Jeremy Fitzhardinge | 4bac07c | 2007-07-17 18:37:06 -0700 | [diff] [blame] | 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 | #include <linux/unistd.h> |
| 35 | #include <linux/errno.h> |
| 36 | #include <linux/types.h> |
| 37 | #include <linux/uio.h> |
| 38 | #include <linux/kernel.h> |
| 39 | #include <linux/string.h> |
| 40 | #include <linux/err.h> |
| 41 | #include <linux/slab.h> |
| 42 | #include <linux/fcntl.h> |
| 43 | #include <linux/kthread.h> |
| 44 | #include <linux/rwsem.h> |
| 45 | #include <linux/module.h> |
| 46 | #include <linux/mutex.h> |
| 47 | #include <xen/xenbus.h> |
| 48 | #include "xenbus_comms.h" |
| 49 | |
| 50 | struct xs_stored_msg { |
| 51 | struct list_head list; |
| 52 | |
| 53 | struct xsd_sockmsg hdr; |
| 54 | |
| 55 | union { |
| 56 | /* Queued replies. */ |
| 57 | struct { |
| 58 | char *body; |
| 59 | } reply; |
| 60 | |
| 61 | /* Queued watch events. */ |
| 62 | struct { |
| 63 | struct xenbus_watch *handle; |
| 64 | char **vec; |
| 65 | unsigned int vec_size; |
| 66 | } watch; |
| 67 | } u; |
| 68 | }; |
| 69 | |
| 70 | struct xs_handle { |
| 71 | /* A list of replies. Currently only one will ever be outstanding. */ |
| 72 | struct list_head reply_list; |
| 73 | spinlock_t reply_lock; |
| 74 | wait_queue_head_t reply_waitq; |
| 75 | |
| 76 | /* |
| 77 | * Mutex ordering: transaction_mutex -> watch_mutex -> request_mutex. |
| 78 | * response_mutex is never taken simultaneously with the other three. |
| 79 | */ |
| 80 | |
| 81 | /* One request at a time. */ |
| 82 | struct mutex request_mutex; |
| 83 | |
| 84 | /* Protect xenbus reader thread against save/restore. */ |
| 85 | struct mutex response_mutex; |
| 86 | |
| 87 | /* Protect transactions against save/restore. */ |
| 88 | struct rw_semaphore transaction_mutex; |
| 89 | |
| 90 | /* Protect watch (de)register against save/restore. */ |
| 91 | struct rw_semaphore watch_mutex; |
| 92 | }; |
| 93 | |
| 94 | static struct xs_handle xs_state; |
| 95 | |
| 96 | /* List of registered watches, and a lock to protect it. */ |
| 97 | static LIST_HEAD(watches); |
| 98 | static DEFINE_SPINLOCK(watches_lock); |
| 99 | |
| 100 | /* List of pending watch callback events, and a lock to protect it. */ |
| 101 | static LIST_HEAD(watch_events); |
| 102 | static DEFINE_SPINLOCK(watch_events_lock); |
| 103 | |
| 104 | /* |
| 105 | * Details of the xenwatch callback kernel thread. The thread waits on the |
| 106 | * watch_events_waitq for work to do (queued on watch_events list). When it |
| 107 | * wakes up it acquires the xenwatch_mutex before reading the list and |
| 108 | * carrying out work. |
| 109 | */ |
| 110 | static pid_t xenwatch_pid; |
| 111 | static DEFINE_MUTEX(xenwatch_mutex); |
| 112 | static DECLARE_WAIT_QUEUE_HEAD(watch_events_waitq); |
| 113 | |
| 114 | static int get_error(const char *errorstring) |
| 115 | { |
| 116 | unsigned int i; |
| 117 | |
| 118 | for (i = 0; strcmp(errorstring, xsd_errors[i].errstring) != 0; i++) { |
| 119 | if (i == ARRAY_SIZE(xsd_errors) - 1) { |
| 120 | printk(KERN_WARNING |
| 121 | "XENBUS xen store gave: unknown error %s", |
| 122 | errorstring); |
| 123 | return EINVAL; |
| 124 | } |
| 125 | } |
| 126 | return xsd_errors[i].errnum; |
