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/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include "compat.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 /* Mask out flags that are inappropriate for the given type of inode. */
  54 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  55 {
  56         if (S_ISDIR(mode))
  57                 return flags;
  58         else if (S_ISREG(mode))
  59                 return flags & ~FS_DIRSYNC_FL;
  60         else
  61                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  62 }
  63
  64 /*
  65  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  66  */
  67 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  68 {
  69         unsigned int iflags = 0;
  70
  71         if (flags & BTRFS_INODE_SYNC)
  72                 iflags |= FS_SYNC_FL;
  73         if (flags & BTRFS_INODE_IMMUTABLE)
  74                 iflags |= FS_IMMUTABLE_FL;
  75         if (flags & BTRFS_INODE_APPEND)
  76                 iflags |= FS_APPEND_FL;
  77         if (flags & BTRFS_INODE_NODUMP)
  78                 iflags |= FS_NODUMP_FL;
  79         if (flags & BTRFS_INODE_NOATIME)
  80                 iflags |= FS_NOATIME_FL;
  81         if (flags & BTRFS_INODE_DIRSYNC)
  82                 iflags |= FS_DIRSYNC_FL;
  83
  84         return iflags;
  85 }
  86
  87 /*
  88  * Update inode->i_flags based on the btrfs internal flags.
  89  */
  90 void btrfs_update_iflags(struct inode *inode)
  91 {
  92         struct btrfs_inode *ip = BTRFS_I(inode);
  93
  94         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
  95
  96         if (ip->flags & BTRFS_INODE_SYNC)
  97                 inode->i_flags |= S_SYNC;
  98         if (ip->flags & BTRFS_INODE_IMMUTABLE)
  99                 inode->i_flags |= S_IMMUTABLE;
 100         if (ip->flags & BTRFS_INODE_APPEND)
 101                 inode->i_flags |= S_APPEND;
 102         if (ip->flags & BTRFS_INODE_NOATIME)
 103                 inode->i_flags |= S_NOATIME;
 104         if (ip->flags & BTRFS_INODE_DIRSYNC)
 105                 inode->i_flags |= S_DIRSYNC;
 106 }
 107
 108 /*
 109  * Inherit flags from the parent inode.
 110  *
 111  * Unlike extN we don't have any flags we don't want to inherit currently.
 112  */
 113 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 114 {
 115         unsigned int flags;
 116
 117         if (!dir)
 118                 return;
 119
 120         flags = BTRFS_I(dir)->flags;
 121
 122         if (S_ISREG(inode->i_mode))
 123                 flags &= ~BTRFS_INODE_DIRSYNC;
 124         else if (!S_ISDIR(inode->i_mode))
 125                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
 126
 127         BTRFS_I(inode)->flags = flags;
 128         btrfs_update_iflags(inode);
 129 }
 130
 131 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 132 {
 133         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 134         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 135
 136         if (copy_to_user(arg, &flags, sizeof(flags)))
 137                 return -EFAULT;
 138         return 0;
 139 }
 140
 141 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 142 {
 143         struct inode *inode = file->f_path.dentry->d_inode;
 144         struct btrfs_inode *ip = BTRFS_I(inode);
 145         struct btrfs_root *root = ip->root;
 146         struct btrfs_trans_handle *trans;
 147         unsigned int flags, oldflags;
 148         int ret;
 149
 150         if (copy_from_user(&flags, arg, sizeof(flags)))
 151                 return -EFAULT;
 152
 153         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 154                       FS_NOATIME_FL | FS_NODUMP_FL | \
 155                       FS_SYNC_FL | FS_DIRSYNC_FL))
 156                 return -EOPNOTSUPP;
 157
 158         if (!is_owner_or_cap(inode))
 159                 return -EACCES;
 160
 161         mutex_lock(&inode->i_mutex);
 162
 163         flags = btrfs_mask_flags(inode->i_mode, flags);
 164         oldflags = btrfs_flags_to_ioctl(ip->flags);
 165         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 166                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 167                         ret = -EPERM;
 168                         goto out_unlock;
 169                 }
 170         }
 171
 172         ret = mnt_want_write(file->f_path.mnt);
 173         if (ret)
 174                 goto out_unlock;
 175
 176         if (flags & FS_SYNC_FL)
 177                 ip->flags |= BTRFS_INODE_SYNC;
 178         else
 179                 ip->flags &= ~BTRFS_INODE_SYNC;
 180         if (flags & FS_IMMUTABLE_FL)
 181                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 182         else
 183                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 184         if (flags & FS_APPEND_FL)
 185                 ip->flags |= BTRFS_INODE_APPEND;
 186         else
 187                 ip->flags &= ~BTRFS_INODE_APPEND;
 188         if (flags & FS_NODUMP_FL)
 189                 ip->flags |= BTRFS_INODE_NODUMP;
 190         else
 191                 ip->flags &= ~BTRFS_INODE_NODUMP;
 192         if (flags & FS_NOATIME_FL)
 193                 ip->flags |= BTRFS_INODE_NOATIME;
 194         else
 195                 ip->flags &= ~BTRFS_INODE_NOATIME;
 196         if (flags & FS_DIRSYNC_FL)
 197                 ip->flags |= BTRFS_INODE_DIRSYNC;
 198         else
 199                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 200
 201
 202         trans = btrfs_join_transaction(root, 1);
 203         BUG_ON(!trans);
 204
 205         ret = btrfs_update_inode(trans, root, inode);
 206         BUG_ON(ret);
 207
 208         btrfs_update_iflags(inode);
 209         inode->i_ctime = CURRENT_TIME;
 210         btrfs_end_transaction(trans, root);
 211
 212         mnt_drop_write(file->f_path.mnt);
 213  out_unlock:
 214         mutex_unlock(&inode->i_mutex);
 215         return 0;
 216 }
 217
 218 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 219 {
 220         struct inode *inode = file->f_path.dentry->d_inode;
 221
 222         return put_user(inode->i_generation, arg);
 223 }
 224
 225 static noinline int create_subvol(struct btrfs_root *root,
 226                                   struct dentry *dentry,
 227                                   char *name, int namelen)
 228 {
 229         struct btrfs_trans_handle *trans;
 230         struct btrfs_key key;
 231         struct btrfs_root_item root_item;
 232         struct btrfs_inode_item *inode_item;
 233         struct extent_buffer *leaf;
 234         struct btrfs_root *new_root;
 235         struct inode *dir = dentry->d_parent->d_inode;
 236         int ret;
 237         int err;
 238         u64 objectid;
 239         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 240         u64 index = 0;
 241
 242         ret = btrfs_find_free_objectid(NULL, root->fs_info->tree_root,
 243                                        0, &objectid);
 244         if (ret)
 245                 return ret;
 246         /*
 247          * 1 - inode item
 248          * 2 - refs
 249          * 1 - root item
 250          * 2 - dir items
 251          */
 252         trans = btrfs_start_transaction(root, 6);
 253         if (IS_ERR(trans))
 254                 return PTR_ERR(trans);
 255
 256         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 257                                       0, objectid, NULL, 0, 0, 0);
 258         if (IS_ERR(leaf)) {
 259                 ret = PTR_ERR(leaf);
 260                 goto fail;
 261         }
 262
 263         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 264         btrfs_set_header_bytenr(leaf, leaf->start);
 265         btrfs_set_header_generation(leaf, trans->transid);
 266         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 267         btrfs_set_header_owner(leaf, objectid);
 268
 269         write_extent_buffer(leaf, root->fs_info->fsid,
 270                             (unsigned long)btrfs_header_fsid(leaf),
 271                             BTRFS_FSID_SIZE);
 272         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 273                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 274                             BTRFS_UUID_SIZE);
 275         btrfs_mark_buffer_dirty(leaf);
 276
 277         inode_item = &root_item.inode;
 278         memset(inode_item, 0, sizeof(*inode_item));
 279         inode_item->generation = cpu_to_le64(1);
 280         inode_item->size = cpu_to_le64(3);
 281         inode_item->nlink = cpu_to_le32(1);
