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 <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "compat.h"
  48 #include "ctree.h"
  49 #include "disk-io.h"
  50 #include "transaction.h"
  51 #include "btrfs_inode.h"
  52 #include "print-tree.h"
  53 #include "volumes.h"
  54 #include "locking.h"
  55 #include "inode-map.h"
  56 #include "backref.h"
  57 #include "rcu-string.h"
  58 #include "send.h"
  59 #include "dev-replace.h"
  60
  61 static int btrfs_clone(struct inode *src, struct inode *inode,
  62                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  63
  64 /* Mask out flags that are inappropriate for the given type of inode. */
  65 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  66 {
  67         if (S_ISDIR(mode))
  68                 return flags;
  69         else if (S_ISREG(mode))
  70                 return flags & ~FS_DIRSYNC_FL;
  71         else
  72                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  73 }
  74
  75 /*
  76  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  77  */
  78 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  79 {
  80         unsigned int iflags = 0;
  81
  82         if (flags & BTRFS_INODE_SYNC)
  83                 iflags |= FS_SYNC_FL;
  84         if (flags & BTRFS_INODE_IMMUTABLE)
  85                 iflags |= FS_IMMUTABLE_FL;
  86         if (flags & BTRFS_INODE_APPEND)
  87                 iflags |= FS_APPEND_FL;
  88         if (flags & BTRFS_INODE_NODUMP)
  89                 iflags |= FS_NODUMP_FL;
  90         if (flags & BTRFS_INODE_NOATIME)
  91                 iflags |= FS_NOATIME_FL;
  92         if (flags & BTRFS_INODE_DIRSYNC)
  93                 iflags |= FS_DIRSYNC_FL;
  94         if (flags & BTRFS_INODE_NODATACOW)
  95                 iflags |= FS_NOCOW_FL;
  96
  97         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
  98                 iflags |= FS_COMPR_FL;
  99         else if (flags & BTRFS_INODE_NOCOMPRESS)
 100                 iflags |= FS_NOCOMP_FL;
 101
 102         return iflags;
 103 }
 104
 105 /*
 106  * Update inode->i_flags based on the btrfs internal flags.
 107  */
 108 void btrfs_update_iflags(struct inode *inode)
 109 {
 110         struct btrfs_inode *ip = BTRFS_I(inode);
 111
 112         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 113
 114         if (ip->flags & BTRFS_INODE_SYNC)
 115                 inode->i_flags |= S_SYNC;
 116         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 117                 inode->i_flags |= S_IMMUTABLE;
 118         if (ip->flags & BTRFS_INODE_APPEND)
 119                 inode->i_flags |= S_APPEND;
 120         if (ip->flags & BTRFS_INODE_NOATIME)
 121                 inode->i_flags |= S_NOATIME;
 122         if (ip->flags & BTRFS_INODE_DIRSYNC)
 123                 inode->i_flags |= S_DIRSYNC;
 124 }
 125
 126 /*
 127  * Inherit flags from the parent inode.
 128  *
 129  * Currently only the compression flags and the cow flags are inherited.
 130  */
 131 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 132 {
 133         unsigned int flags;
 134
 135         if (!dir)
 136                 return;
 137
 138         flags = BTRFS_I(dir)->flags;
 139
 140         if (flags & BTRFS_INODE_NOCOMPRESS) {
 141                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 142                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 143         } else if (flags & BTRFS_INODE_COMPRESS) {
 144                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 145                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 146         }
 147
 148         if (flags & BTRFS_INODE_NODATACOW) {
 149                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 150                 if (S_ISREG(inode->i_mode))
 151                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 152         }
 153
 154         btrfs_update_iflags(inode);
 155 }
 156
 157 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 158 {
 159         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 160         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 161
 162         if (copy_to_user(arg, &flags, sizeof(flags)))
 163                 return -EFAULT;
 164         return 0;
 165 }
 166
 167 static int check_flags(unsigned int flags)
 168 {
 169         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 170                       FS_NOATIME_FL | FS_NODUMP_FL | \
 171                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 172                       FS_NOCOMP_FL | FS_COMPR_FL |
 173                       FS_NOCOW_FL))
 174                 return -EOPNOTSUPP;
 175
 176         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 177                 return -EINVAL;
 178
 179         return 0;
 180 }
 181
 182 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 183 {
 184         struct inode *inode = file_inode(file);
 185         struct btrfs_inode *ip = BTRFS_I(inode);
 186         struct btrfs_root *root = ip->root;
 187         struct btrfs_trans_handle *trans;
 188         unsigned int flags, oldflags;
 189         int ret;
 190         u64 ip_oldflags;
 191         unsigned int i_oldflags;
 192         umode_t mode;
 193
 194         if (btrfs_root_readonly(root))
 195                 return -EROFS;
 196
 197         if (copy_from_user(&flags, arg, sizeof(flags)))
 198                 return -EFAULT;
 199
 200         ret = check_flags(flags);
 201         if (ret)
 202                 return ret;
 203
 204         if (!inode_owner_or_capable(inode))
 205                 return -EACCES;
 206
 207         ret = mnt_want_write_file(file);
 208         if (ret)
 209                 return ret;
 210
 211         mutex_lock(&inode->i_mutex);
 212
 213         ip_oldflags = ip->flags;
 214         i_oldflags = inode->i_flags;
 215         mode = inode->i_mode;
 216
 217         flags = btrfs_mask_flags(inode->i_mode, flags);
 218         oldflags = btrfs_flags_to_ioctl(ip->flags);
 219         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 220                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 221                         ret = -EPERM;
 222                         goto out_unlock;
 223                 }
 224         }
 225
 226         if (flags & FS_SYNC_FL)
 227                 ip->flags |= BTRFS_INODE_SYNC;
 228         else
 229                 ip->flags &= ~BTRFS_INODE_SYNC;
 230         if (flags & FS_IMMUTABLE_FL)
 231                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 232         else
 233                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 234         if (flags & FS_APPEND_FL)
 235                 ip->flags |= BTRFS_INODE_APPEND;
 236         else
 237                 ip->flags &= ~BTRFS_INODE_APPEND;
 238         if (flags & FS_NODUMP_FL)
 239                 ip->flags |= BTRFS_INODE_NODUMP;
 240         else
 241                 ip->flags &= ~BTRFS_INODE_NODUMP;
 242         if (flags & FS_NOATIME_FL)
 243                 ip->flags |= BTRFS_INODE_NOATIME;
 244         else
 245                 ip->flags &= ~BTRFS_INODE_NOATIME;
 246         if (flags & FS_DIRSYNC_FL)
 247                 ip->flags |= BTRFS_INODE_DIRSYNC;
 248         else
 249                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 250         if (flags & FS_NOCOW_FL) {
 251                 if (S_ISREG(mode)) {
 252                         /*
 253                          * It's safe to turn csums off here, no extents exist.
 254                          * Otherwise we want the flag to reflect the real COW
 255                          * status of the file and will not set it.
 256                          */
 257                         if (inode->i_size == 0)
 258                                 ip->flags |= BTRFS_INODE_NODATACOW
 259                                            | BTRFS_INODE_NODATASUM;
 260                 } else {
 261                         ip->flags |= BTRFS_INODE_NODATACOW;
 262                 }
 263         } else {
 264                 /*
 265                  * Revert back under same assuptions as above
 266                  */
 267                 if (S_ISREG(mode)) {
 268                         if (inode->i_size == 0)
 269                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 270                                              | BTRFS_INODE_NODATASUM);
 271                 } else {
 272                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 273                 }
 274         }
 275
 276         /*
 277          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 278          * flag may be changed automatically if compression code won't make
 279          * things smaller.
 280          */
 281         if (flags & FS_NOCOMP_FL) {
 282                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 283                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 284         } else if (flags & FS_COMPR_FL) {
 285                 ip->flags |= BTRFS_INODE_COMPRESS;
 286                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 287         } else {
 288                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 289         }
 290
 291         trans = btrfs_start_transaction(root, 1);
 292         if (IS_ERR(trans)) {
 293                 ret = PTR_ERR(trans);
 294                 goto out_drop;
 295         }
 296
 297         btrfs_update_iflags(inode);
 298         inode_inc_iversion(inode);
 299         inode->i_ctime = CURRENT_TIME;
 300         ret = btrfs_update_inode(trans, root, inode);
 301
 302         btrfs_end_transaction(trans, root);
 303  out_drop:
 304         if (ret) {
 305                 ip->flags = ip_oldflags;
 306                 inode->i_flags = i_oldflags;
 307         }
 308
 309  out_unlock:
 310         mutex_unlock(&inode->i_mutex);
 311         mnt_drop_write_file(file);
 312         return ret;
 313 }
 314
 315 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 316 {
 317         struct inode *inode = file_inode(file);
 318
 319         return put_user(inode->i_generation, arg);
 320 }
 321
 322 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 323 {
 324         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
 325         struct btrfs_device *device;
 326         struct request_queue *q;
 327         struct fstrim_range range;
 328         u64 minlen = ULLONG_MAX;
 329         u64 num_devices = 0;
 330         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 331         int ret;
 332
 333         if (!capable(CAP_SYS_ADMIN))
 334                 return -EPERM;
 335
 336         rcu_read_lock();
 337         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 338                                 dev_list) {
 339                 if (!device->bdev)
 340                         continue;
 341                 q = bdev_get_queue(device->bdev);
 342                 if (blk_queue_discard(q)) {
 343                         num_devices++;
 344                         minlen = min((u64)q->limits.discard_granularity,
 345                                      minlen);
 346                 }
 347         }
 348         rcu_read_unlock();
 349
 350         if (!num_devices)
 351                 return -EOPNOTSUPP;
 352         if (copy_from_user(&range, arg, sizeof(range)))
 353                 return -EFAULT;
 354         if (range.start > total_bytes ||
 355             range.len < fs_info->sb->s_blocksize)
 356                 return -EINVAL;
 357
 358         range.len = min(range.len, total_bytes - range.start);
 359         range.minlen = max(range.minlen, minlen);
 360         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 361         if (ret < 0)
 362                 return ret;
 363
 364         if (copy_to_user(arg, &range, sizeof(range)))
 365                 return -EFAULT;
 366
 367         return 0;
 368 }
 369
 370 int btrfs_is_empty_uuid(u8 *uuid)
 371 {
 372         static char empty_uuid[BTRFS_UUID_SIZE] = {0};
 373
 374         return !memcmp(uuid, empty_uuid, BTRFS_UUID_SIZE);
 375 }
 376
 377 static noinline int create_subvol(struct inode *dir,
 378                                   struct dentry *dentry,
 379                                   char *name, int namelen,
 380                                   u64 *async_transid,
 381                                   struct btrfs_qgroup_inherit *inherit)
 382 {
 383         struct btrfs_trans_handle *trans;
 384         struct btrfs_key key;
 385         struct btrfs_root_item root_item;
 386         struct btrfs_inode_item *inode_item;
 387         struct extent_buffer *leaf;
 388         struct btrfs_root *root = BTRFS_I(dir)->root;
 389         struct btrfs_root *new_root;
 390         struct btrfs_block_rsv block_rsv;
 391         struct timespec cur_time = CURRENT_TIME;
 392         int ret;
 393         int err;
 394         u64 objectid;
 395         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 396         u64 index = 0;
 397         u64 qgroup_reserved;
 398         uuid_le new_uuid;
 399
 400         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 401         if (ret)
 402                 return ret;
 403
 404         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 405         /*
 406          * The same as the snapshot creation, please see the comment
 407          * of create_snapshot().
