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 (btrfs_root_readonly(root))
 151                 return -EROFS;
 152
 153         if (copy_from_user(&flags, arg, sizeof(flags)))
 154                 return -EFAULT;
 155
 156         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 157                       FS_NOATIME_FL | FS_NODUMP_FL | \
 158                       FS_SYNC_FL | FS_DIRSYNC_FL))
 159                 return -EOPNOTSUPP;
 160
 161         if (!is_owner_or_cap(inode))
 162                 return -EACCES;
 163
 164         mutex_lock(&inode->i_mutex);
 165
 166         flags = btrfs_mask_flags(inode->i_mode, flags);
 167         oldflags = btrfs_flags_to_ioctl(ip->flags);
 168         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 169                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 170                         ret = -EPERM;
 171                         goto out_unlock;
 172                 }
 173         }
 174
 175         ret = mnt_want_write(file->f_path.mnt);
 176         if (ret)
 177                 goto out_unlock;
 178
 179         if (flags & FS_SYNC_FL)
 180                 ip->flags |= BTRFS_INODE_SYNC;
 181         else
 182                 ip->flags &= ~BTRFS_INODE_SYNC;
 183         if (flags & FS_IMMUTABLE_FL)
 184                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 185         else
 186                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 187         if (flags & FS_APPEND_FL)
 188                 ip->flags |= BTRFS_INODE_APPEND;
 189         else
 190                 ip->flags &= ~BTRFS_INODE_APPEND;
 191         if (flags & FS_NODUMP_FL)
 192                 ip->flags |= BTRFS_INODE_NODUMP;
 193         else
 194                 ip->flags &= ~BTRFS_INODE_NODUMP;
 195         if (flags & FS_NOATIME_FL)
 196                 ip->flags |= BTRFS_INODE_NOATIME;
 197         else
 198                 ip->flags &= ~BTRFS_INODE_NOATIME;
 199         if (flags & FS_DIRSYNC_FL)
 200                 ip->flags |= BTRFS_INODE_DIRSYNC;
 201         else
 202                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 203
 204
 205         trans = btrfs_join_transaction(root, 1);
 206         BUG_ON(!trans);
 207
 208         ret = btrfs_update_inode(trans, root, inode);
 209         BUG_ON(ret);
 210
 211         btrfs_update_iflags(inode);
 212         inode->i_ctime = CURRENT_TIME;
 213         btrfs_end_transaction(trans, root);
 214
 215         mnt_drop_write(file->f_path.mnt);
 216  out_unlock:
 217         mutex_unlock(&inode->i_mutex);
 218         return 0;
 219 }
 220
 221 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 222 {
 223         struct inode *inode = file->f_path.dentry->d_inode;
 224
 225         return put_user(inode->i_generation, arg);
 226 }
 227
 228 static noinline int create_subvol(struct btrfs_root *root,
 229                                   struct dentry *dentry,
 230                                   char *name, int namelen,
 231                                   u64 *async_transid)
 232 {
 233         struct btrfs_trans_handle *trans;
 234         struct btrfs_key key;
 235         struct btrfs_root_item root_item;
 236         struct btrfs_inode_item *inode_item;
 237         struct extent_buffer *leaf;
 238         struct btrfs_root *new_root;
 239         struct dentry *parent = dget_parent(dentry);
 240         struct inode *dir;
 241         int ret;
 242         int err;
 243         u64 objectid;
 244         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 245         u64 index = 0;
 246
 247         ret = btrfs_find_free_objectid(NULL, root->fs_info->tree_root,
 248                                        0, &objectid);
 249         if (ret) {
 250                 dput(parent);
 251                 return ret;
 252         }
 253
 254         dir = parent->d_inode;
 255
 256         /*
 257          * 1 - inode item
 258          * 2 - refs
 259          * 1 - root item
 260          * 2 - dir items
 261          */
 262         trans = btrfs_start_transaction(root, 6);
 263         if (IS_ERR(trans)) {
 264                 dput(parent);
 265                 return PTR_ERR(trans);
 266         }
 267
 268         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 269                                       0, objectid, NULL, 0, 0, 0);
 270         if (IS_ERR(leaf)) {
 271                 ret = PTR_ERR(leaf);
 272                 goto fail;
 273         }
 274
 275         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 276         btrfs_set_header_bytenr(leaf, leaf->start);
 277         btrfs_set_header_generation(leaf, trans->transid);
 278         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 279         btrfs_set_header_owner(leaf, objectid);
 280
 281         write_extent_buffer(leaf, root->fs_info->fsid,
 282                             (unsigned long)btrfs_header_fsid(leaf),
 283                             BTRFS_FSID_SIZE);
 284         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 285                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 286                             BTRFS_UUID_SIZE);
 287         btrfs_mark_buffer_dirty(leaf);
 288
 289         inode_item = &root_item.inode;
 290         memset(inode_item, 0, sizeof(*inode_item));
 291         inode_item->generation = cpu_to_le64(1);
 292         inode_item->size = cpu_to_le64(3);
 293         inode_item->nlink = cpu_to_le32(1);
 294         inode_item->nbytes = cpu_to_le64(root->leafsize);
 295         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 296
 297         btrfs_set_root_bytenr(&root_item, leaf->start);
 298         btrfs_set_root_generation(&root_item, trans->transid);
 299         btrfs_set_root_level(&root_item, 0);
 300         btrfs_set_root_refs(&root_item, 1);
 301         btrfs_set_root_used(&root_item, leaf->len);
 302         btrfs_set_root_last_snapshot(&root_item, 0);
 303
 304         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 305         root_item.drop_level = 0;
 306
 307         btrfs_tree_unlock(leaf);
 308         free_extent_buffer(leaf);
 309         leaf = NULL;
 310
 311         btrfs_set_root_dirid(&root_item, new_dirid);
 312
 313         key.objectid = objectid;
 314         key.offset = 0;
 315         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 316         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 317                                 &root_item);
 318         if (ret)
 319                 goto fail;
 320
 321         key.offset = (u64)-1;
 322         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 323         BUG_ON(IS_ERR(new_root));
 324
 325         btrfs_record_root_in_trans(trans, new_root);
 326
 327         ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
 328                                        BTRFS_I(dir)->block_group);
 329         /*
 330          * insert the directory item
 331          */
 332         ret = btrfs_set_inode_index(dir, &index);
 333         BUG_ON(ret);
 334
 335         ret = btrfs_insert_dir_item(trans, root,
 336                                     name, namelen, dir->i_ino, &key,
 337                                     BTRFS_FT_DIR, index);
 338         if (ret)
 339                 goto fail;
 340
 341         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 342         ret = btrfs_update_inode(trans, root, dir);
 343         BUG_ON(ret);
 344
 345         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 346                                  objectid, root->root_key.objectid,
 347                                  dir->i_ino, index, name, namelen);
 348
 349         BUG_ON(ret);
 350
 351         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 352 fail:
 353         dput(parent);
 354         if (async_transid) {
 355                 *async_transid = trans->transid;
 356                 err = btrfs_commit_transaction_async(trans, root, 1);
 357         } else {
 358                 err = btrfs_commit_transaction(trans, root);
 359         }
 360         if (err && !ret)
 361                 ret = err;
 362         return ret;
 363 }
 364
 365 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 366                            char *name, int namelen, u64 *async_transid,
 367                            bool readonly)
 368 {
 369         struct inode *inode;
 370         struct dentry *parent;
 371         struct btrfs_pending_snapshot *pending_snapshot;
 372         struct btrfs_trans_handle *trans;
 373         int ret;
 374
 375         if (!root->ref_cows)
 376                 return -EINVAL;
 377
 378         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 379         if (!pending_snapshot)
 380                 return -ENOMEM;
 381
 382         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
 383         pending_snapshot->dentry = dentry;
 384         pending_snapshot->root = root;
 385         pending_snapshot->readonly = readonly;
 386
 387         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
 388         if (IS_ERR(trans)) {
 389                 ret = PTR_ERR(trans);
 390                 goto fail;
 391         }
 392
 393         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 394         BUG_ON(ret);
 395
 396         list_add(&pending_snapshot->list,
 397                  &trans->transaction->pending_snapshots);
 398         if (async_transid) {
 399                 *async_transid = trans->transid;
 400                 ret = btrfs_commit_transaction_async(trans,
 401                                      root->fs_info->extent_root, 1);
 402         } else {
 403                 ret = btrfs_commit_transaction(trans,
 404                                                root->fs_info->extent_root);
 405         }
 406         BUG_ON(ret);
 407
 408         ret = pending_snapshot->error;
 409         if (ret)
 410                 goto fail;
 411
 412         btrfs_orphan_cleanup(pending_snapshot->snap);
 413
 414         parent = dget_parent(dentry);
 415         inode = btrfs_lookup_dentry(parent->d_inode, dentry);
 416         dput(parent);
 417         if (IS_ERR(inode)) {
 418                 ret = PTR_ERR(inode);
 419                 goto fail;
 420         }
 421         BUG_ON(!inode);
 422         d_instantiate(dentry, inode);
 423         ret = 0;
 424 fail:
 425         kfree(pending_snapshot);
 426         return ret;
 427 }
 428
 429 /*  copy of check_sticky in fs/namei.c()
 430 * It's inline, so penalty for filesystems that don't use sticky bit is
 431 * minimal.