 282         inode_item->nbytes = cpu_to_le64(root->leafsize);
 283         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 284
 285         btrfs_set_root_bytenr(&root_item, leaf->start);
 286         btrfs_set_root_generation(&root_item, trans->transid);
 287         btrfs_set_root_level(&root_item, 0);
 288         btrfs_set_root_refs(&root_item, 1);
 289         btrfs_set_root_used(&root_item, leaf->len);
 290         btrfs_set_root_last_snapshot(&root_item, 0);
 291
 292         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 293         root_item.drop_level = 0;
 294
 295         btrfs_tree_unlock(leaf);
 296         free_extent_buffer(leaf);
 297         leaf = NULL;
 298
 299         btrfs_set_root_dirid(&root_item, new_dirid);
 300
 301         key.objectid = objectid;
 302         key.offset = 0;
 303         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 304         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 305                                 &root_item);
 306         if (ret)
 307                 goto fail;
 308
 309         key.offset = (u64)-1;
 310         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 311         BUG_ON(IS_ERR(new_root));
 312
 313         btrfs_record_root_in_trans(trans, new_root);
 314
 315         ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
 316                                        BTRFS_I(dir)->block_group);
 317         /*
 318          * insert the directory item
 319          */
 320         ret = btrfs_set_inode_index(dir, &index);
 321         BUG_ON(ret);
 322
 323         ret = btrfs_insert_dir_item(trans, root,
 324                                     name, namelen, dir->i_ino, &key,
 325                                     BTRFS_FT_DIR, index);
 326         if (ret)
 327                 goto fail;
 328
 329         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 330         ret = btrfs_update_inode(trans, root, dir);
 331         BUG_ON(ret);
 332
 333         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 334                                  objectid, root->root_key.objectid,
 335                                  dir->i_ino, index, name, namelen);
 336
 337         BUG_ON(ret);
 338
 339         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 340 fail:
 341         err = btrfs_commit_transaction(trans, root);
 342         if (err && !ret)
 343                 ret = err;
 344         return ret;
 345 }
 346
 347 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry)
 348 {
 349         struct inode *inode;
 350         struct btrfs_pending_snapshot *pending_snapshot;
 351         struct btrfs_trans_handle *trans;
 352         int ret;
 353
 354         if (!root->ref_cows)
 355                 return -EINVAL;
 356
 357         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 358         if (!pending_snapshot)
 359                 return -ENOMEM;
 360
 361         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
 362         pending_snapshot->dentry = dentry;
 363         pending_snapshot->root = root;
 364
 365         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
 366         if (IS_ERR(trans)) {
 367                 ret = PTR_ERR(trans);
 368                 goto fail;
 369         }
 370
 371         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 372         BUG_ON(ret);
 373
 374         list_add(&pending_snapshot->list,
 375                  &trans->transaction->pending_snapshots);
 376         ret = btrfs_commit_transaction(trans, root->fs_info->extent_root);
 377         BUG_ON(ret);
 378
 379         ret = pending_snapshot->error;
 380         if (ret)
 381                 goto fail;
 382
 383         btrfs_orphan_cleanup(pending_snapshot->snap);
 384
 385         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 386         if (IS_ERR(inode)) {
 387                 ret = PTR_ERR(inode);
 388                 goto fail;
 389         }
 390         BUG_ON(!inode);
 391         d_instantiate(dentry, inode);
 392         ret = 0;
 393 fail:
 394         kfree(pending_snapshot);
 395         return ret;
 396 }
 397
 398 /* copy of may_create in fs/namei.c() */
 399 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 400 {
 401         if (child->d_inode)
 402                 return -EEXIST;
 403         if (IS_DEADDIR(dir))
 404                 return -ENOENT;
 405         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 406 }
 407
 408 /*
 409  * Create a new subvolume below @parent.  This is largely modeled after
 410  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 411  * inside this filesystem so it's quite a bit simpler.
 412  */
 413 static noinline int btrfs_mksubvol(struct path *parent,
 414                                    char *name, int namelen,
 415                                    struct btrfs_root *snap_src)
 416 {
 417         struct inode *dir  = parent->dentry->d_inode;
 418         struct dentry *dentry;
 419         int error;
 420
 421         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 422
 423         dentry = lookup_one_len(name, parent->dentry, namelen);
 424         error = PTR_ERR(dentry);
 425         if (IS_ERR(dentry))
 426                 goto out_unlock;
 427
 428         error = -EEXIST;
 429         if (dentry->d_inode)
 430                 goto out_dput;
 431
 432         error = mnt_want_write(parent->mnt);
 433         if (error)
 434                 goto out_dput;
 435
 436         error = btrfs_may_create(dir, dentry);
 437         if (error)
 438                 goto out_drop_write;
 439
 440         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 441
 442         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 443                 goto out_up_read;
 444
 445         if (snap_src) {
 446                 error = create_snapshot(snap_src, dentry);
 447         } else {
 448                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 449                                       name, namelen);
 450         }
 451         if (!error)
 452                 fsnotify_mkdir(dir, dentry);
 453 out_up_read:
 454         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 455 out_drop_write:
 456         mnt_drop_write(parent->mnt);
 457 out_dput:
 458         dput(dentry);
 459 out_unlock:
 460         mutex_unlock(&dir->i_mutex);
 461         return error;
 462 }
 463
 464 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
 465                                int thresh, u64 *last_len, u64 *skip,
 466                                u64 *defrag_end)
 467 {
 468         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 469         struct extent_map *em = NULL;
 470         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 471         int ret = 1;
 472
 473
 474         if (thresh == 0)
 475                 thresh = 256 * 1024;
 476
 477         /*
 478          * make sure that once we start defragging and extent, we keep on
 479          * defragging it
 480          */
 481         if (start < *defrag_end)
 482                 return 1;
 483
 484         *skip = 0;
 485
 486         /*
 487          * hopefully we have this extent in the tree already, try without
 488          * the full extent lock
 489          */
 490         read_lock(&em_tree->lock);
 491         em = lookup_extent_mapping(em_tree, start, len);
 492         read_unlock(&em_tree->lock);
 493
 494         if (!em) {
 495                 /* get the big lock and read metadata off disk */
 496                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 497                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 498                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 499
 500                 if (IS_ERR(em))
 501                         return 0;
 502         }
 503
 504         /* this will cover holes, and inline extents */
 505         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
 506                 ret = 0;
 507
 508         /*
 509          * we hit a real extent, if it is big don't bother defragging it again
 510          */
 511         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
 512                 ret = 0;
 513
 514         /*
 515          * last_len ends up being a counter of how many bytes we've defragged.
 516          * every time we choose not to defrag an extent, we reset *last_len
 517          * so that the next tiny extent will force a defrag.
 518          *
 519          * The end result of this is that tiny extents before a single big
 520          * extent will force at least part of that big extent to be defragged.
