drivers/xen/xenfs/xenbus.c

   1 /*
   2  * Driver giving user-space access to the kernel's xenbus connection
   3  * to xenstore.
   4  *
   5  * Copyright (c) 2005, Christian Limpach
   6  * Copyright (c) 2005, Rusty Russell, IBM Corporation
   7  *
   8  * This program is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU General Public License version 2
  10  * as published by the Free Software Foundation; or, when distributed
  11  * separately from the Linux kernel or incorporated into other
  12  * software packages, subject to the following license:
  13  *
  14  * Permission is hereby granted, free of charge, to any person obtaining a copy
  15  * of this source file (the "Software"), to deal in the Software without
  16  * restriction, including without limitation the rights to use, copy, modify,
  17  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  18  * and to permit persons to whom the Software is furnished to do so, subject to
  19  * the following conditions:
  20  *
  21  * The above copyright notice and this permission notice shall be included in
  22  * all copies or substantial portions of the Software.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  27  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  29  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  30  * IN THE SOFTWARE.
  31  *
  32  * Changes:
  33  * 2008-10-07  Alex Zeffertt    Replaced /proc/xen/xenbus with xenfs filesystem
  34  *                              and /proc/xen compatibility mount point.
  35  *                              Turned xenfs into a loadable module.
  36  */
  37
  38 #include <linux/kernel.h>
  39 #include <linux/errno.h>
  40 #include <linux/uio.h>
  41 #include <linux/notifier.h>
  42 #include <linux/wait.h>
  43 #include <linux/fs.h>
  44 #include <linux/poll.h>
  45 #include <linux/mutex.h>
  46 #include <linux/sched.h>
  47 #include <linux/spinlock.h>
  48 #include <linux/mount.h>
  49 #include <linux/pagemap.h>
  50 #include <linux/uaccess.h>
  51 #include <linux/init.h>
  52 #include <linux/namei.h>
  53 #include <linux/string.h>
  54 #include <linux/slab.h>
  55
  56 #include "xenfs.h"
  57 #include "../xenbus/xenbus_comms.h"
  58
  59 #include <xen/xenbus.h>
  60 #include <asm/xen/hypervisor.h>
  61
  62 /*
  63  * An element of a list of outstanding transactions, for which we're
  64  * still waiting a reply.
  65  */
  66 struct xenbus_transaction_holder {
  67         struct list_head list;
  68         struct xenbus_transaction handle;
  69 };
  70
  71 /*
  72  * A buffer of data on the queue.
  73  */
  74 struct read_buffer {
  75         struct list_head list;
  76         unsigned int cons;
  77         unsigned int len;
  78         char msg[];
  79 };
  80
  81 struct xenbus_file_priv {
  82         /*
  83          * msgbuffer_mutex is held while partial requests are built up
  84          * and complete requests are acted on.  It therefore protects
  85          * the "transactions" and "watches" lists, and the partial
  86          * request length and buffer.
  87          *
  88          * reply_mutex protects the reply being built up to return to
  89          * usermode.  It nests inside msgbuffer_mutex but may be held
  90          * alone during a watch callback.
  91          */
  92         struct mutex msgbuffer_mutex;
  93
  94         /* In-progress transactions */
  95         struct list_head transactions;
  96
  97         /* Active watches. */
  98         struct list_head watches;
  99
 100         /* Partial request. */
 101         unsigned int len;
 102         union {
 103                 struct xsd_sockmsg msg;
 104                 char buffer[PAGE_SIZE];
 105         } u;
 106
 107         /* Response queue. */
 108         struct mutex reply_mutex;
 109         struct list_head read_buffers;
 110         wait_queue_head_t read_waitq;
 111
 112 };
 113
 114 /* Read out any raw xenbus messages queued up. */
 115 static ssize_t xenbus_file_read(struct file *filp,
 116                                char __user *ubuf,
 117                                size_t len, loff_t *ppos)
 118 {
 119         struct xenbus_file_priv *u = filp->private_data;
 120         struct read_buffer *rb;
 121         unsigned i;
 122         int ret;
 123
 124         mutex_lock(&u->reply_mutex);
 125         while (list_empty(&u->read_buffers)) {
 126                 mutex_unlock(&u->reply_mutex);
 127                 ret = wait_event_interruptible(u->read_waitq,
 128                                                !list_empty(&u->read_buffers));
 129                 if (ret)
 130                         return ret;
 131                 mutex_lock(&u->reply_mutex);
 132         }
 133
 134         rb = list_entry(u->read_buffers.next, struct read_buffer, list);
 135         i = 0;
 136         while (i < len) {
 137                 unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
 138
 139                 ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
 140
 141                 i += sz - ret;
 142                 rb->cons += sz - ret;
 143
 144                 if (ret != sz) {
 145                         if (i == 0)
 146                                 i = -EFAULT;
 147                         goto out;
 148                 }
 149
 150                 /* Clear out buffer if it has been consumed */
 151                 if (rb->cons == rb->len) {
 152                         list_del(&rb->list);
 153                         kfree(rb);
 154                         if (list_empty(&u->read_buffers))
 155                                 break;
 156                         rb = list_entry(u->read_buffers.next,
 157                                         struct read_buffer, list);
 158                 }
 159         }
 160
 161 out:
 162         mutex_unlock(&u->reply_mutex);
 163         return i;
 164 }
 165
 166 /*
 167  * Add a buffer to the queue.  Caller must hold the appropriate lock
 168  * if the queue is not local.  (Commonly the caller will build up
 169  * multiple queued buffers on a temporary local list, and then add it
 170  * to the appropriate list under lock once all the buffers have een
 171  * successfully allocated.)