 432 */
 433 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 434 {
 435         uid_t fsuid = current_fsuid();
 436
 437         if (!(dir->i_mode & S_ISVTX))
 438                 return 0;
 439         if (inode->i_uid == fsuid)
 440                 return 0;
 441         if (dir->i_uid == fsuid)
 442                 return 0;
 443         return !capable(CAP_FOWNER);
 444 }
 445
 446 /*  copy of may_delete in fs/namei.c()
 447  *      Check whether we can remove a link victim from directory dir, check
 448  *  whether the type of victim is right.
 449  *  1. We can't do it if dir is read-only (done in permission())
 450  *  2. We should have write and exec permissions on dir
 451  *  3. We can't remove anything from append-only dir
 452  *  4. We can't do anything with immutable dir (done in permission())
 453  *  5. If the sticky bit on dir is set we should either
 454  *      a. be owner of dir, or
 455  *      b. be owner of victim, or
 456  *      c. have CAP_FOWNER capability
 457  *  6. If the victim is append-only or immutable we can't do antyhing with
 458  *     links pointing to it.
 459  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 460  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 461  *  9. We can't remove a root or mountpoint.
 462  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 463  *     nfs_async_unlink().
 464  */
 465
 466 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 467 {
 468         int error;
 469
 470         if (!victim->d_inode)
 471                 return -ENOENT;
 472
 473         BUG_ON(victim->d_parent->d_inode != dir);
 474         audit_inode_child(victim, dir);
 475
 476         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 477         if (error)
 478                 return error;
 479         if (IS_APPEND(dir))
 480                 return -EPERM;
 481         if (btrfs_check_sticky(dir, victim->d_inode)||
 482                 IS_APPEND(victim->d_inode)||
 483             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 484                 return -EPERM;
 485         if (isdir) {
 486                 if (!S_ISDIR(victim->d_inode->i_mode))
 487                         return -ENOTDIR;
 488                 if (IS_ROOT(victim))
 489                         return -EBUSY;
 490         } else if (S_ISDIR(victim->d_inode->i_mode))
 491                 return -EISDIR;
 492         if (IS_DEADDIR(dir))
 493                 return -ENOENT;
 494         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 495                 return -EBUSY;
 496         return 0;
 497 }
 498
 499 /* copy of may_create in fs/namei.c() */
 500 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 501 {
 502         if (child->d_inode)
 503                 return -EEXIST;
 504         if (IS_DEADDIR(dir))
 505                 return -ENOENT;
 506         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 507 }
 508
 509 /*
 510  * Create a new subvolume below @parent.  This is largely modeled after
 511  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 512  * inside this filesystem so it's quite a bit simpler.
 513  */
 514 static noinline int btrfs_mksubvol(struct path *parent,
 515                                    char *name, int namelen,
 516                                    struct btrfs_root *snap_src,
 517                                    u64 *async_transid, bool readonly)
 518 {
 519         struct inode *dir  = parent->dentry->d_inode;
 520         struct dentry *dentry;
 521         int error;
 522
 523         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 524
 525         dentry = lookup_one_len(name, parent->dentry, namelen);
 526         error = PTR_ERR(dentry);
 527         if (IS_ERR(dentry))
 528                 goto out_unlock;
 529
 530         error = -EEXIST;
 531         if (dentry->d_inode)
 532                 goto out_dput;
 533
 534         error = mnt_want_write(parent->mnt);
 535         if (error)
 536                 goto out_dput;
 537
 538         error = btrfs_may_create(dir, dentry);
 539         if (error)
 540                 goto out_drop_write;
 541
 542         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 543
 544         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 545                 goto out_up_read;
 546
 547         if (snap_src) {
 548                 error = create_snapshot(snap_src, dentry,
 549                                         name, namelen, async_transid, readonly);
 550         } else {
 551                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 552                                       name, namelen, async_transid);
 553         }
 554         if (!error)
 555                 fsnotify_mkdir(dir, dentry);
 556 out_up_read:
 557         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 558 out_drop_write:
 559         mnt_drop_write(parent->mnt);
 560 out_dput:
 561         dput(dentry);
 562 out_unlock:
 563         mutex_unlock(&dir->i_mutex);
 564         return error;
 565 }
 566
 567 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
 568                                int thresh, u64 *last_len, u64 *skip,
 569                                u64 *defrag_end)
 570 {
 571         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 572         struct extent_map *em = NULL;
 573         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 574         int ret = 1;
 575
 576
 577         if (thresh == 0)
 578                 thresh = 256 * 1024;
 579
 580         /*
 581          * make sure that once we start defragging and extent, we keep on
 582          * defragging it
 583          */
 584         if (start < *defrag_end)
 585                 return 1;
 586
 587         *skip = 0;
 588
 589         /*
 590          * hopefully we have this extent in the tree already, try without
 591          * the full extent lock
 592          */
 593         read_lock(&em_tree->lock);
 594         em = lookup_extent_mapping(em_tree, start, len);
 595         read_unlock(&em_tree->lock);
 596
 597         if (!em) {
 598                 /* get the big lock and read metadata off disk */
 599                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 600                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 601                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 602
 603                 if (IS_ERR(em))
 604                         return 0;
 605         }
 606
 607         /* this will cover holes, and inline extents */
 608         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
 609                 ret = 0;
 610
 611         /*
 612          * we hit a real extent, if it is big don't bother defragging it again
 613          */
 614         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
 615                 ret = 0;
 616
 617         /*
 618          * last_len ends up being a counter of how many bytes we've defragged.
 619          * every time we choose not to defrag an extent, we reset *last_len
 620          * so that the next tiny extent will force a defrag.
 621          *
 622          * The end result of this is that tiny extents before a single big
 623          * extent will force at least part of that big extent to be defragged.
 624          */
 625         if (ret) {
 626                 *last_len += len;
 627                 *defrag_end = extent_map_end(em);
 628         } else {
 629                 *last_len = 0;
 630                 *skip = extent_map_end(em);
 631                 *defrag_end = 0;
 632         }
 633
 634         free_extent_map(em);
 635         return ret;
 636 }
 637
 638 static int btrfs_defrag_file(struct file *file,
 639                              struct btrfs_ioctl_defrag_range_args *range)
 640 {
 641         struct inode *inode = fdentry(file)->d_inode;
 642         struct btrfs_root *root = BTRFS_I(inode)->root;
 643         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 644         struct btrfs_ordered_extent *ordered;
 645         struct page *page;
 646         struct btrfs_super_block *disk_super;
 647         unsigned long last_index;
 648         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 649         unsigned long total_read = 0;
 650         u64 features;
 651         u64 page_start;
 652         u64 page_end;
 653         u64 last_len = 0;
 654         u64 skip = 0;
 655         u64 defrag_end = 0;
 656         unsigned long i;
 657         int ret;
 658         int compress_type = BTRFS_COMPRESS_ZLIB;
 659
 660         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
 661                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
 662                         return -EINVAL;
 663                 if (range->compress_type)
 664                         compress_type = range->compress_type;
 665         }
 666
 667         if (inode->i_size == 0)
 668                 return 0;
 669
 670         if (range->start + range->len > range->start) {
 671                 last_index = min_t(u64, inode->i_size - 1,
 672                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
 673         } else {
 674                 last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
 675         }
 676
 677         i = range->start >> PAGE_CACHE_SHIFT;
 678         while (i <= last_index) {
 679                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
 680                                         PAGE_CACHE_SIZE,
 681                                         range->extent_thresh,
 682                                         &last_len, &skip,
 683                                         &defrag_end)) {
 684                         unsigned long next;
 685                         /*
 686                          * the should_defrag function tells us how much to skip
 687                          * bump our counter by the suggested amount
 688                          */
 689                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 690                         i = max(i + 1, next);
 691                         continue;
 692                 }
 693
 694                 if (total_read % ra_pages == 0) {
 695                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 696                                        min(last_index, i + ra_pages - 1));
 697                 }
 698                 total_read++;
 699                 mutex_lock(&inode->i_mutex);
 700                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
 701                         BTRFS_I(inode)->force_compress = compress_type;
 702
 703                 ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
 704                 if (ret)
 705                         goto err_unlock;
 706 again:
 707                 if (inode->i_size == 0 ||
 708                     i > ((inode->i_size - 1) >> PAGE_CACHE_SHIFT)) {
 709                         ret = 0;
 710                         goto err_reservations;
 711                 }
 712
 713                 page = grab_cache_page(inode->i_mapping, i);
 714                 if (!page) {
 715                         ret = -ENOMEM;
 716                         goto err_reservations;
 717                 }
 718
 719                 if (!PageUptodate(page)) {
 720                         btrfs_readpage(NULL, page);
 721                         lock_page(page);
 722                         if (!PageUptodate(page)) {
 723                                 unlock_page(page);
 724                                 page_cache_release(page);
 725                                 ret = -EIO;
 726                                 goto err_reservations;