| 127 | } |
| 128 | |
| 129 | static void *read_reply(enum xsd_sockmsg_type *type, unsigned int *len) |
| 130 | { |
| 131 | struct xs_stored_msg *msg; |
| 132 | char *body; |
| 133 | |
| 134 | spin_lock(&xs_state.reply_lock); |
| 135 | |
| 136 | while (list_empty(&xs_state.reply_list)) { |
| 137 | spin_unlock(&xs_state.reply_lock); |
| 138 | /* XXX FIXME: Avoid synchronous wait for response here. */ |
| 139 | wait_event(xs_state.reply_waitq, |
| 140 | !list_empty(&xs_state.reply_list)); |
| 141 | spin_lock(&xs_state.reply_lock); |
| 142 | } |
| 143 | |
| 144 | msg = list_entry(xs_state.reply_list.next, |
| 145 | struct xs_stored_msg, list); |
| 146 | list_del(&msg->list); |
| 147 | |
| 148 | spin_unlock(&xs_state.reply_lock); |
| 149 | |
| 150 | *type = msg->hdr.type; |
| 151 | if (len) |
| 152 | *len = msg->hdr.len; |
| 153 | body = msg->u.reply.body; |
| 154 | |
| 155 | kfree(msg); |
| 156 | |
| 157 | return body; |
| 158 | } |
| 159 | |
| 160 | void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg) |
| 161 | { |
| 162 | void *ret; |
| 163 | struct xsd_sockmsg req_msg = *msg; |
| 164 | int err; |
| 165 | |
| 166 | if (req_msg.type == XS_TRANSACTION_START) |
| 167 | down_read(&xs_state.transaction_mutex); |
| 168 | |
| 169 | mutex_lock(&xs_state.request_mutex); |
| 170 | |
| 171 | err = xb_write(msg, sizeof(*msg) + msg->len); |
| 172 | if (err) { |
| 173 | msg->type = XS_ERROR; |
| 174 | ret = ERR_PTR(err); |
| 175 | } else |
| 176 | ret = read_reply(&msg->type, &msg->len); |
| 177 | |
| 178 | mutex_unlock(&xs_state.request_mutex); |
| 179 | |
| 180 | if ((msg->type == XS_TRANSACTION_END) || |
| 181 | ((req_msg.type == XS_TRANSACTION_START) && |
| 182 | (msg->type == XS_ERROR))) |
| 183 | up_read(&xs_state.transaction_mutex); |
| 184 | |
| 185 | return ret; |
| 186 | } |
| 187 | |
| 188 | /* Send message to xs, get kmalloc'ed reply. ERR_PTR() on error. */ |
| 189 | static void *xs_talkv(struct xenbus_transaction t, |
| 190 | enum xsd_sockmsg_type type, |
| 191 | const struct kvec *iovec, |
| 192 | unsigned int num_vecs, |
| 193 | unsigned int *len) |
| 194 | { |
| 195 | struct xsd_sockmsg msg; |
| 196 | void *ret = NULL; |
| 197 | unsigned int i; |
| 198 | int err; |
| 199 | |
| 200 | msg.tx_id = t.id; |
| 201 | msg.req_id = 0; |
| 202 | msg.type = type; |
| 203 | msg.len = 0; |
| 204 | for (i = 0; i < num_vecs; i++) |
| 205 | msg.len += iovec[i].iov_len; |
| 206 | |
| 207 | mutex_lock(&xs_state.request_mutex); |
| 208 | |
| 209 | err = xb_write(&msg, sizeof(msg)); |
| 210 | if (err) { |
| 211 | mutex_unlock(&xs_state.request_mutex); |
| 212 | return ERR_PTR(err); |
| 213 | } |
| 214 | |
| 215 | for (i = 0; i < num_vecs; i++) { |
| 216 | err = xb_write(iovec[i].iov_base, iovec[i].iov_len); |
| 217 | if (err) { |
| 218 | mutex_unlock(&xs_state.request_mutex); |
| 219 | return ERR_PTR(err); |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | ret = read_reply(&msg.type, len); |
| 224 | |
| 225 | mutex_unlock(&xs_state.request_mutex); |
| 226 | |
| 227 | if (IS_ERR(ret)) |
| 228 | return ret; |
| 229 | |
| 230 | if (msg.type == XS_ERROR) { |
| 231 | err = get_error(ret); |
| 232 | kfree(ret); |
| 233 | return ERR_PTR(-err); |
| 234 | } |
| 235 | |