 521          */
 522         if (ret) {
 523                 *last_len += len;
 524                 *defrag_end = extent_map_end(em);
 525         } else {
 526                 *last_len = 0;
 527                 *skip = extent_map_end(em);
 528                 *defrag_end = 0;
 529         }
 530
 531         free_extent_map(em);
 532         return ret;
 533 }
 534
 535 static int btrfs_defrag_file(struct file *file,
 536                              struct btrfs_ioctl_defrag_range_args *range)
 537 {
 538         struct inode *inode = fdentry(file)->d_inode;
 539         struct btrfs_root *root = BTRFS_I(inode)->root;
 540         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 541         struct btrfs_ordered_extent *ordered;
 542         struct page *page;
 543         unsigned long last_index;
 544         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 545         unsigned long total_read = 0;
 546         u64 page_start;
 547         u64 page_end;
 548         u64 last_len = 0;
 549         u64 skip = 0;
 550         u64 defrag_end = 0;
 551         unsigned long i;
 552         int ret;
 553
 554         if (inode->i_size == 0)
 555                 return 0;
 556
 557         if (range->start + range->len > range->start) {
 558                 last_index = min_t(u64, inode->i_size - 1,
 559                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
 560         } else {
 561                 last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
 562         }
 563
 564         i = range->start >> PAGE_CACHE_SHIFT;
 565         while (i <= last_index) {
 566                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
 567                                         PAGE_CACHE_SIZE,
 568                                         range->extent_thresh,
 569                                         &last_len, &skip,
 570                                         &defrag_end)) {
 571                         unsigned long next;
 572                         /*
 573                          * the should_defrag function tells us how much to skip
 574                          * bump our counter by the suggested amount
 575                          */
 576                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 577                         i = max(i + 1, next);
 578                         continue;
 579                 }
 580
 581                 if (total_read % ra_pages == 0) {
 582                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 583                                        min(last_index, i + ra_pages - 1));
 584                 }
 585                 total_read++;
 586                 mutex_lock(&inode->i_mutex);
 587                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
 588                         BTRFS_I(inode)->force_compress = 1;
 589
 590                 ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
 591                 if (ret)
 592                         goto err_unlock;
 593 again:
 594                 if (inode->i_size == 0 ||
 595                     i > ((inode->i_size - 1) >> PAGE_CACHE_SHIFT)) {
 596                         ret = 0;
 597                         goto err_reservations;
 598                 }
 599
 600                 page = grab_cache_page(inode->i_mapping, i);
 601                 if (!page) {
 602                         ret = -ENOMEM;
 603                         goto err_reservations;
 604                 }
 605
 606                 if (!PageUptodate(page)) {
 607                         btrfs_readpage(NULL, page);
 608                         lock_page(page);
 609                         if (!PageUptodate(page)) {
 610                                 unlock_page(page);
 611                                 page_cache_release(page);
 612                                 ret = -EIO;
 613                                 goto err_reservations;
 614                         }
 615                 }
 616
 617                 if (page->mapping != inode->i_mapping) {
 618                         unlock_page(page);
 619                         page_cache_release(page);
 620                         goto again;
 621                 }
 622
 623                 wait_on_page_writeback(page);
 624
 625                 if (PageDirty(page)) {
 626                         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 627                         goto loop_unlock;
 628                 }
 629
 630                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 631                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 632                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 633
 634                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 635                 if (ordered) {
 636                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 637                         unlock_page(page);
 638                         page_cache_release(page);
 639                         btrfs_start_ordered_extent(inode, ordered, 1);
 640                         btrfs_put_ordered_extent(ordered);
 641                         goto again;
 642                 }
 643                 set_page_extent_mapped(page);
 644
 645                 /*
 646                  * this makes sure page_mkwrite is called on the
 647                  * page if it is dirtied again later
 648                  */
 649                 clear_page_dirty_for_io(page);
 650                 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start,
 651                                   page_end, EXTENT_DIRTY | EXTENT_DELALLOC |
 652                                   EXTENT_DO_ACCOUNTING, GFP_NOFS);
 653
 654                 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
 655                 ClearPageChecked(page);
 656                 set_page_dirty(page);
 657                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 658
 659 loop_unlock:
 660                 unlock_page(page);
 661                 page_cache_release(page);
 662                 mutex_unlock(&inode->i_mutex);
 663
 664                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 665                 i++;
 666         }
 667
 668         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
 669                 filemap_flush(inode->i_mapping);
 670
 671         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
 672                 /* the filemap_flush will queue IO into the worker threads, but
 673                  * we have to make sure the IO is actually started and that
 674                  * ordered extents get created before we return
 675                  */
 676                 atomic_inc(&root->fs_info->async_submit_draining);
 677                 while (atomic_read(&root->fs_info->nr_async_submits) ||
 678                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 679                         wait_event(root->fs_info->async_submit_wait,
 680                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
 681                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
 682                 }
 683                 atomic_dec(&root->fs_info->async_submit_draining);
 684
 685                 mutex_lock(&inode->i_mutex);
 686                 BTRFS_I(inode)->force_compress = 0;
 687                 mutex_unlock(&inode->i_mutex);
 688         }
 689
 690         return 0;
 691
 692 err_reservations:
 693         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 694 err_unlock:
 695         mutex_unlock(&inode->i_mutex);
 696         return ret;
 697 }
 698
 699 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
 700                                         void __user *arg)
 701 {
 702         u64 new_size;
 703         u64 old_size;
 704         u64 devid = 1;
 705         struct btrfs_ioctl_vol_args *vol_args;
 706         struct btrfs_trans_handle *trans;
 707         struct btrfs_device *device = NULL;
 708         char *sizestr;
 709         char *devstr = NULL;
 710         int ret = 0;
 711         int mod = 0;
 712
 713         if (root->fs_info->sb->s_flags & MS_RDONLY)
 714                 return -EROFS;
 715
 716         if (!capable(CAP_SYS_ADMIN))
 717                 return -EPERM;
 718
 719         vol_args = memdup_user(arg, sizeof(*vol_args));
 720         if (IS_ERR(vol_args))
 721                 return PTR_ERR(vol_args);
 722
 723         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 724
 725         mutex_lock(&root->fs_info->volume_mutex);
 726         sizestr = vol_args->name;
 727         devstr = strchr(sizestr, ':');
 728         if (devstr) {
 729                 char *end;
 730                 sizestr = devstr + 1;
 731                 *devstr = '\0';
 732                 devstr = vol_args->name;
 733                 devid = simple_strtoull(devstr, &end, 10);
 734                 printk(KERN_INFO "resizing devid %llu\n",
 735                        (unsigned long long)devid);
 736         }
 737         device = btrfs_find_device(root, devid, NULL, NULL);
 738         if (!device) {
 739                 printk(KERN_INFO "resizer unable to find device %llu\n",
 740                        (unsigned long long)devid);
 741                 ret = -EINVAL;
 742                 goto out_unlock;
 743         }
 744         if (!strcmp(sizestr, "max"))
 745                 new_size = device->bdev->bd_inode->i_size;
 746         else {
 747                 if (sizestr[0] == '-') {
 748                         mod = -1;
 749                         sizestr++;
 750                 } else if (sizestr[0] == '+') {
 751                         mod = 1;
 752                         sizestr++;
 753                 }
 754                 new_size = memparse(sizestr, NULL);
 755                 if (new_size == 0) {
 756                         ret = -EINVAL;
 757                         goto out_unlock;
 758                 }
 759         }
 760
 761         old_size = device->total_bytes;
 762
 763         if (mod < 0) {
 764                 if (new_size > old_size) {
 765                         ret = -EINVAL;
 766                         goto out_unlock;
 767                 }
 768                 new_size = old_size - new_size;
 769         } else if (mod > 0) {
 770                 new_size = old_size + new_size;
 771         }
 772
 773         if (new_size < 256 * 1024 * 1024) {
 774                 ret = -EINVAL;
 775                 goto out_unlock;
 776         }
 777         if (new_size > device->bdev->bd_inode->i_size) {
 778                 ret = -EFBIG;
 779                 goto out_unlock;
 780         }
 781
 782         do_div(new_size, root->sectorsize);
 783         new_size *= root->sectorsize;
 784
 785         printk(KERN_INFO "new size for %s is %llu\n",
 786                 device->name, (unsigned long long)new_size);
 787
