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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 39 40 #include <linux/kernel.h> 41 #include <linux/errno.h> 42 #include <linux/uio.h> 43 #include <linux/notifier.h> 44 #include <linux/wait.h> 45 #include <linux/fs.h> 46 #include <linux/poll.h> 47 #include <linux/mutex.h> 48 #include <linux/sched.h> 49 #include <linux/spinlock.h> 50 #include <linux/mount.h> 51 #include <linux/pagemap.h> 52 #include <linux/uaccess.h> 53 #include <linux/init.h> 54 #include <linux/namei.h> 55 #include <linux/string.h> 56 #include <linux/slab.h> 57 #include <linux/miscdevice.h> 58 #include <linux/init.h> 59 60 #include "xenbus_comms.h" 61 62 #include <xen/xenbus.h> 63 #include <xen/xen.h> 64 #include <asm/xen/hypervisor.h> 65 66 /* 67 * An element of a list of outstanding transactions, for which we're 68 * still waiting a reply. 69 */ 70 struct xenbus_transaction_holder { 71 struct list_head list; 72 struct xenbus_transaction handle; 73 }; 74 75 /* 76 * A buffer of data on the queue. 77 */ 78 struct read_buffer { 79 struct list_head list; 80 unsigned int cons; 81 unsigned int len; 82 char msg[]; 83 }; 84 85 struct xenbus_file_priv { 86 /* 87 * msgbuffer_mutex is held while partial requests are built up 88 * and complete requests are acted on. It therefore protects 89 * the "transactions" and "watches" lists, and the partial 90 * request length and buffer. 91 * 92 * reply_mutex protects the reply being built up to return to 93 * usermode. It nests inside msgbuffer_mutex but may be held 94 * alone during a watch callback. 95 */ 96 struct mutex msgbuffer_mutex; 97 98 /* In-progress transactions */ 99 struct list_head transactions; 100 101 /* Active watches. */ 102 struct list_head watches; 103 104 /* Partial request. */ 105 unsigned int len; 106 union { 107 struct xsd_sockmsg msg; 108 char buffer[XENSTORE_PAYLOAD_MAX]; 109 } u; 110 111 /* Response queue. */ 112 struct mutex reply_mutex; 113 struct list_head read_buffers; 114 wait_queue_head_t read_waitq; 115 116 }; 117 118 /* Read out any raw xenbus messages queued up. */ 119 static ssize_t xenbus_file_read(struct file *filp, 120 char __user *ubuf, 121 size_t len, loff_t *ppos) 122 { 123 struct xenbus_file_priv *u = filp->private_data; 124 struct read_buffer *rb; 125 unsigned i; 126 int ret; 127 128 mutex_lock(&u->reply_mutex); 129 again: 130 while (list_empty(&u->read_buffers)) { 131 mutex_unlock(&u->reply_mutex); 132 if (filp->f_flags & O_NONBLOCK) 133 return -EAGAIN; 134 135 ret = wait_event_interruptible(u->read_waitq, 136 !list_empty(&u->read_buffers)); 137 if (ret) 138 return ret; 139 mutex_lock(&u->reply_mutex); 140 } 141 142 rb = list_entry(u->read_buffers.next, struct read_buffer, list); 143 i = 0; 144 while (i < len) { 145 unsigned sz = min((unsigned)len - i, rb->len - rb->cons); 146 147 ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz); 148 149 i += sz - ret; 150 rb->cons += sz - ret; 151 152 if (ret != 0) { 153 if (i == 0) 154 i = -EFAULT; 155 goto out; 156 } 157 158 /* Clear out buffer if it has been consumed */ 159 if (rb->cons == rb->len) { 160 list_del(&rb->list); 161 kfree(rb); 162 if (list_empty(&u->read_buffers)) 163 break; 164 rb = list_entry(u->read_buffers.next, 165 struct read_buffer, list); 166 } 167 } 168 if (i == 0) 169 goto again; 170 171 out: 172 mutex_unlock(&u->reply_mutex); 173 return i; 174 } 175 176 /* 177 * Add a buffer to the queue. Caller must hold the appropriate lock 178 * if the queue is not local. (Commonly the caller will build up 179 * multiple queued buffers on a temporary local list, and then add it 180 * to the appropriate list under lock once all the buffers have een 181 * successfully allocated.) 