fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/version.h>
  38 #include <linux/xattr.h>
  39 #include <linux/vmalloc.h>
  40 #include "ctree.h"
  41 #include "disk-io.h"
  42 #include "transaction.h"
  43 #include "btrfs_inode.h"
  44 #include "ioctl.h"
  45 #include "print-tree.h"
  46 #include "volumes.h"
  47 #include "locking.h"
  48
  49
  50
  51 static noinline int create_subvol(struct btrfs_root *root, char *name,
  52                                   int namelen)
  53 {
  54         struct btrfs_trans_handle *trans;
  55         struct btrfs_key key;
  56         struct btrfs_root_item root_item;
  57         struct btrfs_inode_item *inode_item;
  58         struct extent_buffer *leaf;
  59         struct btrfs_root *new_root = root;
  60         struct inode *dir;
  61         int ret;
  62         int err;
  63         u64 objectid;
  64         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
  65         unsigned long nr = 1;
  66
  67         ret = btrfs_check_free_space(root, 1, 0);
  68         if (ret)
  69                 goto fail_commit;
  70
  71         trans = btrfs_start_transaction(root, 1);
  72         BUG_ON(!trans);
  73
  74         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
  75                                        0, &objectid);
  76         if (ret)
  77                 goto fail;
  78
  79         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
  80                                       objectid, trans->transid, 0, 0, 0);
  81         if (IS_ERR(leaf)) {
  82                 ret = PTR_ERR(leaf);
  83                 goto fail;
  84         }
  85
  86         btrfs_set_header_nritems(leaf, 0);
  87         btrfs_set_header_level(leaf, 0);
  88         btrfs_set_header_bytenr(leaf, leaf->start);
  89         btrfs_set_header_generation(leaf, trans->transid);
  90         btrfs_set_header_owner(leaf, objectid);
  91
  92         write_extent_buffer(leaf, root->fs_info->fsid,
  93                             (unsigned long)btrfs_header_fsid(leaf),
  94                             BTRFS_FSID_SIZE);
  95         btrfs_mark_buffer_dirty(leaf);
  96
  97         inode_item = &root_item.inode;
  98         memset(inode_item, 0, sizeof(*inode_item));
  99         inode_item->generation = cpu_to_le64(1);
 100         inode_item->size = cpu_to_le64(3);
 101         inode_item->nlink = cpu_to_le32(1);
 102         inode_item->nbytes = cpu_to_le64(root->leafsize);
 103         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 104
 105         btrfs_set_root_bytenr(&root_item, leaf->start);
 106         btrfs_set_root_level(&root_item, 0);
 107         btrfs_set_root_refs(&root_item, 1);
 108         btrfs_set_root_used(&root_item, 0);
 109
 110         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 111         root_item.drop_level = 0;
 112
 113         btrfs_tree_unlock(leaf);
 114         free_extent_buffer(leaf);
 115         leaf = NULL;
 116
 117         btrfs_set_root_dirid(&root_item, new_dirid);
 118
 119         key.objectid = objectid;
 120         key.offset = 1;
 121         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 122         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 123                                 &root_item);
 124         if (ret)
 125                 goto fail;
 126
 127         /*
 128          * insert the directory item
 129          */
 130         key.offset = (u64)-1;
 131         dir = root->fs_info->sb->s_root->d_inode;
 132         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 133                                     name, namelen, dir->i_ino, &key,
 134                                     BTRFS_FT_DIR, 0);
 135         if (ret)
 136                 goto fail;
 137
 138         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 139                              name, namelen, objectid,
 140                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 141         if (ret)
 142                 goto fail;
 143
 144         ret = btrfs_commit_transaction(trans, root);
 145         if (ret)
 146                 goto fail_commit;
 147
 148         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
 149         BUG_ON(!new_root);
 150
 151         trans = btrfs_start_transaction(new_root, 1);
 152         BUG_ON(!trans);
 153
 154         ret = btrfs_create_subvol_root(new_root, trans, new_dirid,
 155                                        BTRFS_I(dir)->block_group);
 156         if (ret)
 157                 goto fail;
 158
 159         /* Invalidate existing dcache entry for new subvolume. */
 160         btrfs_invalidate_dcache_root(root, name, namelen);
 161
 162 fail:
 163         nr = trans->blocks_used;
 164         err = btrfs_commit_transaction(trans, new_root);
 165         if (err && !ret)
 166                 ret = err;
 167 fail_commit:
 168         btrfs_btree_balance_dirty(root, nr);
 169         return ret;
 170 }
 171