 788         if (new_size > old_size) {
 789                 trans = btrfs_start_transaction(root, 0);
 790                 ret = btrfs_grow_device(trans, device, new_size);
 791                 btrfs_commit_transaction(trans, root);
 792         } else {
 793                 ret = btrfs_shrink_device(device, new_size);
 794         }
 795
 796 out_unlock:
 797         mutex_unlock(&root->fs_info->volume_mutex);
 798         kfree(vol_args);
 799         return ret;
 800 }
 801
 802 static noinline int btrfs_ioctl_snap_create(struct file *file,
 803                                             void __user *arg, int subvol)
 804 {
 805         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 806         struct btrfs_ioctl_vol_args *vol_args;
 807         struct file *src_file;
 808         int namelen;
 809         int ret = 0;
 810
 811         if (root->fs_info->sb->s_flags & MS_RDONLY)
 812                 return -EROFS;
 813
 814         vol_args = memdup_user(arg, sizeof(*vol_args));
 815         if (IS_ERR(vol_args))
 816                 return PTR_ERR(vol_args);
 817
 818         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 819         namelen = strlen(vol_args->name);
 820         if (strchr(vol_args->name, '/')) {
 821                 ret = -EINVAL;
 822                 goto out;
 823         }
 824
 825         if (subvol) {
 826                 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
 827                                      NULL);
 828         } else {
 829                 struct inode *src_inode;
 830                 src_file = fget(vol_args->fd);
 831                 if (!src_file) {
 832                         ret = -EINVAL;
 833                         goto out;
 834                 }
 835
 836                 src_inode = src_file->f_path.dentry->d_inode;
 837                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 838                         printk(KERN_INFO "btrfs: Snapshot src from "
 839                                "another FS\n");
 840                         ret = -EINVAL;
 841                         fput(src_file);
 842                         goto out;
 843                 }
 844                 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
 845                                      BTRFS_I(src_inode)->root);
 846                 fput(src_file);
 847         }
 848 out:
 849         kfree(vol_args);
 850         return ret;
 851 }
 852
 853 /*
 854  * helper to check if the subvolume references other subvolumes
 855  */
 856 static noinline int may_destroy_subvol(struct btrfs_root *root)
 857 {
 858         struct btrfs_path *path;
 859         struct btrfs_key key;
 860         int ret;
 861
 862         path = btrfs_alloc_path();
 863         if (!path)
 864                 return -ENOMEM;
 865
 866         key.objectid = root->root_key.objectid;
 867         key.type = BTRFS_ROOT_REF_KEY;
 868         key.offset = (u64)-1;
 869
 870         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
 871                                 &key, path, 0, 0);
 872         if (ret < 0)
 873                 goto out;
 874         BUG_ON(ret == 0);
 875
 876         ret = 0;
 877         if (path->slots[0] > 0) {
 878                 path->slots[0]--;
 879                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 880                 if (key.objectid == root->root_key.objectid &&
 881                     key.type == BTRFS_ROOT_REF_KEY)
 882                         ret = -ENOTEMPTY;
 883         }
 884 out:
 885         btrfs_free_path(path);
 886         return ret;
 887 }
 888
 889 static noinline int key_in_sk(struct btrfs_key *key,
 890                               struct btrfs_ioctl_search_key *sk)
 891 {
 892         struct btrfs_key test;
 893         int ret;
 894
 895         test.objectid = sk->min_objectid;
 896         test.type = sk->min_type;
 897         test.offset = sk->min_offset;
 898
 899         ret = btrfs_comp_cpu_keys(key, &test);
 900         if (ret < 0)
 901                 return 0;
 902
 903         test.objectid = sk->max_objectid;
 904         test.type = sk->max_type;
 905         test.offset = sk->max_offset;
 906
 907         ret = btrfs_comp_cpu_keys(key, &test);
 908         if (ret > 0)
 909                 return 0;
 910         return 1;
 911 }
 912
 913 static noinline int copy_to_sk(struct btrfs_root *root,
 914                                struct btrfs_path *path,
 915                                struct btrfs_key *key,
 916                                struct btrfs_ioctl_search_key *sk,
 917                                char *buf,
 918                                unsigned long *sk_offset,
 919                                int *num_found)
 920 {
 921         u64 found_transid;
 922         struct extent_buffer *leaf;
 923         struct btrfs_ioctl_search_header sh;
 924         unsigned long item_off;
 925         unsigned long item_len;
 926         int nritems;
 927         int i;
 928         int slot;
 929         int found = 0;
 930         int ret = 0;
 931
 932         leaf = path->nodes[0];
 933         slot = path->slots[0];
 934         nritems = btrfs_header_nritems(leaf);
 935
 936         if (btrfs_header_generation(leaf) > sk->max_transid) {
 937                 i = nritems;
 938                 goto advance_key;
 939         }
 940         found_transid = btrfs_header_generation(leaf);
 941
 942         for (i = slot; i < nritems; i++) {
 943                 item_off = btrfs_item_ptr_offset(leaf, i);
 944                 item_len = btrfs_item_size_nr(leaf, i);
 945
 946                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
 947                         item_len = 0;
 948
 949                 if (sizeof(sh) + item_len + *sk_offset >
 950                     BTRFS_SEARCH_ARGS_BUFSIZE) {
 951                         ret = 1;
 952                         goto overflow;
 953                 }
 954
 955                 btrfs_item_key_to_cpu(leaf, key, i);
 956                 if (!key_in_sk(key, sk))
 957                         continue;
 958
 959                 sh.objectid = key->objectid;
 960                 sh.offset = key->offset;
 961                 sh.type = key->type;
 962                 sh.len = item_len;
 963                 sh.transid = found_transid;
 964
 965                 /* copy search result header */
 966                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
 967                 *sk_offset += sizeof(sh);
 968
 969                 if (item_len) {
 970                         char *p = buf + *sk_offset;
 971                         /* copy the item */
 972                         read_extent_buffer(leaf, p,
 973                                            item_off, item_len);
 974                         *sk_offset += item_len;
 975                 }
 976                 found++;
 977
 978                 if (*num_found >= sk->nr_items)
 979                         break;
 980         }
 981 advance_key:
 982         ret = 0;
 983         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
 984                 key->offset++;
 985         else if (key->type < (u8)-1 && key->type < sk->max_type) {
 986                 key->offset = 0;
 987                 key->type++;
 988         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
 989                 key->offset = 0;
 990                 key->type = 0;
 991                 key->objectid++;
 992         } else
 993                 ret = 1;
 994 overflow:
 995         *num_found += found;
 996         return ret;
 997 }
 998
 999 static noinline int search_ioctl(struct inode *inode,
1000                                  struct btrfs_ioctl_search_args *args)
1001 {
1002         struct btrfs_root *root;
1003         struct btrfs_key key;
1004         struct btrfs_key max_key;
1005         struct btrfs_path *path;
1006         struct btrfs_ioctl_search_key *sk = &args->key;
1007         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1008         int ret;
1009         int num_found = 0;
1010         unsigned long sk_offset = 0;
1011
1012         path = btrfs_alloc_path();
1013         if (!path)
1014                 return -ENOMEM;
1015
1016         if (sk->tree_id == 0) {
1017                 /* search the root of the inode that was passed */
1018                 root = BTRFS_I(inode)->root;
1019         } else {
1020                 key.objectid = sk->tree_id;
1021                 key.type = BTRFS_ROOT_ITEM_KEY;
1022                 key.offset = (u64)-1;
1023                 root = btrfs_read_fs_root_no_name(info, &key);
1024                 if (IS_ERR(root)) {
1025                         printk(KERN_ERR "could not find root %llu\n",
1026                                sk->tree_id);
1027                         btrfs_free_path(path);
1028                         return -ENOENT;
1029                 }
1030         }
1031
1032         key.objectid = sk->min_objectid;
1033         key.type = sk->min_type;
1034         key.offset = sk->min_offset;
1035
1036         max_key.objectid = sk->max_objectid;
1037         max_key.type = sk->max_type;
1038         max_key.offset = sk->max_offset;
1039
1040         path->keep_locks = 1;
1041
1042         while(1) {
1043                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1044                                            sk->min_transid);
1045                 if (ret != 0) {
1046                         if (ret > 0)
1047                                 ret = 0;
1048                         goto err;
1049                 }
1050                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1051                                  &sk_offset, &num_found);
1052                 btrfs_release_path(root, path);
1053                 if (ret || num_found >= sk->nr_items)
1054                         break;
1055
1056         }
1057         ret = 0;
1058 err:
1059         sk->nr_items = num_found;
1060         btrfs_free_path(path);
1061         return ret;
1062 }
1063
1064 static noinline int btrfs_ioctl_tree_search(struct file *file,
1065                                            void __user *argp)
1066 {
1067          struct btrfs_ioctl_search_args *args;
1068          struct inode *inode;
1069          int ret;
1070
1071         if (!capable(CAP_SYS_ADMIN))
1072                 return -EPERM;
1073
1074         args = memdup_user(argp, sizeof(*args));
1075         if (IS_ERR(args))
1076                 return PTR_ERR(args);
1077
1078         inode = fdentry(file)->d_inode;
1079         ret = search_ioctl(inode, args);
1080         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1081                 ret = -EFAULT;
1082         kfree(args);
1083         return ret;
1084 }
1085
1086 /*
1087  * Search INODE_REFs to identify path name of 'dirid' directory
1088  * in a 'tree_id' tree. and sets path name to 'name'.