 408          */
 409         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 410                                                8, &qgroup_reserved, false);
 411         if (ret)
 412                 return ret;
 413
 414         trans = btrfs_start_transaction(root, 0);
 415         if (IS_ERR(trans)) {
 416                 ret = PTR_ERR(trans);
 417                 goto out;
 418         }
 419         trans->block_rsv = &block_rsv;
 420         trans->bytes_reserved = block_rsv.size;
 421
 422         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 423         if (ret)
 424                 goto fail;
 425
 426         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 427                                       0, objectid, NULL, 0, 0, 0);
 428         if (IS_ERR(leaf)) {
 429                 ret = PTR_ERR(leaf);
 430                 goto fail;
 431         }
 432
 433         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 434         btrfs_set_header_bytenr(leaf, leaf->start);
 435         btrfs_set_header_generation(leaf, trans->transid);
 436         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 437         btrfs_set_header_owner(leaf, objectid);
 438
 439         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 440                             BTRFS_FSID_SIZE);
 441         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 442                             btrfs_header_chunk_tree_uuid(leaf),
 443                             BTRFS_UUID_SIZE);
 444         btrfs_mark_buffer_dirty(leaf);
 445
 446         memset(&root_item, 0, sizeof(root_item));
 447
 448         inode_item = &root_item.inode;
 449         btrfs_set_stack_inode_generation(inode_item, 1);
 450         btrfs_set_stack_inode_size(inode_item, 3);
 451         btrfs_set_stack_inode_nlink(inode_item, 1);
 452         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
 453         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 454
 455         btrfs_set_root_flags(&root_item, 0);
 456         btrfs_set_root_limit(&root_item, 0);
 457         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 458
 459         btrfs_set_root_bytenr(&root_item, leaf->start);
 460         btrfs_set_root_generation(&root_item, trans->transid);
 461         btrfs_set_root_level(&root_item, 0);
 462         btrfs_set_root_refs(&root_item, 1);
 463         btrfs_set_root_used(&root_item, leaf->len);
 464         btrfs_set_root_last_snapshot(&root_item, 0);
 465
 466         btrfs_set_root_generation_v2(&root_item,
 467                         btrfs_root_generation(&root_item));
 468         uuid_le_gen(&new_uuid);
 469         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 470         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 471         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 472         root_item.ctime = root_item.otime;
 473         btrfs_set_root_ctransid(&root_item, trans->transid);
 474         btrfs_set_root_otransid(&root_item, trans->transid);
 475
 476         btrfs_tree_unlock(leaf);
 477         free_extent_buffer(leaf);
 478         leaf = NULL;
 479
 480         btrfs_set_root_dirid(&root_item, new_dirid);
 481
 482         key.objectid = objectid;
 483         key.offset = 0;
 484         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 485         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 486                                 &root_item);
 487         if (ret)
 488                 goto fail;
 489
 490         key.offset = (u64)-1;
 491         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 492         if (IS_ERR(new_root)) {
 493                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 494                 ret = PTR_ERR(new_root);
 495                 goto fail;
 496         }
 497
 498         btrfs_record_root_in_trans(trans, new_root);
 499
 500         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
 501         if (ret) {
 502                 /* We potentially lose an unused inode item here */
 503                 btrfs_abort_transaction(trans, root, ret);
 504                 goto fail;
 505         }
 506
 507         /*
 508          * insert the directory item
 509          */
 510         ret = btrfs_set_inode_index(dir, &index);
 511         if (ret) {
 512                 btrfs_abort_transaction(trans, root, ret);
 513                 goto fail;
 514         }
 515
 516         ret = btrfs_insert_dir_item(trans, root,
 517                                     name, namelen, dir, &key,
 518                                     BTRFS_FT_DIR, index);
 519         if (ret) {
 520                 btrfs_abort_transaction(trans, root, ret);
 521                 goto fail;
 522         }
 523
 524         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 525         ret = btrfs_update_inode(trans, root, dir);
 526         BUG_ON(ret);
 527
 528         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 529                                  objectid, root->root_key.objectid,
 530                                  btrfs_ino(dir), index, name, namelen);
 531         BUG_ON(ret);
 532
 533         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 534                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 535                                   objectid);
 536         if (ret)
 537                 btrfs_abort_transaction(trans, root, ret);
 538
 539 fail:
 540         trans->block_rsv = NULL;
 541         trans->bytes_reserved = 0;
 542         if (async_transid) {
 543                 *async_transid = trans->transid;
 544                 err = btrfs_commit_transaction_async(trans, root, 1);
 545                 if (err)
 546                         err = btrfs_commit_transaction(trans, root);
 547         } else {
 548                 err = btrfs_commit_transaction(trans, root);
 549         }
 550         if (err && !ret)
 551                 ret = err;
 552
 553         if (!ret)
 554                 d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 555 out:
 556         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 557         return ret;
 558 }
 559
 560 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 561                            struct dentry *dentry, char *name, int namelen,
 562                            u64 *async_transid, bool readonly,
 563                            struct btrfs_qgroup_inherit *inherit)
 564 {
 565         struct inode *inode;
 566         struct btrfs_pending_snapshot *pending_snapshot;
 567         struct btrfs_trans_handle *trans;
 568         int ret;
 569
 570         if (!root->ref_cows)
 571                 return -EINVAL;
 572
 573         ret = btrfs_start_delalloc_inodes(root, 0);
 574         if (ret)
 575                 return ret;
 576
 577         btrfs_wait_ordered_extents(root);
 578
 579         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 580         if (!pending_snapshot)
 581                 return -ENOMEM;
 582
 583         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 584                              BTRFS_BLOCK_RSV_TEMP);
 585         /*
 586          * 1 - parent dir inode
 587          * 2 - dir entries
 588          * 1 - root item
 589          * 2 - root ref/backref
 590          * 1 - root of snapshot
 591          * 1 - UUID item
 592          */
 593         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 594                                         &pending_snapshot->block_rsv, 8,
 595                                         &pending_snapshot->qgroup_reserved,
 596                                         false);
 597         if (ret)
 598                 goto out;
 599
 600         pending_snapshot->dentry = dentry;
 601         pending_snapshot->root = root;
 602         pending_snapshot->readonly = readonly;
 603         pending_snapshot->dir = dir;
 604         pending_snapshot->inherit = inherit;
 605
 606         trans = btrfs_start_transaction(root, 0);
 607         if (IS_ERR(trans)) {
 608                 ret = PTR_ERR(trans);
 609                 goto fail;
 610         }
 611
 612         spin_lock(&root->fs_info->trans_lock);
 613         list_add(&pending_snapshot->list,
 614                  &trans->transaction->pending_snapshots);
 615         spin_unlock(&root->fs_info->trans_lock);
 616         if (async_transid) {
 617                 *async_transid = trans->transid;
 618                 ret = btrfs_commit_transaction_async(trans,
 619                                      root->fs_info->extent_root, 1);
 620                 if (ret)
 621                         ret = btrfs_commit_transaction(trans, root);
 622         } else {
 623                 ret = btrfs_commit_transaction(trans,
 624                                                root->fs_info->extent_root);
 625         }
 626         if (ret)
 627                 goto fail;
 628
 629         ret = pending_snapshot->error;
 630         if (ret)
 631                 goto fail;
 632
 633         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 634         if (ret)
 635                 goto fail;
 636
 637         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 638         if (IS_ERR(inode)) {
 639                 ret = PTR_ERR(inode);
 640                 goto fail;
 641         }
 642         BUG_ON(!inode);
 643         d_instantiate(dentry, inode);
 644         ret = 0;
 645 fail:
 646         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 647                                          &pending_snapshot->block_rsv,
 648                                          pending_snapshot->qgroup_reserved);
 649 out:
 650         kfree(pending_snapshot);
 651         return ret;
 652 }
 653
 654 /*  copy of check_sticky in fs/namei.c()
 655 * It's inline, so penalty for filesystems that don't use sticky bit is
 656 * minimal.
 657 */
 658 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 659 {
 660         kuid_t fsuid = current_fsuid();
 661
 662         if (!(dir->i_mode & S_ISVTX))
 663                 return 0;
 664         if (uid_eq(inode->i_uid, fsuid))
 665                 return 0;
 666         if (uid_eq(dir->i_uid, fsuid))
 667                 return 0;
 668         return !capable(CAP_FOWNER);
 669 }
 670
 671 /*  copy of may_delete in fs/namei.c()
 672  *      Check whether we can remove a link victim from directory dir, check
 673  *  whether the type of victim is right.
 674  *  1. We can't do it if dir is read-only (done in permission())
 675  *  2. We should have write and exec permissions on dir
 676  *  3. We can't remove anything from append-only dir
 677  *  4. We can't do anything with immutable dir (done in permission())
 678  *  5. If the sticky bit on dir is set we should either
 679  *      a. be owner of dir, or
 680  *      b. be owner of victim, or
 681  *      c. have CAP_FOWNER capability
 682  *  6. If the victim is append-only or immutable we can't do antyhing with
 683  *     links pointing to it.
 684  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 685  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 686  *  9. We can't remove a root or mountpoint.
 687  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 688  *     nfs_async_unlink().
 689  */
 690
 691 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 692 {
 693         int error;
 694
 695         if (!victim->d_inode)
 696                 return -ENOENT;
 697
 698         BUG_ON(victim->d_parent->d_inode != dir);
 699         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 700
 701         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 702         if (error)
 703                 return error;
 704         if (IS_APPEND(dir))
 705                 return -EPERM;
 706         if (btrfs_check_sticky(dir, victim->d_inode)||
 707                 IS_APPEND(victim->d_inode)||
 708             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 709                 return -EPERM;
 710         if (isdir) {
 711                 if (!S_ISDIR(victim->d_inode->i_mode))
 712                         return -ENOTDIR;
 713                 if (IS_ROOT(victim))
 714                         return -EBUSY;
 715         } else if (S_ISDIR(victim->d_inode->i_mode))
 716                 return -EISDIR;
 717         if (IS_DEADDIR(dir))
 718                 return -ENOENT;
 719         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 720                 return -EBUSY;
 721         return 0;
 722 }
 723
 724 /* copy of may_create in fs/namei.c() */
 725 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 726 {
 727         if (child->d_inode)
 728                 return -EEXIST;
 729         if (IS_DEADDIR(dir))
 730                 return -ENOENT;
 731         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 732 }
 733
 734 /*
 735  * Create a new subvolume below @parent.  This is largely modeled after
 736  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 737  * inside this filesystem so it's quite a bit simpler.
 738  */
 739 static noinline int btrfs_mksubvol(struct path *parent,
 740                                    char *name, int namelen,
 741                                    struct btrfs_root *snap_src,
 742                                    u64 *async_transid, bool readonly,
 743                                    struct btrfs_qgroup_inherit *inherit)
 744 {
 745         struct inode *dir  = parent->dentry->d_inode;
 746         struct dentry *dentry;
 747         int error;
 748
 749         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 750         if (error == -EINTR)
 751                 return error;
 752
 753         dentry = lookup_one_len(name, parent->dentry, namelen);
 754         error = PTR_ERR(dentry);
 755         if (IS_ERR(dentry))
 756                 goto out_unlock;
 757
 758         error = -EEXIST;
 759         if (dentry->d_inode)
 760                 goto out_dput;
 761
 762         error = btrfs_may_create(dir, dentry);
 763         if (error)
 764                 goto out_dput;
 765
 766         /*
 767          * even if this name doesn't exist, we may get hash collisions.
 768          * check for them now when we can safely fail
 769          */
 770         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 771                                                dir->i_ino, name,
 772                                                namelen);
 773         if (error)
 774                 goto out_dput;
 775
 776         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 777
 778         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 779                 goto out_up_read;
 780
 781         if (snap_src) {
 782                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 783                                         async_transid, readonly, inherit);
 784         } else {
 785                 error = create_subvol(dir, dentry, name, namelen,
 786                                       async_transid, inherit);
 787         }
 788         if (!error)
 789                 fsnotify_mkdir(dir, dentry);
 790 out_up_read:
 791         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 792 out_dput:
 793         dput(dentry);
 794 out_unlock:
 795         mutex_unlock(&dir->i_mutex);
 796         return error;
 797 }
 798
 799 /*
 800  * When we're defragging a range, we don't want to kick it off again
 801  * if it is really just waiting for delalloc to send it down.
 802  * If we find a nice big extent or delalloc range for the bytes in the
 803  * file you want to defrag, we return 0 to let you know to skip this
 804  * part of the file
 805  */
 806 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 807 {
 808         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 809         struct extent_map *em = NULL;
 810         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 811         u64 end;
 812
 813         read_lock(&em_tree->lock);
 814         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 815         read_unlock(&em_tree->lock);
 816
 817         if (em) {
 818                 end = extent_map_end(em);
 819                 free_extent_map(em);
 820                 if (end - offset > thresh)
 821                         return 0;
 822         }
 823         /* if we already have a nice delalloc here, just stop */
 824         thresh /= 2;
 825         end = count_range_bits(io_tree, &offset, offset + thresh,
 826                                thresh, EXTENT_DELALLOC, 1);
 827         if (end >= thresh)
 828                 return 0;
 829         return 1;
 830 }
 831
 832 /*
 833  * helper function to walk through a file and find extents
 834  * newer than a specific transid, and smaller than thresh.