 727                         }
 728                 }
 729
 730                 if (page->mapping != inode->i_mapping) {
 731                         unlock_page(page);
 732                         page_cache_release(page);
 733                         goto again;
 734                 }
 735
 736                 wait_on_page_writeback(page);
 737
 738                 if (PageDirty(page)) {
 739                         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 740                         goto loop_unlock;
 741                 }
 742
 743                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 744                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 745                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 746
 747                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 748                 if (ordered) {
 749                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 750                         unlock_page(page);
 751                         page_cache_release(page);
 752                         btrfs_start_ordered_extent(inode, ordered, 1);
 753                         btrfs_put_ordered_extent(ordered);
 754                         goto again;
 755                 }
 756                 set_page_extent_mapped(page);
 757
 758                 /*
 759                  * this makes sure page_mkwrite is called on the
 760                  * page if it is dirtied again later
 761                  */
 762                 clear_page_dirty_for_io(page);
 763                 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start,
 764                                   page_end, EXTENT_DIRTY | EXTENT_DELALLOC |
 765                                   EXTENT_DO_ACCOUNTING, GFP_NOFS);
 766
 767                 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
 768                 ClearPageChecked(page);
 769                 set_page_dirty(page);
 770                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 771
 772 loop_unlock:
 773                 unlock_page(page);
 774                 page_cache_release(page);
 775                 mutex_unlock(&inode->i_mutex);
 776
 777                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 778                 i++;
 779         }
 780
 781         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
 782                 filemap_flush(inode->i_mapping);
 783
 784         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
 785                 /* the filemap_flush will queue IO into the worker threads, but
 786                  * we have to make sure the IO is actually started and that
 787                  * ordered extents get created before we return
 788                  */
 789                 atomic_inc(&root->fs_info->async_submit_draining);
 790                 while (atomic_read(&root->fs_info->nr_async_submits) ||
 791                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 792                         wait_event(root->fs_info->async_submit_wait,
 793                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
 794                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
 795                 }
 796                 atomic_dec(&root->fs_info->async_submit_draining);
 797
 798                 mutex_lock(&inode->i_mutex);
 799                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
 800                 mutex_unlock(&inode->i_mutex);
 801         }
 802
 803         disk_super = &root->fs_info->super_copy;
 804         features = btrfs_super_incompat_flags(disk_super);
 805         if (range->compress_type == BTRFS_COMPRESS_LZO) {
 806                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
 807                 btrfs_set_super_incompat_flags(disk_super, features);
 808         }
 809
 810         return 0;
 811
 812 err_reservations:
 813         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 814 err_unlock:
 815         mutex_unlock(&inode->i_mutex);
 816         return ret;
 817 }
 818
 819 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
 820                                         void __user *arg)
 821 {
 822         u64 new_size;
 823         u64 old_size;
 824         u64 devid = 1;
 825         struct btrfs_ioctl_vol_args *vol_args;
 826         struct btrfs_trans_handle *trans;
 827         struct btrfs_device *device = NULL;
 828         char *sizestr;
 829         char *devstr = NULL;
 830         int ret = 0;
 831         int mod = 0;
 832
 833         if (root->fs_info->sb->s_flags & MS_RDONLY)
 834                 return -EROFS;
 835
 836         if (!capable(CAP_SYS_ADMIN))
 837                 return -EPERM;
 838
 839         vol_args = memdup_user(arg, sizeof(*vol_args));
 840         if (IS_ERR(vol_args))
 841                 return PTR_ERR(vol_args);
 842
 843         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 844
 845         mutex_lock(&root->fs_info->volume_mutex);
 846         sizestr = vol_args->name;
 847         devstr = strchr(sizestr, ':');
 848         if (devstr) {
 849                 char *end;
 850                 sizestr = devstr + 1;
 851                 *devstr = '\0';
 852                 devstr = vol_args->name;
 853                 devid = simple_strtoull(devstr, &end, 10);
 854                 printk(KERN_INFO "resizing devid %llu\n",
 855                        (unsigned long long)devid);
 856         }
 857         device = btrfs_find_device(root, devid, NULL, NULL);
 858         if (!device) {
 859                 printk(KERN_INFO "resizer unable to find device %llu\n",
 860                        (unsigned long long)devid);
 861                 ret = -EINVAL;
 862                 goto out_unlock;
 863         }
 864         if (!strcmp(sizestr, "max"))
 865                 new_size = device->bdev->bd_inode->i_size;
 866         else {
 867                 if (sizestr[0] == '-') {
 868                         mod = -1;
 869                         sizestr++;
 870                 } else if (sizestr[0] == '+') {
 871                         mod = 1;
 872                         sizestr++;
 873                 }
 874                 new_size = memparse(sizestr, NULL);
 875                 if (new_size == 0) {
 876                         ret = -EINVAL;
 877                         goto out_unlock;
 878                 }
 879         }
 880
 881         old_size = device->total_bytes;
 882
 883         if (mod < 0) {
 884                 if (new_size > old_size) {
 885                         ret = -EINVAL;
 886                         goto out_unlock;
 887                 }
 888                 new_size = old_size - new_size;
 889         } else if (mod > 0) {
 890                 new_size = old_size + new_size;
 891         }
 892
 893         if (new_size < 256 * 1024 * 1024) {
 894                 ret = -EINVAL;
 895                 goto out_unlock;
 896         }
 897         if (new_size > device->bdev->bd_inode->i_size) {
 898                 ret = -EFBIG;
 899                 goto out_unlock;
 900         }
 901
 902         do_div(new_size, root->sectorsize);
 903         new_size *= root->sectorsize;
 904
 905         printk(KERN_INFO "new size for %s is %llu\n",
 906                 device->name, (unsigned long long)new_size);
 907
 908         if (new_size > old_size) {
 909                 trans = btrfs_start_transaction(root, 0);
 910                 ret = btrfs_grow_device(trans, device, new_size);
 911                 btrfs_commit_transaction(trans, root);
 912         } else {
 913                 ret = btrfs_shrink_device(device, new_size);
 914         }
 915
 916 out_unlock:
 917         mutex_unlock(&root->fs_info->volume_mutex);
 918         kfree(vol_args);
 919         return ret;
 920 }
 921
 922 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
 923                                                     char *name,
 924                                                     unsigned long fd,
 925                                                     int subvol,
 926                                                     u64 *transid,
 927                                                     bool readonly)
 928 {
 929         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 930         struct file *src_file;
 931         int namelen;
 932         int ret = 0;
 933
 934         if (root->fs_info->sb->s_flags & MS_RDONLY)
 935                 return -EROFS;
 936
 937         namelen = strlen(name);
 938         if (strchr(name, '/')) {
 939                 ret = -EINVAL;
 940                 goto out;
 941         }
 942
 943         if (subvol) {
 944                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
 945                                      NULL, transid, readonly);
 946         } else {
 947                 struct inode *src_inode;
 948                 src_file = fget(fd);
 949                 if (!src_file) {
 950                         ret = -EINVAL;
 951                         goto out;
 952                 }
 953
 954                 src_inode = src_file->f_path.dentry->d_inode;
 955                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 956                         printk(KERN_INFO "btrfs: Snapshot src from "
 957                                "another FS\n");
 958                         ret = -EINVAL;
 959                         fput(src_file);
 960                         goto out;
 961                 }
 962                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
 963                                      BTRFS_I(src_inode)->root,
 964                                      transid, readonly);
 965                 fput(src_file);
 966         }
 967 out:
 968         return ret;
 969 }
 970
 971 static noinline int btrfs_ioctl_snap_create(struct file *file,
 972                                             void __user *arg, int subvol)
 973 {
 974         struct btrfs_ioctl_vol_args *vol_args;
 975         int ret;
 976
 977         vol_args = memdup_user(arg, sizeof(*vol_args));
 978         if (IS_ERR(vol_args))
 979                 return PTR_ERR(vol_args);
 980         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 981
 982         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
 983                                               vol_args->fd, subvol,
 984                                               NULL, false);
 985
 986         kfree(vol_args);
 987         return ret;
 988 }
 989
 990 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
 991                                                void __user *arg, int subvol)
 992 {
 993         struct btrfs_ioctl_vol_args_v2 *vol_args;
 994         int ret;
 995         u64 transid = 0;
 996         u64 *ptr = NULL;
 997         bool readonly = false;
 998
 999         vol_args = memdup_user(arg, sizeof(*vol_args));
1000         if (IS_ERR(vol_args))
1001                 return PTR_ERR(vol_args);
1002         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1003
1004         if (vol_args->flags &
1005             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1006                 ret = -EOPNOTSUPP;
1007                 goto out;
1008         }
1009
1010         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1011                 ptr = &transid;
1012         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1013                 readonly = true;
1014
1015         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1016                                               vol_args->fd, subvol,