1089  */
1090 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1091                                 u64 tree_id, u64 dirid, char *name)
1092 {
1093         struct btrfs_root *root;
1094         struct btrfs_key key;
1095         char *ptr;
1096         int ret = -1;
1097         int slot;
1098         int len;
1099         int total_len = 0;
1100         struct btrfs_inode_ref *iref;
1101         struct extent_buffer *l;
1102         struct btrfs_path *path;
1103
1104         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1105                 name[0]='\0';
1106                 return 0;
1107         }
1108
1109         path = btrfs_alloc_path();
1110         if (!path)
1111                 return -ENOMEM;
1112
1113         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1114
1115         key.objectid = tree_id;
1116         key.type = BTRFS_ROOT_ITEM_KEY;
1117         key.offset = (u64)-1;
1118         root = btrfs_read_fs_root_no_name(info, &key);
1119         if (IS_ERR(root)) {
1120                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1121                 ret = -ENOENT;
1122                 goto out;
1123         }
1124
1125         key.objectid = dirid;
1126         key.type = BTRFS_INODE_REF_KEY;
1127         key.offset = (u64)-1;
1128
1129         while(1) {
1130                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1131                 if (ret < 0)
1132                         goto out;
1133
1134                 l = path->nodes[0];
1135                 slot = path->slots[0];
1136                 if (ret > 0 && slot > 0)
1137                         slot--;
1138                 btrfs_item_key_to_cpu(l, &key, slot);
1139
1140                 if (ret > 0 && (key.objectid != dirid ||
1141                                 key.type != BTRFS_INODE_REF_KEY)) {
1142                         ret = -ENOENT;
1143                         goto out;
1144                 }
1145
1146                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1147                 len = btrfs_inode_ref_name_len(l, iref);
1148                 ptr -= len + 1;
1149                 total_len += len + 1;
1150                 if (ptr < name)
1151                         goto out;
1152
1153                 *(ptr + len) = '/';
1154                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1155
1156                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1157                         break;
1158
1159                 btrfs_release_path(root, path);
1160                 key.objectid = key.offset;
1161                 key.offset = (u64)-1;
1162                 dirid = key.objectid;
1163
1164         }
1165         if (ptr < name)
1166                 goto out;
1167         memcpy(name, ptr, total_len);
1168         name[total_len]='\0';
1169         ret = 0;
1170 out:
1171         btrfs_free_path(path);
1172         return ret;
1173 }
1174
1175 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1176                                            void __user *argp)
1177 {
1178          struct btrfs_ioctl_ino_lookup_args *args;
1179          struct inode *inode;
1180          int ret;
1181
1182         if (!capable(CAP_SYS_ADMIN))
1183                 return -EPERM;
1184
1185         args = memdup_user(argp, sizeof(*args));
1186         if (IS_ERR(args))
1187                 return PTR_ERR(args);
1188
1189         inode = fdentry(file)->d_inode;
1190
1191         if (args->treeid == 0)
1192                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1193
1194         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1195                                         args->treeid, args->objectid,
1196                                         args->name);
1197
1198         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1199                 ret = -EFAULT;
1200
1201         kfree(args);
1202         return ret;
1203 }
1204
1205 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1206                                              void __user *arg)
1207 {
1208         struct dentry *parent = fdentry(file);
1209         struct dentry *dentry;
1210         struct inode *dir = parent->d_inode;
1211         struct inode *inode;
1212         struct btrfs_root *root = BTRFS_I(dir)->root;
1213         struct btrfs_root *dest = NULL;
1214         struct btrfs_ioctl_vol_args *vol_args;
1215         struct btrfs_trans_handle *trans;
1216         int namelen;
1217         int ret;
1218         int err = 0;
1219
1220         if (!capable(CAP_SYS_ADMIN))
1221                 return -EPERM;
1222
1223         vol_args = memdup_user(arg, sizeof(*vol_args));
1224         if (IS_ERR(vol_args))
1225                 return PTR_ERR(vol_args);
1226
1227         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1228         namelen = strlen(vol_args->name);
1229         if (strchr(vol_args->name, '/') ||
1230             strncmp(vol_args->name, "..", namelen) == 0) {
1231                 err = -EINVAL;
1232                 goto out;
1233         }
1234
1235         err = mnt_want_write(file->f_path.mnt);
1236         if (err)
1237                 goto out;
1238
1239         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1240         dentry = lookup_one_len(vol_args->name, parent, namelen);
1241         if (IS_ERR(dentry)) {
1242                 err = PTR_ERR(dentry);
1243                 goto out_unlock_dir;
1244         }
1245
1246         if (!dentry->d_inode) {
1247                 err = -ENOENT;
1248                 goto out_dput;
1249         }
1250
1251         inode = dentry->d_inode;
1252         if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
1253                 err = -EINVAL;
1254                 goto out_dput;
1255         }
1256
1257         dest = BTRFS_I(inode)->root;
1258
1259         mutex_lock(&inode->i_mutex);
1260         err = d_invalidate(dentry);
1261         if (err)
1262                 goto out_unlock;
1263
1264         down_write(&root->fs_info->subvol_sem);
1265
1266         err = may_destroy_subvol(dest);
1267         if (err)
1268                 goto out_up_write;
1269
1270         trans = btrfs_start_transaction(root, 0);
1271         if (IS_ERR(trans)) {
1272                 err = PTR_ERR(trans);
1273                 goto out_up_write;
1274         }
1275         trans->block_rsv = &root->fs_info->global_block_rsv;
1276
1277         ret = btrfs_unlink_subvol(trans, root, dir,
1278                                 dest->root_key.objectid,
1279                                 dentry->d_name.name,
1280                                 dentry->d_name.len);
1281         BUG_ON(ret);
1282
1283         btrfs_record_root_in_trans(trans, dest);
1284
1285         memset(&dest->root_item.drop_progress, 0,
1286                 sizeof(dest->root_item.drop_progress));
1287         dest->root_item.drop_level = 0;
1288         btrfs_set_root_refs(&dest->root_item, 0);
1289
1290         if (!xchg(&dest->orphan_item_inserted, 1)) {
1291                 ret = btrfs_insert_orphan_item(trans,
1292                                         root->fs_info->tree_root,
1293                                         dest->root_key.objectid);
1294                 BUG_ON(ret);
1295         }
1296
1297         ret = btrfs_commit_transaction(trans, root);
1298         BUG_ON(ret);
1299         inode->i_flags |= S_DEAD;
1300 out_up_write:
1301         up_write(&root->fs_info->subvol_sem);
1302 out_unlock:
1303         mutex_unlock(&inode->i_mutex);
1304         if (!err) {
1305                 shrink_dcache_sb(root->fs_info->sb);
1306                 btrfs_invalidate_inodes(dest);
1307                 d_delete(dentry);
1308         }
1309 out_dput:
1310         dput(dentry);
1311 out_unlock_dir:
1312         mutex_unlock(&dir->i_mutex);
1313         mnt_drop_write(file->f_path.mnt);
1314 out:
1315         kfree(vol_args);
1316         return err;
1317 }
1318
1319 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1320 {
1321         struct inode *inode = fdentry(file)->d_inode;
1322         struct btrfs_root *root = BTRFS_I(inode)->root;
1323         struct btrfs_ioctl_defrag_range_args *range;
1324         int ret;
1325
1326         ret = mnt_want_write(file->f_path.mnt);
1327         if (ret)
1328                 return ret;
1329
1330         switch (inode->i_mode & S_IFMT) {
1331         case S_IFDIR:
1332                 if (!capable(CAP_SYS_ADMIN)) {
1333                         ret = -EPERM;
1334                         goto out;
1335                 }
1336                 ret = btrfs_defrag_root(root, 0);
1337                 if (ret)
1338                         goto out;
1339                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
1340                 break;
1341         case S_IFREG:
1342                 if (!(file->f_mode & FMODE_WRITE)) {
1343                         ret = -EINVAL;
1344                         goto out;
1345                 }
1346
1347                 range = kzalloc(sizeof(*range), GFP_KERNEL);
1348                 if (!range) {
1349                         ret = -ENOMEM;
1350                         goto out;
1351                 }
1352
1353                 if (argp) {
1354                         if (copy_from_user(range, argp,
1355                                            sizeof(*range))) {
1356                                 ret = -EFAULT;
1357                                 kfree(range);
1358                                 goto out;
1359                         }
1360                         /* compression requires us to start the IO */
1361                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1362                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
1363                                 range->extent_thresh = (u32)-1;
1364                         }
1365                 } else {
1366                         /* the rest are all set to zero by kzalloc */
1367                         range->len = (u64)-1;
1368                 }
1369                 ret = btrfs_defrag_file(file, range);
1370                 kfree(range);
1371                 break;
1372         default:
1373                 ret = -EINVAL;
1374         }
1375 out:
1376         mnt_drop_write(file->f_path.mnt);
1377         return ret;
1378 }
1379
1380 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
1381 {
1382         struct btrfs_ioctl_vol_args *vol_args;
1383         int ret;
1384
1385         if (!capable(CAP_SYS_ADMIN))
1386                 return -EPERM;
1387
1388         vol_args = memdup_user(arg, sizeof(*vol_args));
1389         if (IS_ERR(vol_args))
1390                 return PTR_ERR(vol_args);
1391
1392         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1393         ret = btrfs_init_new_device(root, vol_args->name);
1394
1395         kfree(vol_args);
1396         return ret;
1397 }
1398
1399 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
1400 {
1401         struct btrfs_ioctl_vol_args *vol_args;
1402         int ret;
1403
1404         if (!capable(CAP_SYS_ADMIN))
1405                 return -EPERM;
1406
1407         if (root->fs_info->sb->s_flags & MS_RDONLY)
1408                 return -EROFS;
1409
1410         vol_args = memdup_user(arg, sizeof(*vol_args));
1411         if (IS_ERR(vol_args))
1412                 return PTR_ERR(vol_args);
1413
1414         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1415         ret = btrfs_rm_device(root, vol_args->name);
1416
1417         kfree(vol_args);
1418         return ret;
1419 }
1420
1421 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
1422                                        u64 off, u64 olen, u64 destoff)
1423 {
1424         struct inode *inode = fdentry(file)->d_inode;
1425         struct btrfs_root *root = BTRFS_I(inode)->root;
1426         struct file *src_file;
1427         struct inode *src;
1428         struct btrfs_trans_handle *trans;
1429         struct btrfs_path *path;
1430         struct extent_buffer *leaf;
1431         char *buf;
1432         struct btrfs_key key;
1433         u32 nritems;
1434         int slot;
1435         int ret;
1436         u64 len = olen;
1437         u64 bs = root->fs_info->sb->s_blocksize;
1438         u64 hint_byte;
1439
1440         /*
1441          * TODO:
1442          * - split compressed inline extents.  annoying: we need to
1443          *   decompress into destination's address_space (the file offset
1444          *   may change, so source mapping won't do), then recompress (or
1445          *   otherwise reinsert) a subrange.