 835  *
 836  * This is used by the defragging code to find new and small
 837  * extents
 838  */
 839 static int find_new_extents(struct btrfs_root *root,
 840                             struct inode *inode, u64 newer_than,
 841                             u64 *off, int thresh)
 842 {
 843         struct btrfs_path *path;
 844         struct btrfs_key min_key;
 845         struct extent_buffer *leaf;
 846         struct btrfs_file_extent_item *extent;
 847         int type;
 848         int ret;
 849         u64 ino = btrfs_ino(inode);
 850
 851         path = btrfs_alloc_path();
 852         if (!path)
 853                 return -ENOMEM;
 854
 855         min_key.objectid = ino;
 856         min_key.type = BTRFS_EXTENT_DATA_KEY;
 857         min_key.offset = *off;
 858
 859         path->keep_locks = 1;
 860
 861         while(1) {
 862                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 863                 if (ret != 0)
 864                         goto none;
 865                 if (min_key.objectid != ino)
 866                         goto none;
 867                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 868                         goto none;
 869
 870                 leaf = path->nodes[0];
 871                 extent = btrfs_item_ptr(leaf, path->slots[0],
 872                                         struct btrfs_file_extent_item);
 873
 874                 type = btrfs_file_extent_type(leaf, extent);
 875                 if (type == BTRFS_FILE_EXTENT_REG &&
 876                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 877                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 878                         *off = min_key.offset;
 879                         btrfs_free_path(path);
 880                         return 0;
 881                 }
 882
 883                 if (min_key.offset == (u64)-1)
 884                         goto none;
 885
 886                 min_key.offset++;
 887                 btrfs_release_path(path);
 888         }
 889 none:
 890         btrfs_free_path(path);
 891         return -ENOENT;
 892 }
 893
 894 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 895 {
 896         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 897         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 898         struct extent_map *em;
 899         u64 len = PAGE_CACHE_SIZE;
 900
 901         /*
 902          * hopefully we have this extent in the tree already, try without
 903          * the full extent lock
 904          */
 905         read_lock(&em_tree->lock);
 906         em = lookup_extent_mapping(em_tree, start, len);
 907         read_unlock(&em_tree->lock);
 908
 909         if (!em) {
 910                 /* get the big lock and read metadata off disk */
 911                 lock_extent(io_tree, start, start + len - 1);
 912                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 913                 unlock_extent(io_tree, start, start + len - 1);
 914
 915                 if (IS_ERR(em))
 916                         return NULL;
 917         }
 918
 919         return em;
 920 }
 921
 922 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 923 {
 924         struct extent_map *next;
 925         bool ret = true;
 926
 927         /* this is the last extent */
 928         if (em->start + em->len >= i_size_read(inode))
 929                 return false;
 930
 931         next = defrag_lookup_extent(inode, em->start + em->len);
 932         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 933                 ret = false;
 934
 935         free_extent_map(next);
 936         return ret;
 937 }
 938
 939 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
 940                                u64 *last_len, u64 *skip, u64 *defrag_end,
 941                                int compress)
 942 {
 943         struct extent_map *em;
 944         int ret = 1;
 945         bool next_mergeable = true;
 946
 947         /*
 948          * make sure that once we start defragging an extent, we keep on
 949          * defragging it
 950          */
 951         if (start < *defrag_end)
 952                 return 1;
 953
 954         *skip = 0;
 955
 956         em = defrag_lookup_extent(inode, start);
 957         if (!em)
 958                 return 0;
 959
 960         /* this will cover holes, and inline extents */
 961         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 962                 ret = 0;
 963                 goto out;
 964         }
 965
 966         next_mergeable = defrag_check_next_extent(inode, em);
 967
 968         /*
 969          * we hit a real extent, if it is big or the next extent is not a
 970          * real extent, don't bother defragging it
 971          */
 972         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
 973             (em->len >= thresh || !next_mergeable))
 974                 ret = 0;
 975 out:
 976         /*
 977          * last_len ends up being a counter of how many bytes we've defragged.
 978          * every time we choose not to defrag an extent, we reset *last_len
 979          * so that the next tiny extent will force a defrag.
 980          *
 981          * The end result of this is that tiny extents before a single big
 982          * extent will force at least part of that big extent to be defragged.
 983          */
 984         if (ret) {
 985                 *defrag_end = extent_map_end(em);
 986         } else {
 987                 *last_len = 0;
 988                 *skip = extent_map_end(em);
 989                 *defrag_end = 0;
 990         }
 991
 992         free_extent_map(em);
 993         return ret;
 994 }
 995
 996 /*
 997  * it doesn't do much good to defrag one or two pages
 998  * at a time.  This pulls in a nice chunk of pages
 999  * to COW and defrag.
1000  *
1001  * It also makes sure the delalloc code has enough
1002  * dirty data to avoid making new small extents as part
1003  * of the defrag
1004  *
1005  * It's a good idea to start RA on this range
1006  * before calling this.
1007  */
1008 static int cluster_pages_for_defrag(struct inode *inode,
1009                                     struct page **pages,
1010                                     unsigned long start_index,
1011                                     int num_pages)
1012 {
1013         unsigned long file_end;
1014         u64 isize = i_size_read(inode);
1015         u64 page_start;
1016         u64 page_end;
1017         u64 page_cnt;
1018         int ret;
1019         int i;
1020         int i_done;
1021         struct btrfs_ordered_extent *ordered;
1022         struct extent_state *cached_state = NULL;
1023         struct extent_io_tree *tree;
1024         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1025
1026         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1027         if (!isize || start_index > file_end)
1028                 return 0;
1029
1030         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1031
1032         ret = btrfs_delalloc_reserve_space(inode,
1033                                            page_cnt << PAGE_CACHE_SHIFT);
1034         if (ret)
1035                 return ret;
1036         i_done = 0;
1037         tree = &BTRFS_I(inode)->io_tree;
1038
1039         /* step one, lock all the pages */
1040         for (i = 0; i < page_cnt; i++) {
1041                 struct page *page;
1042 again:
1043                 page = find_or_create_page(inode->i_mapping,
1044                                            start_index + i, mask);
1045                 if (!page)
1046                         break;
1047
1048                 page_start = page_offset(page);
1049                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1050                 while (1) {
1051                         lock_extent(tree, page_start, page_end);
1052                         ordered = btrfs_lookup_ordered_extent(inode,
1053                                                               page_start);
1054                         unlock_extent(tree, page_start, page_end);
1055                         if (!ordered)
1056                                 break;
1057
1058                         unlock_page(page);
1059                         btrfs_start_ordered_extent(inode, ordered, 1);
1060                         btrfs_put_ordered_extent(ordered);
1061                         lock_page(page);
1062                         /*
1063                          * we unlocked the page above, so we need check if
1064                          * it was released or not.
1065                          */
1066                         if (page->mapping != inode->i_mapping) {
1067                                 unlock_page(page);
1068                                 page_cache_release(page);
1069                                 goto again;
1070                         }
1071                 }
1072
1073                 if (!PageUptodate(page)) {
1074                         btrfs_readpage(NULL, page);
1075                         lock_page(page);
1076                         if (!PageUptodate(page)) {
1077                                 unlock_page(page);
1078                                 page_cache_release(page);
1079                                 ret = -EIO;
1080                                 break;
1081                         }
1082                 }
1083
1084                 if (page->mapping != inode->i_mapping) {
1085                         unlock_page(page);
1086                         page_cache_release(page);
1087                         goto again;
1088                 }
1089
1090                 pages[i] = page;
1091                 i_done++;
1092         }
1093         if (!i_done || ret)
1094                 goto out;
1095
1096         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1097                 goto out;
1098
1099         /*
1100          * so now we have a nice long stream of locked
1101          * and up to date pages, lets wait on them
1102          */
1103         for (i = 0; i < i_done; i++)
1104                 wait_on_page_writeback(pages[i]);
1105
1106         page_start = page_offset(pages[0]);
1107         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1108
1109         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1110                          page_start, page_end - 1, 0, &cached_state);
1111         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1112                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1113                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1114                           &cached_state, GFP_NOFS);
1115
1116         if (i_done != page_cnt) {
1117                 spin_lock(&BTRFS_I(inode)->lock);
1118                 BTRFS_I(inode)->outstanding_extents++;
1119                 spin_unlock(&BTRFS_I(inode)->lock);
1120                 btrfs_delalloc_release_space(inode,
1121                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1122         }
1123
1124
1125         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1126                           &cached_state, GFP_NOFS);
1127
1128         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1129                              page_start, page_end - 1, &cached_state,
1130                              GFP_NOFS);
1131
1132         for (i = 0; i < i_done; i++) {
1133                 clear_page_dirty_for_io(pages[i]);
1134                 ClearPageChecked(pages[i]);
1135                 set_page_extent_mapped(pages[i]);
1136                 set_page_dirty(pages[i]);
1137                 unlock_page(pages[i]);
1138                 page_cache_release(pages[i]);
1139         }
1140         return i_done;
1141 out:
1142         for (i = 0; i < i_done; i++) {
1143                 unlock_page(pages[i]);
1144                 page_cache_release(pages[i]);
1145         }
1146         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1147         return ret;
1148
1149 }
1150
1151 int btrfs_defrag_file(struct inode *inode, struct file *file,
1152                       struct btrfs_ioctl_defrag_range_args *range,
1153                       u64 newer_than, unsigned long max_to_defrag)
1154 {
1155         struct btrfs_root *root = BTRFS_I(inode)->root;
1156         struct file_ra_state *ra = NULL;
1157         unsigned long last_index;
1158         u64 isize = i_size_read(inode);
1159         u64 last_len = 0;
1160         u64 skip = 0;
1161         u64 defrag_end = 0;
1162         u64 newer_off = range->start;
1163         unsigned long i;
1164         unsigned long ra_index = 0;
1165         int ret;
1166         int defrag_count = 0;
1167         int compress_type = BTRFS_COMPRESS_ZLIB;
1168         int extent_thresh = range->extent_thresh;
1169         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1170         int cluster = max_cluster;
1171         u64 new_align = ~((u64)128 * 1024 - 1);
1172         struct page **pages = NULL;
1173
1174         if (isize == 0)
1175                 return 0;
1176
1177         if (range->start >= isize)
1178                 return -EINVAL;
1179
1180         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1181                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1182                         return -EINVAL;
1183                 if (range->compress_type)
1184                         compress_type = range->compress_type;
1185         }
1186
1187         if (extent_thresh == 0)
1188                 extent_thresh = 256 * 1024;
1189
1190         /*
1191          * if we were not given a file, allocate a readahead
1192          * context
1193          */
1194         if (!file) {
1195                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1196                 if (!ra)
1197                         return -ENOMEM;
1198                 file_ra_state_init(ra, inode->i_mapping);
1199         } else {
1200                 ra = &file->f_ra;
1201         }
1202
1203         pages = kmalloc(sizeof(struct page *) * max_cluster,
1204                         GFP_NOFS);
1205         if (!pages) {
1206                 ret = -ENOMEM;
1207                 goto out_ra;
1208         }
1209
1210         /* find the last page to defrag */
1211         if (range->start + range->len > range->start) {
1212                 last_index = min_t(u64, isize - 1,
1213                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1214         } else {
1215                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1216         }
1217
1218         if (newer_than) {
1219                 ret = find_new_extents(root, inode, newer_than,
1220                                        &newer_off, 64 * 1024);
1221                 if (!ret) {
1222                         range->start = newer_off;
1223                         /*
1224                          * we always align our defrag to help keep
1225                          * the extents in the file evenly spaced
1226                          */
1227                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1228                 } else
1229                         goto out_ra;
1230         } else {
1231                 i = range->start >> PAGE_CACHE_SHIFT;
1232         }
1233         if (!max_to_defrag)
1234                 max_to_defrag = last_index + 1;
1235
1236         /*
1237          * make writeback starts from i, so the defrag range can be
1238          * written sequentially.
