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/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/mount.h>
  33 #include <linux/mpage.h>
  34 #include <linux/namei.h>
  35 #include <linux/swap.h>
  36 #include <linux/writeback.h>
  37 #include <linux/statfs.h>
  38 #include <linux/compat.h>
  39 #include <linux/bit_spinlock.h>
  40 #include <linux/security.h>
  41 #include <linux/version.h>
  42 #include <linux/xattr.h>
  43 #include <linux/vmalloc.h>
  44 #include "ctree.h"
  45 #include "disk-io.h"
  46 #include "transaction.h"
  47 #include "btrfs_inode.h"
  48 #include "ioctl.h"
  49 #include "print-tree.h"
  50 #include "volumes.h"
  51 #include "locking.h"
  52
  53
  54
  55 static noinline int create_subvol(struct btrfs_root *root,
  56                                   struct dentry *dentry,
  57                                   char *name, int namelen)
  58 {
  59         struct btrfs_trans_handle *trans;
  60         struct btrfs_key key;
  61         struct btrfs_root_item root_item;
  62         struct btrfs_inode_item *inode_item;
  63         struct extent_buffer *leaf;
  64         struct btrfs_root *new_root = root;
  65         struct inode *dir;
  66         int ret;
  67         int err;
  68         u64 objectid;
  69         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
  70         unsigned long nr = 1;
  71
  72         ret = btrfs_check_free_space(root, 1, 0);
  73         if (ret)
  74                 goto fail_commit;
  75
  76         trans = btrfs_start_transaction(root, 1);
  77         BUG_ON(!trans);
  78
  79         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
  80                                        0, &objectid);
  81         if (ret)
  82                 goto fail;
  83
  84         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
  85                                       objectid, trans->transid, 0, 0, 0);
  86         if (IS_ERR(leaf)) {
  87                 ret = PTR_ERR(leaf);
  88                 goto fail;
  89         }
  90
  91         btrfs_set_header_nritems(leaf, 0);
  92         btrfs_set_header_level(leaf, 0);
  93         btrfs_set_header_bytenr(leaf, leaf->start);
  94         btrfs_set_header_generation(leaf, trans->transid);
  95         btrfs_set_header_owner(leaf, objectid);
  96
  97         write_extent_buffer(leaf, root->fs_info->fsid,
  98                             (unsigned long)btrfs_header_fsid(leaf),
  99                             BTRFS_FSID_SIZE);
 100         btrfs_mark_buffer_dirty(leaf);
 101
 102         inode_item = &root_item.inode;
 103         memset(inode_item, 0, sizeof(*inode_item));
 104         inode_item->generation = cpu_to_le64(1);
 105         inode_item->size = cpu_to_le64(3);
 106         inode_item->nlink = cpu_to_le32(1);
 107         inode_item->nbytes = cpu_to_le64(root->leafsize);
 108         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 109
 110         btrfs_set_root_bytenr(&root_item, leaf->start);
 111         btrfs_set_root_level(&root_item, 0);
 112         btrfs_set_root_refs(&root_item, 1);
 113         btrfs_set_root_used(&root_item, 0);
 114
 115         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 116         root_item.drop_level = 0;
 117
 118         btrfs_tree_unlock(leaf);
 119         free_extent_buffer(leaf);
 120         leaf = NULL;
 121
 122         btrfs_set_root_dirid(&root_item, new_dirid);
 123
 124         key.objectid = objectid;
 125         key.offset = 1;
 126         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 127         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 128                                 &root_item);
 129         if (ret)
 130                 goto fail;
 131
 132         /*
 133          * insert the directory item
 134          */
 135         key.offset = (u64)-1;
 136         dir = root->fs_info->sb->s_root->d_inode;
 137         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 138                                     name, namelen, dir->i_ino, &key,
 139                                     BTRFS_FT_DIR, 0);
 140         if (ret)
 141                 goto fail;
 142
 143         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 144                              name, namelen, objectid,