1446          * - allow ranges within the same file to be cloned (provided
1447          *   they don't overlap)?
1448          */
1449
1450         /* the destination must be opened for writing */
1451         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
1452                 return -EINVAL;
1453
1454         ret = mnt_want_write(file->f_path.mnt);
1455         if (ret)
1456                 return ret;
1457
1458         src_file = fget(srcfd);
1459         if (!src_file) {
1460                 ret = -EBADF;
1461                 goto out_drop_write;
1462         }
1463
1464         src = src_file->f_dentry->d_inode;
1465
1466         ret = -EINVAL;
1467         if (src == inode)
1468                 goto out_fput;
1469
1470         /* the src must be open for reading */
1471         if (!(src_file->f_mode & FMODE_READ))
1472                 goto out_fput;
1473
1474         ret = -EISDIR;
1475         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
1476                 goto out_fput;
1477
1478         ret = -EXDEV;
1479         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
1480                 goto out_fput;
1481
1482         ret = -ENOMEM;
1483         buf = vmalloc(btrfs_level_size(root, 0));
1484         if (!buf)
1485                 goto out_fput;
1486
1487         path = btrfs_alloc_path();
1488         if (!path) {
1489                 vfree(buf);
1490                 goto out_fput;
1491         }
1492         path->reada = 2;
1493
1494         if (inode < src) {
1495                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
1496                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
1497         } else {
1498                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
1499                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1500         }
1501
1502         /* determine range to clone */
1503         ret = -EINVAL;
1504         if (off + len > src->i_size || off + len < off)
1505                 goto out_unlock;
1506         if (len == 0)
1507                 olen = len = src->i_size - off;
1508         /* if we extend to eof, continue to block boundary */
1509         if (off + len == src->i_size)
1510                 len = ((src->i_size + bs-1) & ~(bs-1))
1511                         - off;
1512
1513         /* verify the end result is block aligned */
1514         if ((off & (bs-1)) ||
1515             ((off + len) & (bs-1)))
1516                 goto out_unlock;
1517
1518         /* do any pending delalloc/csum calc on src, one way or
1519            another, and lock file content */
1520         while (1) {
1521                 struct btrfs_ordered_extent *ordered;
1522                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1523                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
1524                 if (!ordered &&
1525                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
1526                                    EXTENT_DELALLOC, 0, NULL))
1527                         break;
1528                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1529                 if (ordered)
1530                         btrfs_put_ordered_extent(ordered);
1531                 btrfs_wait_ordered_range(src, off, len);
1532         }
1533
1534         /* clone data */
1535         key.objectid = src->i_ino;
1536         key.type = BTRFS_EXTENT_DATA_KEY;
1537         key.offset = 0;
1538
1539         while (1) {
1540                 /*
1541                  * note the key will change type as we walk through the
1542                  * tree.
1543                  */
1544                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1545                 if (ret < 0)
1546                         goto out;
1547
1548                 nritems = btrfs_header_nritems(path->nodes[0]);
1549                 if (path->slots[0] >= nritems) {
1550                         ret = btrfs_next_leaf(root, path);
1551                         if (ret < 0)
1552                                 goto out;
1553                         if (ret > 0)
1554                                 break;
1555                         nritems = btrfs_header_nritems(path->nodes[0]);
1556                 }
1557                 leaf = path->nodes[0];
1558                 slot = path->slots[0];
1559
1560                 btrfs_item_key_to_cpu(leaf, &key, slot);
1561                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1562                     key.objectid != src->i_ino)
1563                         break;
1564
1565                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1566                         struct btrfs_file_extent_item *extent;
1567                         int type;
1568                         u32 size;
1569                         struct btrfs_key new_key;
1570                         u64 disko = 0, diskl = 0;
1571                         u64 datao = 0, datal = 0;
1572                         u8 comp;
1573                         u64 endoff;
1574
1575                         size = btrfs_item_size_nr(leaf, slot);
1576                         read_extent_buffer(leaf, buf,
1577                                            btrfs_item_ptr_offset(leaf, slot),
1578                                            size);
1579
1580                         extent = btrfs_item_ptr(leaf, slot,
1581                                                 struct btrfs_file_extent_item);
1582                         comp = btrfs_file_extent_compression(leaf, extent);
1583                         type = btrfs_file_extent_type(leaf, extent);
1584                         if (type == BTRFS_FILE_EXTENT_REG ||
1585                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1586                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1587                                                                       extent);
1588                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1589                                                                  extent);
1590                                 datao = btrfs_file_extent_offset(leaf, extent);
1591                                 datal = btrfs_file_extent_num_bytes(leaf,
1592                                                                     extent);
1593                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1594                                 /* take upper bound, may be compressed */
1595                                 datal = btrfs_file_extent_ram_bytes(leaf,
1596                                                                     extent);
1597                         }
1598                         btrfs_release_path(root, path);
1599
1600                         if (key.offset + datal <= off ||
1601                             key.offset >= off+len)
1602                                 goto next;
1603
1604                         memcpy(&new_key, &key, sizeof(new_key));
1605                         new_key.objectid = inode->i_ino;
1606                         new_key.offset = key.offset + destoff - off;
1607
1608                         trans = btrfs_start_transaction(root, 1);
1609                         if (IS_ERR(trans)) {
1610                                 ret = PTR_ERR(trans);
1611                                 goto out;
1612                         }
1613
1614                         if (type == BTRFS_FILE_EXTENT_REG ||
1615                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1616                                 if (off > key.offset) {
1617                                         datao += off - key.offset;
1618                                         datal -= off - key.offset;
1619                                 }
1620
1621                                 if (key.offset + datal > off + len)
1622                                         datal = off + len - key.offset;
1623
1624                                 ret = btrfs_drop_extents(trans, inode,
1625                                                          new_key.offset,
1626                                                          new_key.offset + datal,
1627                                                          &hint_byte, 1);
1628                                 BUG_ON(ret);
1629
1630                                 ret = btrfs_insert_empty_item(trans, root, path,
1631                                                               &new_key, size);
1632                                 BUG_ON(ret);
1633
1634                                 leaf = path->nodes[0];
1635                                 slot = path->slots[0];
1636                                 write_extent_buffer(leaf, buf,
1637                                             btrfs_item_ptr_offset(leaf, slot),
1638                                             size);
1639
1640                                 extent = btrfs_item_ptr(leaf, slot,
1641                                                 struct btrfs_file_extent_item);
1642
1643                                 /* disko == 0 means it's a hole */
1644                                 if (!disko)
1645                                         datao = 0;
1646
1647                                 btrfs_set_file_extent_offset(leaf, extent,
1648                                                              datao);
1649                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1650                                                                 datal);
1651                                 if (disko) {
1652                                         inode_add_bytes(inode, datal);