1239          */
1240         if (i < inode->i_mapping->writeback_index)
1241                 inode->i_mapping->writeback_index = i;
1242
1243         while (i <= last_index && defrag_count < max_to_defrag &&
1244                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1245                 PAGE_CACHE_SHIFT)) {
1246                 /*
1247                  * make sure we stop running if someone unmounts
1248                  * the FS
1249                  */
1250                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1251                         break;
1252
1253                 if (btrfs_defrag_cancelled(root->fs_info)) {
1254                         printk(KERN_DEBUG "btrfs: defrag_file cancelled\n");
1255                         ret = -EAGAIN;
1256                         break;
1257                 }
1258
1259                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1260                                          extent_thresh, &last_len, &skip,
1261                                          &defrag_end, range->flags &
1262                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1263                         unsigned long next;
1264                         /*
1265                          * the should_defrag function tells us how much to skip
1266                          * bump our counter by the suggested amount
1267                          */
1268                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1269                         i = max(i + 1, next);
1270                         continue;
1271                 }
1272
1273                 if (!newer_than) {
1274                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1275                                    PAGE_CACHE_SHIFT) - i;
1276                         cluster = min(cluster, max_cluster);
1277                 } else {
1278                         cluster = max_cluster;
1279                 }
1280
1281                 if (i + cluster > ra_index) {
1282                         ra_index = max(i, ra_index);
1283                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1284                                        cluster);
1285                         ra_index += max_cluster;
1286                 }
1287
1288                 mutex_lock(&inode->i_mutex);
1289                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1290                         BTRFS_I(inode)->force_compress = compress_type;
1291                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1292                 if (ret < 0) {
1293                         mutex_unlock(&inode->i_mutex);
1294                         goto out_ra;
1295                 }
1296
1297                 defrag_count += ret;
1298                 balance_dirty_pages_ratelimited(inode->i_mapping);
1299                 mutex_unlock(&inode->i_mutex);
1300
1301                 if (newer_than) {
1302                         if (newer_off == (u64)-1)
1303                                 break;
1304
1305                         if (ret > 0)
1306                                 i += ret;
1307
1308                         newer_off = max(newer_off + 1,
1309                                         (u64)i << PAGE_CACHE_SHIFT);
1310
1311                         ret = find_new_extents(root, inode,
1312                                                newer_than, &newer_off,
1313                                                64 * 1024);
1314                         if (!ret) {
1315                                 range->start = newer_off;
1316                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1317                         } else {
1318                                 break;
1319                         }
1320                 } else {
1321                         if (ret > 0) {
1322                                 i += ret;
1323                                 last_len += ret << PAGE_CACHE_SHIFT;
1324                         } else {
1325                                 i++;
1326                                 last_len = 0;
1327                         }
1328                 }
1329         }
1330
1331         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1332                 filemap_flush(inode->i_mapping);
1333
1334         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1335                 /* the filemap_flush will queue IO into the worker threads, but
1336                  * we have to make sure the IO is actually started and that
1337                  * ordered extents get created before we return
1338                  */
1339                 atomic_inc(&root->fs_info->async_submit_draining);
1340                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1341                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1342                         wait_event(root->fs_info->async_submit_wait,
1343                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1344                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1345                 }
1346                 atomic_dec(&root->fs_info->async_submit_draining);
1347         }
1348
1349         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1350                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1351         }
1352
1353         ret = defrag_count;
1354
1355 out_ra:
1356         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1357                 mutex_lock(&inode->i_mutex);
1358                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1359                 mutex_unlock(&inode->i_mutex);
1360         }
1361         if (!file)
1362                 kfree(ra);
1363         kfree(pages);
1364         return ret;
1365 }
1366
1367 static noinline int btrfs_ioctl_resize(struct file *file,
1368                                         void __user *arg)
1369 {
1370         u64 new_size;
1371         u64 old_size;
1372         u64 devid = 1;
1373         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1374         struct btrfs_ioctl_vol_args *vol_args;
1375         struct btrfs_trans_handle *trans;
1376         struct btrfs_device *device = NULL;
1377         char *sizestr;
1378         char *devstr = NULL;
1379         int ret = 0;
1380         int mod = 0;
1381
1382         if (!capable(CAP_SYS_ADMIN))
1383                 return -EPERM;
1384
1385         ret = mnt_want_write_file(file);
1386         if (ret)
1387                 return ret;
1388
1389         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1390                         1)) {
1391                 mnt_drop_write_file(file);
1392                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1393         }
1394
1395         mutex_lock(&root->fs_info->volume_mutex);
1396         vol_args = memdup_user(arg, sizeof(*vol_args));
1397         if (IS_ERR(vol_args)) {
1398                 ret = PTR_ERR(vol_args);
1399                 goto out;
1400         }
1401
1402         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1403
1404         sizestr = vol_args->name;
1405         devstr = strchr(sizestr, ':');
1406         if (devstr) {
1407                 char *end;
1408                 sizestr = devstr + 1;
1409                 *devstr = '\0';
1410                 devstr = vol_args->name;
1411                 devid = simple_strtoull(devstr, &end, 10);
1412                 if (!devid) {
1413                         ret = -EINVAL;
1414                         goto out_free;
1415                 }
1416                 printk(KERN_INFO "btrfs: resizing devid %llu\n", devid);
1417         }
1418
1419         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1420         if (!device) {
1421                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1422                        devid);
1423                 ret = -ENODEV;
1424                 goto out_free;
1425         }
1426
1427         if (!device->writeable) {
1428                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1429                        "readonly device %llu\n",
1430                        devid);
1431                 ret = -EPERM;
1432                 goto out_free;
1433         }
1434
1435         if (!strcmp(sizestr, "max"))
1436                 new_size = device->bdev->bd_inode->i_size;
1437         else {
1438                 if (sizestr[0] == '-') {
1439                         mod = -1;
1440                         sizestr++;
1441                 } else if (sizestr[0] == '+') {
1442                         mod = 1;
1443                         sizestr++;
1444                 }
1445                 new_size = memparse(sizestr, NULL);
1446                 if (new_size == 0) {
1447                         ret = -EINVAL;
1448                         goto out_free;
1449                 }
1450         }
1451
1452         if (device->is_tgtdev_for_dev_replace) {
1453                 ret = -EPERM;
1454                 goto out_free;
1455         }
1456
1457         old_size = device->total_bytes;
1458
1459         if (mod < 0) {
1460                 if (new_size > old_size) {
1461                         ret = -EINVAL;
1462                         goto out_free;
1463                 }
1464                 new_size = old_size - new_size;
1465         } else if (mod > 0) {
1466                 new_size = old_size + new_size;
1467         }
1468
1469         if (new_size < 256 * 1024 * 1024) {
1470                 ret = -EINVAL;
1471                 goto out_free;
1472         }
1473         if (new_size > device->bdev->bd_inode->i_size) {
1474                 ret = -EFBIG;
1475                 goto out_free;
1476         }
1477
1478         do_div(new_size, root->sectorsize);
1479         new_size *= root->sectorsize;
1480
1481         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1482                       rcu_str_deref(device->name), new_size);
1483
1484         if (new_size > old_size) {
1485                 trans = btrfs_start_transaction(root, 0);
1486                 if (IS_ERR(trans)) {
1487                         ret = PTR_ERR(trans);
1488                         goto out_free;
1489                 }
1490                 ret = btrfs_grow_device(trans, device, new_size);
1491                 btrfs_commit_transaction(trans, root);
1492         } else if (new_size < old_size) {
1493                 ret = btrfs_shrink_device(device, new_size);
1494         } /* equal, nothing need to do */
1495
1496 out_free:
1497         kfree(vol_args);
1498 out:
1499         mutex_unlock(&root->fs_info->volume_mutex);
1500         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1501         mnt_drop_write_file(file);
1502         return ret;
1503 }
1504
1505 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1506                                 char *name, unsigned long fd, int subvol,
1507                                 u64 *transid, bool readonly,
1508                                 struct btrfs_qgroup_inherit *inherit)
1509 {
1510         int namelen;
1511         int ret = 0;
1512
1513         ret = mnt_want_write_file(file);
1514         if (ret)
1515                 goto out;
1516
1517         namelen = strlen(name);
1518         if (strchr(name, '/')) {
1519                 ret = -EINVAL;
1520                 goto out_drop_write;
1521         }
1522
1523         if (name[0] == '.' &&
1524            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1525                 ret = -EEXIST;
1526                 goto out_drop_write;
1527         }
1528
1529         if (subvol) {
1530                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1531                                      NULL, transid, readonly, inherit);
1532         } else {
1533                 struct fd src = fdget(fd);
1534                 struct inode *src_inode;
1535                 if (!src.file) {
1536                         ret = -EINVAL;
1537                         goto out_drop_write;
1538                 }
1539
1540                 src_inode = file_inode(src.file);
1541                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1542                         printk(KERN_INFO "btrfs: Snapshot src from "
1543                                "another FS\n");
1544                         ret = -EINVAL;
1545                 } else {
1546                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1547                                              BTRFS_I(src_inode)->root,
1548                                              transid, readonly, inherit);
1549                 }
1550                 fdput(src);
1551         }
1552 out_drop_write:
1553         mnt_drop_write_file(file);
1554 out:
1555         return ret;
1556 }
1557
1558 static noinline int btrfs_ioctl_snap_create(struct file *file,
1559                                             void __user *arg, int subvol)
1560 {
1561         struct btrfs_ioctl_vol_args *vol_args;
1562         int ret;
1563
1564         vol_args = memdup_user(arg, sizeof(*vol_args));
1565         if (IS_ERR(vol_args))
1566                 return PTR_ERR(vol_args);
1567         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1568
1569         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1570                                               vol_args->fd, subvol,
1571                                               NULL, false, NULL);
1572
1573         kfree(vol_args);
1574         return ret;
1575 }
1576
1577 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1578                                                void __user *arg, int subvol)
1579 {
1580         struct btrfs_ioctl_vol_args_v2 *vol_args;
1581         int ret;
1582         u64 transid = 0;
1583         u64 *ptr = NULL;
1584         bool readonly = false;
1585         struct btrfs_qgroup_inherit *inherit = NULL;
1586
1587         vol_args = memdup_user(arg, sizeof(*vol_args));
1588         if (IS_ERR(vol_args))
1589                 return PTR_ERR(vol_args);
1590         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1591
1592         if (vol_args->flags &
1593             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1594               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1595                 ret = -EOPNOTSUPP;
1596                 goto out;
1597         }
1598
1599         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1600                 ptr = &transid;
1601         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1602                 readonly = true;
1603         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1604                 if (vol_args->size > PAGE_CACHE_SIZE) {
1605                         ret = -EINVAL;
1606                         goto out;
1607                 }
1608                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1609                 if (IS_ERR(inherit)) {
1610                         ret = PTR_ERR(inherit);
1611                         goto out;
1612                 }
1613         }
1614
1615         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1616                                               vol_args->fd, subvol, ptr,
1617                                               readonly, inherit);
1618
1619         if (ret == 0 && ptr &&
1620             copy_to_user(arg +
1621                          offsetof(struct btrfs_ioctl_vol_args_v2,
1622                                   transid), ptr, sizeof(*ptr)))
1623                 ret = -EFAULT;
1624 out:
1625         kfree(vol_args);
1626         kfree(inherit);
1627         return ret;
1628 }
1629
1630 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1631                                                 void __user *arg)
1632 {
1633         struct inode *inode = file_inode(file);
1634         struct btrfs_root *root = BTRFS_I(inode)->root;
1635         int ret = 0;
1636         u64 flags = 0;
1637
1638         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1639                 return -EINVAL;
1640
1641         down_read(&root->fs_info->subvol_sem);
1642         if (btrfs_root_readonly(root))
1643                 flags |= BTRFS_SUBVOL_RDONLY;
1644         up_read(&root->fs_info->subvol_sem);
1645
1646         if (copy_to_user(arg, &flags, sizeof(flags)))
1647                 ret = -EFAULT;
1648
1649         return ret;
1650 }
1651
1652 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1653                                               void __user *arg)
1654 {
1655         struct inode *inode = file_inode(file);
1656         struct btrfs_root *root = BTRFS_I(inode)->root;
1657         struct btrfs_trans_handle *trans;
1658         u64 root_flags;
1659         u64 flags;
1660         int ret = 0;
1661
1662         ret = mnt_want_write_file(file);
1663         if (ret)
1664                 goto out;
1665
1666         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1667                 ret = -EINVAL;
1668                 goto out_drop_write;
