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         mode &= (S_IRWXUGO|S_ISVTX);
 258         error = security_inode_mkdir(parent->dentry->d_inode, dentry, mode);
 259         if (error)
 260                 goto out_drop_write;
 261
 262         /*
 263          * Actually perform the low-level subvolume creation after all
 264          * this VFS fuzz.
 265          *
 266          * Eventually we want to pass in an inode under which we create this
 267          * subvolume, but for now all are under the filesystem root.
 268          *
 269          * Also we should pass on the mode eventually to allow creating new
 270          * subvolume with specific mode bits.
 271          */
 272         error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root, dentry,
 273                               name, namelen);
 274         if (error)
 275                 goto out_drop_write;
 276
 277         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 278 out_drop_write:
 279         mnt_drop_write(parent->mnt);
 280 out_dput:
 281         dput(dentry);
 282 out_unlock:
 283         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 284         return error;
 285 }
 286
 287
 288 int btrfs_defrag_file(struct file *file)
 289 {
 290         struct inode *inode = fdentry(file)->d_inode;
 291         struct btrfs_root *root = BTRFS_I(inode)->root;
 292         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 293         struct btrfs_ordered_extent *ordered;
 294         struct page *page;
 295         unsigned long last_index;
 296         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 297         unsigned long total_read = 0;
 298         u64 page_start;
 299         u64 page_end;
 300         unsigned long i;
 301         int ret;
 302
 303         ret = btrfs_check_free_space(root, inode->i_size, 0);
 304         if (ret)
 305                 return -ENOSPC;
 306
 307         mutex_lock(&inode->i_mutex);
 308         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 309         for (i = 0; i <= last_index; i++) {
 310                 if (total_read % ra_pages == 0) {
 311                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 312                                        min(last_index, i + ra_pages - 1));
 313                 }
 314                 total_read++;
 315 again:
 316                 page = grab_cache_page(inode->i_mapping, i);
 317                 if (!page)
 318                         goto out_unlock;
 319                 if (!PageUptodate(page)) {
 320                         btrfs_readpage(NULL, page);
 321                         lock_page(page);
 322                         if (!PageUptodate(page)) {
 323                                 unlock_page(page);
 324                                 page_cache_release(page);
 325                                 goto out_unlock;
 326                         }
 327                 }
 328
 329                 wait_on_page_writeback(page);
 330
 331                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 332                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 333                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 334
 335                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 336                 if (ordered) {
 337                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 338                         unlock_page(page);
 339                         page_cache_release(page);
 340                         btrfs_start_ordered_extent(inode, ordered, 1);
 341                         btrfs_put_ordered_extent(ordered);
 342                         goto again;
 343                 }
 344                 set_page_extent_mapped(page);
 345
 346                 /*
 347                  * this makes sure page_mkwrite is called on the
 348                  * page if it is dirtied again later
 349                  */
 350                 clear_page_dirty_for_io(page);
 351
 352                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 353
 354                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 355                 set_page_dirty(page);
 356                 unlock_page(page);
 357                 page_cache_release(page);
 358                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 359         }
 360
 361 out_unlock:
 362         mutex_unlock(&inode->i_mutex);
 363         return 0;
 364 }
 365
 366 /*
 367  * Called inside transaction, so use GFP_NOFS
 368  */
 369
 370 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 371 {
 372         u64 new_size;
 373         u64 old_size;
 374         u64 devid = 1;
 375         struct btrfs_ioctl_vol_args *vol_args;
 376         struct btrfs_trans_handle *trans;
 377         struct btrfs_device *device = NULL;
 378         char *sizestr;
 379         char *devstr = NULL;
 380         int ret = 0;
 381         int namelen;
 382         int mod = 0;
 383
 384         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 385
 386         if (!vol_args)
 387                 return -ENOMEM;
 388
 389         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 390                 ret = -EFAULT;
 391                 goto out;
 392         }
 393
 394         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 395         namelen = strlen(vol_args->name);
 396
 397         mutex_lock(&root->fs_info->volume_mutex);
 398         sizestr = vol_args->name;
 399         devstr = strchr(sizestr, ':');
 400         if (devstr) {
 401                 char *end;
 402                 sizestr = devstr + 1;
 403                 *devstr = '\0';
 404                 devstr = vol_args->name;
 405                 devid = simple_strtoull(devstr, &end, 10);
 406                 printk(KERN_INFO "resizing devid %llu\n", devid);
 407         }
 408         device = btrfs_find_device(root, devid, NULL);
 409         if (!device) {
 410                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 411                 ret = -EINVAL;
 412                 goto out_unlock;
 413         }
 414         if (!strcmp(sizestr, "max"))
 415                 new_size = device->bdev->bd_inode->i_size;
 416         else {
 417                 if (sizestr[0] == '-') {
 418                         mod = -1;
 419                         sizestr++;
 420                 } else if (sizestr[0] == '+') {
 421                         mod = 1;
 422                         sizestr++;
 423                 }
 424                 new_size = btrfs_parse_size(sizestr);
 425                 if (new_size == 0) {
 426                         ret = -EINVAL;