| 236 | if (msg.type != type) { |
| 237 | if (printk_ratelimit()) |
| 238 | printk(KERN_WARNING |
| 239 | "XENBUS unexpected type [%d], expected [%d]\n", |
| 240 | msg.type, type); |
| 241 | kfree(ret); |
| 242 | return ERR_PTR(-EINVAL); |
| 243 | } |
| 244 | return ret; |
| 245 | } |
| 246 | |
| 247 | /* Simplified version of xs_talkv: single message. */ |
| 248 | static void *xs_single(struct xenbus_transaction t, |
| 249 | enum xsd_sockmsg_type type, |
| 250 | const char *string, |
| 251 | unsigned int *len) |
| 252 | { |
| 253 | struct kvec iovec; |
| 254 | |
| 255 | iovec.iov_base = (void *)string; |
| 256 | iovec.iov_len = strlen(string) + 1; |
| 257 | return xs_talkv(t, type, &iovec, 1, len); |
| 258 | } |
| 259 | |
| 260 | /* Many commands only need an ack, don't care what it says. */ |
| 261 | static int xs_error(char *reply) |
| 262 | { |
| 263 | if (IS_ERR(reply)) |
| 264 | return PTR_ERR(reply); |
| 265 | kfree(reply); |
| 266 | return 0; |
| 267 | } |
| 268 | |
| 269 | static unsigned int count_strings(const char *strings, unsigned int len) |
| 270 | { |
| 271 | unsigned int num; |
| 272 | const char *p; |
| 273 | |
| 274 | for (p = strings, num = 0; p < strings + len; p += strlen(p) + 1) |
| 275 | num++; |
| 276 | |
| 277 | return num; |
| 278 | } |
| 279 | |
| 280 | /* Return the path to dir with /name appended. Buffer must be kfree()'ed. */ |
| 281 | static char *join(const char *dir, const char *name) |
| 282 | { |
| 283 | char *buffer; |
| 284 | |
| 285 | if (strlen(name) == 0) |
| 286 | buffer = kasprintf(GFP_KERNEL, "%s", dir); |
| 287 | else |
| 288 | buffer = kasprintf(GFP_KERNEL, "%s/%s", dir, name); |
| 289 | return (!buffer) ? ERR_PTR(-ENOMEM) : buffer; |
| 290 | } |
| 291 | |
| 292 | static char **split(char *strings, unsigned int len, unsigned int *num) |
| 293 | { |
| 294 | char *p, **ret; |
| 295 | |
| 296 | /* Count the strings. */ |
| 297 | *num = count_strings(strings, len); |
| 298 | |
| 299 | /* Transfer to one big alloc for easy freeing. */ |
| 300 | ret = kmalloc(*num * sizeof(char *) + len, GFP_KERNEL); |
| 301 | if (!ret) { |
| 302 | kfree(strings); |
| 303 | return ERR_PTR(-ENOMEM); |
| 304 | } |
| 305 | memcpy(&ret[*num], strings, len); |
| 306 | kfree(strings); |
| 307 | |
| 308 | strings = (char *)&ret[*num]; |
| 309 | for (p = strings, *num = 0; p < strings + len; p += strlen(p) + 1) |
| 310 | ret[(*num)++] = p; |
| 311 | |
| 312 | return ret; |
| 313 | } |
| 314 | |
| 315 | char **xenbus_directory(struct xenbus_transaction t, |
| 316 | const char *dir, const char *node, unsigned int *num) |
| 317 | { |
| 318 | char *strings, *path; |
| 319 | unsigned int len; |
| 320 | |
| 321 | path = join(dir, node); |
| 322 | if (IS_ERR(path)) |
| 323 | return (char **)path; |
| 324 | |
| 325 | strings = xs_single(t, XS_DIRECTORY, path, &len); |
| 326 | kfree(path); |
| 327 | if (IS_ERR(strings)) |
| 328 | return (char **)strings; |
| 329 | |
| 330 | return split(strings, len, num); |
| 331 | } |
| 332 | EXPORT_SYMBOL_GPL(xenbus_directory); |
| 333 | |
| 334 | /* Check if a path exists. Return 1 if it does. */ |
| 335 | int xenbus_exists(struct xenbus_transaction t, |
| 336 | const char *dir, const char *node) |
| 337 | { |
| 338 | char **d; |