 172 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
 173 {
 174         struct btrfs_pending_snapshot *pending_snapshot;
 175         struct btrfs_trans_handle *trans;
 176         int ret;
 177         int err;
 178         unsigned long nr = 0;
 179
 180         if (!root->ref_cows)
 181                 return -EINVAL;
 182
 183         ret = btrfs_check_free_space(root, 1, 0);
 184         if (ret)
 185                 goto fail_unlock;
 186
 187         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
 188         if (!pending_snapshot) {
 189                 ret = -ENOMEM;
 190                 goto fail_unlock;
 191         }
 192         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 193         if (!pending_snapshot->name) {
 194                 ret = -ENOMEM;
 195                 kfree(pending_snapshot);
 196                 goto fail_unlock;
 197         }
 198         memcpy(pending_snapshot->name, name, namelen);
 199         pending_snapshot->name[namelen] = '\0';
 200         trans = btrfs_start_transaction(root, 1);
 201         BUG_ON(!trans);
 202         pending_snapshot->root = root;
 203         list_add(&pending_snapshot->list,
 204                  &trans->transaction->pending_snapshots);
 205         ret = btrfs_update_inode(trans, root, root->inode);
 206         err = btrfs_commit_transaction(trans, root);
 207
 208 fail_unlock:
 209         btrfs_btree_balance_dirty(root, nr);
 210         return ret;
 211 }
 212
 213 int btrfs_defrag_file(struct file *file)
 214 {
 215         struct inode *inode = fdentry(file)->d_inode;
 216         struct btrfs_root *root = BTRFS_I(inode)->root;
 217         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 218         struct btrfs_ordered_extent *ordered;
 219         struct page *page;
 220         unsigned long last_index;
 221         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 222         unsigned long total_read = 0;
 223         u64 page_start;
 224         u64 page_end;
 225         unsigned long i;
 226         int ret;
 227
 228         ret = btrfs_check_free_space(root, inode->i_size, 0);
 229         if (ret)
 230                 return -ENOSPC;
 231
 232         mutex_lock(&inode->i_mutex);
 233         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 234         for (i = 0; i <= last_index; i++) {
 235                 if (total_read % ra_pages == 0) {
 236                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 237                                        min(last_index, i + ra_pages - 1));
 238                 }
 239                 total_read++;
 240 again:
 241                 page = grab_cache_page(inode->i_mapping, i);
 242                 if (!page)
 243                         goto out_unlock;
 244                 if (!PageUptodate(page)) {
 245                         btrfs_readpage(NULL, page);
 246                         lock_page(page);
 247                         if (!PageUptodate(page)) {
 248                                 unlock_page(page);
 249                                 page_cache_release(page);
 250                                 goto out_unlock;
 251                         }
 252                 }
 253
 254                 wait_on_page_writeback(page);
 255
 256                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 257                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 258                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 259
 260                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 261                 if (ordered) {
 262                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 263                         unlock_page(page);
 264                         page_cache_release(page);
 265                         btrfs_start_ordered_extent(inode, ordered, 1);
 266                         btrfs_put_ordered_extent(ordered);
 267                         goto again;
 268                 }
 269                 set_page_extent_mapped(page);
 270
 271                 /*
 272                  * this makes sure page_mkwrite is called on the
 273                  * page if it is dirtied again later
 274                  */
 275                 clear_page_dirty_for_io(page);
 276
 277                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 278
 279                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 280                 set_page_dirty(page);
 281                 unlock_page(page);
 282                 page_cache_release(page);
 283                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 284         }
 285
 286 out_unlock:
 287         mutex_unlock(&inode->i_mutex);
 288         return 0;
 289 }
 290
 291 /*
 292  * Called inside transaction, so use GFP_NOFS
 293  */
 294
 295 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 296 {
 297         u64 new_size;
 298         u64 old_size;
 299         u64 devid = 1;