 172  */
 173 static int queue_reply(struct list_head *queue, const void *data, size_t len)
 174 {
 175         struct read_buffer *rb;
 176
 177         if (len == 0)
 178                 return 0;
 179
 180         rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
 181         if (rb == NULL)
 182                 return -ENOMEM;
 183
 184         rb->cons = 0;
 185         rb->len = len;
 186
 187         memcpy(rb->msg, data, len);
 188
 189         list_add_tail(&rb->list, queue);
 190         return 0;
 191 }
 192
 193 /*
 194  * Free all the read_buffer s on a list.
 195  * Caller must have sole reference to list.
 196  */
 197 static void queue_cleanup(struct list_head *list)
 198 {
 199         struct read_buffer *rb;
 200
 201         while (!list_empty(list)) {
 202                 rb = list_entry(list->next, struct read_buffer, list);
 203                 list_del(list->next);
 204                 kfree(rb);
 205         }
 206 }
 207
 208 struct watch_adapter {
 209         struct list_head list;
 210         struct xenbus_watch watch;
 211         struct xenbus_file_priv *dev_data;
 212         char *token;
 213 };
 214
 215 static void free_watch_adapter(struct watch_adapter *watch)
 216 {
 217         kfree(watch->watch.node);
 218         kfree(watch->token);
 219         kfree(watch);
 220 }
 221
 222 static struct watch_adapter *alloc_watch_adapter(const char *path,
 223                                                  const char *token)
 224 {
 225         struct watch_adapter *watch;
 226
 227         watch = kzalloc(sizeof(*watch), GFP_KERNEL);
 228         if (watch == NULL)
 229                 goto out_fail;
 230
 231         watch->watch.node = kstrdup(path, GFP_KERNEL);
 232         if (watch->watch.node == NULL)
 233                 goto out_free;
 234
 235         watch->token = kstrdup(token, GFP_KERNEL);
 236         if (watch->token == NULL)
 237                 goto out_free;
 238
 239         return watch;
 240
 241 out_free:
 242         free_watch_adapter(watch);
 243
 244 out_fail:
 245         return NULL;
 246 }
 247
 248 static void watch_fired(struct xenbus_watch *watch,
 249                         const char **vec,
 250                         unsigned int len)
 251 {
 252         struct watch_adapter *adap;
 253         struct xsd_sockmsg hdr;
 254         const char *path, *token;
 255         int path_len, tok_len, body_len, data_len = 0;
 256         int ret;
 257         LIST_HEAD(staging_q);
 258
 259         adap = container_of(watch, struct watch_adapter, watch);
 260
 261         path = vec[XS_WATCH_PATH];
 262         token = adap->token;
 263
 264         path_len = strlen(path) + 1;
 265         tok_len = strlen(token) + 1;
 266         if (len > 2)
 267                 data_len = vec[len] - vec[2] + 1;
 268         body_len = path_len + tok_len + data_len;
 269
 270         hdr.type = XS_WATCH_EVENT;
 271         hdr.len = body_len;
 272
 273         mutex_lock(&adap->dev_data->reply_mutex);
 274
 275         ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
 276         if (!ret)
 277                 ret = queue_reply(&staging_q, path, path_len);
 278         if (!ret)
 279                 ret = queue_reply(&staging_q, token, tok_len);
 280         if (!ret && len > 2)
 281                 ret = queue_reply(&staging_q, vec[2], data_len);
 282
 283         if (!ret) {
 284                 /* success: pass reply list onto watcher */
 285                 list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
 286                 wake_up(&adap->dev_data->read_waitq);
 287         } else
 288                 queue_cleanup(&staging_q);
 289
 290         mutex_unlock(&adap->dev_data->reply_mutex);
 291 }
 292
 293 static int xenbus_write_transaction(unsigned msg_type,
 294                                     struct xenbus_file_priv *u)
 295 {
 296         int rc;
 297         void *reply;
 298         struct xenbus_transaction_holder *trans = NULL;
 299         LIST_HEAD(staging_q);
 300
 301         if (msg_type == XS_TRANSACTION_START) {
 302                 trans = kmalloc(sizeof(*trans), GFP_KERNEL);
 303                 if (!trans) {
 304                         rc = -ENOMEM;
 305                         goto out;
 306                 }
 307         }
 308
 309         reply = xenbus_dev_request_and_reply(&u->u.msg);
 310         if (IS_ERR(reply)) {
 311                 kfree(trans);
 312                 rc = PTR_ERR(reply);
 313                 goto out;
 314         }
 315
 316         if (msg_type == XS_TRANSACTION_START) {