| 339 | int dir_n; |
| 340 | |
| 341 | d = xenbus_directory(t, dir, node, &dir_n); |
| 342 | if (IS_ERR(d)) |
| 343 | return 0; |
| 344 | kfree(d); |
| 345 | return 1; |
| 346 | } |
| 347 | EXPORT_SYMBOL_GPL(xenbus_exists); |
| 348 | |
| 349 | /* Get the value of a single file. |
| 350 | * Returns a kmalloced value: call free() on it after use. |
| 351 | * len indicates length in bytes. |
| 352 | */ |
| 353 | void *xenbus_read(struct xenbus_transaction t, |
| 354 | const char *dir, const char *node, unsigned int *len) |
| 355 | { |
| 356 | char *path; |
| 357 | void *ret; |
| 358 | |
| 359 | path = join(dir, node); |
| 360 | if (IS_ERR(path)) |
| 361 | return (void *)path; |
| 362 | |
| 363 | ret = xs_single(t, XS_READ, path, len); |
| 364 | kfree(path); |
| 365 | return ret; |
| 366 | } |
| 367 | EXPORT_SYMBOL_GPL(xenbus_read); |
| 368 | |
| 369 | /* Write the value of a single file. |
| 370 | * Returns -err on failure. |
| 371 | */ |
| 372 | int xenbus_write(struct xenbus_transaction t, |
| 373 | const char *dir, const char *node, const char *string) |
| 374 | { |
| 375 | const char *path; |
| 376 | struct kvec iovec[2]; |
| 377 | int ret; |
| 378 | |
| 379 | path = join(dir, node); |
| 380 | if (IS_ERR(path)) |
| 381 | return PTR_ERR(path); |
| 382 | |
| 383 | iovec[0].iov_base = (void *)path; |
| 384 | iovec[0].iov_len = strlen(path) + 1; |
| 385 | iovec[1].iov_base = (void *)string; |
| 386 | iovec[1].iov_len = strlen(string); |
| 387 | |
| 388 | ret = xs_error(xs_talkv(t, XS_WRITE, iovec, ARRAY_SIZE(iovec), NULL)); |
| 389 | kfree(path); |
| 390 | return ret; |
| 391 | } |
| 392 | EXPORT_SYMBOL_GPL(xenbus_write); |
| 393 | |
| 394 | /* Create a new directory. */ |
| 395 | int xenbus_mkdir(struct xenbus_transaction t, |
| 396 | const char *dir, const char *node) |
| 397 | { |
| 398 | char *path; |
| 399 | int ret; |
| 400 | |
| 401 | path = join(dir, node); |
| 402 | if (IS_ERR(path)) |
| 403 | return PTR_ERR(path); |
| 404 | |
| 405 | ret = xs_error(xs_single(t, XS_MKDIR, path, NULL)); |
| 406 | kfree(path); |
| 407 | return ret; |
| 408 | } |
| 409 | EXPORT_SYMBOL_GPL(xenbus_mkdir); |
| 410 | |
| 411 | /* Destroy a file or directory (directories must be empty). */ |
| 412 | int xenbus_rm(struct xenbus_transaction t, const char *dir, const char *node) |
| 413 | { |
| 414 | char *path; |
| 415 | int ret; |
| 416 | |
| 417 | path = join(dir, node); |
| 418 | if (IS_ERR(path)) |
| 419 | return PTR_ERR(path); |
| 420 | |
| 421 | ret = xs_error(xs_single(t, XS_RM, path, NULL)); |
| 422 | kfree(path); |
| 423 | return ret; |
| 424 | } |
| 425 | EXPORT_SYMBOL_GPL(xenbus_rm); |
| 426 | |
| 427 | /* Start a transaction: changes by others will not be seen during this |
| 428 | * transaction, and changes will not be visible to others until end. |
| 429 | */ |
| 430 | int xenbus_transaction_start(struct xenbus_transaction *t) |
| 431 | { |
| 432 | char *id_str; |
| 433 | |
| 434 | down_read(&xs_state.transaction_mutex); |
| 435 | |
| 436 | id_str = xs_single(XBT_NIL, XS_TRANSACTION_START, "", NULL); |
| 437 | if (IS_ERR(id_str)) { |
| 438 | up_read(&xs_state.transaction_mutex); |
| 439 | return PTR_ERR(id_str); |
| 440 | } |
| 441 | |
| 442 | t->id = simple_strtoul(id_str, NULL, 0); |