 300         struct btrfs_ioctl_vol_args *vol_args;
 301         struct btrfs_trans_handle *trans;
 302         struct btrfs_device *device = NULL;
 303         char *sizestr;
 304         char *devstr = NULL;
 305         int ret = 0;
 306         int namelen;
 307         int mod = 0;
 308
 309         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 310
 311         if (!vol_args)
 312                 return -ENOMEM;
 313
 314         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 315                 ret = -EFAULT;
 316                 goto out;
 317         }
 318
 319         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 320         namelen = strlen(vol_args->name);
 321
 322         mutex_lock(&root->fs_info->volume_mutex);
 323         sizestr = vol_args->name;
 324         devstr = strchr(sizestr, ':');
 325         if (devstr) {
 326                 char *end;
 327                 sizestr = devstr + 1;
 328                 *devstr = '\0';
 329                 devstr = vol_args->name;
 330                 devid = simple_strtoull(devstr, &end, 10);
 331                 printk(KERN_INFO "resizing devid %llu\n", devid);
 332         }
 333         device = btrfs_find_device(root, devid, NULL);
 334         if (!device) {
 335                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 336                 ret = -EINVAL;
 337                 goto out_unlock;
 338         }
 339         if (!strcmp(sizestr, "max"))
 340                 new_size = device->bdev->bd_inode->i_size;
 341         else {
 342                 if (sizestr[0] == '-') {
 343                         mod = -1;
 344                         sizestr++;
 345                 } else if (sizestr[0] == '+') {
 346                         mod = 1;
 347                         sizestr++;
 348                 }
 349                 new_size = btrfs_parse_size(sizestr);
 350                 if (new_size == 0) {
 351                         ret = -EINVAL;
 352                         goto out_unlock;
 353                 }
 354         }
 355
 356         old_size = device->total_bytes;
 357
 358         if (mod < 0) {
 359                 if (new_size > old_size) {
 360                         ret = -EINVAL;
 361                         goto out_unlock;
 362                 }
 363                 new_size = old_size - new_size;
 364         } else if (mod > 0) {
 365                 new_size = old_size + new_size;
 366         }
 367
 368         if (new_size < 256 * 1024 * 1024) {
 369                 ret = -EINVAL;
 370                 goto out_unlock;
 371         }
 372         if (new_size > device->bdev->bd_inode->i_size) {
 373                 ret = -EFBIG;
 374                 goto out_unlock;
 375         }
 376
 377         do_div(new_size, root->sectorsize);
 378         new_size *= root->sectorsize;
 379
 380         printk(KERN_INFO "new size for %s is %llu\n",
 381                 device->name, (unsigned long long)new_size);
 382
 383         if (new_size > old_size) {
 384                 trans = btrfs_start_transaction(root, 1);
 385                 ret = btrfs_grow_device(trans, device, new_size);
 386                 btrfs_commit_transaction(trans, root);
 387         } else {
 388                 ret = btrfs_shrink_device(device, new_size);
 389         }
 390
 391 out_unlock:
 392         mutex_unlock(&root->fs_info->volume_mutex);
 393 out:
 394         kfree(vol_args);
 395         return ret;
 396 }
 397
 398 static noinline int btrfs_ioctl_snap_create(struct btrfs_root *root,
 399                                             void __user *arg)
 400 {
 401         struct btrfs_ioctl_vol_args *vol_args;
 402         struct btrfs_dir_item *di;
 403         struct btrfs_path *path;
 404         u64 root_dirid;
 405         int namelen;
 406         int ret;
 407
 408         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 409
 410         if (!vol_args)
 411                 return -ENOMEM;
 412
 413         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 414                 ret = -EFAULT;
 415                 goto out;
 416         }
 417
 418         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 419         namelen = strlen(vol_args->name);
 420         if (strchr(vol_args->name, '/')) {
 421                 ret = -EINVAL;
 422                 goto out;
 423         }
 424
 425         path = btrfs_alloc_path();
 426         if (!path) {
 427                 ret = -ENOMEM;
 428                 goto out;
 429         }
 430
 431         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 432         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 433                             path, root_dirid,
 434                             vol_args->name, namelen, 0);
 435         btrfs_free_path(path);
 436
 437         if (di && !IS_ERR(di)) {
 438                 ret = -EEXIST;
 439                 goto out;
 440         }
 441
 442         if (IS_ERR(di)) {