1017                                               ptr, readonly);
1018
1019         if (ret == 0 && ptr &&
1020             copy_to_user(arg +
1021                          offsetof(struct btrfs_ioctl_vol_args_v2,
1022                                   transid), ptr, sizeof(*ptr)))
1023                 ret = -EFAULT;
1024 out:
1025         kfree(vol_args);
1026         return ret;
1027 }
1028
1029 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1030                                                 void __user *arg)
1031 {
1032         struct inode *inode = fdentry(file)->d_inode;
1033         struct btrfs_root *root = BTRFS_I(inode)->root;
1034         int ret = 0;
1035         u64 flags = 0;
1036
1037         if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID)
1038                 return -EINVAL;
1039
1040         down_read(&root->fs_info->subvol_sem);
1041         if (btrfs_root_readonly(root))
1042                 flags |= BTRFS_SUBVOL_RDONLY;
1043         up_read(&root->fs_info->subvol_sem);
1044
1045         if (copy_to_user(arg, &flags, sizeof(flags)))
1046                 ret = -EFAULT;
1047
1048         return ret;
1049 }
1050
1051 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1052                                               void __user *arg)
1053 {
1054         struct inode *inode = fdentry(file)->d_inode;
1055         struct btrfs_root *root = BTRFS_I(inode)->root;
1056         struct btrfs_trans_handle *trans;
1057         u64 root_flags;
1058         u64 flags;
1059         int ret = 0;
1060
1061         if (root->fs_info->sb->s_flags & MS_RDONLY)
1062                 return -EROFS;
1063
1064         if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID)
1065                 return -EINVAL;
1066
1067         if (copy_from_user(&flags, arg, sizeof(flags)))
1068                 return -EFAULT;
1069
1070         if (flags & ~BTRFS_SUBVOL_CREATE_ASYNC)
1071                 return -EINVAL;
1072
1073         if (flags & ~BTRFS_SUBVOL_RDONLY)
1074                 return -EOPNOTSUPP;
1075
1076         down_write(&root->fs_info->subvol_sem);
1077
1078         /* nothing to do */
1079         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1080                 goto out;
1081
1082         root_flags = btrfs_root_flags(&root->root_item);
1083         if (flags & BTRFS_SUBVOL_RDONLY)
1084                 btrfs_set_root_flags(&root->root_item,
1085                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1086         else
1087                 btrfs_set_root_flags(&root->root_item,
1088                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1089
1090         trans = btrfs_start_transaction(root, 1);
1091         if (IS_ERR(trans)) {
1092                 ret = PTR_ERR(trans);
1093                 goto out_reset;
1094         }
1095
1096         ret = btrfs_update_root(trans, root,
1097                                 &root->root_key, &root->root_item);
1098
1099         btrfs_commit_transaction(trans, root);
1100 out_reset:
1101         if (ret)
1102                 btrfs_set_root_flags(&root->root_item, root_flags);
1103 out:
1104         up_write(&root->fs_info->subvol_sem);
1105         return ret;
1106 }
1107
1108 /*
1109  * helper to check if the subvolume references other subvolumes
1110  */
1111 static noinline int may_destroy_subvol(struct btrfs_root *root)
1112 {
1113         struct btrfs_path *path;
1114         struct btrfs_key key;
1115         int ret;
1116
1117         path = btrfs_alloc_path();
1118         if (!path)
1119                 return -ENOMEM;
1120
1121         key.objectid = root->root_key.objectid;
1122         key.type = BTRFS_ROOT_REF_KEY;
1123         key.offset = (u64)-1;
1124
1125         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1126                                 &key, path, 0, 0);
1127         if (ret < 0)
1128                 goto out;
1129         BUG_ON(ret == 0);
1130
1131         ret = 0;
1132         if (path->slots[0] > 0) {
1133                 path->slots[0]--;
1134                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1135                 if (key.objectid == root->root_key.objectid &&
1136                     key.type == BTRFS_ROOT_REF_KEY)
1137                         ret = -ENOTEMPTY;
1138         }
1139 out:
1140         btrfs_free_path(path);
1141         return ret;
1142 }
1143
1144 static noinline int key_in_sk(struct btrfs_key *key,
1145                               struct btrfs_ioctl_search_key *sk)
1146 {
1147         struct btrfs_key test;
1148         int ret;
1149
1150         test.objectid = sk->min_objectid;
1151         test.type = sk->min_type;
1152         test.offset = sk->min_offset;
1153
1154         ret = btrfs_comp_cpu_keys(key, &test);
1155         if (ret < 0)
1156                 return 0;
1157
1158         test.objectid = sk->max_objectid;
1159         test.type = sk->max_type;
1160         test.offset = sk->max_offset;
1161
1162         ret = btrfs_comp_cpu_keys(key, &test);
1163         if (ret > 0)
1164                 return 0;
1165         return 1;
1166 }
1167
1168 static noinline int copy_to_sk(struct btrfs_root *root,
1169                                struct btrfs_path *path,
1170                                struct btrfs_key *key,
1171                                struct btrfs_ioctl_search_key *sk,
1172                                char *buf,
1173                                unsigned long *sk_offset,
1174                                int *num_found)
1175 {
1176         u64 found_transid;
1177         struct extent_buffer *leaf;
1178         struct btrfs_ioctl_search_header sh;
1179         unsigned long item_off;
1180         unsigned long item_len;
1181         int nritems;
1182         int i;
1183         int slot;
1184         int found = 0;
1185         int ret = 0;
1186
1187         leaf = path->nodes[0];
1188         slot = path->slots[0];
1189         nritems = btrfs_header_nritems(leaf);
1190
1191         if (btrfs_header_generation(leaf) > sk->max_transid) {
1192                 i = nritems;
1193                 goto advance_key;
1194         }
1195         found_transid = btrfs_header_generation(leaf);
1196
1197         for (i = slot; i < nritems; i++) {
1198                 item_off = btrfs_item_ptr_offset(leaf, i);
1199                 item_len = btrfs_item_size_nr(leaf, i);
1200
1201                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1202                         item_len = 0;
1203
1204                 if (sizeof(sh) + item_len + *sk_offset >
1205                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1206                         ret = 1;
1207                         goto overflow;
1208                 }
1209
1210                 btrfs_item_key_to_cpu(leaf, key, i);
1211                 if (!key_in_sk(key, sk))
1212                         continue;
1213
1214                 sh.objectid = key->objectid;
1215                 sh.offset = key->offset;
1216                 sh.type = key->type;
1217                 sh.len = item_len;
1218                 sh.transid = found_transid;
1219
1220                 /* copy search result header */
1221                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1222                 *sk_offset += sizeof(sh);
1223
1224                 if (item_len) {
1225                         char *p = buf + *sk_offset;
1226                         /* copy the item */
1227                         read_extent_buffer(leaf, p,
1228                                            item_off, item_len);
1229                         *sk_offset += item_len;
1230                 }
1231                 found++;
1232
1233                 if (*num_found >= sk->nr_items)
1234                         break;
1235         }
1236 advance_key:
1237         ret = 0;
1238         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1239                 key->offset++;
1240         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1241                 key->offset = 0;
1242                 key->type++;
1243         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1244                 key->offset = 0;
1245                 key->type = 0;
1246                 key->objectid++;
1247         } else
1248                 ret = 1;
1249 overflow:
1250         *num_found += found;
1251         return ret;
1252 }
1253
1254 static noinline int search_ioctl(struct inode *inode,
1255                                  struct btrfs_ioctl_search_args *args)
1256 {
1257         struct btrfs_root *root;
1258         struct btrfs_key key;
1259         struct btrfs_key max_key;
1260         struct btrfs_path *path;
1261         struct btrfs_ioctl_search_key *sk = &args->key;
1262         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1263         int ret;
1264         int num_found = 0;
1265         unsigned long sk_offset = 0;
1266
1267         path = btrfs_alloc_path();
1268         if (!path)
1269                 return -ENOMEM;
1270
1271         if (sk->tree_id == 0) {
1272                 /* search the root of the inode that was passed */
1273                 root = BTRFS_I(inode)->root;
1274         } else {
1275                 key.objectid = sk->tree_id;
1276                 key.type = BTRFS_ROOT_ITEM_KEY;
1277                 key.offset = (u64)-1;
1278                 root = btrfs_read_fs_root_no_name(info, &key);
1279                 if (IS_ERR(root)) {
1280                         printk(KERN_ERR "could not find root %llu\n",
1281                                sk->tree_id);
1282                         btrfs_free_path(path);
1283                         return -ENOENT;
1284                 }
1285         }
1286
1287         key.objectid = sk->min_objectid;
1288         key.type = sk->min_type;
1289         key.offset = sk->min_offset;
1290
1291         max_key.objectid = sk->max_objectid;
1292         max_key.type = sk->max_type;
1293         max_key.offset = sk->max_offset;
1294
1295         path->keep_locks = 1;
1296
1297         while(1) {
1298                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1299                                            sk->min_transid);
1300                 if (ret != 0) {
1301                         if (ret > 0)
1302                                 ret = 0;
1303                         goto err;
1304                 }
1305                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1306                                  &sk_offset, &num_found);
1307                 btrfs_release_path(root, path);
1308                 if (ret || num_found >= sk->nr_items)
1309                         break;
1310
1311         }
1312         ret = 0;
1313 err:
1314         sk->nr_items = num_found;
1315         btrfs_free_path(path);
1316         return ret;
1317 }
1318
1319 static noinline int btrfs_ioctl_tree_search(struct file *file,
1320                                            void __user *argp)
1321 {
1322          struct btrfs_ioctl_search_args *args;
1323          struct inode *inode;
1324          int ret;
1325
1326         if (!capable(CAP_SYS_ADMIN))
1327                 return -EPERM;
1328
1329         args = memdup_user(argp, sizeof(*args));
1330         if (IS_ERR(args))
1331                 return PTR_ERR(args);
1332
1333         inode = fdentry(file)->d_inode;
1334         ret = search_ioctl(inode, args);
1335         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1336                 ret = -EFAULT;
1337         kfree(args);
1338         return ret;
1339 }
1340
1341 /*
1342  * Search INODE_REFs to identify path name of 'dirid' directory
1343  * in a 'tree_id' tree. and sets path name to 'name'.