 427                         goto out_unlock;
 428                 }
 429         }
 430
 431         old_size = device->total_bytes;
 432
 433         if (mod < 0) {
 434                 if (new_size > old_size) {
 435                         ret = -EINVAL;
 436                         goto out_unlock;
 437                 }
 438                 new_size = old_size - new_size;
 439         } else if (mod > 0) {
 440                 new_size = old_size + new_size;
 441         }
 442
 443         if (new_size < 256 * 1024 * 1024) {
 444                 ret = -EINVAL;
 445                 goto out_unlock;
 446         }
 447         if (new_size > device->bdev->bd_inode->i_size) {
 448                 ret = -EFBIG;
 449                 goto out_unlock;
 450         }
 451
 452         do_div(new_size, root->sectorsize);
 453         new_size *= root->sectorsize;
 454
 455         printk(KERN_INFO "new size for %s is %llu\n",
 456                 device->name, (unsigned long long)new_size);
 457
 458         if (new_size > old_size) {
 459                 trans = btrfs_start_transaction(root, 1);
 460                 ret = btrfs_grow_device(trans, device, new_size);
 461                 btrfs_commit_transaction(trans, root);
 462         } else {
 463                 ret = btrfs_shrink_device(device, new_size);
 464         }
 465
 466 out_unlock:
 467         mutex_unlock(&root->fs_info->volume_mutex);
 468 out:
 469         kfree(vol_args);
 470         return ret;
 471 }
 472
 473 static noinline int btrfs_ioctl_snap_create(struct file *file,
 474                                             void __user *arg)
 475 {
 476         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 477         struct btrfs_ioctl_vol_args *vol_args;
 478         struct btrfs_dir_item *di;
 479         struct btrfs_path *path;
 480         u64 root_dirid;
 481         int namelen;
 482         int ret;
 483
 484         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 485
 486         if (!vol_args)
 487                 return -ENOMEM;
 488
 489         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 490                 ret = -EFAULT;
 491                 goto out;
 492         }
 493
 494         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 495         namelen = strlen(vol_args->name);
 496         if (strchr(vol_args->name, '/')) {
 497                 ret = -EINVAL;
 498                 goto out;
 499         }
 500
 501         path = btrfs_alloc_path();
 502         if (!path) {
 503                 ret = -ENOMEM;
 504                 goto out;
 505         }
 506
 507         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 508         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 509                             path, root_dirid,
 510                             vol_args->name, namelen, 0);
 511         btrfs_free_path(path);
 512
 513         if (di && !IS_ERR(di)) {
 514                 ret = -EEXIST;
 515                 goto out;
 516         }
 517
 518         if (IS_ERR(di)) {
 519                 ret = PTR_ERR(di);
 520                 goto out;
 521         }
 522
 523         if (root == root->fs_info->tree_root) {
 524                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 525                                      file->f_path.dentry->d_inode->i_mode,
 526                                      namelen);
 527         } else {
 528                 ret = create_snapshot(root, vol_args->name, namelen);
 529         }
 530
 531 out:
 532         kfree(vol_args);
 533         return ret;
 534 }
 535
 536 static int btrfs_ioctl_defrag(struct file *file)
 537 {
 538         struct inode *inode = fdentry(file)->d_inode;
 539         struct btrfs_root *root = BTRFS_I(inode)->root;
 540
 541         switch (inode->i_mode & S_IFMT) {
 542         case S_IFDIR:
 543                 btrfs_defrag_root(root, 0);
 544                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 545                 break;
 546         case S_IFREG:
 547                 btrfs_defrag_file(file);
 548                 break;
 549         }
 550
 551         return 0;
 552 }
 553
 554 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 555 {
 556         struct btrfs_ioctl_vol_args *vol_args;
 557         int ret;
 558
 559         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 560
 561         if (!vol_args)
 562                 return -ENOMEM;
 563
 564         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 565                 ret = -EFAULT;
 566                 goto out;
 567         }
 568         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 569         ret = btrfs_init_new_device(root, vol_args->name);
 570
 571 out:
 572         kfree(vol_args);
 573         return ret;
 574 }
 575
 576 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 577 {
 578         struct btrfs_ioctl_vol_args *vol_args;
 579         int ret;
 580
 581         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 582
 583         if (!vol_args)
 584                 return -ENOMEM;
 585
 586         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 587                 ret = -EFAULT;
 588                 goto out;
 589         }
 590         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 591         ret = btrfs_rm_device(root, vol_args->name);
 592
 593 out:
 594         kfree(vol_args);
 595         return ret;
 596 }
 597
 598 long btrfs_ioctl_clone(struct file *file, unsigned long src_fd)
 599 {
 600         struct inode *inode = fdentry(file)->d_inode;
 601         struct btrfs_root *root = BTRFS_I(inode)->root;
 602         struct file *src_file;
 603         struct inode *src;
 604         struct btrfs_trans_handle *trans;
 605         struct btrfs_path *path;
 606         struct extent_buffer *leaf;
 607         char *buf;
 608         struct btrfs_key key;
 609         u32 nritems;
 610         int slot;
 611         int ret;
 612
 613         src_file = fget(src_fd);
 614         if (!src_file)
 615                 return -EBADF;
 616         src = src_file->f_dentry->d_inode;
 617
 618         ret = -EISDIR;
 619         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 620                 goto out_fput;
 621
 622         ret = -EXDEV;
 623         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 624                 goto out_fput;
 625
 626         ret = -ENOMEM;
 627         buf = vmalloc(btrfs_level_size(root, 0));
 628         if (!buf)
 629                 goto out_fput;
 630
 631         path = btrfs_alloc_path();
 632         if (!path) {
 633                 vfree(buf);