 443                 ret = PTR_ERR(di);
 444                 goto out;
 445         }
 446
 447         if (root == root->fs_info->tree_root)
 448                 ret = create_subvol(root, vol_args->name, namelen);
 449         else
 450                 ret = create_snapshot(root, vol_args->name, namelen);
 451 out:
 452         kfree(vol_args);
 453         return ret;
 454 }
 455
 456 static int btrfs_ioctl_defrag(struct file *file)
 457 {
 458         struct inode *inode = fdentry(file)->d_inode;
 459         struct btrfs_root *root = BTRFS_I(inode)->root;
 460
 461         switch (inode->i_mode & S_IFMT) {
 462         case S_IFDIR:
 463                 btrfs_defrag_root(root, 0);
 464                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 465                 break;
 466         case S_IFREG:
 467                 btrfs_defrag_file(file);
 468                 break;
 469         }
 470
 471         return 0;
 472 }
 473
 474 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 475 {
 476         struct btrfs_ioctl_vol_args *vol_args;
 477         int ret;
 478
 479         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 480
 481         if (!vol_args)
 482                 return -ENOMEM;
 483
 484         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 485                 ret = -EFAULT;
 486                 goto out;
 487         }
 488         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 489         ret = btrfs_init_new_device(root, vol_args->name);
 490
 491 out:
 492         kfree(vol_args);
 493         return ret;
 494 }
 495
 496 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 497 {
 498         struct btrfs_ioctl_vol_args *vol_args;
 499         int ret;
 500
 501         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 502
 503         if (!vol_args)
 504                 return -ENOMEM;
 505
 506         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 507                 ret = -EFAULT;
 508                 goto out;
 509         }
 510         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 511         ret = btrfs_rm_device(root, vol_args->name);
 512
 513 out:
 514         kfree(vol_args);
 515         return ret;
 516 }
 517
 518 long btrfs_ioctl_clone(struct file *file, unsigned long src_fd)
 519 {
 520         struct inode *inode = fdentry(file)->d_inode;
 521         struct btrfs_root *root = BTRFS_I(inode)->root;
 522         struct file *src_file;
 523         struct inode *src;
 524         struct btrfs_trans_handle *trans;
 525         struct btrfs_path *path;
 526         struct extent_buffer *leaf;
 527         char *buf;
 528         struct btrfs_key key;
 529         u32 nritems;
 530         int slot;
 531         int ret;
 532
 533         src_file = fget(src_fd);
 534         if (!src_file)
 535                 return -EBADF;
 536         src = src_file->f_dentry->d_inode;
 537
 538         ret = -EISDIR;
 539         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 540                 goto out_fput;
 541
 542         ret = -EXDEV;
 543         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 544                 goto out_fput;
 545
 546         ret = -ENOMEM;
 547         buf = vmalloc(btrfs_level_size(root, 0));
 548         if (!buf)
 549                 goto out_fput;
 550
 551         path = btrfs_alloc_path();
 552         if (!path) {
 553                 vfree(buf);
 554                 goto out_fput;
 555         }
 556         path->reada = 2;
 557
 558         if (inode < src) {
 559                 mutex_lock(&inode->i_mutex);
 560                 mutex_lock(&src->i_mutex);
 561         } else {
 562                 mutex_lock(&src->i_mutex);
 563                 mutex_lock(&inode->i_mutex);
 564         }
 565
 566         ret = -ENOTEMPTY;
 567         if (inode->i_size)
 568                 goto out_unlock;
 569
 570         /* do any pending delalloc/csum calc on src, one way or
 571            another, and lock file content */
 572         while (1) {
 573                 struct btrfs_ordered_extent *ordered;
 574                 lock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 575                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
 576                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 577                         break;
 578                 unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 579                 if (ordered)
 580                         btrfs_put_ordered_extent(ordered);
 581                 btrfs_wait_ordered_range(src, 0, (u64)-1);
 582         }
 583
 584         trans = btrfs_start_transaction(root, 1);
 585         BUG_ON(!trans);
 586
 587         key.objectid = src->i_ino;
 588         key.type = BTRFS_EXTENT_DATA_KEY;
 589         key.offset = 0;
 590
 591         while (1) {
 592                 /*
 593                  * note the key will change type as we walk through the
 594                  * tree.