1669         }
1670
1671         if (copy_from_user(&flags, arg, sizeof(flags))) {
1672                 ret = -EFAULT;
1673                 goto out_drop_write;
1674         }
1675
1676         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1677                 ret = -EINVAL;
1678                 goto out_drop_write;
1679         }
1680
1681         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1682                 ret = -EOPNOTSUPP;
1683                 goto out_drop_write;
1684         }
1685
1686         if (!inode_owner_or_capable(inode)) {
1687                 ret = -EACCES;
1688                 goto out_drop_write;
1689         }
1690
1691         down_write(&root->fs_info->subvol_sem);
1692
1693         /* nothing to do */
1694         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1695                 goto out_drop_sem;
1696
1697         root_flags = btrfs_root_flags(&root->root_item);
1698         if (flags & BTRFS_SUBVOL_RDONLY)
1699                 btrfs_set_root_flags(&root->root_item,
1700                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1701         else
1702                 btrfs_set_root_flags(&root->root_item,
1703                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1704
1705         trans = btrfs_start_transaction(root, 1);
1706         if (IS_ERR(trans)) {
1707                 ret = PTR_ERR(trans);
1708                 goto out_reset;
1709         }
1710
1711         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1712                                 &root->root_key, &root->root_item);
1713
1714         btrfs_commit_transaction(trans, root);
1715 out_reset:
1716         if (ret)
1717                 btrfs_set_root_flags(&root->root_item, root_flags);
1718 out_drop_sem:
1719         up_write(&root->fs_info->subvol_sem);
1720 out_drop_write:
1721         mnt_drop_write_file(file);
1722 out:
1723         return ret;
1724 }
1725
1726 /*
1727  * helper to check if the subvolume references other subvolumes
1728  */
1729 static noinline int may_destroy_subvol(struct btrfs_root *root)
1730 {
1731         struct btrfs_path *path;
1732         struct btrfs_dir_item *di;
1733         struct btrfs_key key;
1734         u64 dir_id;
1735         int ret;
1736
1737         path = btrfs_alloc_path();
1738         if (!path)
1739                 return -ENOMEM;
1740
1741         /* Make sure this root isn't set as the default subvol */
1742         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1743         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1744                                    dir_id, "default", 7, 0);
1745         if (di && !IS_ERR(di)) {
1746                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1747                 if (key.objectid == root->root_key.objectid) {
1748                         ret = -ENOTEMPTY;
1749                         goto out;
1750                 }
1751                 btrfs_release_path(path);
1752         }
1753
1754         key.objectid = root->root_key.objectid;
1755         key.type = BTRFS_ROOT_REF_KEY;
1756         key.offset = (u64)-1;
1757
1758         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1759                                 &key, path, 0, 0);
1760         if (ret < 0)
1761                 goto out;
1762         BUG_ON(ret == 0);
1763
1764         ret = 0;
1765         if (path->slots[0] > 0) {
1766                 path->slots[0]--;
1767                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1768                 if (key.objectid == root->root_key.objectid &&
1769                     key.type == BTRFS_ROOT_REF_KEY)
1770                         ret = -ENOTEMPTY;
1771         }
1772 out:
1773         btrfs_free_path(path);
1774         return ret;
1775 }
1776
1777 static noinline int key_in_sk(struct btrfs_key *key,
1778                               struct btrfs_ioctl_search_key *sk)
1779 {
1780         struct btrfs_key test;
1781         int ret;
1782
1783         test.objectid = sk->min_objectid;
1784         test.type = sk->min_type;
1785         test.offset = sk->min_offset;
1786
1787         ret = btrfs_comp_cpu_keys(key, &test);
1788         if (ret < 0)
1789                 return 0;
1790
1791         test.objectid = sk->max_objectid;
1792         test.type = sk->max_type;
1793         test.offset = sk->max_offset;
1794
1795         ret = btrfs_comp_cpu_keys(key, &test);
1796         if (ret > 0)
1797                 return 0;
1798         return 1;
1799 }
1800
1801 static noinline int copy_to_sk(struct btrfs_root *root,
1802                                struct btrfs_path *path,
1803                                struct btrfs_key *key,
1804                                struct btrfs_ioctl_search_key *sk,
1805                                char *buf,
1806                                unsigned long *sk_offset,
1807                                int *num_found)
1808 {
1809         u64 found_transid;
1810         struct extent_buffer *leaf;
1811         struct btrfs_ioctl_search_header sh;
1812         unsigned long item_off;
1813         unsigned long item_len;
1814         int nritems;
1815         int i;
1816         int slot;
1817         int ret = 0;
1818
1819         leaf = path->nodes[0];
1820         slot = path->slots[0];
1821         nritems = btrfs_header_nritems(leaf);
1822
1823         if (btrfs_header_generation(leaf) > sk->max_transid) {
1824                 i = nritems;
1825                 goto advance_key;
1826         }
1827         found_transid = btrfs_header_generation(leaf);
1828
1829         for (i = slot; i < nritems; i++) {
1830                 item_off = btrfs_item_ptr_offset(leaf, i);
1831                 item_len = btrfs_item_size_nr(leaf, i);
1832
1833                 btrfs_item_key_to_cpu(leaf, key, i);
1834                 if (!key_in_sk(key, sk))
1835                         continue;
1836
1837                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1838                         item_len = 0;
1839
1840                 if (sizeof(sh) + item_len + *sk_offset >
1841                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1842                         ret = 1;
1843                         goto overflow;
1844                 }
1845
1846                 sh.objectid = key->objectid;
1847                 sh.offset = key->offset;
1848                 sh.type = key->type;
1849                 sh.len = item_len;
1850                 sh.transid = found_transid;
1851
1852                 /* copy search result header */
1853                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1854                 *sk_offset += sizeof(sh);
1855
1856                 if (item_len) {
1857                         char *p = buf + *sk_offset;
1858                         /* copy the item */
1859                         read_extent_buffer(leaf, p,
1860                                            item_off, item_len);
1861                         *sk_offset += item_len;
1862                 }
1863                 (*num_found)++;
1864
1865                 if (*num_found >= sk->nr_items)
1866                         break;
1867         }
1868 advance_key:
1869         ret = 0;
1870         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1871                 key->offset++;
1872         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1873                 key->offset = 0;
1874                 key->type++;
1875         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1876                 key->offset = 0;
1877                 key->type = 0;
1878                 key->objectid++;
1879         } else
1880                 ret = 1;
1881 overflow:
1882         return ret;
1883 }
1884
1885 static noinline int search_ioctl(struct inode *inode,
1886                                  struct btrfs_ioctl_search_args *args)
1887 {
1888         struct btrfs_root *root;
1889         struct btrfs_key key;
1890         struct btrfs_path *path;
1891         struct btrfs_ioctl_search_key *sk = &args->key;
1892         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1893         int ret;
1894         int num_found = 0;
1895         unsigned long sk_offset = 0;
1896
1897         path = btrfs_alloc_path();
1898         if (!path)
1899                 return -ENOMEM;
1900
1901         if (sk->tree_id == 0) {
1902                 /* search the root of the inode that was passed */
1903                 root = BTRFS_I(inode)->root;
1904         } else {
1905                 key.objectid = sk->tree_id;
1906                 key.type = BTRFS_ROOT_ITEM_KEY;
1907                 key.offset = (u64)-1;
1908                 root = btrfs_read_fs_root_no_name(info, &key);
1909                 if (IS_ERR(root)) {
1910                         printk(KERN_ERR "could not find root %llu\n",
1911                                sk->tree_id);
1912                         btrfs_free_path(path);
1913                         return -ENOENT;
1914                 }
1915         }
1916
1917         key.objectid = sk->min_objectid;
1918         key.type = sk->min_type;
1919         key.offset = sk->min_offset;
1920
1921         path->keep_locks = 1;
1922
1923         while(1) {
1924                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
1925                 if (ret != 0) {
1926                         if (ret > 0)
1927                                 ret = 0;
1928                         goto err;
1929                 }
1930                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1931                                  &sk_offset, &num_found);
1932                 btrfs_release_path(path);
1933                 if (ret || num_found >= sk->nr_items)
1934                         break;
1935
1936         }
1937         ret = 0;
1938 err:
1939         sk->nr_items = num_found;
1940         btrfs_free_path(path);
1941         return ret;
1942 }
1943
1944 static noinline int btrfs_ioctl_tree_search(struct file *file,
1945                                            void __user *argp)
1946 {
1947          struct btrfs_ioctl_search_args *args;
1948          struct inode *inode;
1949          int ret;
1950
1951         if (!capable(CAP_SYS_ADMIN))
1952                 return -EPERM;
1953
1954         args = memdup_user(argp, sizeof(*args));
1955         if (IS_ERR(args))
1956                 return PTR_ERR(args);
1957
1958         inode = file_inode(file);
1959         ret = search_ioctl(inode, args);
1960         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1961                 ret = -EFAULT;
1962         kfree(args);
1963         return ret;
1964 }
1965
1966 /*
1967  * Search INODE_REFs to identify path name of 'dirid' directory
1968  * in a 'tree_id' tree. and sets path name to 'name'.
1969  */
1970 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1971                                 u64 tree_id, u64 dirid, char *name)
1972 {
1973         struct btrfs_root *root;
1974         struct btrfs_key key;
1975         char *ptr;
1976         int ret = -1;
1977         int slot;
1978         int len;
1979         int total_len = 0;
1980         struct btrfs_inode_ref *iref;
1981         struct extent_buffer *l;
1982         struct btrfs_path *path;
1983
1984         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1985                 name[0]='\0';
1986                 return 0;
1987         }
1988
1989         path = btrfs_alloc_path();
1990         if (!path)
1991                 return -ENOMEM;
1992
1993         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1994
1995         key.objectid = tree_id;
1996         key.type = BTRFS_ROOT_ITEM_KEY;
1997         key.offset = (u64)-1;
1998         root = btrfs_read_fs_root_no_name(info, &key);
1999         if (IS_ERR(root)) {
2000                 printk(KERN_ERR "could not find root %llu\n", tree_id);
2001                 ret = -ENOENT;
2002                 goto out;
2003         }
2004
2005         key.objectid = dirid;
2006         key.type = BTRFS_INODE_REF_KEY;
2007         key.offset = (u64)-1;
2008
2009         while(1) {
2010                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2011                 if (ret < 0)
2012                         goto out;
2013                 else if (ret > 0) {
2014                         ret = btrfs_previous_item(root, path, dirid,
2015                                                   BTRFS_INODE_REF_KEY);
2016                         if (ret < 0)
2017                                 goto out;
2018                         else if (ret > 0) {
2019                                 ret = -ENOENT;
2020                                 goto out;
2021                         }
2022                 }
2023
2024                 l = path->nodes[0];
2025                 slot = path->slots[0];
2026                 btrfs_item_key_to_cpu(l, &key, slot);
2027
2028                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2029                 len = btrfs_inode_ref_name_len(l, iref);
2030                 ptr -= len + 1;
2031                 total_len += len + 1;
2032                 if (ptr < name) {
2033                         ret = -ENAMETOOLONG;
2034                         goto out;
2035                 }
2036
2037                 *(ptr + len) = '/';
2038                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
2039
2040                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2041                         break;
2042
2043                 btrfs_release_path(path);
2044                 key.objectid = key.offset;
2045                 key.offset = (u64)-1;
2046                 dirid = key.objectid;
2047         }
2048         memmove(name, ptr, total_len);
2049         name[total_len]='\0';
2050         ret = 0;
2051 out:
2052         btrfs_free_path(path);
2053         return ret;
2054 }
2055
2056 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2057                                            void __user *argp)
2058 {
2059          struct btrfs_ioctl_ino_lookup_args *args;
2060          struct inode *inode;
2061          int ret;
2062
2063         if (!capable(CAP_SYS_ADMIN))
2064                 return -EPERM;
2065
2066         args = memdup_user(argp, sizeof(*args));
2067         if (IS_ERR(args))
2068                 return PTR_ERR(args);
2069
2070         inode = file_inode(file);
2071
2072         if (args->treeid == 0)
2073                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2074
2075         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2076                                         args->treeid, args->objectid,
2077                                         args->name);
2078
2079         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2080                 ret = -EFAULT;
2081
2082         kfree(args);
2083         return ret;
2084 }
2085
2086 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2087                                              void __user *arg)
2088 {
2089         struct dentry *parent = fdentry(file);
2090         struct dentry *dentry;
2091         struct inode *dir = parent->d_inode;
2092         struct inode *inode;
2093         struct btrfs_root *root = BTRFS_I(dir)->root;
2094         struct btrfs_root *dest = NULL;
2095         struct btrfs_ioctl_vol_args *vol_args;
2096         struct btrfs_trans_handle *trans;
2097         struct btrfs_block_rsv block_rsv;
2098         u64 qgroup_reserved;
2099         int namelen;
2100         int ret;
2101         int err = 0;
2102
2103         vol_args = memdup_user(arg, sizeof(*vol_args));
2104         if (IS_ERR(vol_args))
2105                 return PTR_ERR(vol_args);
2106
2107         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2108         namelen = strlen(vol_args->name);
2109         if (strchr(vol_args->name, '/') ||
2110             strncmp(vol_args->name, "..", namelen) == 0) {
2111                 err = -EINVAL;
2112                 goto out;
2113         }
2114
2115         err = mnt_want_write_file(file);
2116         if (err)
2117                 goto out;
2118
2119         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2120         if (err == -EINTR)
2121                 goto out;
2122         dentry = lookup_one_len(vol_args->name, parent, namelen);
2123         if (IS_ERR(dentry)) {
2124                 err = PTR_ERR(dentry);
2125                 goto out_unlock_dir;
2126         }
2127
2128         if (!dentry->d_inode) {
2129                 err = -ENOENT;
2130                 goto out_dput;
2131         }
2132
2133         inode = dentry->d_inode;
2134         dest = BTRFS_I(inode)->root;
2135         if (!capable(CAP_SYS_ADMIN)){
2136                 /*
2137                  * Regular user.  Only allow this with a special mount
2138                  * option, when the user has write+exec access to the
2139                  * subvol root, and when rmdir(2) would have been
2140                  * allowed.