 145                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 146         if (ret)
 147                 goto fail;
 148
 149         ret = btrfs_commit_transaction(trans, root);
 150         if (ret)
 151                 goto fail_commit;
 152
 153         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
 154         BUG_ON(!new_root);
 155
 156         trans = btrfs_start_transaction(new_root, 1);
 157         BUG_ON(!trans);
 158
 159         ret = btrfs_create_subvol_root(new_root, dentry, trans, new_dirid,
 160                                        BTRFS_I(dir)->block_group);
 161         if (ret)
 162                 goto fail;
 163
 164 fail:
 165         nr = trans->blocks_used;
 166         err = btrfs_commit_transaction(trans, new_root);
 167         if (err && !ret)
 168                 ret = err;
 169 fail_commit:
 170         btrfs_btree_balance_dirty(root, nr);
 171         return ret;
 172 }
 173
 174 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
 175 {
 176         struct btrfs_pending_snapshot *pending_snapshot;
 177         struct btrfs_trans_handle *trans;
 178         int ret;
 179         int err;
 180         unsigned long nr = 0;
 181
 182         if (!root->ref_cows)
 183                 return -EINVAL;
 184
 185         ret = btrfs_check_free_space(root, 1, 0);
 186         if (ret)
 187                 goto fail_unlock;
 188
 189         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
 190         if (!pending_snapshot) {
 191                 ret = -ENOMEM;
 192                 goto fail_unlock;
 193         }
 194         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 195         if (!pending_snapshot->name) {
 196                 ret = -ENOMEM;
 197                 kfree(pending_snapshot);
 198                 goto fail_unlock;
 199         }
 200         memcpy(pending_snapshot->name, name, namelen);
 201         pending_snapshot->name[namelen] = '\0';
 202         trans = btrfs_start_transaction(root, 1);
 203         BUG_ON(!trans);
 204         pending_snapshot->root = root;
 205         list_add(&pending_snapshot->list,
 206                  &trans->transaction->pending_snapshots);
 207         ret = btrfs_update_inode(trans, root, root->inode);
 208         err = btrfs_commit_transaction(trans, root);
 209
 210 fail_unlock:
 211         btrfs_btree_balance_dirty(root, nr);
 212         return ret;
 213 }
 214
 215 /* copy of may_create in fs/namei.c() */
 216 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 217 {
 218         if (child->d_inode)
 219                 return -EEXIST;
 220         if (IS_DEADDIR(dir))
 221                 return -ENOENT;
 222         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 223 }
 224
 225 /*
 226  * Create a new subvolume below @parent.  This is largely modeled after
 227  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 228  * inside this filesystem so it's quite a bit simpler.
 229  */
 230 static noinline int btrfs_mksubvol(struct path *parent, char *name,
 231                                    int mode, int namelen)
 232 {
 233         struct dentry *dentry;
 234         int error;
 235
 236         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 237
 238         dentry = lookup_one_len(name, parent->dentry, namelen);
 239         error = PTR_ERR(dentry);
 240         if (IS_ERR(dentry))
 241                 goto out_unlock;
 242
 243         error = -EEXIST;
 244         if (dentry->d_inode)
 245                 goto out_dput;
 246
 247         if (!IS_POSIXACL(parent->dentry->d_inode))
 248                 mode &= ~current->fs->umask;
 249         error = mnt_want_write(parent->mnt);
 250         if (error)
 251                 goto out_dput;
 252
 253         error = btrfs_may_create(parent->dentry->d_inode, dentry);
 254         if (error)
 255                 goto out_drop_write;
 256
 257         /*
 258          * Actually perform the low-level subvolume creation after all
 259          * this VFS fuzz.
 260          *
 261          * Eventually we want to pass in an inode under which we create this
 262          * subvolume, but for now all are under the filesystem root.
 263          *
 264          * Also we should pass on the mode eventually to allow creating new
 265          * subvolume with specific mode bits.