1653                                         ret = btrfs_inc_extent_ref(trans, root,
1654                                                         disko, diskl, 0,
1655                                                         root->root_key.objectid,
1656                                                         inode->i_ino,
1657                                                         new_key.offset - datao);
1658                                         BUG_ON(ret);
1659                                 }
1660                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1661                                 u64 skip = 0;
1662                                 u64 trim = 0;
1663                                 if (off > key.offset) {
1664                                         skip = off - key.offset;
1665                                         new_key.offset += skip;
1666                                 }
1667
1668                                 if (key.offset + datal > off+len)
1669                                         trim = key.offset + datal - (off+len);
1670
1671                                 if (comp && (skip || trim)) {
1672                                         ret = -EINVAL;
1673                                         btrfs_end_transaction(trans, root);
1674                                         goto out;
1675                                 }
1676                                 size -= skip + trim;
1677                                 datal -= skip + trim;
1678
1679                                 ret = btrfs_drop_extents(trans, inode,
1680                                                          new_key.offset,
1681                                                          new_key.offset + datal,
1682                                                          &hint_byte, 1);
1683                                 BUG_ON(ret);
1684
1685                                 ret = btrfs_insert_empty_item(trans, root, path,
1686                                                               &new_key, size);
1687                                 BUG_ON(ret);
1688
1689                                 if (skip) {
1690                                         u32 start =
1691                                           btrfs_file_extent_calc_inline_size(0);
1692                                         memmove(buf+start, buf+start+skip,
1693                                                 datal);
1694                                 }
1695
1696                                 leaf = path->nodes[0];
1697                                 slot = path->slots[0];
1698                                 write_extent_buffer(leaf, buf,
1699                                             btrfs_item_ptr_offset(leaf, slot),
1700                                             size);
1701                                 inode_add_bytes(inode, datal);
1702                         }
1703
1704                         btrfs_mark_buffer_dirty(leaf);
1705                         btrfs_release_path(root, path);
1706
1707                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1708
1709                         /*
1710                          * we round up to the block size at eof when
1711                          * determining which extents to clone above,
1712                          * but shouldn't round up the file size
1713                          */
1714                         endoff = new_key.offset + datal;
1715                         if (endoff > off+olen)
1716                                 endoff = off+olen;
1717                         if (endoff > inode->i_size)
1718                                 btrfs_i_size_write(inode, endoff);
1719
1720                         BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1721                         ret = btrfs_update_inode(trans, root, inode);
1722                         BUG_ON(ret);
1723                         btrfs_end_transaction(trans, root);
1724                 }
1725 next:
1726                 btrfs_release_path(root, path);
1727                 key.offset++;
1728         }
1729         ret = 0;
1730 out:
1731         btrfs_release_path(root, path);
1732         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1733 out_unlock:
1734         mutex_unlock(&src->i_mutex);
1735         mutex_unlock(&inode->i_mutex);
1736         vfree(buf);
1737         btrfs_free_path(path);
1738 out_fput:
1739         fput(src_file);
1740 out_drop_write:
1741         mnt_drop_write(file->f_path.mnt);
1742         return ret;
1743 }
1744
1745 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1746 {
1747         struct btrfs_ioctl_clone_range_args args;
1748
1749         if (copy_from_user(&args, argp, sizeof(args)))
1750                 return -EFAULT;
1751         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1752                                  args.src_length, args.dest_offset);
1753 }
1754
1755 /*
1756  * there are many ways the trans_start and trans_end ioctls can lead
1757  * to deadlocks.  They should only be used by applications that
1758  * basically own the machine, and have a very in depth understanding
1759  * of all the possible deadlocks and enospc problems.
1760  */
1761 static long btrfs_ioctl_trans_start(struct file *file)
1762 {
1763         struct inode *inode = fdentry(file)->d_inode;
1764         struct btrfs_root *root = BTRFS_I(inode)->root;
1765         struct btrfs_trans_handle *trans;
1766         int ret;
1767
1768         ret = -EPERM;
1769         if (!capable(CAP_SYS_ADMIN))
1770                 goto out;
1771
1772         ret = -EINPROGRESS;
1773         if (file->private_data)
1774                 goto out;
1775
1776         ret = mnt_want_write(file->f_path.mnt);
1777         if (ret)
1778                 goto out;
1779
1780         mutex_lock(&root->fs_info->trans_mutex);
1781         root->fs_info->open_ioctl_trans++;
1782         mutex_unlock(&root->fs_info->trans_mutex);
1783
1784         ret = -ENOMEM;
1785         trans = btrfs_start_ioctl_transaction(root, 0);
1786         if (!trans)
1787                 goto out_drop;
1788
1789         file->private_data = trans;
1790         return 0;
1791
1792 out_drop:
1793         mutex_lock(&root->fs_info->trans_mutex);
1794         root->fs_info->open_ioctl_trans--;
1795         mutex_unlock(&root->fs_info->trans_mutex);
1796         mnt_drop_write(file->f_path.mnt);
1797 out:
1798         return ret;
1799 }
1800
1801 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
1802 {
1803         struct inode *inode = fdentry(file)->d_inode;
1804         struct btrfs_root *root = BTRFS_I(inode)->root;
1805         struct btrfs_root *new_root;
1806         struct btrfs_dir_item *di;
1807         struct btrfs_trans_handle *trans;
1808         struct btrfs_path *path;
1809         struct btrfs_key location;
1810         struct btrfs_disk_key disk_key;
1811         struct btrfs_super_block *disk_super;
1812         u64 features;
1813         u64 objectid = 0;
1814         u64 dir_id;
1815
1816         if (!capable(CAP_SYS_ADMIN))
1817                 return -EPERM;
1818
1819         if (copy_from_user(&objectid, argp, sizeof(objectid)))
1820                 return -EFAULT;
1821
1822         if (!objectid)
1823                 objectid = root->root_key.objectid;
1824
1825         location.objectid = objectid;
1826         location.type = BTRFS_ROOT_ITEM_KEY;
1827         location.offset = (u64)-1;
1828
1829         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
1830         if (IS_ERR(new_root))
1831                 return PTR_ERR(new_root);
1832
1833         if (btrfs_root_refs(&new_root->root_item) == 0)
1834                 return -ENOENT;
1835
1836         path = btrfs_alloc_path();
1837         if (!path)
1838                 return -ENOMEM;
1839         path->leave_spinning = 1;
1840
1841         trans = btrfs_start_transaction(root, 1);
1842         if (!trans) {
1843                 btrfs_free_path(path);
1844                 return -ENOMEM;
1845         }
1846
1847         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
1848         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
1849                                    dir_id, "default", 7, 1);
1850         if (IS_ERR_OR_NULL(di)) {
1851                 btrfs_free_path(path);
1852                 btrfs_end_transaction(trans, root);
1853                 printk(KERN_ERR "Umm, you don't have the default dir item, "
1854                        "this isn't going to work\n");
1855                 return -ENOENT;
1856         }
1857
1858         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
1859         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
1860         btrfs_mark_buffer_dirty(path->nodes[0]);
1861         btrfs_free_path(path);
1862
1863         disk_super = &root->fs_info->super_copy;
1864         features = btrfs_super_incompat_flags(disk_super);
1865         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
1866                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
1867                 btrfs_set_super_incompat_flags(disk_super, features);
1868         }
1869         btrfs_end_transaction(trans, root);
1870
1871         return 0;
1872 }
1873
1874 static void get_block_group_info(struct list_head *groups_list,
1875                                  struct btrfs_ioctl_space_info *space)
1876 {
1877         struct btrfs_block_group_cache *block_group;
1878
1879         space->total_bytes = 0;
1880         space->used_bytes = 0;