1344  */
1345 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1346                                 u64 tree_id, u64 dirid, char *name)
1347 {
1348         struct btrfs_root *root;
1349         struct btrfs_key key;
1350         char *ptr;
1351         int ret = -1;
1352         int slot;
1353         int len;
1354         int total_len = 0;
1355         struct btrfs_inode_ref *iref;
1356         struct extent_buffer *l;
1357         struct btrfs_path *path;
1358
1359         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1360                 name[0]='\0';
1361                 return 0;
1362         }
1363
1364         path = btrfs_alloc_path();
1365         if (!path)
1366                 return -ENOMEM;
1367
1368         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1369
1370         key.objectid = tree_id;
1371         key.type = BTRFS_ROOT_ITEM_KEY;
1372         key.offset = (u64)-1;
1373         root = btrfs_read_fs_root_no_name(info, &key);
1374         if (IS_ERR(root)) {
1375                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1376                 ret = -ENOENT;
1377                 goto out;
1378         }
1379
1380         key.objectid = dirid;
1381         key.type = BTRFS_INODE_REF_KEY;
1382         key.offset = (u64)-1;
1383
1384         while(1) {
1385                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1386                 if (ret < 0)
1387                         goto out;
1388
1389                 l = path->nodes[0];
1390                 slot = path->slots[0];
1391                 if (ret > 0 && slot > 0)
1392                         slot--;
1393                 btrfs_item_key_to_cpu(l, &key, slot);
1394
1395                 if (ret > 0 && (key.objectid != dirid ||
1396                                 key.type != BTRFS_INODE_REF_KEY)) {
1397                         ret = -ENOENT;
1398                         goto out;
1399                 }
1400
1401                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1402                 len = btrfs_inode_ref_name_len(l, iref);
1403                 ptr -= len + 1;
1404                 total_len += len + 1;
1405                 if (ptr < name)
1406                         goto out;
1407
1408                 *(ptr + len) = '/';
1409                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1410
1411                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1412                         break;
1413
1414                 btrfs_release_path(root, path);
1415                 key.objectid = key.offset;
1416                 key.offset = (u64)-1;
1417                 dirid = key.objectid;
1418
1419         }
1420         if (ptr < name)
1421                 goto out;
1422         memcpy(name, ptr, total_len);
1423         name[total_len]='\0';
1424         ret = 0;
1425 out:
1426         btrfs_free_path(path);
1427         return ret;
1428 }
1429
1430 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1431                                            void __user *argp)
1432 {
1433          struct btrfs_ioctl_ino_lookup_args *args;
1434          struct inode *inode;
1435          int ret;
1436
1437         if (!capable(CAP_SYS_ADMIN))
1438                 return -EPERM;
1439
1440         args = memdup_user(argp, sizeof(*args));
1441         if (IS_ERR(args))
1442                 return PTR_ERR(args);
1443
1444         inode = fdentry(file)->d_inode;
1445
1446         if (args->treeid == 0)
1447                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1448
1449         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1450                                         args->treeid, args->objectid,
1451                                         args->name);
1452
1453         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1454                 ret = -EFAULT;
1455
1456         kfree(args);
1457         return ret;
1458 }
1459
1460 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1461                                              void __user *arg)
1462 {
1463         struct dentry *parent = fdentry(file);
1464         struct dentry *dentry;
1465         struct inode *dir = parent->d_inode;
1466         struct inode *inode;
1467         struct btrfs_root *root = BTRFS_I(dir)->root;
1468         struct btrfs_root *dest = NULL;
1469         struct btrfs_ioctl_vol_args *vol_args;
1470         struct btrfs_trans_handle *trans;
1471         int namelen;
1472         int ret;
1473         int err = 0;
1474
1475         vol_args = memdup_user(arg, sizeof(*vol_args));
1476         if (IS_ERR(vol_args))
1477                 return PTR_ERR(vol_args);
1478
1479         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1480         namelen = strlen(vol_args->name);
1481         if (strchr(vol_args->name, '/') ||
1482             strncmp(vol_args->name, "..", namelen) == 0) {
1483                 err = -EINVAL;
1484                 goto out;
1485         }
1486
1487         err = mnt_want_write(file->f_path.mnt);
1488         if (err)
1489                 goto out;
1490
1491         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1492         dentry = lookup_one_len(vol_args->name, parent, namelen);
1493         if (IS_ERR(dentry)) {
1494                 err = PTR_ERR(dentry);
1495                 goto out_unlock_dir;
1496         }
1497
1498         if (!dentry->d_inode) {
1499                 err = -ENOENT;
1500                 goto out_dput;
1501         }
1502
1503         inode = dentry->d_inode;
1504         dest = BTRFS_I(inode)->root;
1505         if (!capable(CAP_SYS_ADMIN)){
1506                 /*
1507                  * Regular user.  Only allow this with a special mount
1508                  * option, when the user has write+exec access to the
1509                  * subvol root, and when rmdir(2) would have been
1510                  * allowed.
1511                  *
1512                  * Note that this is _not_ check that the subvol is
1513                  * empty or doesn't contain data that we wouldn't
1514                  * otherwise be able to delete.
1515                  *
1516                  * Users who want to delete empty subvols should try
1517                  * rmdir(2).
1518                  */
1519                 err = -EPERM;
1520                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1521                         goto out_dput;
1522
1523                 /*
1524                  * Do not allow deletion if the parent dir is the same
1525                  * as the dir to be deleted.  That means the ioctl
1526                  * must be called on the dentry referencing the root
1527                  * of the subvol, not a random directory contained
1528                  * within it.
1529                  */
1530                 err = -EINVAL;
1531                 if (root == dest)
1532                         goto out_dput;
1533
1534                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1535                 if (err)
1536                         goto out_dput;
1537
1538                 /* check if subvolume may be deleted by a non-root user */
1539                 err = btrfs_may_delete(dir, dentry, 1);
1540                 if (err)
1541                         goto out_dput;
1542         }
1543
1544         if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
1545                 err = -EINVAL;
1546                 goto out_dput;
1547         }
1548
1549         mutex_lock(&inode->i_mutex);
1550         err = d_invalidate(dentry);
1551         if (err)
1552                 goto out_unlock;
1553
1554         down_write(&root->fs_info->subvol_sem);
1555
1556         err = may_destroy_subvol(dest);
1557         if (err)
1558                 goto out_up_write;
1559
1560         trans = btrfs_start_transaction(root, 0);
1561         if (IS_ERR(trans)) {
1562                 err = PTR_ERR(trans);
1563                 goto out_up_write;
1564         }
1565         trans->block_rsv = &root->fs_info->global_block_rsv;
1566
1567         ret = btrfs_unlink_subvol(trans, root, dir,
1568                                 dest->root_key.objectid,
1569                                 dentry->d_name.name,
1570                                 dentry->d_name.len);
1571         BUG_ON(ret);
1572
1573         btrfs_record_root_in_trans(trans, dest);
1574
1575         memset(&dest->root_item.drop_progress, 0,
1576                 sizeof(dest->root_item.drop_progress));
1577         dest->root_item.drop_level = 0;
1578         btrfs_set_root_refs(&dest->root_item, 0);
1579
1580         if (!xchg(&dest->orphan_item_inserted, 1)) {
1581                 ret = btrfs_insert_orphan_item(trans,
1582                                         root->fs_info->tree_root,
1583                                         dest->root_key.objectid);
1584                 BUG_ON(ret);
1585         }
1586
1587         ret = btrfs_end_transaction(trans, root);
1588         BUG_ON(ret);