 266          */
 267         error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root, dentry,
 268                               name, namelen);
 269         if (error)
 270                 goto out_drop_write;
 271
 272         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 273 out_drop_write:
 274         mnt_drop_write(parent->mnt);
 275 out_dput:
 276         dput(dentry);
 277 out_unlock:
 278         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 279         return error;
 280 }
 281
 282
 283 int btrfs_defrag_file(struct file *file)
 284 {
 285         struct inode *inode = fdentry(file)->d_inode;
 286         struct btrfs_root *root = BTRFS_I(inode)->root;
 287         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 288         struct btrfs_ordered_extent *ordered;
 289         struct page *page;
 290         unsigned long last_index;
 291         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 292         unsigned long total_read = 0;
 293         u64 page_start;
 294         u64 page_end;
 295         unsigned long i;
 296         int ret;
 297
 298         ret = btrfs_check_free_space(root, inode->i_size, 0);
 299         if (ret)
 300                 return -ENOSPC;
 301
 302         mutex_lock(&inode->i_mutex);
 303         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 304         for (i = 0; i <= last_index; i++) {
 305                 if (total_read % ra_pages == 0) {
 306                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 307                                        min(last_index, i + ra_pages - 1));
 308                 }
 309                 total_read++;
 310 again:
 311                 page = grab_cache_page(inode->i_mapping, i);
 312                 if (!page)
 313                         goto out_unlock;
 314                 if (!PageUptodate(page)) {
 315                         btrfs_readpage(NULL, page);
 316                         lock_page(page);
 317                         if (!PageUptodate(page)) {
 318                                 unlock_page(page);
 319                                 page_cache_release(page);
 320                                 goto out_unlock;
 321                         }
 322                 }
 323
 324                 wait_on_page_writeback(page);
 325
 326                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 327                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 328                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 329
 330                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 331                 if (ordered) {
 332                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 333                         unlock_page(page);
 334                         page_cache_release(page);
 335                         btrfs_start_ordered_extent(inode, ordered, 1);
 336                         btrfs_put_ordered_extent(ordered);
 337                         goto again;
 338                 }
 339                 set_page_extent_mapped(page);
 340
 341                 /*
 342                  * this makes sure page_mkwrite is called on the
 343                  * page if it is dirtied again later
 344                  */
 345                 clear_page_dirty_for_io(page);
 346
 347                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 348
 349                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 350                 set_page_dirty(page);
 351                 unlock_page(page);
 352                 page_cache_release(page);
 353                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 354         }
 355
 356 out_unlock:
 357         mutex_unlock(&inode->i_mutex);
 358         return 0;
 359 }
 360
 361 /*
 362  * Called inside transaction, so use GFP_NOFS
 363  */
 364
 365 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 366 {
 367         u64 new_size;
 368         u64 old_size;
 369         u64 devid = 1;
 370         struct btrfs_ioctl_vol_args *vol_args;
 371         struct btrfs_trans_handle *trans;
 372         struct btrfs_device *device = NULL;
 373         char *sizestr;
 374         char *devstr = NULL;
 375         int ret = 0;
 376         int namelen;
 377         int mod = 0;
 378
 379         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 380
 381         if (!vol_args)
 382                 return -ENOMEM;
 383
 384         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 385                 ret = -EFAULT;
 386                 goto out;
 387         }
 388
 389         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 390         namelen = strlen(vol_args->name);
 391
 392         mutex_lock(&root->fs_info->volume_mutex);
 393         sizestr = vol_args->name;
 394         devstr = strchr(sizestr, ':');
 395         if (devstr) {
 396                 char *end;
 397                 sizestr = devstr + 1;
 398                 *devstr = '\0';
 399                 devstr = vol_args->name;
 400                 devid = simple_strtoull(devstr, &end, 10);
 401                 printk(KERN_INFO "resizing devid %llu\n", devid);
 402         }
 403         device = btrfs_find_device(root, devid, NULL);
 404         if (!device) {
 405                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 406                 ret = -EINVAL;
 407                 goto out_unlock;
 408         }
 409         if (!strcmp(sizestr, "max"))
 410                 new_size = device->bdev->bd_inode->i_size;
 411         else {
 412                 if (sizestr[0] == '-') {
 413                         mod = -1;
 414                         sizestr++;
 415                 } else if (sizestr[0] == '+') {
 416                         mod = 1;
 417                         sizestr++;
 418                 }
 419                 new_size = btrfs_parse_size(sizestr);
 420                 if (new_size == 0) {
 421                         ret = -EINVAL;
 422                         goto out_unlock;
 423                 }
 424         }
 425
 426         old_size = device->total_bytes;
 427
 428         if (mod < 0) {
 429                 if (new_size > old_size) {
 430                         ret = -EINVAL;
 431                         goto out_unlock;
 432                 }