182 */ 183 static int queue_reply(struct list_head *queue, const void *data, size_t len) 184 { 185 struct read_buffer *rb; 186 187 if (len == 0) 188 return 0; 189 if (len > XENSTORE_PAYLOAD_MAX) 190 return -EINVAL; 191 192 rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL); 193 if (rb == NULL) 194 return -ENOMEM; 195 196 rb->cons = 0; 197 rb->len = len; 198 199 memcpy(rb->msg, data, len); 200 201 list_add_tail(&rb->list, queue); 202 return 0; 203 } 204 205 /* 206 * Free all the read_buffer s on a list. 207 * Caller must have sole reference to list. 208 */ 209 static void queue_cleanup(struct list_head *list) 210 { 211 struct read_buffer *rb; 212 213 while (!list_empty(list)) { 214 rb = list_entry(list->next, struct read_buffer, list); 215 list_del(list->next); 216 kfree(rb); 217 } 218 } 219 220 struct watch_adapter { 221 struct list_head list; 222 struct xenbus_watch watch; 223 struct xenbus_file_priv *dev_data; 224 char *token; 225 }; 226 227 static void free_watch_adapter(struct watch_adapter *watch) 228 { 229 kfree(watch->watch.node); 230 kfree(watch->token); 231 kfree(watch); 232 } 233 234 static struct watch_adapter *alloc_watch_adapter(const char *path, 235 const char *token) 236 { 237 struct watch_adapter *watch; 238 239 watch = kzalloc(sizeof(*watch), GFP_KERNEL); 240 if (watch == NULL) 241 goto out_fail; 242 243 watch->watch.node = kstrdup(path, GFP_KERNEL); 244 if (watch->watch.node == NULL) 245 goto out_free; 246 247 watch->token = kstrdup(token, GFP_KERNEL); 248 if (watch->token == NULL) 249 goto out_free; 250 251 return watch; 252 253 out_free: 254 free_watch_adapter(watch); 255 256 out_fail: 257 return NULL; 258 } 259 260 static void watch_fired(struct xenbus_watch *watch, 261 const char **vec, 262 unsigned int len) 263 { 264 struct watch_adapter *adap; 265 struct xsd_sockmsg hdr; 266 const char *path, *token; 267 int path_len, tok_len, body_len, data_len = 0; 268 int ret; 269 LIST_HEAD(staging_q); 270 271 adap = container_of(watch, struct watch_adapter, watch); 272 273 path = vec[XS_WATCH_PATH]; 274 token = adap->token; 275 276 path_len = strlen(path) + 1; 277 tok_len = strlen(token) + 1; 278 if (len > 2) 279 data_len = vec[len] - vec[2] + 1; 280 body_len = path_len + tok_len + data_len; 281 282 hdr.type = XS_WATCH_EVENT; 283 hdr.len = body_len; 284 285 mutex_lock(&adap->dev_data->reply_mutex); 286 287 ret = queue_reply(&staging_q, &hdr, sizeof(hdr)); 288 if (!ret) 289 ret = queue_reply(&staging_q, path, path_len); 290 if (!ret) 291 ret = queue_reply(&staging_q, token, tok_len); 292 if (!ret && len > 2) 293 ret = queue_reply(&staging_q, vec[2], data_len); 294 295 if (!ret) { 296 /* success: pass reply list onto watcher */ 297 list_splice_tail(&staging_q, &adap->dev_data->read_buffers); 298 wake_up(&adap->dev_data->read_waitq); 299 } else 300 queue_cleanup(&staging_q); 301 302 mutex_unlock(&adap->dev_data->reply_mutex); 303 } 304 305 static int xenbus_command_reply(struct xenbus_file_priv *u, 306 unsigned int msg_type, const char *reply) 307 { 308 struct { 309 struct xsd_sockmsg hdr; 310 const char body[16]; 311 } msg; 312 int rc; 313 314 msg.hdr = u->u.msg; 315 msg.hdr.type = msg_type; 316 msg.hdr.len = strlen(reply) + 1; 317 if (msg.hdr.len > sizeof(msg.body)) 318 return -E2BIG; 319 320 mutex_lock(&u->reply_mutex); 321 rc = queue_reply(&u->read_buffers, &msg, sizeof(msg.hdr) + msg.hdr.len); 322 wake_up(&u->read_waitq); 323 mutex_unlock(&u->reply_mutex); 324 325 return rc; 326 } 327 328 static int xenbus_write_transaction(unsigned msg_type, 329 struct xenbus_file_priv *u) 330 { 331 int rc; 332 void *reply; 333 struct xenbus_transaction_holder *trans = NULL; 334 LIST_HEAD(staging_q); 335 336 if (msg_type == XS_TRANSACTION_START) { 337 trans = kmalloc(sizeof(*trans), GFP_KERNEL); 338 if (!trans) { 339 