 595                  */
 596                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 597                 if (ret < 0)
 598                         goto out;
 599
 600                 nritems = btrfs_header_nritems(path->nodes[0]);
 601                 if (path->slots[0] >= nritems) {
 602                         ret = btrfs_next_leaf(root, path);
 603                         if (ret < 0)
 604                                 goto out;
 605                         if (ret > 0)
 606                                 break;
 607                         nritems = btrfs_header_nritems(path->nodes[0]);
 608                 }
 609                 leaf = path->nodes[0];
 610                 slot = path->slots[0];
 611
 612                 btrfs_item_key_to_cpu(leaf, &key, slot);
 613                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
 614                     key.objectid != src->i_ino)
 615                         break;
 616
 617                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY ||
 618                     btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
 619                         u32 size;
 620                         struct btrfs_key new_key;
 621
 622                         size = btrfs_item_size_nr(leaf, slot);
 623                         read_extent_buffer(leaf, buf,
 624                                            btrfs_item_ptr_offset(leaf, slot),
 625                                            size);
 626                         btrfs_release_path(root, path);
 627
 628                         memcpy(&new_key, &key, sizeof(new_key));
 629                         new_key.objectid = inode->i_ino;
 630                         ret = btrfs_insert_empty_item(trans, root, path,
 631                                                       &new_key, size);
 632                         if (ret)
 633                                 goto out;
 634
 635                         leaf = path->nodes[0];
 636                         slot = path->slots[0];
 637                         write_extent_buffer(leaf, buf,
 638                                             btrfs_item_ptr_offset(leaf, slot),
 639                                             size);
 640                         btrfs_mark_buffer_dirty(leaf);
 641                 }
 642
 643                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 644                         struct btrfs_file_extent_item *extent;
 645                         int found_type;
 646
 647                         extent = btrfs_item_ptr(leaf, slot,
 648                                                 struct btrfs_file_extent_item);
 649                         found_type = btrfs_file_extent_type(leaf, extent);
 650                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 651                                 u64 ds = btrfs_file_extent_disk_bytenr(leaf,
 652                                                                        extent);
 653                                 u64 dl = btrfs_file_extent_disk_num_bytes(leaf,
 654                                                                  extent);
 655                                 /* ds == 0 means there's a hole */
 656                                 if (ds != 0) {
 657                                         ret = btrfs_inc_extent_ref(trans, root,
 658                                                      ds, dl, leaf->start,
 659                                                      root->root_key.objectid,
 660                                                      trans->transid,
 661                                                      inode->i_ino, key.offset);
 662                                         BUG_ON(ret);
 663                                 }
 664                         }
 665                 }
 666                 btrfs_release_path(root, path);
 667                 key.offset++;
 668         }
 669         ret = 0;
 670 out:
 671         btrfs_release_path(root, path);
 672         if (ret == 0) {
 673                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 674                 inode_set_bytes(inode, inode_get_bytes(src));
 675                 btrfs_i_size_write(inode, src->i_size);
 676                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
 677                 ret = btrfs_update_inode(trans, root, inode);
 678         }
 679         btrfs_end_transaction(trans, root);
 680         unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 681         if (ret)
 682                 vmtruncate(inode, 0);
 683 out_unlock:
 684         mutex_unlock(&src->i_mutex);
 685         mutex_unlock(&inode->i_mutex);
 686         vfree(buf);
 687         btrfs_free_path(path);
 688 out_fput:
 689         fput(src_file);
 690         return ret;
 691 }
 692
 693 /*
 694  * there are many ways the trans_start and trans_end ioctls can lead
 695  * to deadlocks.  They should only be used by applications that
 696  * basically own the machine, and have a very in depth understanding
 697  * of all the possible deadlocks and enospc problems.