 433                 new_size = old_size - new_size;
 434         } else if (mod > 0) {
 435                 new_size = old_size + new_size;
 436         }
 437
 438         if (new_size < 256 * 1024 * 1024) {
 439                 ret = -EINVAL;
 440                 goto out_unlock;
 441         }
 442         if (new_size > device->bdev->bd_inode->i_size) {
 443                 ret = -EFBIG;
 444                 goto out_unlock;
 445         }
 446
 447         do_div(new_size, root->sectorsize);
 448         new_size *= root->sectorsize;
 449
 450         printk(KERN_INFO "new size for %s is %llu\n",
 451                 device->name, (unsigned long long)new_size);
 452
 453         if (new_size > old_size) {
 454                 trans = btrfs_start_transaction(root, 1);
 455                 ret = btrfs_grow_device(trans, device, new_size);
 456                 btrfs_commit_transaction(trans, root);
 457         } else {
 458                 ret = btrfs_shrink_device(device, new_size);
 459         }
 460
 461 out_unlock:
 462         mutex_unlock(&root->fs_info->volume_mutex);
 463 out:
 464         kfree(vol_args);
 465         return ret;
 466 }
 467
 468 static noinline int btrfs_ioctl_snap_create(struct file *file,
 469                                             void __user *arg)
 470 {
 471         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 472         struct btrfs_ioctl_vol_args *vol_args;
 473         struct btrfs_dir_item *di;
 474         struct btrfs_path *path;
 475         u64 root_dirid;
 476         int namelen;
 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
 489         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 490         namelen = strlen(vol_args->name);
 491         if (strchr(vol_args->name, '/')) {
 492                 ret = -EINVAL;
 493                 goto out;
 494         }
 495
 496         path = btrfs_alloc_path();
 497         if (!path) {
 498                 ret = -ENOMEM;
 499                 goto out;
 500         }
 501
 502         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 503         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 504                             path, root_dirid,
 505                             vol_args->name, namelen, 0);
 506         btrfs_free_path(path);
 507
 508         if (di && !IS_ERR(di)) {
 509                 ret = -EEXIST;
 510                 goto out;
 511         }
 512
 513         if (IS_ERR(di)) {
 514                 ret = PTR_ERR(di);
 515                 goto out;
 516         }
 517
 518         if (root == root->fs_info->tree_root) {
 519                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 520                                      file->f_path.dentry->d_inode->i_mode,
 521                                      namelen);
 522         } else {
 523                 ret = create_snapshot(root, vol_args->name, namelen);
 524         }
 525
 526 out:
 527         kfree(vol_args);
 528         return ret;
 529 }
 530
 531 static int btrfs_ioctl_defrag(struct file *file)
 532 {
 533         struct inode *inode = fdentry(file)->d_inode;
 534         struct btrfs_root *root = BTRFS_I(inode)->root;
 535
 536         switch (inode->i_mode & S_IFMT) {
 537         case S_IFDIR:
 538                 btrfs_defrag_root(root, 0);
 539                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 540                 break;
 541         case S_IFREG:
 542                 btrfs_defrag_file(file);
 543                 break;
 544         }
 545
 546         return 0;
 547 }
 548
 549 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 550 {
 551         struct btrfs_ioctl_vol_args *vol_args;
 552         int ret;
 553
 554         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 555
 556         if (!vol_args)
 557                 return -ENOMEM;
 558
 559         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 560                 ret = -EFAULT;
 561                 goto out;
 562         }
 563         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 564         ret = btrfs_init_new_device(root, vol_args->name);
 565
 566 out:
 567         kfree(vol_args);
 568         return ret;
 569 }
 570
 571 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 572 {
 573         struct btrfs_ioctl_vol_args *vol_args;
 574         int ret;
 575
 576         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 577
 578         if (!vol_args)
 579                 return -ENOMEM;
 580
 581         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 582                 ret = -EFAULT;
 583                 goto out;
 584         }
 585         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 586         ret = btrfs_rm_device(root, vol_args->name);
 587
 588 out:
 589         kfree(vol_args);
 590         return ret;
 591 }
 592
 593 long btrfs_ioctl_clone(struct file *file, unsigned long src_fd)
 594 {
 595         struct inode *inode = fdentry(file)->d_inode;
 596         struct btrfs_root *root = BTRFS_I(inode)->root;
 597         struct file *src_file;
 598         struct inode *src;
 599         struct btrfs_trans_handle *trans;
 600         struct btrfs_path *path;
 601         struct extent_buffer *leaf;
 602         char *buf;
 603         struct btrfs_key key;
 604         u32 nritems;
 605         int slot;
 606         int ret;
 607
 608         src_file = fget(src_fd);
 609         if (!src_file)
 610                 return -EBADF;
 611         src = src_file->f_dentry->d_inode;
 612
 613         ret = -EISDIR;
 614         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 615                 goto out_fput;
 616
 617         ret = -EXDEV;
 618         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 619                 goto out_fput;
 620
 621         ret = -ENOMEM;
 622         buf = vmalloc(btrfs_level_size(root, 0));
 623         if (!buf)
 624                 goto out_fput;
 625
 626         path = btrfs_alloc_path();
 627         if (!path) {
 628                 vfree(buf);
 629                 goto out_fput;
 630         }
 631         path->reada = 2;
 632
 633         if (inode < src) {
 634                 mutex_lock(&inode->i_mutex);
 635                 mutex_lock(&src->i_mutex);
 636         } else {
 637                 mutex_lock(&src->i_mutex);