2141                  *
2142                  * Note that this is _not_ check that the subvol is
2143                  * empty or doesn't contain data that we wouldn't
2144                  * otherwise be able to delete.
2145                  *
2146                  * Users who want to delete empty subvols should try
2147                  * rmdir(2).
2148                  */
2149                 err = -EPERM;
2150                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2151                         goto out_dput;
2152
2153                 /*
2154                  * Do not allow deletion if the parent dir is the same
2155                  * as the dir to be deleted.  That means the ioctl
2156                  * must be called on the dentry referencing the root
2157                  * of the subvol, not a random directory contained
2158                  * within it.
2159                  */
2160                 err = -EINVAL;
2161                 if (root == dest)
2162                         goto out_dput;
2163
2164                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2165                 if (err)
2166                         goto out_dput;
2167         }
2168
2169         /* check if subvolume may be deleted by a user */
2170         err = btrfs_may_delete(dir, dentry, 1);
2171         if (err)
2172                 goto out_dput;
2173
2174         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2175                 err = -EINVAL;
2176                 goto out_dput;
2177         }
2178
2179         mutex_lock(&inode->i_mutex);
2180         err = d_invalidate(dentry);
2181         if (err)
2182                 goto out_unlock;
2183
2184         down_write(&root->fs_info->subvol_sem);
2185
2186         err = may_destroy_subvol(dest);
2187         if (err)
2188                 goto out_up_write;
2189
2190         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2191         /*
2192          * One for dir inode, two for dir entries, two for root
2193          * ref/backref.
2194          */
2195         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2196                                                5, &qgroup_reserved, true);
2197         if (err)
2198                 goto out_up_write;
2199
2200         trans = btrfs_start_transaction(root, 0);
2201         if (IS_ERR(trans)) {
2202                 err = PTR_ERR(trans);
2203                 goto out_release;
2204         }
2205         trans->block_rsv = &block_rsv;
2206         trans->bytes_reserved = block_rsv.size;
2207
2208         ret = btrfs_unlink_subvol(trans, root, dir,
2209                                 dest->root_key.objectid,
2210                                 dentry->d_name.name,
2211                                 dentry->d_name.len);
2212         if (ret) {
2213                 err = ret;
2214                 btrfs_abort_transaction(trans, root, ret);
2215                 goto out_end_trans;
2216         }
2217
2218         btrfs_record_root_in_trans(trans, dest);
2219
2220         memset(&dest->root_item.drop_progress, 0,
2221                 sizeof(dest->root_item.drop_progress));
2222         dest->root_item.drop_level = 0;
2223         btrfs_set_root_refs(&dest->root_item, 0);
2224
2225         if (!xchg(&dest->orphan_item_inserted, 1)) {
2226                 ret = btrfs_insert_orphan_item(trans,
2227                                         root->fs_info->tree_root,
2228                                         dest->root_key.objectid);
2229                 if (ret) {
2230                         btrfs_abort_transaction(trans, root, ret);
2231                         err = ret;
2232                         goto out_end_trans;
2233                 }
2234         }
2235
2236         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2237                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2238                                   dest->root_key.objectid);
2239         if (ret && ret != -ENOENT) {
2240                 btrfs_abort_transaction(trans, root, ret);
2241                 err = ret;
2242                 goto out_end_trans;
2243         }
2244         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2245                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2246                                           dest->root_item.received_uuid,
2247                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2248                                           dest->root_key.objectid);
2249                 if (ret && ret != -ENOENT) {
2250                         btrfs_abort_transaction(trans, root, ret);
2251                         err = ret;
2252                         goto out_end_trans;
2253                 }
2254         }
2255
2256 out_end_trans:
2257         trans->block_rsv = NULL;
2258         trans->bytes_reserved = 0;
2259         ret = btrfs_end_transaction(trans, root);
2260         if (ret && !err)
2261                 err = ret;
2262         inode->i_flags |= S_DEAD;
2263 out_release:
2264         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2265 out_up_write:
2266         up_write(&root->fs_info->subvol_sem);
2267 out_unlock:
2268         mutex_unlock(&inode->i_mutex);
2269         if (!err) {
2270                 shrink_dcache_sb(root->fs_info->sb);
2271                 btrfs_invalidate_inodes(dest);
2272                 d_delete(dentry);
2273
2274                 /* the last ref */
2275                 if (dest->cache_inode) {
2276                         iput(dest->cache_inode);
2277                         dest->cache_inode = NULL;
2278                 }
2279         }
2280 out_dput:
2281         dput(dentry);
2282 out_unlock_dir:
2283         mutex_unlock(&dir->i_mutex);
2284         mnt_drop_write_file(file);
2285 out:
2286         kfree(vol_args);
2287         return err;
2288 }
2289
2290 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2291 {
2292         struct inode *inode = file_inode(file);
2293         struct btrfs_root *root = BTRFS_I(inode)->root;
2294         struct btrfs_ioctl_defrag_range_args *range;
2295         int ret;
2296
2297         ret = mnt_want_write_file(file);
2298         if (ret)
2299                 return ret;
2300
2301         if (btrfs_root_readonly(root)) {
2302                 ret = -EROFS;
2303                 goto out;
2304         }
2305
2306         switch (inode->i_mode & S_IFMT) {
2307         case S_IFDIR:
2308                 if (!capable(CAP_SYS_ADMIN)) {
2309                         ret = -EPERM;
2310                         goto out;
2311                 }
2312                 ret = btrfs_defrag_root(root);
2313                 if (ret)
2314                         goto out;
2315                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2316                 break;
2317         case S_IFREG:
2318                 if (!(file->f_mode & FMODE_WRITE)) {
2319                         ret = -EINVAL;
2320                         goto out;
2321                 }
2322
2323                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2324                 if (!range) {
2325                         ret = -ENOMEM;
2326                         goto out;
2327                 }
2328
2329                 if (argp) {
2330                         if (copy_from_user(range, argp,
2331                                            sizeof(*range))) {
2332                                 ret = -EFAULT;
2333                                 kfree(range);
2334                                 goto out;
2335                         }
2336                         /* compression requires us to start the IO */
2337                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2338                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2339                                 range->extent_thresh = (u32)-1;
2340                         }
2341                 } else {
2342                         /* the rest are all set to zero by kzalloc */
2343                         range->len = (u64)-1;
2344                 }
2345                 ret = btrfs_defrag_file(file_inode(file), file,
2346                                         range, 0, 0);
2347                 if (ret > 0)
2348                         ret = 0;
2349                 kfree(range);
2350                 break;
2351         default:
2352                 ret = -EINVAL;
2353         }
2354 out:
2355         mnt_drop_write_file(file);
2356         return ret;
2357 }
2358
2359 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2360 {
2361         struct btrfs_ioctl_vol_args *vol_args;
2362         int ret;
2363
2364         if (!capable(CAP_SYS_ADMIN))
2365                 return -EPERM;
2366
2367         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2368                         1)) {
2369                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2370         }
2371
2372         mutex_lock(&root->fs_info->volume_mutex);
2373         vol_args = memdup_user(arg, sizeof(*vol_args));
2374         if (IS_ERR(vol_args)) {
2375                 ret = PTR_ERR(vol_args);
2376                 goto out;
2377         }
2378
2379         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2380         ret = btrfs_init_new_device(root, vol_args->name);
2381
2382         kfree(vol_args);
2383 out:
2384         mutex_unlock(&root->fs_info->volume_mutex);
2385         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2386         return ret;
2387 }
2388
2389 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2390 {
2391         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2392         struct btrfs_ioctl_vol_args *vol_args;
2393         int ret;
2394
2395         if (!capable(CAP_SYS_ADMIN))
2396                 return -EPERM;
2397
2398         ret = mnt_want_write_file(file);
2399         if (ret)
2400                 return ret;
2401
2402         vol_args = memdup_user(arg, sizeof(*vol_args));
2403         if (IS_ERR(vol_args)) {
2404                 ret = PTR_ERR(vol_args);
2405                 goto out;
2406         }
2407
2408         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2409
2410         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2411                         1)) {
2412                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2413                 goto out;
2414         }
2415
2416         mutex_lock(&root->fs_info->volume_mutex);
2417         ret = btrfs_rm_device(root, vol_args->name);
2418         mutex_unlock(&root->fs_info->volume_mutex);
2419         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2420
2421 out:
2422         kfree(vol_args);
2423         mnt_drop_write_file(file);
2424         return ret;
2425 }
2426
2427 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2428 {
2429         struct btrfs_ioctl_fs_info_args *fi_args;
2430         struct btrfs_device *device;
2431         struct btrfs_device *next;
2432         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2433         int ret = 0;
2434
2435         if (!capable(CAP_SYS_ADMIN))
2436                 return -EPERM;
2437
2438         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2439         if (!fi_args)
2440                 return -ENOMEM;
2441
2442         mutex_lock(&fs_devices->device_list_mutex);
2443         fi_args->num_devices = fs_devices->num_devices;
2444         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2445
2446         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2447                 if (device->devid > fi_args->max_id)
2448                         fi_args->max_id = device->devid;
2449         }
2450         mutex_unlock(&fs_devices->device_list_mutex);
2451
2452         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2453                 ret = -EFAULT;
2454
2455         kfree(fi_args);
2456         return ret;
2457 }
2458
2459 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2460 {
2461         struct btrfs_ioctl_dev_info_args *di_args;
2462         struct btrfs_device *dev;
2463         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2464         int ret = 0;
2465         char *s_uuid = NULL;
2466
2467         if (!capable(CAP_SYS_ADMIN))
2468                 return -EPERM;
2469
2470         di_args = memdup_user(arg, sizeof(*di_args));
2471         if (IS_ERR(di_args))
2472                 return PTR_ERR(di_args);
2473
2474         if (!btrfs_is_empty_uuid(di_args->uuid))
2475                 s_uuid = di_args->uuid;
2476
2477         mutex_lock(&fs_devices->device_list_mutex);
2478         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2479
2480         if (!dev) {
2481                 ret = -ENODEV;
2482                 goto out;
2483         }
2484
2485         di_args->devid = dev->devid;
2486         di_args->bytes_used = dev->bytes_used;
2487         di_args->total_bytes = dev->total_bytes;
2488         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2489         if (dev->name) {
2490                 struct rcu_string *name;
2491
2492                 rcu_read_lock();
2493                 name = rcu_dereference(dev->name);
2494                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2495                 rcu_read_unlock();
2496                 di_args->path[sizeof(di_args->path) - 1] = 0;
2497         } else {
2498                 di_args->path[0] = '\0';
2499         }
2500
2501 out:
2502         mutex_unlock(&fs_devices->device_list_mutex);
2503         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2504                 ret = -EFAULT;
2505
2506         kfree(di_args);
2507         return ret;
2508 }
2509
2510 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2511 {
2512         struct page *page;
2513         pgoff_t index;
2514         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2515
2516         index = off >> PAGE_CACHE_SHIFT;
2517
2518         page = grab_cache_page(inode->i_mapping, index);
2519         if (!page)
2520                 return NULL;
2521
2522         if (!PageUptodate(page)) {
2523                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2524                                                  0))
2525                         return NULL;
2526                 lock_page(page);
2527                 if (!PageUptodate(page)) {
2528                         unlock_page(page);
2529                         page_cache_release(page);
2530                         return NULL;
2531                 }
2532         }
2533         unlock_page(page);
2534
2535         return page;
2536 }
2537
2538 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2539 {
2540         /* do any pending delalloc/csum calc on src, one way or
2541            another, and lock file content */
2542         while (1) {
2543                 struct btrfs_ordered_extent *ordered;
2544                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2545                 ordered = btrfs_lookup_first_ordered_extent(inode,
2546                                                             off + len - 1);
2547                 if (!ordered &&
2548                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2549                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2550                         break;
2551                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2552                 if (ordered)
2553                         btrfs_put_ordered_extent(ordered);
2554                 btrfs_wait_ordered_range(inode, off, len);
2555         }
2556 }
2557
2558 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2559                                 struct inode *inode2, u64 loff2, u64 len)
2560 {
2561         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2562         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2563
2564         mutex_unlock(&inode1->i_mutex);
2565         mutex_unlock(&inode2->i_mutex);
2566 }
2567
2568 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2569                               struct inode *inode2, u64 loff2, u64 len)
2570 {
2571         if (inode1 < inode2) {
2572                 swap(inode1, inode2);
2573                 swap(loff1, loff2);
2574         }
2575
2576         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2577         lock_extent_range(inode1, loff1, len);
2578         if (inode1 != inode2) {
2579                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2580                 lock_extent_range(inode2, loff2, len);
2581         }
2582 }
2583
2584 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2585                           u64 dst_loff, u64 len)
2586 {
2587         int ret = 0;
2588         struct page *src_page, *dst_page;
2589         unsigned int cmp_len = PAGE_CACHE_SIZE;
2590         void *addr, *dst_addr;
2591
2592         while (len) {
2593                 if (len < PAGE_CACHE_SIZE)
2594                         cmp_len = len;
2595
2596                 src_page = extent_same_get_page(src, loff);
2597                 if (!src_page)
2598                         return -EINVAL;
2599                 dst_page = extent_same_get_page(dst, dst_loff);
2600                 if (!dst_page) {
2601                         page_cache_release(src_page);
2602                         return -EINVAL;
2603                 }
2604                 addr = kmap_atomic(src_page);
2605                 dst_addr = kmap_atomic(dst_page);
2606
2607                 flush_dcache_page(src_page);
2608                 flush_dcache_page(dst_page);
2609
2610                 if (memcmp(addr, dst_addr, cmp_len))
2611                         ret = BTRFS_SAME_DATA_DIFFERS;
2612
2613                 kunmap_atomic(addr);
2614                 kunmap_atomic(dst_addr);
2615                 page_cache_release(src_page);
2616                 page_cache_release(dst_page);
2617
2618                 if (ret)
2619                         break;
2620
2621                 loff += cmp_len;
2622                 dst_loff += cmp_len;
2623                 len -= cmp_len;
2624         }
2625
2626         return ret;
2627 }
2628
2629 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2630 {
2631         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2632
2633         if (off + len > inode->i_size || off + len < off)
2634                 return -EINVAL;
2635         /* Check that we are block aligned - btrfs_clone() requires this */
2636         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2637                 return -EINVAL;
2638
2639         return 0;
2640 }
2641
2642 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2643                              struct inode *dst, u64 dst_loff)
2644 {
2645         int ret;
2646
2647         /*
2648          * btrfs_clone() can't handle extents in the same file
2649          * yet. Once that works, we can drop this check and replace it
2650          * with a check for the same inode, but overlapping extents.