1881         space->flags = 0;
1882         list_for_each_entry(block_group, groups_list, list) {
1883                 space->flags = block_group->flags;
1884                 space->total_bytes += block_group->key.offset;
1885                 space->used_bytes +=
1886                         btrfs_block_group_used(&block_group->item);
1887         }
1888 }
1889
1890 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
1891 {
1892         struct btrfs_ioctl_space_args space_args;
1893         struct btrfs_ioctl_space_info space;
1894         struct btrfs_ioctl_space_info *dest;
1895         struct btrfs_ioctl_space_info *dest_orig;
1896         struct btrfs_ioctl_space_info *user_dest;
1897         struct btrfs_space_info *info;
1898         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
1899                        BTRFS_BLOCK_GROUP_SYSTEM,
1900                        BTRFS_BLOCK_GROUP_METADATA,
1901                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
1902         int num_types = 4;
1903         int alloc_size;
1904         int ret = 0;
1905         int slot_count = 0;
1906         int i, c;
1907
1908         if (copy_from_user(&space_args,
1909                            (struct btrfs_ioctl_space_args __user *)arg,
1910                            sizeof(space_args)))
1911                 return -EFAULT;
1912
1913         for (i = 0; i < num_types; i++) {
1914                 struct btrfs_space_info *tmp;
1915
1916                 info = NULL;
1917                 rcu_read_lock();
1918                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
1919                                         list) {
1920                         if (tmp->flags == types[i]) {
1921                                 info = tmp;
1922                                 break;
1923                         }
1924                 }
1925                 rcu_read_unlock();
1926
1927                 if (!info)
1928                         continue;
1929
1930                 down_read(&info->groups_sem);
1931                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
1932                         if (!list_empty(&info->block_groups[c]))
1933                                 slot_count++;
1934                 }
1935                 up_read(&info->groups_sem);
1936         }
1937
1938         /* space_slots == 0 means they are asking for a count */
1939         if (space_args.space_slots == 0) {
1940                 space_args.total_spaces = slot_count;
1941                 goto out;
1942         }
1943
1944         slot_count = min_t(int, space_args.space_slots, slot_count);
1945
1946         alloc_size = sizeof(*dest) * slot_count;
1947
1948         /* we generally have at most 6 or so space infos, one for each raid
1949          * level.  So, a whole page should be more than enough for everyone
1950          */
1951         if (alloc_size > PAGE_CACHE_SIZE)
1952                 return -ENOMEM;
1953
1954         space_args.total_spaces = 0;
1955         dest = kmalloc(alloc_size, GFP_NOFS);
1956         if (!dest)
1957                 return -ENOMEM;
1958         dest_orig = dest;
1959
1960         /* now we have a buffer to copy into */
1961         for (i = 0; i < num_types; i++) {
1962                 struct btrfs_space_info *tmp;
1963
1964                 info = NULL;
1965                 rcu_read_lock();
1966                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
1967                                         list) {
1968                         if (tmp->flags == types[i]) {
1969                                 info = tmp;
1970                                 break;
1971                         }
1972                 }
1973                 rcu_read_unlock();
1974
1975                 if (!info)
1976                         continue;
1977                 down_read(&info->groups_sem);
1978                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
1979                         if (!list_empty(&info->block_groups[c])) {
1980                                 get_block_group_info(&info->block_groups[c],
1981                                                      &space);
1982                                 memcpy(dest, &space, sizeof(space));
1983                                 dest++;
1984                                 space_args.total_spaces++;
1985                         }
1986                 }
1987                 up_read(&info->groups_sem);
1988         }
1989
1990         user_dest = (struct btrfs_ioctl_space_info *)
1991                 (arg + sizeof(struct btrfs_ioctl_space_args));
1992
1993         if (copy_to_user(user_dest, dest_orig, alloc_size))
1994                 ret = -EFAULT;
1995
1996         kfree(dest_orig);
1997 out:
1998         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
1999                 ret = -EFAULT;
2000
2001         return ret;
2002 }
2003
2004 /*
2005  * there are many ways the trans_start and trans_end ioctls can lead
2006  * to deadlocks.  They should only be used by applications that
2007  * basically own the machine, and have a very in depth understanding
2008  * of all the possible deadlocks and enospc problems.
2009  */
2010 long btrfs_ioctl_trans_end(struct file *file)
2011 {
2012         struct inode *inode = fdentry(file)->d_inode;
2013         struct btrfs_root *root = BTRFS_I(inode)->root;
2014         struct btrfs_trans_handle *trans;
2015
2016         trans = file->private_data;
2017         if (!trans)
2018                 return -EINVAL;
2019         file->private_data = NULL;
2020
2021         btrfs_end_transaction(trans, root);
2022
2023         mutex_lock(&root->fs_info->trans_mutex);
2024         root->fs_info->open_ioctl_trans--;
2025         mutex_unlock(&root->fs_info->trans_mutex);
2026
2027         mnt_drop_write(file->f_path.mnt);
2028         return 0;
2029 }
2030
2031 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2032 {
2033         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2034         struct btrfs_trans_handle *trans;
2035         u64 transid;
2036
2037         trans = btrfs_start_transaction(root, 0);
2038         transid = trans->transid;
2039         btrfs_commit_transaction_async(trans, root, 0);
2040
2041         if (argp)
2042                 if (copy_to_user(argp, &transid, sizeof(transid)))
2043                         return -EFAULT;
2044         return 0;
2045 }
2046
2047 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2048 {
2049         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2050         u64 transid;
2051
2052         if (argp) {
2053                 if (copy_from_user(&transid, argp, sizeof(transid)))
2054                         return -EFAULT;
2055         } else {
2056                 transid = 0;  /* current trans */
2057         }
2058         return btrfs_wait_for_commit(root, transid);
2059 }
2060
2061 long btrfs_ioctl(struct file *file, unsigned int
2062                 cmd, unsigned long arg)
2063 {
2064         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2065         void __user *argp = (void __user *)arg;
2066
2067         switch (cmd) {
2068         case FS_IOC_GETFLAGS:
2069                 return btrfs_ioctl_getflags(file, argp);
2070         case FS_IOC_SETFLAGS:
2071                 return btrfs_ioctl_setflags(file, argp);
2072         case FS_IOC_GETVERSION:
2073                 return btrfs_ioctl_getversion(file, argp);
2074         case BTRFS_IOC_SNAP_CREATE:
2075                 return btrfs_ioctl_snap_create(file, argp, 0);
2076         case BTRFS_IOC_SUBVOL_CREATE:
2077                 return btrfs_ioctl_snap_create(file, argp, 1);
2078         case BTRFS_IOC_SNAP_DESTROY:
2079                 return btrfs_ioctl_snap_destroy(file, argp);
2080         case BTRFS_IOC_DEFAULT_SUBVOL:
2081                 return btrfs_ioctl_default_subvol(file, argp);
2082         case BTRFS_IOC_DEFRAG:
2083                 return btrfs_ioctl_defrag(file, NULL);
2084         case BTRFS_IOC_DEFRAG_RANGE:
2085                 return btrfs_ioctl_defrag(file, argp);
2086         case BTRFS_IOC_RESIZE:
2087                 return btrfs_ioctl_resize(root, argp);
2088         case BTRFS_IOC_ADD_DEV:
2089                 return btrfs_ioctl_add_dev(root, argp);
2090         case BTRFS_IOC_RM_DEV:
2091                 return btrfs_ioctl_rm_dev(root, argp);
2092         case BTRFS_IOC_BALANCE:
2093                 return btrfs_balance(root->fs_info->dev_root);
2094         case BTRFS_IOC_CLONE:
2095                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
2096         case BTRFS_IOC_CLONE_RANGE:
2097                 return btrfs_ioctl_clone_range(file, argp);
2098         case BTRFS_IOC_TRANS_START:
2099                 return btrfs_ioctl_trans_start(file);
2100         case BTRFS_IOC_TRANS_END:
2101                 return btrfs_ioctl_trans_end(file);
2102         case BTRFS_IOC_TREE_SEARCH:
2103                 return btrfs_ioctl_tree_search(file, argp);
2104         case BTRFS_IOC_INO_LOOKUP:
2105                 return btrfs_ioctl_ino_lookup(file, argp);
2106         case BTRFS_IOC_SPACE_INFO:
2107                 return btrfs_ioctl_space_info(root, argp);
2108         case BTRFS_IOC_SYNC:
2109                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
2110                 return 0;
2111         case BTRFS_IOC_START_SYNC:
2112                 return btrfs_ioctl_start_sync(file, argp);
2113         case BTRFS_IOC_WAIT_SYNC:
2114                 return btrfs_ioctl_wait_sync(file, argp);
2115         }
2116
2117         return -ENOTTY;
2118 }