 638                 mutex_lock(&inode->i_mutex);
 639         }
 640
 641         ret = -ENOTEMPTY;
 642         if (inode->i_size)
 643                 goto out_unlock;
 644
 645         /* do any pending delalloc/csum calc on src, one way or
 646            another, and lock file content */
 647         while (1) {
 648                 struct btrfs_ordered_extent *ordered;
 649                 lock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 650                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
 651                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 652                         break;
 653                 unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 654                 if (ordered)
 655                         btrfs_put_ordered_extent(ordered);
 656                 btrfs_wait_ordered_range(src, 0, (u64)-1);
 657         }
 658
 659         trans = btrfs_start_transaction(root, 1);
 660         BUG_ON(!trans);
 661
 662         key.objectid = src->i_ino;
 663         key.type = BTRFS_EXTENT_DATA_KEY;
 664         key.offset = 0;
 665
 666         while (1) {
 667                 /*
 668                  * note the key will change type as we walk through the
 669                  * tree.
 670                  */
 671                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 672                 if (ret < 0)
 673                         goto out;
 674
 675                 nritems = btrfs_header_nritems(path->nodes[0]);
 676                 if (path->slots[0] >= nritems) {
 677                         ret = btrfs_next_leaf(root, path);
 678                         if (ret < 0)
 679                                 goto out;
 680                         if (ret > 0)
 681                                 break;
 682                         nritems = btrfs_header_nritems(path->nodes[0]);
 683                 }
 684                 leaf = path->nodes[0];
 685                 slot = path->slots[0];
 686
 687                 btrfs_item_key_to_cpu(leaf, &key, slot);
 688                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
 689                     key.objectid != src->i_ino)
 690                         break;
 691
 692                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY ||
 693                     btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
 694                         u32 size;
 695                         struct btrfs_key new_key;
 696
 697                         size = btrfs_item_size_nr(leaf, slot);
 698                         read_extent_buffer(leaf, buf,
 699                                            btrfs_item_ptr_offset(leaf, slot),
 700                                            size);
 701                         btrfs_release_path(root, path);
 702
 703                         memcpy(&new_key, &key, sizeof(new_key));
 704                         new_key.objectid = inode->i_ino;
 705                         ret = btrfs_insert_empty_item(trans, root, path,
 706                                                       &new_key, size);
 707                         if (ret)
 708                                 goto out;
 709
 710                         leaf = path->nodes[0];
 711                         slot = path->slots[0];
 712                         write_extent_buffer(leaf, buf,
 713                                             btrfs_item_ptr_offset(leaf, slot),
 714                                             size);
 715                         btrfs_mark_buffer_dirty(leaf);
 716                 }
 717
 718                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 719                         struct btrfs_file_extent_item *extent;
 720                         int found_type;
 721
 722                         extent = btrfs_item_ptr(leaf, slot,
 723                                                 struct btrfs_file_extent_item);
 724                         found_type = btrfs_file_extent_type(leaf, extent);
 725                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 726                                 u64 ds = btrfs_file_extent_disk_bytenr(leaf,
 727                                                                        extent);
 728                                 u64 dl = btrfs_file_extent_disk_num_bytes(leaf,
 729                                                                  extent);
 730                                 /* ds == 0 means there's a hole */
 731                                 if (ds != 0) {
 732                                         ret = btrfs_inc_extent_ref(trans, root,
 733                                                      ds, dl, leaf->start,
 734                                                      root->root_key.objectid,
 735                                                      trans->transid,
 736                                                      inode->i_ino);
 737                                         BUG_ON(ret);
 738                                 }
 739                         }
 740                 }
 741                 btrfs_release_path(root, path);
 742                 key.offset++;
 743         }
 744         ret = 0;
 745 out:
 746         btrfs_release_path(root, path);
 747         if (ret == 0) {
 748                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 749                 inode_set_bytes(inode, inode_get_bytes(src));
 750                 btrfs_i_size_write(inode, src->i_size);
 751                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
 752                 ret = btrfs_update_inode(trans, root, inode);
 753         }
 754         btrfs_end_transaction(trans, root);
 755         unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 756         if (ret)
 757                 vmtruncate(inode, 0);
 758 out_unlock:
 759         mutex_unlock(&src->i_mutex);
 760         mutex_unlock(&inode->i_mutex);
 761         vfree(buf);
 762         btrfs_free_path(path);
 763 out_fput:
 764         fput(src_file);
 765         return ret;
 766 }
 767
 768 /*
 769  * there are many ways the trans_start and trans_end ioctls can lead
 770  * to deadlocks.  They should only be used by applications that
 771  * basically own the machine, and have a very in depth understanding
 772  * of all the possible deadlocks and enospc problems.