| 443 | kfree(id_str); |
| 444 | return 0; |
| 445 | } |
| 446 | EXPORT_SYMBOL_GPL(xenbus_transaction_start); |
| 447 | |
| 448 | /* End a transaction. |
| 449 | * If abandon is true, transaction is discarded instead of committed. |
| 450 | */ |
| 451 | int xenbus_transaction_end(struct xenbus_transaction t, int abort) |
| 452 | { |
| 453 | char abortstr[2]; |
| 454 | int err; |
| 455 | |
| 456 | if (abort) |
| 457 | strcpy(abortstr, "F"); |
| 458 | else |
| 459 | strcpy(abortstr, "T"); |
| 460 | |
| 461 | err = xs_error(xs_single(t, XS_TRANSACTION_END, abortstr, NULL)); |
| 462 | |
| 463 | up_read(&xs_state.transaction_mutex); |
| 464 | |
| 465 | return err; |
| 466 | } |
| 467 | EXPORT_SYMBOL_GPL(xenbus_transaction_end); |
| 468 | |
| 469 | /* Single read and scanf: returns -errno or num scanned. */ |
| 470 | int xenbus_scanf(struct xenbus_transaction t, |
| 471 | const char *dir, const char *node, const char *fmt, ...) |
| 472 | { |
| 473 | va_list ap; |
| 474 | int ret; |
| 475 | char *val; |
| 476 | |
| 477 | val = xenbus_read(t, dir, node, NULL); |
| 478 | if (IS_ERR(val)) |
| 479 | return PTR_ERR(val); |
| 480 | |
| 481 | va_start(ap, fmt); |
| 482 | ret = vsscanf(val, fmt, ap); |
| 483 | va_end(ap); |
| 484 | kfree(val); |
| 485 | /* Distinctive errno. */ |
| 486 | if (ret == 0) |
| 487 | return -ERANGE; |
| 488 | return ret; |
| 489 | } |
| 490 | EXPORT_SYMBOL_GPL(xenbus_scanf); |
| 491 | |
| 492 | /* Single printf and write: returns -errno or 0. */ |
| 493 | int xenbus_printf(struct xenbus_transaction t, |
| 494 | const char *dir, const char *node, const char *fmt, ...) |
| 495 | { |
| 496 | va_list ap; |
| 497 | int ret; |
| 498 | #define PRINTF_BUFFER_SIZE 4096 |
| 499 | char *printf_buffer; |
| 500 | |
| 501 | printf_buffer = kmalloc(PRINTF_BUFFER_SIZE, GFP_KERNEL); |
| 502 | if (printf_buffer == NULL) |
| 503 | return -ENOMEM; |
| 504 | |
| 505 | va_start(ap, fmt); |
| 506 | ret = vsnprintf(printf_buffer, PRINTF_BUFFER_SIZE, fmt, ap); |
| 507 | va_end(ap); |
| 508 | |
| 509 | BUG_ON(ret > PRINTF_BUFFER_SIZE-1); |
| 510 | ret = xenbus_write(t, dir, node, printf_buffer); |
| 511 | |
| 512 | kfree(printf_buffer); |
| 513 | |
| 514 | return ret; |
| 515 | } |
| 516 | EXPORT_SYMBOL_GPL(xenbus_printf); |
| 517 | |
| 518 | /* Takes tuples of names, scanf-style args, and void **, NULL terminated. */ |
| 519 | int xenbus_gather(struct xenbus_transaction t, const char *dir, ...) |
| 520 | { |
| 521 | va_list ap; |
| 522 | const char *name; |
| 523 | int ret = 0; |
| 524 | |
| 525 | va_start(ap, dir); |
| 526 | while (ret == 0 && (name = va_arg(ap, char *)) != NULL) { |
| 527 | const char *fmt = va_arg(ap, char *); |
| 528 | void *result = va_arg(ap, void *); |
| 529 | char *p; |
| 530 | |
| 531 | p = xenbus_read(t, dir, name, NULL); |
| 532 | if (IS_ERR(p)) { |
| 533 | ret = PTR_ERR(p); |
| 534 | break; |
| 535 | } |
| 536 | if (fmt) { |
| 537 | if (sscanf(p, fmt, result) == 0) |
| 538 | ret = -EINVAL; |
| 539 | kfree(p); |
| 540 | } else |
| 541 | *(char **)result = p; |
| 542 | } |
| 543 | va_end(ap); |
| 544 | return ret; |
| 545 | } |
| 546 | EXPORT_SYMBOL_GPL(xenbus_gather); |
| 547 | |
| 548 | static int xs_watch(const char *path, const char *token) |