2651          */
2652         if (src == dst)
2653                 return -EINVAL;
2654
2655         btrfs_double_lock(src, loff, dst, dst_loff, len);
2656
2657         ret = extent_same_check_offsets(src, loff, len);
2658         if (ret)
2659                 goto out_unlock;
2660
2661         ret = extent_same_check_offsets(dst, dst_loff, len);
2662         if (ret)
2663                 goto out_unlock;
2664
2665         /* don't make the dst file partly checksummed */
2666         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2667             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2668                 ret = -EINVAL;
2669                 goto out_unlock;
2670         }
2671
2672         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2673         if (ret == 0)
2674                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2675
2676 out_unlock:
2677         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2678
2679         return ret;
2680 }
2681
2682 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2683
2684 static long btrfs_ioctl_file_extent_same(struct file *file,
2685                                          void __user *argp)
2686 {
2687         struct btrfs_ioctl_same_args tmp;
2688         struct btrfs_ioctl_same_args *same;
2689         struct btrfs_ioctl_same_extent_info *info;
2690         struct inode *src = file->f_dentry->d_inode;
2691         struct file *dst_file = NULL;
2692         struct inode *dst;
2693         u64 off;
2694         u64 len;
2695         int i;
2696         int ret;
2697         unsigned long size;
2698         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2699         bool is_admin = capable(CAP_SYS_ADMIN);
2700
2701         if (!(file->f_mode & FMODE_READ))
2702                 return -EINVAL;
2703
2704         ret = mnt_want_write_file(file);
2705         if (ret)
2706                 return ret;
2707
2708         if (copy_from_user(&tmp,
2709                            (struct btrfs_ioctl_same_args __user *)argp,
2710                            sizeof(tmp))) {
2711                 ret = -EFAULT;
2712                 goto out;
2713         }
2714
2715         size = sizeof(tmp) +
2716                 tmp.dest_count * sizeof(struct btrfs_ioctl_same_extent_info);
2717
2718         same = kmalloc(size, GFP_NOFS);
2719         if (!same) {
2720                 ret = -EFAULT;
2721                 goto out;
2722         }
2723
2724         if (copy_from_user(same,
2725                            (struct btrfs_ioctl_same_args __user *)argp, size)) {
2726                 ret = -EFAULT;
2727                 goto out;
2728         }
2729
2730         off = same->logical_offset;
2731         len = same->length;
2732
2733         /*
2734          * Limit the total length we will dedupe for each operation.
2735          * This is intended to bound the total time spent in this
2736          * ioctl to something sane.
2737          */
2738         if (len > BTRFS_MAX_DEDUPE_LEN)
2739                 len = BTRFS_MAX_DEDUPE_LEN;
2740
2741         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2742                 /*
2743                  * Btrfs does not support blocksize < page_size. As a
2744                  * result, btrfs_cmp_data() won't correctly handle
2745                  * this situation without an update.
2746                  */
2747                 ret = -EINVAL;
2748                 goto out;
2749         }
2750
2751         ret = -EISDIR;
2752         if (S_ISDIR(src->i_mode))
2753                 goto out;
2754
2755         ret = -EACCES;
2756         if (!S_ISREG(src->i_mode))
2757                 goto out;
2758
2759         /* pre-format output fields to sane values */
2760         for (i = 0; i < same->dest_count; i++) {
2761                 same->info[i].bytes_deduped = 0ULL;
2762                 same->info[i].status = 0;
2763         }
2764
2765         ret = 0;
2766         for (i = 0; i < same->dest_count; i++) {
2767                 info = &same->info[i];
2768
2769                 dst_file = fget(info->fd);
2770                 if (!dst_file) {
2771                         info->status = -EBADF;
2772                         goto next;
2773                 }
2774
2775                 if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2776                         info->status = -EINVAL;
2777                         goto next;
2778                 }
2779
2780                 info->status = -EXDEV;
2781                 if (file->f_path.mnt != dst_file->f_path.mnt)
2782                         goto next;
2783
2784                 dst = dst_file->f_dentry->d_inode;
2785                 if (src->i_sb != dst->i_sb)
2786                         goto next;
2787
2788                 if (S_ISDIR(dst->i_mode)) {
2789                         info->status = -EISDIR;
2790                         goto next;
2791                 }
2792
2793                 if (!S_ISREG(dst->i_mode)) {
2794                         info->status = -EACCES;
2795                         goto next;
2796                 }
2797
2798                 info->status = btrfs_extent_same(src, off, len, dst,
2799                                                 info->logical_offset);
2800                 if (info->status == 0)
2801                         info->bytes_deduped += len;
2802
2803 next:
2804                 if (dst_file)
2805                         fput(dst_file);
2806         }
2807
2808         ret = copy_to_user(argp, same, size);
2809         if (ret)
2810                 ret = -EFAULT;
2811
2812 out:
2813         mnt_drop_write_file(file);
2814         return ret;
2815 }
2816
2817 /**
2818  * btrfs_clone() - clone a range from inode file to another
2819  *
2820  * @src: Inode to clone from
2821  * @inode: Inode to clone to
2822  * @off: Offset within source to start clone from
2823  * @olen: Original length, passed by user, of range to clone
2824  * @olen_aligned: Block-aligned value of olen, extent_same uses
2825  *               identical values here
2826  * @destoff: Offset within @inode to start clone
2827  */
2828 static int btrfs_clone(struct inode *src, struct inode *inode,
2829                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2830 {
2831         struct btrfs_root *root = BTRFS_I(inode)->root;
2832         struct btrfs_path *path = NULL;
2833         struct extent_buffer *leaf;
2834         struct btrfs_trans_handle *trans;
2835         char *buf = NULL;
2836         struct btrfs_key key;
2837         u32 nritems;
2838         int slot;
2839         int ret;
2840         u64 len = olen_aligned;
2841
2842         ret = -ENOMEM;
2843         buf = vmalloc(btrfs_level_size(root, 0));
2844         if (!buf)
2845                 return ret;
2846
2847         path = btrfs_alloc_path();
2848         if (!path) {
2849                 vfree(buf);
2850                 return ret;
2851         }
2852
2853         path->reada = 2;
2854         /* clone data */
2855         key.objectid = btrfs_ino(src);
2856         key.type = BTRFS_EXTENT_DATA_KEY;
2857         key.offset = 0;
2858
2859         while (1) {
2860                 /*
2861                  * note the key will change type as we walk through the
2862                  * tree.
2863                  */
2864                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2865                                 0, 0);
2866                 if (ret < 0)
2867                         goto out;
2868
2869                 nritems = btrfs_header_nritems(path->nodes[0]);
2870                 if (path->slots[0] >= nritems) {
2871                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2872                         if (ret < 0)
2873                                 goto out;
2874                         if (ret > 0)
2875                                 break;
2876                         nritems = btrfs_header_nritems(path->nodes[0]);
2877                 }
2878                 leaf = path->nodes[0];
2879                 slot = path->slots[0];
2880
2881                 btrfs_item_key_to_cpu(leaf, &key, slot);
2882                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2883                     key.objectid != btrfs_ino(src))
2884                         break;
2885
2886                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2887                         struct btrfs_file_extent_item *extent;
2888                         int type;
2889                         u32 size;
2890                         struct btrfs_key new_key;
2891                         u64 disko = 0, diskl = 0;
2892                         u64 datao = 0, datal = 0;
2893                         u8 comp;
2894                         u64 endoff;
2895
2896                         size = btrfs_item_size_nr(leaf, slot);
2897                         read_extent_buffer(leaf, buf,
2898                                            btrfs_item_ptr_offset(leaf, slot),
2899                                            size);
2900
2901                         extent = btrfs_item_ptr(leaf, slot,
2902                                                 struct btrfs_file_extent_item);
2903                         comp = btrfs_file_extent_compression(leaf, extent);
2904                         type = btrfs_file_extent_type(leaf, extent);
2905                         if (type == BTRFS_FILE_EXTENT_REG ||
2906                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2907                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2908                                                                       extent);
2909                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2910                                                                  extent);
2911                                 datao = btrfs_file_extent_offset(leaf, extent);
2912                                 datal = btrfs_file_extent_num_bytes(leaf,
2913                                                                     extent);
2914                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2915                                 /* take upper bound, may be compressed */
2916                                 datal = btrfs_file_extent_ram_bytes(leaf,
2917                                                                     extent);
2918                         }
2919                         btrfs_release_path(path);
2920
2921                         if (key.offset + datal <= off ||
2922                             key.offset >= off + len - 1)
2923                                 goto next;
2924
2925                         memcpy(&new_key, &key, sizeof(new_key));
2926                         new_key.objectid = btrfs_ino(inode);
2927                         if (off <= key.offset)
2928                                 new_key.offset = key.offset + destoff - off;
2929                         else
2930                                 new_key.offset = destoff;
2931
2932                         /*
2933                          * 1 - adjusting old extent (we may have to split it)
2934                          * 1 - add new extent
2935                          * 1 - inode update
2936                          */
2937                         trans = btrfs_start_transaction(root, 3);
2938                         if (IS_ERR(trans)) {
2939                                 ret = PTR_ERR(trans);
2940                                 goto out;
2941                         }
2942
2943                         if (type == BTRFS_FILE_EXTENT_REG ||
2944                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2945                                 /*
2946                                  *    a  | --- range to clone ---|  b
2947                                  * | ------------- extent ------------- |
2948                                  */
2949
2950                                 /* substract range b */
2951                                 if (key.offset + datal > off + len)
2952                                         datal = off + len - key.offset;
2953
2954                                 /* substract range a */
2955                                 if (off > key.offset) {
2956                                         datao += off - key.offset;
2957                                         datal -= off - key.offset;
2958                                 }
2959
2960                                 ret = btrfs_drop_extents(trans, root, inode,
2961                                                          new_key.offset,
2962                                                          new_key.offset + datal,
2963                                                          1);
2964                                 if (ret) {
2965                                         btrfs_abort_transaction(trans, root,
2966                                                                 ret);
2967                                         btrfs_end_transaction(trans, root);
2968                                         goto out;
2969                                 }
2970
2971                                 ret = btrfs_insert_empty_item(trans, root, path,
2972                                                               &new_key, size);
2973                                 if (ret) {
2974                                         btrfs_abort_transaction(trans, root,
2975                                                                 ret);
2976                                         btrfs_end_transaction(trans, root);
2977                                         goto out;
2978                                 }
2979
2980                                 leaf = path->nodes[0];
2981                                 slot = path->slots[0];
2982                                 write_extent_buffer(leaf, buf,
2983                                             btrfs_item_ptr_offset(leaf, slot),
2984                                             size);
2985
2986                                 extent = btrfs_item_ptr(leaf, slot,
2987                                                 struct btrfs_file_extent_item);
2988
2989                                 /* disko == 0 means it's a hole */
2990                                 if (!disko)
2991                                         datao = 0;
2992
2993                                 btrfs_set_file_extent_offset(leaf, extent,
2994                                                              datao);
2995                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2996                                                                 datal);
2997                                 if (disko) {
2998                                         inode_add_bytes(inode, datal);
2999                                         ret = btrfs_inc_extent_ref(trans, root,
3000                                                         disko, diskl, 0,
3001                                                         root->root_key.objectid,
3002                                                         btrfs_ino(inode),
3003                                                         new_key.offset - datao,
3004                                                         0);
3005                                         if (ret) {
3006                                                 btrfs_abort_transaction(trans,
3007                                                                         root,