rc = -ENOMEM; 340 goto out; 341 } 342 } else if (u->u.msg.tx_id != 0) { 343 list_for_each_entry(trans, &u->transactions, list) 344 if (trans->handle.id == u->u.msg.tx_id) 345 break; 346 if (&trans->list == &u->transactions) 347 return xenbus_command_reply(u, XS_ERROR, "ENOENT"); 348 } 349 350 reply = xenbus_dev_request_and_reply(&u->u.msg); 351 if (IS_ERR(reply)) { 352 if (msg_type == XS_TRANSACTION_START) 353 kfree(trans); 354 rc = PTR_ERR(reply); 355 goto out; 356 } 357 358 if (msg_type == XS_TRANSACTION_START) { 359 if (u->u.msg.type == XS_ERROR) 360 kfree(trans); 361 else { 362 trans->handle.id = simple_strtoul(reply, NULL, 0); 363 list_add(&trans->list, &u->transactions); 364 } 365 } else if (u->u.msg.type == XS_TRANSACTION_END) { 366 list_del(&trans->list); 367 kfree(trans); 368 } 369 370 mutex_lock(&u->reply_mutex); 371 rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg)); 372 if (!rc) 373 rc = queue_reply(&staging_q, reply, u->u.msg.len); 374 if (!rc) { 375 list_splice_tail(&staging_q, &u->read_buffers); 376 wake_up(&u->read_waitq); 377 } else { 378 queue_cleanup(&staging_q); 379 } 380 mutex_unlock(&u->reply_mutex); 381 382 kfree(reply); 383 384 out: 385 return rc; 386 } 387 388 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u) 389 { 390 struct watch_adapter *watch; 391 char *path, *token; 392 int err, rc; 393 LIST_HEAD(staging_q); 394 395 path = u->u.buffer + sizeof(u->u.msg); 396 token = memchr(path, 0, u->u.msg.len); 397 if (token == NULL) { 398 rc = xenbus_command_reply(u, XS_ERROR, "EINVAL"); 399 goto out; 400 } 401 token++; 402 if (memchr(token, 0, u->u.msg.len - (token - path)) == NULL) { 403 rc = xenbus_command_reply(u, XS_ERROR, "EINVAL"); 404 goto out; 405 } 406 407 if (msg_type == XS_WATCH) { 408 watch = alloc_watch_adapter(path, token); 409 if (watch == NULL) { 410 rc = -ENOMEM; 411 goto out; 412 } 413 414 watch->watch.callback = watch_fired; 415 watch->dev_data = u; 416 417 err = register_xenbus_watch(&watch->watch); 418 if (err) { 419 free_watch_adapter(watch); 420 rc = err; 421 goto out; 422 } 423 list_add(&watch->list, &u->watches); 424 } else { 425 list_for_each_entry(watch, &u->watches, list) { 426 if (!strcmp(watch->token, token) && 427 !strcmp(watch->watch.node, path)) { 428 unregister_xenbus_watch(&watch->watch); 429 list_del(&watch->list); 430 free_watch_adapter(watch); 431 break; 432 } 433 } 434 } 435 436 /* Success. Synthesize a reply to say all is OK. */ 437 rc = xenbus_command_reply(u, msg_type, "OK"); 438 439 out: 440 return rc; 441 } 442 443 static ssize_t xenbus_file_write(struct file *filp, 444 const char __user *ubuf, 445 size_t len, loff_t *ppos) 446 { 447 struct xenbus_file_priv *u = filp->private_data; 448 uint32_t msg_type; 449 int rc = len; 450 int ret; 451 LIST_HEAD(staging_q); 452 453 /* 454 * We're expecting usermode to be writing properly formed 455 * xenbus messages. If they write an incomplete message we 456 * buffer it up. Once it is complete, we act on it. 457 */ 458 459 /* 460 * Make sure concurrent writers can't stomp all over each 461 * other's messages and make a mess of our partial message 462 * buffer. We don't make any attemppt to stop multiple 463 * writers from making a mess of each other's incomplete 464 * messages; we're just trying to guarantee our own internal 465 * consistency and make sure that single writes are handled 466 * atomically. 