| 549 | { |
| 550 | struct kvec iov[2]; |
| 551 | |
| 552 | iov[0].iov_base = (void *)path; |
| 553 | iov[0].iov_len = strlen(path) + 1; |
| 554 | iov[1].iov_base = (void *)token; |
| 555 | iov[1].iov_len = strlen(token) + 1; |
| 556 | |
| 557 | return xs_error(xs_talkv(XBT_NIL, XS_WATCH, iov, |
| 558 | ARRAY_SIZE(iov), NULL)); |
| 559 | } |
| 560 | |
| 561 | static int xs_unwatch(const char *path, const char *token) |
| 562 | { |
| 563 | struct kvec iov[2]; |
| 564 | |
| 565 | iov[0].iov_base = (char *)path; |
| 566 | iov[0].iov_len = strlen(path) + 1; |
| 567 | iov[1].iov_base = (char *)token; |
| 568 | iov[1].iov_len = strlen(token) + 1; |
| 569 | |
| 570 | return xs_error(xs_talkv(XBT_NIL, XS_UNWATCH, iov, |
| 571 | ARRAY_SIZE(iov), NULL)); |
| 572 | } |
| 573 | |
| 574 | static struct xenbus_watch *find_watch(const char *token) |
| 575 | { |
| 576 | struct xenbus_watch *i, *cmp; |
| 577 | |
| 578 | cmp = (void *)simple_strtoul(token, NULL, 16); |
| 579 | |
| 580 | list_for_each_entry(i, &watches, list) |
| 581 | if (i == cmp) |
| 582 | return i; |
| 583 | |
| 584 | return NULL; |
| 585 | } |
| 586 | |
| 587 | /* Register callback to watch this node. */ |
| 588 | int register_xenbus_watch(struct xenbus_watch *watch) |
| 589 | { |
| 590 | /* Pointer in ascii is the token. */ |
| 591 | char token[sizeof(watch) * 2 + 1]; |
| 592 | int err; |
| 593 | |
| 594 | sprintf(token, "%lX", (long)watch); |
| 595 | |
| 596 | down_read(&xs_state.watch_mutex); |
| 597 | |
| 598 | spin_lock(&watches_lock); |
| 599 | BUG_ON(find_watch(token)); |
| 600 | list_add(&watch->list, &watches); |
| 601 | spin_unlock(&watches_lock); |
| 602 | |
| 603 | err = xs_watch(watch->node, token); |
| 604 | |
| 605 | /* Ignore errors due to multiple registration. */ |
| 606 | if ((err != 0) && (err != -EEXIST)) { |
| 607 | spin_lock(&watches_lock); |
| 608 | list_del(&watch->list); |
| 609 | spin_unlock(&watches_lock); |
| 610 | } |
| 611 | |
| 612 | up_read(&xs_state.watch_mutex); |
| 613 | |
| 614 | return err; |
| 615 | } |
| 616 | EXPORT_SYMBOL_GPL(register_xenbus_watch); |
| 617 | |
| 618 | void unregister_xenbus_watch(struct xenbus_watch *watch) |
| 619 | { |
| 620 | struct xs_stored_msg *msg, *tmp; |
| 621 | char token[sizeof(watch) * 2 + 1]; |
| 622 | int err; |
| 623 | |
| 624 | sprintf(token, "%lX", (long)watch); |
| 625 | |
| 626 | down_read(&xs_state.watch_mutex); |
| 627 | |
| 628 | spin_lock(&watches_lock); |
| 629 | BUG_ON(!find_watch(token)); |
| 630 | list_del(&watch->list); |
| 631 | spin_unlock(&watches_lock); |
| 632 | |
| 633 | err = xs_unwatch(watch->node, token); |
| 634 | if (err) |
| 635 | printk(KERN_WARNING |
| 636 | "XENBUS Failed to release watch %s: %i\n", |
| 637 | watch->node, err); |
| 638 | |
| 639 | up_read(&xs_state.watch_mutex); |
| 640 | |
| 641 | /* Make sure there are no callbacks running currently (unless |
| 642 | its us) */ |
| 643 | if (current->pid != xenwatch_pid) |
| 644 | mutex_lock(&xenwatch_mutex); |
| 645 | |
| 646 | /* Cancel pending watch events. */ |
| 647 | spin_lock(&watch_events_lock); |
| 648 | list_for_each_entry_safe(msg, tmp, &watch_events, list) { |
| 649 | if (msg->u.watch.handle != watch) |
| 650 | continue; |