 634                 goto out_fput;
 635         }
 636         path->reada = 2;
 637
 638         if (inode < src) {
 639                 mutex_lock(&inode->i_mutex);
 640                 mutex_lock(&src->i_mutex);
 641         } else {
 642                 mutex_lock(&src->i_mutex);
 643                 mutex_lock(&inode->i_mutex);
 644         }
 645
 646         ret = -ENOTEMPTY;
 647         if (inode->i_size)
 648                 goto out_unlock;
 649
 650         /* do any pending delalloc/csum calc on src, one way or
 651            another, and lock file content */
 652         while (1) {
 653                 struct btrfs_ordered_extent *ordered;
 654                 lock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 655                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
 656                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 657                         break;
 658                 unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 659                 if (ordered)
 660                         btrfs_put_ordered_extent(ordered);
 661                 btrfs_wait_ordered_range(src, 0, (u64)-1);
 662         }
 663
 664         trans = btrfs_start_transaction(root, 1);
 665         BUG_ON(!trans);
 666
 667         key.objectid = src->i_ino;
 668         key.type = BTRFS_EXTENT_DATA_KEY;
 669         key.offset = 0;
 670
 671         while (1) {
 672                 /*
 673                  * note the key will change type as we walk through the
 674                  * tree.
 675                  */
 676                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 677                 if (ret < 0)
 678                         goto out;
 679
 680                 nritems = btrfs_header_nritems(path->nodes[0]);
 681                 if (path->slots[0] >= nritems) {
 682                         ret = btrfs_next_leaf(root, path);
 683                         if (ret < 0)
 684                                 goto out;
 685                         if (ret > 0)
 686                                 break;
 687                         nritems = btrfs_header_nritems(path->nodes[0]);
 688                 }
 689                 leaf = path->nodes[0];
 690                 slot = path->slots[0];
 691
 692                 btrfs_item_key_to_cpu(leaf, &key, slot);
 693                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
 694                     key.objectid != src->i_ino)
 695                         break;
 696
 697                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY ||
 698                     btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
 699                         u32 size;
 700                         struct btrfs_key new_key;
 701
 702                         size = btrfs_item_size_nr(leaf, slot);
 703                         read_extent_buffer(leaf, buf,
 704                                            btrfs_item_ptr_offset(leaf, slot),
 705                                            size);
 706                         btrfs_release_path(root, path);
 707
 708                         memcpy(&new_key, &key, sizeof(new_key));
 709                         new_key.objectid = inode->i_ino;
 710                         ret = btrfs_insert_empty_item(trans, root, path,
 711                                                       &new_key, size);
 712                         if (ret)
 713                                 goto out;
 714
 715                         leaf = path->nodes[0];
 716                         slot = path->slots[0];
 717                         write_extent_buffer(leaf, buf,
 718                                             btrfs_item_ptr_offset(leaf, slot),
 719                                             size);
 720                         btrfs_mark_buffer_dirty(leaf);
 721                 }
 722
 723                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 724                         struct btrfs_file_extent_item *extent;
 725                         int found_type;
 726
 727                         extent = btrfs_item_ptr(leaf, slot,
 728                                                 struct btrfs_file_extent_item);
 729                         found_type = btrfs_file_extent_type(leaf, extent);
 730                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 731                                 u64 ds = btrfs_file_extent_disk_bytenr(leaf,
 732                                                                        extent);
 733                                 u64 dl = btrfs_file_extent_disk_num_bytes(leaf,
 734                                                                  extent);
 735                                 /* ds == 0 means there's a hole */
 736                                 if (ds != 0) {
 737                                         ret = btrfs_inc_extent_ref(trans, root,
 738                                                      ds, dl, leaf->start,
 739                                                      root->root_key.objectid,
 740                                                      trans->transid,
 741                                                      inode->i_ino);
 742                                         BUG_ON(ret);
 743                                 }
 744                         }
 745                 }
 746                 btrfs_release_path(root, path);
 747                 key.offset++;
 748         }
 749         ret = 0;
 750 out:
 751         btrfs_release_path(root, path);
 752         if (ret == 0) {
 753                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 754                 inode_set_bytes(inode, inode_get_bytes(src));
 755                 btrfs_i_size_write(inode, src->i_size);
 756                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
 757                 ret = btrfs_update_inode(trans, root, inode);
 758         }
 759         btrfs_end_transaction(trans, root);
 760         unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
 761         if (ret)
 762                 vmtruncate(inode, 0);
 763 out_unlock:
 764         mutex_unlock(&src->i_mutex);
 765         mutex_unlock(&inode->i_mutex);
 766         vfree(buf);
 767         btrfs_free_path(path);
 768 out_fput:
 769         fput(src_file);
 770         return ret;
 771 }
 772
 773 /*
 774  * there are many ways the trans_start and trans_end ioctls can lead
 775  * to deadlocks.  They should only be used by applications that
 776  * basically own the machine, and have a very in depth understanding
 777  * of all the possible deadlocks and enospc problems.