1589         inode->i_flags |= S_DEAD;
1590 out_up_write:
1591         up_write(&root->fs_info->subvol_sem);
1592 out_unlock:
1593         mutex_unlock(&inode->i_mutex);
1594         if (!err) {
1595                 shrink_dcache_sb(root->fs_info->sb);
1596                 btrfs_invalidate_inodes(dest);
1597                 d_delete(dentry);
1598         }
1599 out_dput:
1600         dput(dentry);
1601 out_unlock_dir:
1602         mutex_unlock(&dir->i_mutex);
1603         mnt_drop_write(file->f_path.mnt);
1604 out:
1605         kfree(vol_args);
1606         return err;
1607 }
1608
1609 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1610 {
1611         struct inode *inode = fdentry(file)->d_inode;
1612         struct btrfs_root *root = BTRFS_I(inode)->root;
1613         struct btrfs_ioctl_defrag_range_args *range;
1614         int ret;
1615
1616         if (btrfs_root_readonly(root))
1617                 return -EROFS;
1618
1619         ret = mnt_want_write(file->f_path.mnt);
1620         if (ret)
1621                 return ret;
1622
1623         switch (inode->i_mode & S_IFMT) {
1624         case S_IFDIR:
1625                 if (!capable(CAP_SYS_ADMIN)) {
1626                         ret = -EPERM;
1627                         goto out;
1628                 }
1629                 ret = btrfs_defrag_root(root, 0);
1630                 if (ret)
1631                         goto out;
1632                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
1633                 break;
1634         case S_IFREG:
1635                 if (!(file->f_mode & FMODE_WRITE)) {
1636                         ret = -EINVAL;
1637                         goto out;
1638                 }
1639
1640                 range = kzalloc(sizeof(*range), GFP_KERNEL);
1641                 if (!range) {
1642                         ret = -ENOMEM;
1643                         goto out;
1644                 }
1645
1646                 if (argp) {
1647                         if (copy_from_user(range, argp,
1648                                            sizeof(*range))) {
1649                                 ret = -EFAULT;
1650                                 kfree(range);
1651                                 goto out;
1652                         }
1653                         /* compression requires us to start the IO */
1654                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1655                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
1656                                 range->extent_thresh = (u32)-1;
1657                         }
1658                 } else {
1659                         /* the rest are all set to zero by kzalloc */
1660                         range->len = (u64)-1;
1661                 }
1662                 ret = btrfs_defrag_file(file, range);
1663                 kfree(range);
1664                 break;
1665         default:
1666                 ret = -EINVAL;
1667         }
1668 out:
1669         mnt_drop_write(file->f_path.mnt);
1670         return ret;
1671 }
1672
1673 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
1674 {
1675         struct btrfs_ioctl_vol_args *vol_args;
1676         int ret;
1677
1678         if (!capable(CAP_SYS_ADMIN))
1679                 return -EPERM;
1680
1681         vol_args = memdup_user(arg, sizeof(*vol_args));
1682         if (IS_ERR(vol_args))
1683                 return PTR_ERR(vol_args);
1684
1685         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1686         ret = btrfs_init_new_device(root, vol_args->name);
1687
1688         kfree(vol_args);
1689         return ret;
1690 }
1691
1692 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
1693 {
1694         struct btrfs_ioctl_vol_args *vol_args;
1695         int ret;
1696
1697         if (!capable(CAP_SYS_ADMIN))
1698                 return -EPERM;
1699
1700         if (root->fs_info->sb->s_flags & MS_RDONLY)
1701                 return -EROFS;
1702
1703         vol_args = memdup_user(arg, sizeof(*vol_args));
1704         if (IS_ERR(vol_args))
1705                 return PTR_ERR(vol_args);
1706
1707         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1708         ret = btrfs_rm_device(root, vol_args->name);
1709
1710         kfree(vol_args);
1711         return ret;
1712 }
1713
1714 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
1715                                        u64 off, u64 olen, u64 destoff)
1716 {
1717         struct inode *inode = fdentry(file)->d_inode;
1718         struct btrfs_root *root = BTRFS_I(inode)->root;
1719         struct file *src_file;
1720         struct inode *src;
1721         struct btrfs_trans_handle *trans;
1722         struct btrfs_path *path;
1723         struct extent_buffer *leaf;
1724         char *buf;
1725         struct btrfs_key key;
1726         u32 nritems;
1727         int slot;
1728         int ret;
1729         u64 len = olen;
1730         u64 bs = root->fs_info->sb->s_blocksize;
1731         u64 hint_byte;
1732
1733         /*
1734          * TODO:
1735          * - split compressed inline extents.  annoying: we need to
1736          *   decompress into destination's address_space (the file offset
1737          *   may change, so source mapping won't do), then recompress (or
1738          *   otherwise reinsert) a subrange.
1739          * - allow ranges within the same file to be cloned (provided
1740          *   they don't overlap)?
1741          */
1742
1743         /* the destination must be opened for writing */
1744         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
1745                 return -EINVAL;
1746
1747         if (btrfs_root_readonly(root))
1748                 return -EROFS;
1749
1750         ret = mnt_want_write(file->f_path.mnt);
1751         if (ret)
1752                 return ret;
1753
1754         src_file = fget(srcfd);
1755         if (!src_file) {
1756                 ret = -EBADF;
1757                 goto out_drop_write;
1758         }
1759
1760         src = src_file->f_dentry->d_inode;
1761
1762         ret = -EINVAL;
1763         if (src == inode)
1764                 goto out_fput;
1765
1766         /* the src must be open for reading */
1767         if (!(src_file->f_mode & FMODE_READ))
1768                 goto out_fput;
1769
1770         ret = -EISDIR;
1771         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
1772                 goto out_fput;
1773
1774         ret = -EXDEV;
1775         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
1776                 goto out_fput;
1777
1778         ret = -ENOMEM;
1779         buf = vmalloc(btrfs_level_size(root, 0));
1780         if (!buf)
1781                 goto out_fput;
1782
1783         path = btrfs_alloc_path();
1784         if (!path) {
1785                 vfree(buf);
1786                 goto out_fput;
1787         }
1788         path->reada = 2;
1789
1790         if (inode < src) {
1791                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
1792                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
1793         } else {
1794                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
1795                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1796         }
1797
1798         /* determine range to clone */
1799         ret = -EINVAL;
1800         if (off + len > src->i_size || off + len < off)
1801                 goto out_unlock;
1802         if (len == 0)
1803                 olen = len = src->i_size - off;
1804         /* if we extend to eof, continue to block boundary */
1805         if (off + len == src->i_size)
1806                 len = ALIGN(src->i_size, bs) - off;
1807
1808         /* verify the end result is block aligned */
1809         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
1810             !IS_ALIGNED(destoff, bs))
1811                 goto out_unlock;
1812
1813         /* do any pending delalloc/csum calc on src, one way or
1814            another, and lock file content */
1815         while (1) {
1816                 struct btrfs_ordered_extent *ordered;
1817                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1818                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
1819                 if (!ordered &&
1820                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
1821                                    EXTENT_DELALLOC, 0, NULL))
1822                         break;
1823                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1824                 if (ordered)
1825                         btrfs_put_ordered_extent(ordered);
1826                 btrfs_wait_ordered_range(src, off, len);
1827         }
1828
1829         /* clone data */
1830         key.objectid = src->i_ino;
1831         key.type = BTRFS_EXTENT_DATA_KEY;
1832         key.offset = 0;
1833
1834         while (1) {
1835                 /*
1836                  * note the key will change type as we walk through the
1837                  * tree.