 773  */
 774 long btrfs_ioctl_trans_start(struct file *file)
 775 {
 776         struct inode *inode = fdentry(file)->d_inode;
 777         struct btrfs_root *root = BTRFS_I(inode)->root;
 778         struct btrfs_trans_handle *trans;
 779         int ret = 0;
 780
 781         if (!capable(CAP_SYS_ADMIN))
 782                 return -EPERM;
 783
 784         if (file->private_data) {
 785                 ret = -EINPROGRESS;
 786                 goto out;
 787         }
 788
 789         mutex_lock(&root->fs_info->trans_mutex);
 790         root->fs_info->open_ioctl_trans++;
 791         mutex_unlock(&root->fs_info->trans_mutex);
 792
 793         trans = btrfs_start_ioctl_transaction(root, 0);
 794         if (trans)
 795                 file->private_data = trans;
 796         else
 797                 ret = -ENOMEM;
 798         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
 799 out:
 800         return ret;
 801 }
 802
 803 /*
 804  * there are many ways the trans_start and trans_end ioctls can lead
 805  * to deadlocks.  They should only be used by applications that
 806  * basically own the machine, and have a very in depth understanding
 807  * of all the possible deadlocks and enospc problems.
 808  */
 809 long btrfs_ioctl_trans_end(struct file *file)
 810 {
 811         struct inode *inode = fdentry(file)->d_inode;
 812         struct btrfs_root *root = BTRFS_I(inode)->root;
 813         struct btrfs_trans_handle *trans;
 814         int ret = 0;
 815
 816         trans = file->private_data;
 817         if (!trans) {
 818                 ret = -EINVAL;
 819                 goto out;
 820         }
 821         btrfs_end_transaction(trans, root);
 822         file->private_data = NULL;
 823
 824         mutex_lock(&root->fs_info->trans_mutex);
 825         root->fs_info->open_ioctl_trans--;
 826         mutex_unlock(&root->fs_info->trans_mutex);
 827
 828 out:
 829         return ret;
 830 }
 831
 832 long btrfs_ioctl(struct file *file, unsigned int
 833                 cmd, unsigned long arg)
 834 {
 835         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 836
 837         switch (cmd) {
 838         case BTRFS_IOC_SNAP_CREATE:
 839                 return btrfs_ioctl_snap_create(file, (void __user *)arg);
 840         case BTRFS_IOC_DEFRAG:
 841                 return btrfs_ioctl_defrag(file);
 842         case BTRFS_IOC_RESIZE:
 843                 return btrfs_ioctl_resize(root, (void __user *)arg);
 844         case BTRFS_IOC_ADD_DEV:
 845                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
 846         case BTRFS_IOC_RM_DEV:
 847                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
 848         case BTRFS_IOC_BALANCE:
 849                 return btrfs_balance(root->fs_info->dev_root);
 850         case BTRFS_IOC_CLONE:
 851                 return btrfs_ioctl_clone(file, arg);
 852         case BTRFS_IOC_TRANS_START:
 853                 return btrfs_ioctl_trans_start(file);
 854         case BTRFS_IOC_TRANS_END:
 855                 return btrfs_ioctl_trans_end(file);
 856         case BTRFS_IOC_SYNC:
 857                 btrfs_start_delalloc_inodes(root);
 858                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
 859                 return 0;
 860         }
 861
 862         return -ENOTTY;
 863 }