| 651 | list_del(&msg->list); |
| 652 | kfree(msg->u.watch.vec); |
| 653 | kfree(msg); |
| 654 | } |
| 655 | spin_unlock(&watch_events_lock); |
| 656 | |
| 657 | if (current->pid != xenwatch_pid) |
| 658 | mutex_unlock(&xenwatch_mutex); |
| 659 | } |
| 660 | EXPORT_SYMBOL_GPL(unregister_xenbus_watch); |
| 661 | |
| 662 | void xs_suspend(void) |
| 663 | { |
| 664 | down_write(&xs_state.transaction_mutex); |
| 665 | down_write(&xs_state.watch_mutex); |
| 666 | mutex_lock(&xs_state.request_mutex); |
| 667 | mutex_lock(&xs_state.response_mutex); |
| 668 | } |
| 669 | |
| 670 | void xs_resume(void) |
| 671 | { |
| 672 | struct xenbus_watch *watch; |
| 673 | char token[sizeof(watch) * 2 + 1]; |
| 674 | |
| 675 | mutex_unlock(&xs_state.response_mutex); |
| 676 | mutex_unlock(&xs_state.request_mutex); |
| 677 | up_write(&xs_state.transaction_mutex); |
| 678 | |
| 679 | /* No need for watches_lock: the watch_mutex is sufficient. */ |
| 680 | list_for_each_entry(watch, &watches, list) { |
| 681 | sprintf(token, "%lX", (long)watch); |
| 682 | xs_watch(watch->node, token); |
| 683 | } |
| 684 | |
| 685 | up_write(&xs_state.watch_mutex); |
| 686 | } |
| 687 | |
| 688 | void xs_suspend_cancel(void) |
| 689 | { |
| 690 | mutex_unlock(&xs_state.response_mutex); |
| 691 | mutex_unlock(&xs_state.request_mutex); |
| 692 | up_write(&xs_state.watch_mutex); |
| 693 | up_write(&xs_state.transaction_mutex); |
| 694 | } |
| 695 | |
| 696 | static int xenwatch_thread(void *unused) |
| 697 | { |
| 698 | struct list_head *ent; |
| 699 | struct xs_stored_msg *msg; |
| 700 | |
| 701 | for (;;) { |
| 702 | wait_event_interruptible(watch_events_waitq, |
| 703 | !list_empty(&watch_events)); |
| 704 | |
| 705 | if (kthread_should_stop()) |
| 706 | break; |
| 707 | |
| 708 | mutex_lock(&xenwatch_mutex); |
| 709 | |
| 710 | spin_lock(&watch_events_lock); |
| 711 | ent = watch_events.next; |
| 712 | if (ent != &watch_events) |
| 713 | list_del(ent); |
| 714 | spin_unlock(&watch_events_lock); |
| 715 | |
| 716 | if (ent != &watch_events) { |
| 717 | msg = list_entry(ent, struct xs_stored_msg, list); |
| 718 | msg->u.watch.handle->callback( |
| 719 | msg->u.watch.handle, |
| 720 | (const char **)msg->u.watch.vec, |
| 721 | msg->u.watch.vec_size); |
| 722 | kfree(msg->u.watch.vec); |
| 723 | kfree(msg); |
| 724 | } |
| 725 | |
| 726 | mutex_unlock(&xenwatch_mutex); |
| 727 | } |
| 728 | |
| 729 | return 0; |
| 730 | } |
| 731 | |
| 732 | static int process_msg(void) |
| 733 | { |
| 734 | struct xs_stored_msg *msg; |
| 735 | char *body; |
| 736 | int err; |
| 737 | |
| 738 | /* |
| 739 | * We must disallow save/restore while reading a xenstore message. |
| 740 | * A partial read across s/r leaves us out of sync with xenstored. |
| 741 | */ |
| 742 | for (;;) { |
| 743 | err = xb_wait_for_data_to_read(); |
| 744 | if (err) |
| 745 | return err; |
| 746 | mutex_lock(&xs_state.response_mutex); |
| 747 | if (xb_data_to_read()) |
| 748 | break; |
| 749 | /* We raced with save/restore: pending data 'disappeared'. */ |
| 750 | mutex_unlock(&xs_state.response_mutex); |
| 751 | } |
| 752 | |
| 753 | |
| 754 | msg = kmalloc(sizeof(*msg), GFP_KERNEL); |
| 755 | if (msg == NULL) { |
| 756 | err = -ENOMEM; |