1838                  */
1839                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1840                 if (ret < 0)
1841                         goto out;
1842
1843                 nritems = btrfs_header_nritems(path->nodes[0]);
1844                 if (path->slots[0] >= nritems) {
1845                         ret = btrfs_next_leaf(root, path);
1846                         if (ret < 0)
1847                                 goto out;
1848                         if (ret > 0)
1849                                 break;
1850                         nritems = btrfs_header_nritems(path->nodes[0]);
1851                 }
1852                 leaf = path->nodes[0];
1853                 slot = path->slots[0];
1854
1855                 btrfs_item_key_to_cpu(leaf, &key, slot);
1856                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1857                     key.objectid != src->i_ino)
1858                         break;
1859
1860                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1861                         struct btrfs_file_extent_item *extent;
1862                         int type;
1863                         u32 size;
1864                         struct btrfs_key new_key;
1865                         u64 disko = 0, diskl = 0;
1866                         u64 datao = 0, datal = 0;
1867                         u8 comp;
1868                         u64 endoff;
1869
1870                         size = btrfs_item_size_nr(leaf, slot);
1871                         read_extent_buffer(leaf, buf,
1872                                            btrfs_item_ptr_offset(leaf, slot),
1873                                            size);
1874
1875                         extent = btrfs_item_ptr(leaf, slot,
1876                                                 struct btrfs_file_extent_item);
1877                         comp = btrfs_file_extent_compression(leaf, extent);
1878                         type = btrfs_file_extent_type(leaf, extent);
1879                         if (type == BTRFS_FILE_EXTENT_REG ||
1880                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1881                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1882                                                                       extent);
1883                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1884                                                                  extent);
1885                                 datao = btrfs_file_extent_offset(leaf, extent);
1886                                 datal = btrfs_file_extent_num_bytes(leaf,
1887                                                                     extent);
1888                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1889                                 /* take upper bound, may be compressed */
1890                                 datal = btrfs_file_extent_ram_bytes(leaf,
1891                                                                     extent);
1892                         }
1893                         btrfs_release_path(root, path);
1894
1895                         if (key.offset + datal <= off ||
1896                             key.offset >= off+len)
1897                                 goto next;
1898
1899                         memcpy(&new_key, &key, sizeof(new_key));
1900                         new_key.objectid = inode->i_ino;
1901                         new_key.offset = key.offset + destoff - off;
1902
1903                         trans = btrfs_start_transaction(root, 1);
1904                         if (IS_ERR(trans)) {
1905                                 ret = PTR_ERR(trans);
1906                                 goto out;
1907                         }
1908
1909                         if (type == BTRFS_FILE_EXTENT_REG ||
1910                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1911                                 if (off > key.offset) {
1912                                         datao += off - key.offset;
1913                                         datal -= off - key.offset;
1914                                 }
1915
1916                                 if (key.offset + datal > off + len)
1917                                         datal = off + len - key.offset;
1918
1919                                 ret = btrfs_drop_extents(trans, inode,
1920                                                          new_key.offset,
1921                                                          new_key.offset + datal,
1922                                                          &hint_byte, 1);
1923                                 BUG_ON(ret);
1924
1925                                 ret = btrfs_insert_empty_item(trans, root, path,
1926                                                               &new_key, size);
1927                                 BUG_ON(ret);
1928
1929                                 leaf = path->nodes[0];
1930                                 slot = path->slots[0];
1931                                 write_extent_buffer(leaf, buf,
1932                                             btrfs_item_ptr_offset(leaf, slot),
1933                                             size);
1934
1935                                 extent = btrfs_item_ptr(leaf, slot,
1936                                                 struct btrfs_file_extent_item);
1937
1938                                 /* disko == 0 means it's a hole */
1939                                 if (!disko)
1940                                         datao = 0;
1941
1942                                 btrfs_set_file_extent_offset(leaf, extent,
1943                                                              datao);
1944                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1945                                                                 datal);
1946                                 if (disko) {
1947                                         inode_add_bytes(inode, datal);
1948                                         ret = btrfs_inc_extent_ref(trans, root,
1949                                                         disko, diskl, 0,
1950                                                         root->root_key.objectid,
1951                                                         inode->i_ino,
1952                                                         new_key.offset - datao);
1953                                         BUG_ON(ret);
1954                                 }
1955                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1956                                 u64 skip = 0;
1957                                 u64 trim = 0;
1958                                 if (off > key.offset) {
1959                                         skip = off - key.offset;
1960                                         new_key.offset += skip;
1961                                 }
1962
1963                                 if (key.offset + datal > off+len)
1964                                         trim = key.offset + datal - (off+len);
1965
1966                                 if (comp && (skip || trim)) {
1967                                         ret = -EINVAL;
1968                                         btrfs_end_transaction(trans, root);
1969                                         goto out;
1970                                 }
1971                                 size -= skip + trim;
1972                                 datal -= skip + trim;
1973
1974                                 ret = btrfs_drop_extents(trans, inode,
1975                                                          new_key.offset,
1976                                                          new_key.offset + datal,
1977                                                          &hint_byte, 1);
1978                                 BUG_ON(ret);
1979
1980                                 ret = btrfs_insert_empty_item(trans, root, path,
1981                                                               &new_key, size);
1982                                 BUG_ON(ret);
1983
1984                                 if (skip) {
1985                                         u32 start =
1986                                           btrfs_file_extent_calc_inline_size(0);
1987                                         memmove(buf+start, buf+start+skip,
1988                                                 datal);
1989                                 }
1990
1991                                 leaf = path->nodes[0];
1992                                 slot = path->slots[0];
1993                                 write_extent_buffer(leaf, buf,
1994                                             btrfs_item_ptr_offset(leaf, slot),
1995                                             size);
1996                                 inode_add_bytes(inode, datal);
1997                         }
1998
1999                         btrfs_mark_buffer_dirty(leaf);
2000                         btrfs_release_path(root, path);
2001
2002                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2003
2004                         /*
2005                          * we round up to the block size at eof when
2006                          * determining which extents to clone above,
2007                          * but shouldn't round up the file size
2008                          */
2009                         endoff = new_key.offset + datal;
2010                         if (endoff > destoff+olen)
2011                                 endoff = destoff+olen;
2012                         if (endoff > inode->i_size)
2013                                 btrfs_i_size_write(inode, endoff);
2014
2015                         BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
2016                         ret = btrfs_update_inode(trans, root, inode);
2017                         BUG_ON(ret);
2018                         btrfs_end_transaction(trans, root);
2019                 }
2020 next:
2021                 btrfs_release_path(root, path);
2022                 key.offset++;
2023         }
2024         ret = 0;
2025 out:
2026         btrfs_release_path(root, path);
2027         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2028 out_unlock:
2029         mutex_unlock(&src->i_mutex);
2030         mutex_unlock(&inode->i_mutex);
2031         vfree(buf);
2032         btrfs_free_path(path);
2033 out_fput:
2034         fput(src_file);
2035 out_drop_write:
2036         mnt_drop_write(file->f_path.mnt);
2037         return ret;
2038 }
2039
2040 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2041 {
2042         struct btrfs_ioctl_clone_range_args args;
2043
2044         if (copy_from_user(&args, argp, sizeof(args)))
2045                 return -EFAULT;
2046         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2047                                  args.src_length, args.dest_offset);
2048 }
2049
2050 /*
2051  * there are many ways the trans_start and trans_end ioctls can lead
2052  * to deadlocks.  They should only be used by applications that
2053  * basically own the machine, and have a very in depth understanding
2054  * of all the possible deadlocks and enospc problems.
2055  */
2056 static long btrfs_ioctl_trans_start(struct file *file)
2057 {
2058         struct inode *inode = fdentry(file)->d_inode;
2059         struct btrfs_root *root = BTRFS_I(inode)->root;
2060         struct btrfs_trans_handle *trans;
2061         int ret;
2062
2063         ret = -EPERM;
2064         if (!capable(CAP_SYS_ADMIN))
2065                 goto out;
2066
2067         ret = -EINPROGRESS;
2068         if (file->private_data)
2069                 goto out;
2070
2071         ret = -EROFS;
2072         if (btrfs_root_readonly(root))
2073                 goto out;
2074
2075         ret = mnt_want_write(file->f_path.mnt);
2076         if (ret)
2077                 goto out;
2078
2079         mutex_lock(&root->fs_info->trans_mutex);
2080         root->fs_info->open_ioctl_trans++;
2081         mutex_unlock(&root->fs_info->trans_mutex);
2082
2083         ret = -ENOMEM;
2084         trans = btrfs_start_ioctl_transaction(root, 0);
2085         if (!trans)
2086                 goto out_drop;
2087
2088         file->private_data = trans;
2089         return 0;
2090
2091 out_drop:
2092         mutex_lock(&root->fs_info->trans_mutex);
2093         root->fs_info->open_ioctl_trans--;
2094         mutex_unlock(&root->fs_info->trans_mutex);
2095         mnt_drop_write(file->f_path.mnt);
2096 out:
2097         return ret;
2098 }
2099
2100 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2101 {
2102         struct inode *inode = fdentry(file)->d_inode;
2103         struct btrfs_root *root = BTRFS_I(inode)->root;