 317                 trans->handle.id = simple_strtoul(reply, NULL, 0);
 318
 319                 list_add(&trans->list, &u->transactions);
 320         } else if (msg_type == XS_TRANSACTION_END) {
 321                 list_for_each_entry(trans, &u->transactions, list)
 322                         if (trans->handle.id == u->u.msg.tx_id)
 323                                 break;
 324                 BUG_ON(&trans->list == &u->transactions);
 325                 list_del(&trans->list);
 326
 327                 kfree(trans);
 328         }
 329
 330         mutex_lock(&u->reply_mutex);
 331         rc = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
 332         if (!rc)
 333                 rc = queue_reply(&staging_q, reply, u->u.msg.len);
 334         if (!rc) {
 335                 list_splice_tail(&staging_q, &u->read_buffers);
 336                 wake_up(&u->read_waitq);
 337         } else {
 338                 queue_cleanup(&staging_q);
 339         }
 340         mutex_unlock(&u->reply_mutex);
 341
 342         kfree(reply);
 343
 344 out:
 345         return rc;
 346 }
 347
 348 static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
 349 {
 350         struct watch_adapter *watch, *tmp_watch;
 351         char *path, *token;
 352         int err, rc;
 353         LIST_HEAD(staging_q);
 354
 355         path = u->u.buffer + sizeof(u->u.msg);
 356         token = memchr(path, 0, u->u.msg.len);
 357         if (token == NULL) {
 358                 rc = -EILSEQ;
 359                 goto out;
 360         }
 361         token++;
 362
 363         if (msg_type == XS_WATCH) {
 364                 watch = alloc_watch_adapter(path, token);
 365                 if (watch == NULL) {
 366                         rc = -ENOMEM;
 367                         goto out;
 368                 }
 369
 370                 watch->watch.callback = watch_fired;
 371                 watch->dev_data = u;
 372
 373                 err = register_xenbus_watch(&watch->watch);
 374                 if (err) {
 375                         free_watch_adapter(watch);
 376                         rc = err;
 377                         goto out;
 378                 }
 379                 list_add(&watch->list, &u->watches);
 380         } else {
 381                 list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
 382                         if (!strcmp(watch->token, token) &&
 383                             !strcmp(watch->watch.node, path)) {
 384                                 unregister_xenbus_watch(&watch->watch);
 385                                 list_del(&watch->list);
 386                                 free_watch_adapter(watch);
 387                                 break;
 388                         }
 389                 }
 390         }
 391
 392         /* Success.  Synthesize a reply to say all is OK. */
 393         {
 394                 struct {
 395                         struct xsd_sockmsg hdr;
 396                         char body[3];
 397                 } __packed reply = {
 398                         {
 399                                 .type = msg_type,
 400                                 .len = sizeof(reply.body)
 401                         },
 402                         "OK"
 403                 };
 404
 405                 mutex_lock(&u->reply_mutex);
 406                 rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
 407                 mutex_unlock(&u->reply_mutex);
 408         }
 409
 410 out:
 411         return rc;
 412 }
 413
 414 static ssize_t xenbus_file_write(struct file *filp,
 415                                 const char __user *ubuf,
 416                                 size_t len, loff_t *ppos)
 417 {
 418         struct xenbus_file_priv *u = filp->private_data;
 419         uint32_t msg_type;
 420         int rc = len;
 421         int ret;
 422         LIST_HEAD(staging_q);
 423
 424         /*
 425          * We're expecting usermode to be writing properly formed
 426          * xenbus messages.  If they write an incomplete message we
 427          * buffer it up.  Once it is complete, we act on it.
 428          */
 429
 430         /*
 431          * Make sure concurrent writers can't stomp all over each
 432          * other's messages and make a mess of our partial message
 433          * buffer.  We don't make any attemppt to stop multiple
 434          * writers from making a mess of each other's incomplete
 435          * messages; we're just trying to guarantee our own internal
 436          * consistency and make sure that single writes are handled
 437          * atomically.