| 757 | goto out; |
| 758 | } |
| 759 | |
| 760 | err = xb_read(&msg->hdr, sizeof(msg->hdr)); |
| 761 | if (err) { |
| 762 | kfree(msg); |
| 763 | goto out; |
| 764 | } |
| 765 | |
| 766 | body = kmalloc(msg->hdr.len + 1, GFP_KERNEL); |
| 767 | if (body == NULL) { |
| 768 | kfree(msg); |
| 769 | err = -ENOMEM; |
| 770 | goto out; |
| 771 | } |
| 772 | |
| 773 | err = xb_read(body, msg->hdr.len); |
| 774 | if (err) { |
| 775 | kfree(body); |
| 776 | kfree(msg); |
| 777 | goto out; |
| 778 | } |
| 779 | body[msg->hdr.len] = '\0'; |
| 780 | |
| 781 | if (msg->hdr.type == XS_WATCH_EVENT) { |
| 782 | msg->u.watch.vec = split(body, msg->hdr.len, |
| 783 | &msg->u.watch.vec_size); |
| 784 | if (IS_ERR(msg->u.watch.vec)) { |
Jeremy Fitzhardinge | 4bac07c | 2007-07-17 18:37:06 -0700 | [diff] [blame] | 785 | err = PTR_ERR(msg->u.watch.vec); |
Adrian Bunk | 98ac0e5 | 2007-07-26 10:41:10 -0700 | [diff] [blame] | 786 | kfree(msg); |
Jeremy Fitzhardinge | 4bac07c | 2007-07-17 18:37:06 -0700 | [diff] [blame] | 787 | goto out; |
| 788 | } |
| 789 | |
| 790 | spin_lock(&watches_lock); |
| 791 | msg->u.watch.handle = find_watch( |
| 792 | msg->u.watch.vec[XS_WATCH_TOKEN]); |
| 793 | if (msg->u.watch.handle != NULL) { |
| 794 | spin_lock(&watch_events_lock); |
| 795 | list_add_tail(&msg->list, &watch_events); |
| 796 | wake_up(&watch_events_waitq); |
| 797 | spin_unlock(&watch_events_lock); |
| 798 | } else { |
| 799 | kfree(msg->u.watch.vec); |
| 800 | kfree(msg); |
| 801 | } |
| 802 | spin_unlock(&watches_lock); |
| 803 | } else { |
| 804 | msg->u.reply.body = body; |
| 805 | spin_lock(&xs_state.reply_lock); |
| 806 | list_add_tail(&msg->list, &xs_state.reply_list); |
| 807 | spin_unlock(&xs_state.reply_lock); |
| 808 | wake_up(&xs_state.reply_waitq); |
| 809 | } |
| 810 | |
| 811 | out: |
| 812 | mutex_unlock(&xs_state.response_mutex); |
| 813 | return err; |
| 814 | } |
| 815 | |
| 816 | static int xenbus_thread(void *unused) |
| 817 | { |
| 818 | int err; |
| 819 | |
| 820 | for (;;) { |
| 821 | err = process_msg(); |
| 822 | if (err) |
| 823 | printk(KERN_WARNING "XENBUS error %d while reading " |
| 824 | "message\n", err); |
| 825 | if (kthread_should_stop()) |
| 826 | break; |
| 827 | } |
| 828 | |
| 829 | return 0; |
| 830 | } |
| 831 | |
| 832 | int xs_init(void) |
| 833 | { |
| 834 | int err; |
| 835 | struct task_struct *task; |
| 836 | |
| 837 | INIT_LIST_HEAD(&xs_state.reply_list); |
| 838 | spin_lock_init(&xs_state.reply_lock); |
| 839 | init_waitqueue_head(&xs_state.reply_waitq); |
| 840 | |
| 841 | mutex_init(&xs_state.request_mutex); |
| 842 | mutex_init(&xs_state.response_mutex); |
| 843 | init_rwsem(&xs_state.transaction_mutex); |
| 844 | init_rwsem(&xs_state.watch_mutex); |
| 845 | |
| 846 | /* Initialize the shared memory rings to talk to xenstored */ |
| 847 | err = xb_init_comms(); |
| 848 | if (err) |
| 849 | return err; |
| 850 | |
| 851 | task = kthread_run(xenwatch_thread, NULL, "xenwatch"); |
| 852 | if (IS_ERR(task)) |
| 853 | return PTR_ERR(task); |
| 854 | xenwatch_pid = task->pid; |
| 855 | |
| 856 | task = kthread_run(xenbus_thread, NULL, "xenbus"); |
| 857 | if (IS_ERR(task)) |
| 858 | return PTR_ERR(task); |
| 859 | |
| 860 | return 0; |
| 861 | } |