watch) * 2 + 1]; | |
831 | ||
832 | xb_init_comms(); | |
833 | ||
834 | mutex_unlock(&xs_response_mutex); | |
835 | ||
836 | xs_suspend_exit(); | |
837 | ||
838 | /* No need for watches_lock: the xs_watch_rwsem is sufficient. */ | |
839 | list_for_each_entry(watch, &watches, list) { | |
840 | sprintf(token, "%lX", (long)watch); | |
841 | xs_watch(watch->node, token); | |
842 | } | |
843 | ||
844 | up_write(&xs_watch_rwsem); | |
845 | } | |
846 | ||
847 | void xs_suspend_cancel(void) | |
848 | { | |
849 | mutex_unlock(&xs_response_mutex); | |
850 | up_write(&xs_watch_rwsem); | |
851 | ||
852 | xs_suspend_exit(); | |
853 | } | |
854 | ||
855 | static int xenwatch_thread(void *unused) | |
856 | { | |
857 | struct list_head *ent; | |
858 | struct xs_watch_event *event; | |
859 | ||
860 | xenwatch_pid = current->pid; | |
861 | ||
862 | for (;;) { | |
863 | wait_event_interruptible(watch_events_waitq, | |
864 | !list_empty(&watch_events)); | |
865 | ||
866 | if (kthread_should_stop()) | |
867 | break; | |
868 | ||
869 | mutex_lock(&xenwatch_mutex); | |
870 | ||
871 | spin_lock(&watch_events_lock); | |
872 | ent = watch_events.next; | |
873 | if (ent != &watch_events) | |
874 | list_del(ent); | |
875 | spin_unlock(&watch_events_lock); | |
876 | ||
877 | if (ent != &watch_events) { | |
878 | event = list_entry(ent, struct xs_watch_event, list); | |
879 | event->handle->callback(event->handle, event->path, | |
880 | event->token); | |
881 | kfree(event); | |
882 | } | |
883 | ||
884 | mutex_unlock(&xenwatch_mutex); | |
885 | } | |
886 | ||
887 | return 0; | |
888 | } | |
889 | ||
890 | /* | |
891 | * Wake up all threads waiting for a xenstore reply. In case of shutdown all | |
892 | * pending replies will be marked as "aborted" in order to let the waiters | |
893 | * return in spite of xenstore possibly no longer being able to reply. This | |
894 | * will avoid blocking shutdown by a thread waiting for xenstore but being | |
895 | * necessary for shutdown processing to proceed. | |
896 | */ | |
897 | static int xs_reboot_notify(struct notifier_block *nb, | |
898 | unsigned long code, void *unused) | |
899 | { | |
900 | struct xb_req_data *req; | |
901 | ||
902 | mutex_lock(&xb_write_mutex); | |
903 | list_for_each_entry(req, &xs_reply_list, list) | |
904 | wake_up(&req->wq); | |
905 | list_for_each_entry(req, &xb_write_list, list) | |
906 | wake_up(&req->wq); | |
907 | mutex_unlock(&xb_write_mutex); | |
908 | return NOTIFY_DONE; | |
909 | } | |
910 | ||
911 | static struct notifier_block xs_reboot_nb = { | |
912 | .notifier_call = xs_reboot_notify, | |
913 | }; | |
914 | ||
915 | int xs_init(void) | |
916 | { | |
917 | int err; | |
918 | struct task_struct *task; | |
919 | ||
920 | register_reboot_notifier(&xs_reboot_nb); | |
921 | ||
922 | /* Initialize the shared memory rings to talk to xenstored */ | |
923 | err = xb_init_comms(); | |
924 | if (err) | |
925 | return err; | |
926 | ||
927 | task = kthread_run(xenwatch_thread, NULL, "xenwatch"); | |
928 | if (IS_ERR(task)) | |
929 | return PTR_ERR(task); | |
930 | ||
931 | /* shutdown watches for kexec boot */ | |
932 | xs_reset_watches(); | |
933 | ||
934 | return 0; | |
935 | } |