 778  */
 779 long btrfs_ioctl_trans_start(struct file *file)
 780 {
 781         struct inode *inode = fdentry(file)->d_inode;
 782         struct btrfs_root *root = BTRFS_I(inode)->root;
 783         struct btrfs_trans_handle *trans;
 784         int ret = 0;
 785
 786         if (!capable(CAP_SYS_ADMIN))
 787                 return -EPERM;
 788
 789         if (file->private_data) {
 790                 ret = -EINPROGRESS;
 791                 goto out;
 792         }
 793
 794         mutex_lock(&root->fs_info->trans_mutex);
 795         root->fs_info->open_ioctl_trans++;
 796         mutex_unlock(&root->fs_info->trans_mutex);
 797
 798         trans = btrfs_start_ioctl_transaction(root, 0);
 799         if (trans)
 800                 file->private_data = trans;
 801         else
 802                 ret = -ENOMEM;
 803         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
 804 out:
 805         return ret;
 806 }
 807
 808 /*
 809  * there are many ways the trans_start and trans_end ioctls can lead
 810  * to deadlocks.  They should only be used by applications that
 811  * basically own the machine, and have a very in depth understanding
 812  * of all the possible deadlocks and enospc problems.
 813  */
 814 long btrfs_ioctl_trans_end(struct file *file)
 815 {
 816         struct inode *inode = fdentry(file)->d_inode;
 817         struct btrfs_root *root = BTRFS_I(inode)->root;
 818         struct btrfs_trans_handle *trans;
 819         int ret = 0;
 820
 821         trans = file->private_data;
 822         if (!trans) {
 823                 ret = -EINVAL;
 824                 goto out;
 825         }
 826         btrfs_end_transaction(trans, root);
 827         file->private_data = NULL;
 828
 829         mutex_lock(&root->fs_info->trans_mutex);
 830         root->fs_info->open_ioctl_trans--;
 831         mutex_unlock(&root->fs_info->trans_mutex);
 832
 833 out:
 834         return ret;
 835 }
 836
 837 long btrfs_ioctl(struct file *file, unsigned int
 838                 cmd, unsigned long arg)
 839 {
 840         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 841
 842         switch (cmd) {
 843         case BTRFS_IOC_SNAP_CREATE:
 844                 return btrfs_ioctl_snap_create(file, (void __user *)arg);
 845         case BTRFS_IOC_DEFRAG:
 846                 return btrfs_ioctl_defrag(file);
 847         case BTRFS_IOC_RESIZE:
 848                 return btrfs_ioctl_resize(root, (void __user *)arg);
 849         case BTRFS_IOC_ADD_DEV:
 850                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
 851         case BTRFS_IOC_RM_DEV:
 852                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
 853         case BTRFS_IOC_BALANCE:
 854                 return btrfs_balance(root->fs_info->dev_root);
 855         case BTRFS_IOC_CLONE:
 856                 return btrfs_ioctl_clone(file, arg);
 857         case BTRFS_IOC_TRANS_START:
 858                 return btrfs_ioctl_trans_start(file);
 859         case BTRFS_IOC_TRANS_END:
 860                 return btrfs_ioctl_trans_end(file);
 861         case BTRFS_IOC_SYNC:
 862                 btrfs_start_delalloc_inodes(root);
 863                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
 864                 return 0;
 865         }
 866
 867         return -ENOTTY;
 868 }