467 */ 468 mutex_lock(&u->msgbuffer_mutex); 469 470 /* Get this out of the way early to avoid confusion */ 471 if (len == 0) 472 goto out; 473 474 /* Can't write a xenbus message larger we can buffer */ 475 if (len > sizeof(u->u.buffer) - u->len) { 476 /* On error, dump existing buffer */ 477 u->len = 0; 478 rc = -EINVAL; 479 goto out; 480 } 481 482 ret = copy_from_user(u->u.buffer + u->len, ubuf, len); 483 484 if (ret != 0) { 485 rc = -EFAULT; 486 goto out; 487 } 488 489 /* Deal with a partial copy. */ 490 len -= ret; 491 rc = len; 492 493 u->len += len; 494 495 /* Return if we haven't got a full message yet */ 496 if (u->len < sizeof(u->u.msg)) 497 goto out; /* not even the header yet */ 498 499 /* If we're expecting a message that's larger than we can 500 possibly send, dump what we have and return an error. */ 501 if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) { 502 rc = -E2BIG; 503 u->len = 0; 504 goto out; 505 } 506 507 if (u->len < (sizeof(u->u.msg) + u->u.msg.len)) 508 goto out; /* incomplete data portion */ 509 510 /* 511 * OK, now we have a complete message. Do something with it. 512 */ 513 514 msg_type = u->u.msg.type; 515 516 switch (msg_type) { 517 case XS_WATCH: 518 case XS_UNWATCH: 519 /* (Un)Ask for some path to be watched for changes */ 520 ret = xenbus_write_watch(msg_type, u); 521 break; 522 523 default: 524 /* Send out a transaction */ 525 ret = xenbus_write_transaction(msg_type, u); 526 break; 527 } 528 if (ret != 0) 529 rc = ret; 530 531 /* Buffered message consumed */ 532 u->len = 0; 533 534 out: 535 mutex_unlock(&u->msgbuffer_mutex); 536 return rc; 537 } 538 539 static int xenbus_file_open(struct inode *inode, struct file *filp) 540 { 541 struct xenbus_file_priv *u; 542 543 if (xen_store_evtchn == 0) 544 return -ENOENT; 545 546 nonseekable_open(inode, filp); 547 548 filp->f_mode &= ~FMODE_ATOMIC_POS; /* cdev-style semantics */ 549 550 u = kzalloc(sizeof(*u), GFP_KERNEL); 551 if (u == NULL) 552 return -ENOMEM; 553 554 INIT_LIST_HEAD(&u->transactions); 555 INIT_LIST_HEAD(&u->watches); 556 INIT_LIST_HEAD(&u->read_buffers); 557 init_waitqueue_head(&u->read_waitq); 558 559 mutex_init(&u->reply_mutex); 560 mutex_init(&u->msgbuffer_mutex); 561 562 filp->private_data = u; 563 564 return 0; 565 } 566 567 static int xenbus_file_release(struct inode *inode, struct file *filp) 568 { 569 struct xenbus_file_priv *u = filp->private_data; 570 struct xenbus_transaction_holder *trans, *tmp; 571 struct watch_adapter *watch, *tmp_watch; 572 struct read_buffer *rb, *tmp_rb; 573 574 /* 575 * No need for locking here because there are no other users, 576 * by definition. 577 */ 578 579 list_for_each_entry_safe(trans, tmp, &u->transactions, list) { 580 xenbus_transaction_end(trans->handle, 1); 581 list_del(&trans->list); 582 kfree(trans); 583 } 584 585 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) { 586 unregister_xenbus_watch(&watch->watch); 587 list_del(&watch->list); 588 free_watch_adapter(watch); 589 } 590 591 list_for_each_entry_safe(rb, tmp_rb, &u->read_buffers, list) { 592 list_del(&rb->list); 593 kfree(rb); 594 } 595 kfree(u); 596 597 return 0; 598 } 599 600 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait) 601 { 602 struct xenbus_file_priv *u = file->private_data; 603 604 poll_wait(file, &u->read_waitq, wait); 605 if (!list_empty(&u->read_buffers)) 606 return POLLIN | POLLRDNORM; 607 return 0; 608 } 609 610 const struct file_operations xen_xenbus_fops = { 611 .read = xenbus_file_read, 612 .write = xenbus_file_write, 613 .open = xenbus_file_open, 614 .release = xenbus_file_release, 615 .poll = xenbus_file_poll, 616 .llseek = no_llseek, 617 }; 618 EXPORT_SYMBOL_GPL(xen_xenbus_fops); 619 620 static struct miscdevice xenbus_dev = { 621 .minor = MISC_DYNAMIC_MINOR, 622 .name = "xen/xenbus", 623 .fops = &xen_xenbus_fops, 624 }; 625 626 static int __init xenbus_init(void) 627 { 628 int err; 629 630 if (!xen_domain()) 631 return -ENODEV; 632 633 err = misc_register(&xenbus_dev); 634 if (err) 635 pr_err("Could not register xenbus frontend device\n"); 636 return err; 637 } 638 device_initcall(xenbus_init); 639