 698  */
 699 long btrfs_ioctl_trans_start(struct file *file)
 700 {
 701         struct inode *inode = fdentry(file)->d_inode;
 702         struct btrfs_root *root = BTRFS_I(inode)->root;
 703         struct btrfs_trans_handle *trans;
 704         int ret = 0;
 705
 706         if (!capable(CAP_SYS_ADMIN))
 707                 return -EPERM;
 708
 709         if (file->private_data) {
 710                 ret = -EINPROGRESS;
 711                 goto out;
 712         }
 713
 714         mutex_lock(&root->fs_info->trans_mutex);
 715         root->fs_info->open_ioctl_trans++;
 716         mutex_unlock(&root->fs_info->trans_mutex);
 717
 718         trans = btrfs_start_ioctl_transaction(root, 0);
 719         if (trans)
 720                 file->private_data = trans;
 721         else
 722                 ret = -ENOMEM;
 723         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
 724 out:
 725         return ret;
 726 }
 727
 728 /*
 729  * there are many ways the trans_start and trans_end ioctls can lead
 730  * to deadlocks.  They should only be used by applications that
 731  * basically own the machine, and have a very in depth understanding
 732  * of all the possible deadlocks and enospc problems.
 733  */
 734 long btrfs_ioctl_trans_end(struct file *file)
 735 {
 736         struct inode *inode = fdentry(file)->d_inode;
 737         struct btrfs_root *root = BTRFS_I(inode)->root;
 738         struct btrfs_trans_handle *trans;
 739         int ret = 0;
 740
 741         trans = file->private_data;
 742         if (!trans) {
 743                 ret = -EINVAL;
 744                 goto out;
 745         }
 746         btrfs_end_transaction(trans, root);
 747         file->private_data = NULL;
 748
 749         mutex_lock(&root->fs_info->trans_mutex);
 750         root->fs_info->open_ioctl_trans--;
 751         mutex_unlock(&root->fs_info->trans_mutex);
 752
 753 out:
 754         return ret;
 755 }
 756
 757 long btrfs_ioctl(struct file *file, unsigned int
 758                 cmd, unsigned long arg)
 759 {
 760         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 761
 762         switch (cmd) {
 763         case BTRFS_IOC_SNAP_CREATE:
 764                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
 765         case BTRFS_IOC_DEFRAG:
 766                 return btrfs_ioctl_defrag(file);
 767         case BTRFS_IOC_RESIZE:
 768                 return btrfs_ioctl_resize(root, (void __user *)arg);
 769         case BTRFS_IOC_ADD_DEV:
 770                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
 771         case BTRFS_IOC_RM_DEV:
 772                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
 773         case BTRFS_IOC_BALANCE:
 774                 return btrfs_balance(root->fs_info->dev_root);
 775         case BTRFS_IOC_CLONE:
 776                 return btrfs_ioctl_clone(file, arg);
 777         case BTRFS_IOC_TRANS_START:
 778                 return btrfs_ioctl_trans_start(file);
 779         case BTRFS_IOC_TRANS_END:
 780                 return btrfs_ioctl_trans_end(file);
 781         case BTRFS_IOC_SYNC:
 782                 btrfs_start_delalloc_inodes(root);
 783                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
 784                 return 0;
 785         }
 786
 787         return -ENOTTY;
 788 }