2104         struct btrfs_root *new_root;
2105         struct btrfs_dir_item *di;
2106         struct btrfs_trans_handle *trans;
2107         struct btrfs_path *path;
2108         struct btrfs_key location;
2109         struct btrfs_disk_key disk_key;
2110         struct btrfs_super_block *disk_super;
2111         u64 features;
2112         u64 objectid = 0;
2113         u64 dir_id;
2114
2115         if (!capable(CAP_SYS_ADMIN))
2116                 return -EPERM;
2117
2118         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2119                 return -EFAULT;
2120
2121         if (!objectid)
2122                 objectid = root->root_key.objectid;
2123
2124         location.objectid = objectid;
2125         location.type = BTRFS_ROOT_ITEM_KEY;
2126         location.offset = (u64)-1;
2127
2128         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2129         if (IS_ERR(new_root))
2130                 return PTR_ERR(new_root);
2131
2132         if (btrfs_root_refs(&new_root->root_item) == 0)
2133                 return -ENOENT;
2134
2135         path = btrfs_alloc_path();
2136         if (!path)
2137                 return -ENOMEM;
2138         path->leave_spinning = 1;
2139
2140         trans = btrfs_start_transaction(root, 1);
2141         if (!trans) {
2142                 btrfs_free_path(path);
2143                 return -ENOMEM;
2144         }
2145
2146         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
2147         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2148                                    dir_id, "default", 7, 1);
2149         if (IS_ERR_OR_NULL(di)) {
2150                 btrfs_free_path(path);
2151                 btrfs_end_transaction(trans, root);
2152                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2153                        "this isn't going to work\n");
2154                 return -ENOENT;
2155         }
2156
2157         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2158         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2159         btrfs_mark_buffer_dirty(path->nodes[0]);
2160         btrfs_free_path(path);
2161
2162         disk_super = &root->fs_info->super_copy;
2163         features = btrfs_super_incompat_flags(disk_super);
2164         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2165                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2166                 btrfs_set_super_incompat_flags(disk_super, features);
2167         }
2168         btrfs_end_transaction(trans, root);
2169
2170         return 0;
2171 }
2172
2173 static void get_block_group_info(struct list_head *groups_list,
2174                                  struct btrfs_ioctl_space_info *space)
2175 {
2176         struct btrfs_block_group_cache *block_group;
2177
2178         space->total_bytes = 0;
2179         space->used_bytes = 0;
2180         space->flags = 0;
2181         list_for_each_entry(block_group, groups_list, list) {
2182                 space->flags = block_group->flags;
2183                 space->total_bytes += block_group->key.offset;
2184                 space->used_bytes +=
2185                         btrfs_block_group_used(&block_group->item);
2186         }
2187 }
2188
2189 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2190 {
2191         struct btrfs_ioctl_space_args space_args;
2192         struct btrfs_ioctl_space_info space;
2193         struct btrfs_ioctl_space_info *dest;
2194         struct btrfs_ioctl_space_info *dest_orig;
2195         struct btrfs_ioctl_space_info *user_dest;
2196         struct btrfs_space_info *info;
2197         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2198                        BTRFS_BLOCK_GROUP_SYSTEM,
2199                        BTRFS_BLOCK_GROUP_METADATA,
2200                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2201         int num_types = 4;
2202         int alloc_size;
2203         int ret = 0;
2204         int slot_count = 0;
2205         int i, c;
2206
2207         if (copy_from_user(&space_args,
2208                            (struct btrfs_ioctl_space_args __user *)arg,
2209                            sizeof(space_args)))
2210                 return -EFAULT;
2211
2212         for (i = 0; i < num_types; i++) {
2213                 struct btrfs_space_info *tmp;
2214
2215                 info = NULL;
2216                 rcu_read_lock();
2217                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2218                                         list) {
2219                         if (tmp->flags == types[i]) {
2220                                 info = tmp;
2221                                 break;
2222                         }
2223                 }
2224                 rcu_read_unlock();
2225
2226                 if (!info)
2227                         continue;
2228
2229                 down_read(&info->groups_sem);
2230                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2231                         if (!list_empty(&info->block_groups[c]))
2232                                 slot_count++;
2233                 }
2234                 up_read(&info->groups_sem);
2235         }
2236
2237         /* space_slots == 0 means they are asking for a count */
2238         if (space_args.space_slots == 0) {
2239                 space_args.total_spaces = slot_count;
2240                 goto out;
2241         }
2242
2243         slot_count = min_t(int, space_args.space_slots, slot_count);
2244
2245         alloc_size = sizeof(*dest) * slot_count;
2246
2247         /* we generally have at most 6 or so space infos, one for each raid
2248          * level.  So, a whole page should be more than enough for everyone
2249          */
2250         if (alloc_size > PAGE_CACHE_SIZE)
2251                 return -ENOMEM;
2252
2253         space_args.total_spaces = 0;
2254         dest = kmalloc(alloc_size, GFP_NOFS);
2255         if (!dest)
2256                 return -ENOMEM;
2257         dest_orig = dest;
2258
2259         /* now we have a buffer to copy into */
2260         for (i = 0; i < num_types; i++) {
2261                 struct btrfs_space_info *tmp;
2262
2263                 info = NULL;
2264                 rcu_read_lock();
2265                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2266                                         list) {
2267                         if (tmp->flags == types[i]) {
2268                                 info = tmp;
2269                                 break;
2270                         }
2271                 }
2272                 rcu_read_unlock();
2273
2274                 if (!info)
2275                         continue;
2276                 down_read(&info->groups_sem);
2277                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2278                         if (!list_empty(&info->block_groups[c])) {
2279                                 get_block_group_info(&info->block_groups[c],
2280                                                      &space);
2281                                 memcpy(dest, &space, sizeof(space));
2282                                 dest++;
2283                                 space_args.total_spaces++;
2284                         }
2285                 }
2286                 up_read(&info->groups_sem);
2287         }
2288
2289         user_dest = (struct btrfs_ioctl_space_info *)
2290                 (arg + sizeof(struct btrfs_ioctl_space_args));
2291
2292         if (copy_to_user(user_dest, dest_orig, alloc_size))
2293                 ret = -EFAULT;
2294
2295         kfree(dest_orig);
2296 out:
2297         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2298                 ret = -EFAULT;
2299
2300         return ret;
2301 }
2302
2303 /*
2304  * there are many ways the trans_start and trans_end ioctls can lead
2305  * to deadlocks.  They should only be used by applications that
2306  * basically own the machine, and have a very in depth understanding
2307  * of all the possible deadlocks and enospc problems.
2308  */
2309 long btrfs_ioctl_trans_end(struct file *file)
2310 {
2311         struct inode *inode = fdentry(file)->d_inode;
2312         struct btrfs_root *root = BTRFS_I(inode)->root;
2313         struct btrfs_trans_handle *trans;
2314
2315         trans = file->private_data;
2316         if (!trans)
2317                 return -EINVAL;
2318         file->private_data = NULL;
2319
2320         btrfs_end_transaction(trans, root);
2321
2322         mutex_lock(&root->fs_info->trans_mutex);
2323         root->fs_info->open_ioctl_trans--;
2324         mutex_unlock(&root->fs_info->trans_mutex);
2325
2326         mnt_drop_write(file->f_path.mnt);
2327         return 0;
2328 }
2329
2330 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2331 {
2332         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2333         struct btrfs_trans_handle *trans;
2334         u64 transid;
2335
2336         trans = btrfs_start_transaction(root, 0);
2337         transid = trans->transid;
2338         btrfs_commit_transaction_async(trans, root, 0);
2339
2340         if (argp)
2341                 if (copy_to_user(argp, &transid, sizeof(transid)))
2342                         return -EFAULT;
2343         return 0;
2344 }
2345
2346 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2347 {
2348         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2349         u64 transid;
2350
2351         if (argp) {
2352                 if (copy_from_user(&transid, argp, sizeof(transid)))
2353                         return -EFAULT;
2354         } else {
2355                 transid = 0;  /* current trans */
2356         }
2357         return btrfs_wait_for_commit(root, transid);
2358 }
2359
2360 long btrfs_ioctl(struct file *file, unsigned int
2361                 cmd, unsigned long arg)
2362 {
2363         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2364         void __user *argp = (void __user *)arg;
2365
2366         switch (cmd) {
2367         case FS_IOC_GETFLAGS:
2368                 return btrfs_ioctl_getflags(file, argp);
2369         case FS_IOC_SETFLAGS:
2370                 return btrfs_ioctl_setflags(file, argp);
2371         case FS_IOC_GETVERSION:
2372                 return btrfs_ioctl_getversion(file, argp);
2373         case BTRFS_IOC_SNAP_CREATE:
2374                 return btrfs_ioctl_snap_create(file, argp, 0);
2375         case BTRFS_IOC_SNAP_CREATE_V2:
2376                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
2377         case BTRFS_IOC_SUBVOL_CREATE:
2378                 return btrfs_ioctl_snap_create(file, argp, 1);
2379         case BTRFS_IOC_SNAP_DESTROY:
2380                 return btrfs_ioctl_snap_destroy(file, argp);
2381         case BTRFS_IOC_SUBVOL_GETFLAGS:
2382                 return btrfs_ioctl_subvol_getflags(file, argp);
2383         case BTRFS_IOC_SUBVOL_SETFLAGS:
2384                 return btrfs_ioctl_subvol_setflags(file, argp);
2385         case BTRFS_IOC_DEFAULT_SUBVOL:
2386                 return btrfs_ioctl_default_subvol(file, argp);
2387         case BTRFS_IOC_DEFRAG:
2388                 return btrfs_ioctl_defrag(file, NULL);
2389         case BTRFS_IOC_DEFRAG_RANGE:
2390                 return btrfs_ioctl_defrag(file, argp);
2391         case BTRFS_IOC_RESIZE:
2392                 return btrfs_ioctl_resize(root, argp);
2393         case BTRFS_IOC_ADD_DEV:
2394                 return btrfs_ioctl_add_dev(root, argp);
2395         case BTRFS_IOC_RM_DEV:
2396                 return btrfs_ioctl_rm_dev(root, argp);
2397         case BTRFS_IOC_BALANCE:
2398                 return btrfs_balance(root->fs_info->dev_root);
2399         case BTRFS_IOC_CLONE:
2400                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
2401         case BTRFS_IOC_CLONE_RANGE:
2402                 return btrfs_ioctl_clone_range(file, argp);
2403         case BTRFS_IOC_TRANS_START:
2404                 return btrfs_ioctl_trans_start(file);
2405         case BTRFS_IOC_TRANS_END:
2406                 return btrfs_ioctl_trans_end(file);
2407         case BTRFS_IOC_TREE_SEARCH:
2408                 return btrfs_ioctl_tree_search(file, argp);
2409         case BTRFS_IOC_INO_LOOKUP:
2410                 return btrfs_ioctl_ino_lookup(file, argp);
2411         case BTRFS_IOC_SPACE_INFO:
2412                 return btrfs_ioctl_space_info(root, argp);
2413         case BTRFS_IOC_SYNC:
2414                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
2415                 return 0;
2416         case BTRFS_IOC_START_SYNC:
2417                 return btrfs_ioctl_start_sync(file, argp);
2418         case BTRFS_IOC_WAIT_SYNC:
2419                 return btrfs_ioctl_wait_sync(file, argp);
2420         }
2421
2422         return -ENOTTY;
2423 }