 438          */
 439         mutex_lock(&u->msgbuffer_mutex);
 440
 441         /* Get this out of the way early to avoid confusion */
 442         if (len == 0)
 443                 goto out;
 444
 445         /* Can't write a xenbus message larger we can buffer */
 446         if ((len + u->len) > sizeof(u->u.buffer)) {
 447                 /* On error, dump existing buffer */
 448                 u->len = 0;
 449                 rc = -EINVAL;
 450                 goto out;
 451         }
 452
 453         ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
 454
 455         if (ret == len) {
 456                 rc = -EFAULT;
 457                 goto out;
 458         }
 459
 460         /* Deal with a partial copy. */
 461         len -= ret;
 462         rc = len;
 463
 464         u->len += len;
 465
 466         /* Return if we haven't got a full message yet */
 467         if (u->len < sizeof(u->u.msg))
 468                 goto out;       /* not even the header yet */
 469
 470         /* If we're expecting a message that's larger than we can
 471            possibly send, dump what we have and return an error. */
 472         if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
 473                 rc = -E2BIG;
 474                 u->len = 0;
 475                 goto out;
 476         }
 477
 478         if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
 479                 goto out;       /* incomplete data portion */
 480
 481         /*
 482          * OK, now we have a complete message.  Do something with it.
 483          */
 484
 485         msg_type = u->u.msg.type;
 486
 487         switch (msg_type) {
 488         case XS_TRANSACTION_START:
 489         case XS_TRANSACTION_END:
 490         case XS_DIRECTORY:
 491         case XS_READ:
 492         case XS_GET_PERMS:
 493         case XS_RELEASE:
 494         case XS_GET_DOMAIN_PATH:
 495         case XS_WRITE:
 496         case XS_MKDIR:
 497         case XS_RM:
 498         case XS_SET_PERMS:
 499                 /* Send out a transaction */
 500                 ret = xenbus_write_transaction(msg_type, u);
 501                 break;
 502
 503         case XS_WATCH:
 504         case XS_UNWATCH:
 505                 /* (Un)Ask for some path to be watched for changes */
 506                 ret = xenbus_write_watch(msg_type, u);
 507                 break;
 508
 509         default:
 510                 ret = -EINVAL;
 511                 break;
 512         }
 513         if (ret != 0)
 514                 rc = ret;
 515
 516         /* Buffered message consumed */
 517         u->len = 0;
 518
 519  out:
 520         mutex_unlock(&u->msgbuffer_mutex);
 521         return rc;
 522 }
 523
 524 static int xenbus_file_open(struct inode *inode, struct file *filp)
 525 {
 526         struct xenbus_file_priv *u;
 527
 528         if (xen_store_evtchn == 0)
 529                 return -ENOENT;
 530
 531         nonseekable_open(inode, filp);
 532
 533         u = kzalloc(sizeof(*u), GFP_KERNEL);
 534         if (u == NULL)
 535                 return -ENOMEM;
 536
 537         INIT_LIST_HEAD(&u->transactions);
 538         INIT_LIST_HEAD(&u->watches);
 539         INIT_LIST_HEAD(&u->read_buffers);
 540         init_waitqueue_head(&u->read_waitq);
 541
 542         mutex_init(&u->reply_mutex);
 543         mutex_init(&u->msgbuffer_mutex);
 544
 545         filp->private_data = u;
 546
 547         return 0;
 548 }
 549
 550 static int xenbus_file_release(struct inode *inode, struct file *filp)
 551 {
 552         struct xenbus_file_priv *u = filp->private_data;
 553         struct xenbus_transaction_holder *trans, *tmp;
 554         struct watch_adapter *watch, *tmp_watch;
 555
 556         /*
 557          * No need for locking here because there are no other users,
 558          * by definition.
 559          */
 560
 561         list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
 562                 xenbus_transaction_end(trans->handle, 1);
 563                 list_del(&trans->list);
 564                 kfree(trans);
 565         }
 566
 567         list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
 568                 unregister_xenbus_watch(&watch->watch);
 569                 list_del(&watch->list);
 570                 free_watch_adapter(watch);
 571         }
 572
 573         kfree(u);
 574
 575         return 0;
 576 }
 577
 578 static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
 579 {
 580         struct xenbus_file_priv *u = file->private_data;
 581
 582         poll_wait(file, &u->read_waitq, wait);
 583         if (!list_empty(&u->read_buffers))
 584                 return POLLIN | POLLRDNORM;
 585         return 0;
 586 }
 587
 588 const struct file_operations xenbus_file_ops = {
 589         .read = xenbus_file_read,
 590         .write = xenbus_file_write,
 591         .open = xenbus_file_open,
 592         .release = xenbus_file_release,
 593         .poll = xenbus_file_poll,
 594 };