3008                                                                         ret);
3009                                                 btrfs_end_transaction(trans,
3010                                                                       root);
3011                                                 goto out;
3012
3013                                         }
3014                                 }
3015                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3016                                 u64 skip = 0;
3017                                 u64 trim = 0;
3018                                 if (off > key.offset) {
3019                                         skip = off - key.offset;
3020                                         new_key.offset += skip;
3021                                 }
3022
3023                                 if (key.offset + datal > off + len)
3024                                         trim = key.offset + datal - (off + len);
3025
3026                                 if (comp && (skip || trim)) {
3027                                         ret = -EINVAL;
3028                                         btrfs_end_transaction(trans, root);
3029                                         goto out;
3030                                 }
3031                                 size -= skip + trim;
3032                                 datal -= skip + trim;
3033
3034                                 ret = btrfs_drop_extents(trans, root, inode,
3035                                                          new_key.offset,
3036                                                          new_key.offset + datal,
3037                                                          1);
3038                                 if (ret) {
3039                                         btrfs_abort_transaction(trans, root,
3040                                                                 ret);
3041                                         btrfs_end_transaction(trans, root);
3042                                         goto out;
3043                                 }
3044
3045                                 ret = btrfs_insert_empty_item(trans, root, path,
3046                                                               &new_key, size);
3047                                 if (ret) {
3048                                         btrfs_abort_transaction(trans, root,
3049                                                                 ret);
3050                                         btrfs_end_transaction(trans, root);
3051                                         goto out;
3052                                 }
3053
3054                                 if (skip) {
3055                                         u32 start =
3056                                           btrfs_file_extent_calc_inline_size(0);
3057                                         memmove(buf+start, buf+start+skip,
3058                                                 datal);
3059                                 }
3060
3061                                 leaf = path->nodes[0];
3062                                 slot = path->slots[0];
3063                                 write_extent_buffer(leaf, buf,
3064                                             btrfs_item_ptr_offset(leaf, slot),
3065                                             size);
3066                                 inode_add_bytes(inode, datal);
3067                         }
3068
3069                         btrfs_mark_buffer_dirty(leaf);
3070                         btrfs_release_path(path);
3071
3072                         inode_inc_iversion(inode);
3073                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3074
3075                         /*
3076                          * we round up to the block size at eof when
3077                          * determining which extents to clone above,
3078                          * but shouldn't round up the file size
3079                          */
3080                         endoff = new_key.offset + datal;
3081                         if (endoff > destoff+olen)
3082                                 endoff = destoff+olen;
3083                         if (endoff > inode->i_size)
3084                                 btrfs_i_size_write(inode, endoff);
3085
3086                         ret = btrfs_update_inode(trans, root, inode);
3087                         if (ret) {
3088                                 btrfs_abort_transaction(trans, root, ret);
3089                                 btrfs_end_transaction(trans, root);
3090                                 goto out;
3091                         }
3092                         ret = btrfs_end_transaction(trans, root);
3093                 }
3094 next:
3095                 btrfs_release_path(path);
3096                 key.offset++;
3097         }
3098         ret = 0;
3099
3100 out:
3101         btrfs_release_path(path);
3102         btrfs_free_path(path);
3103         vfree(buf);
3104         return ret;
3105 }
3106
3107 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3108                                        u64 off, u64 olen, u64 destoff)
3109 {
3110         struct inode *inode = fdentry(file)->d_inode;
3111         struct btrfs_root *root = BTRFS_I(inode)->root;
3112         struct fd src_file;
3113         struct inode *src;
3114         int ret;
3115         u64 len = olen;
3116         u64 bs = root->fs_info->sb->s_blocksize;
3117         int same_inode = 0;
3118
3119         /*
3120          * TODO:
3121          * - split compressed inline extents.  annoying: we need to
3122          *   decompress into destination's address_space (the file offset
3123          *   may change, so source mapping won't do), then recompress (or
3124          *   otherwise reinsert) a subrange.
3125          * - allow ranges within the same file to be cloned (provided
3126          *   they don't overlap)?
3127          */
3128
3129         /* the destination must be opened for writing */
3130         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3131                 return -EINVAL;
3132
3133         if (btrfs_root_readonly(root))
3134                 return -EROFS;
3135
3136         ret = mnt_want_write_file(file);
3137         if (ret)
3138                 return ret;
3139
3140         src_file = fdget(srcfd);
3141         if (!src_file.file) {
3142                 ret = -EBADF;
3143                 goto out_drop_write;
3144         }
3145
3146         ret = -EXDEV;
3147         if (src_file.file->f_path.mnt != file->f_path.mnt)
3148                 goto out_fput;
3149
3150         src = file_inode(src_file.file);
3151
3152         ret = -EINVAL;
3153         if (src == inode)
3154                 same_inode = 1;
3155
3156         /* the src must be open for reading */
3157         if (!(src_file.file->f_mode & FMODE_READ))
3158                 goto out_fput;
3159
3160         /* don't make the dst file partly checksummed */
3161         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3162             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3163                 goto out_fput;
3164
3165         ret = -EISDIR;
3166         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3167                 goto out_fput;
3168
3169         ret = -EXDEV;
3170         if (src->i_sb != inode->i_sb)
3171                 goto out_fput;
3172
3173         if (!same_inode) {
3174                 if (inode < src) {
3175                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3176                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3177                 } else {
3178                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3179                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3180                 }
3181         } else {
3182                 mutex_lock(&src->i_mutex);
3183         }
3184
3185         /* determine range to clone */
3186         ret = -EINVAL;
3187         if (off + len > src->i_size || off + len < off)
3188                 goto out_unlock;
3189         if (len == 0)
3190                 olen = len = src->i_size - off;
3191         /* if we extend to eof, continue to block boundary */
3192         if (off + len == src->i_size)
3193                 len = ALIGN(src->i_size, bs) - off;
3194
3195         /* verify the end result is block aligned */
3196         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3197             !IS_ALIGNED(destoff, bs))
3198                 goto out_unlock;
3199
3200         /* verify if ranges are overlapped within the same file */
3201         if (same_inode) {
3202                 if (destoff + len > off && destoff < off + len)
3203                         goto out_unlock;
3204         }
3205
3206         if (destoff > inode->i_size) {
3207                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3208                 if (ret)
3209                         goto out_unlock;
3210         }
3211
3212         /* truncate page cache pages from target inode range */
3213         truncate_inode_pages_range(&inode->i_data, destoff,
3214                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3215
3216         lock_extent_range(src, off, len);
3217
3218         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3219
3220         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3221 out_unlock:
3222         mutex_unlock(&src->i_mutex);
3223         if (!same_inode)
3224                 mutex_unlock(&inode->i_mutex);
3225 out_fput:
3226         fdput(src_file);
3227 out_drop_write:
3228         mnt_drop_write_file(file);
3229         return ret;
3230 }
3231
3232 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3233 {
3234         struct btrfs_ioctl_clone_range_args args;
3235
3236         if (copy_from_user(&args, argp, sizeof(args)))
3237                 return -EFAULT;
3238         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3239                                  args.src_length, args.dest_offset);
3240 }
3241
3242 /*
3243  * there are many ways the trans_start and trans_end ioctls can lead
3244  * to deadlocks.  They should only be used by applications that
3245  * basically own the machine, and have a very in depth understanding
3246  * of all the possible deadlocks and enospc problems.
3247  */
3248 static long btrfs_ioctl_trans_start(struct file *file)
3249 {
3250         struct inode *inode = file_inode(file);
3251         struct btrfs_root *root = BTRFS_I(inode)->root;
3252         struct btrfs_trans_handle *trans;
3253         int ret;
3254
3255         ret = -EPERM;
3256         if (!capable(CAP_SYS_ADMIN))
3257                 goto out;
3258
3259         ret = -EINPROGRESS;
3260         if (file->private_data)
3261                 goto out;
3262
3263         ret = -EROFS;
3264         if (btrfs_root_readonly(root))
3265                 goto out;
3266
3267         ret = mnt_want_write_file(file);
3268         if (ret)
3269                 goto out;
3270
3271         atomic_inc(&root->fs_info->open_ioctl_trans);
3272
3273         ret = -ENOMEM;
3274         trans = btrfs_start_ioctl_transaction(root);
3275         if (IS_ERR(trans))
3276                 goto out_drop;
3277
3278         file->private_data = trans;
3279         return 0;
3280
3281 out_drop:
3282         atomic_dec(&root->fs_info->open_ioctl_trans);
3283         mnt_drop_write_file(file);
3284 out:
3285         return ret;
3286 }
3287
3288 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3289 {
3290         struct inode *inode = file_inode(file);
3291         struct btrfs_root *root = BTRFS_I(inode)->root;
3292         struct btrfs_root *new_root;
3293         struct btrfs_dir_item *di;
3294         struct btrfs_trans_handle *trans;
3295         struct btrfs_path *path;
3296         struct btrfs_key location;
3297         struct btrfs_disk_key disk_key;
3298         u64 objectid = 0;
3299         u64 dir_id;
3300         int ret;
3301
3302         if (!capable(CAP_SYS_ADMIN))
3303                 return -EPERM;
3304
3305         ret = mnt_want_write_file(file);
3306         if (ret)
3307                 return ret;
3308
3309         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3310                 ret = -EFAULT;
3311                 goto out;
3312         }
3313
3314         if (!objectid)
3315                 objectid = BTRFS_FS_TREE_OBJECTID;
3316
3317         location.objectid = objectid;
3318         location.type = BTRFS_ROOT_ITEM_KEY;
3319         location.offset = (u64)-1;
3320
3321         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3322         if (IS_ERR(new_root)) {
3323                 ret = PTR_ERR(new_root);
3324                 goto out;
3325         }
3326
3327         path = btrfs_alloc_path();
3328         if (!path) {
3329                 ret = -ENOMEM;
3330                 goto out;
3331         }
3332         path->leave_spinning = 1;
3333
3334         trans = btrfs_start_transaction(root, 1);
3335         if (IS_ERR(trans)) {
3336                 btrfs_free_path(path);
3337                 ret = PTR_ERR(trans);
3338                 goto out;
3339         }
3340
3341         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3342         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3343                                    dir_id, "default", 7, 1);
3344         if (IS_ERR_OR_NULL(di)) {
3345                 btrfs_free_path(path);
3346                 btrfs_end_transaction(trans, root);
3347                 printk(KERN_ERR "Umm, you don't have the default dir item, "
3348                        "this isn't going to work\n");
3349                 ret = -ENOENT;
3350                 goto out;
3351         }
3352
3353         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3354         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3355         btrfs_mark_buffer_dirty(path->nodes[0]);
3356         btrfs_free_path(path);
3357
3358         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3359         btrfs_end_transaction(trans, root);
3360 out:
3361         mnt_drop_write_file(file);
3362         return ret;
3363 }
3364
3365 void btrfs_get_block_group_info(struct list_head *groups_list,
3366                                 struct btrfs_ioctl_space_info *space)
3367 {
3368         struct btrfs_block_group_cache *block_group;
3369
3370         space->total_bytes = 0;
3371         space->used_bytes = 0;
3372         space->flags = 0;
3373         list_for_each_entry(block_group, groups_list, list) {
3374                 space->flags = block_group->flags;
3375                 space->total_bytes += block_group->key.offset;
3376                 space->used_bytes +=
3377                         btrfs_block_group_used(&block_group->item);
3378         }
3379 }
3380
3381 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3382 {
3383         struct btrfs_ioctl_space_args space_args;
3384         struct btrfs_ioctl_space_info space;
3385         struct btrfs_ioctl_space_info *dest;
3386         struct btrfs_ioctl_space_info *dest_orig;
3387         struct btrfs_ioctl_space_info __user *user_dest;
3388         struct btrfs_space_info *info;
3389         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3390                        BTRFS_BLOCK_GROUP_SYSTEM,
3391                        BTRFS_BLOCK_GROUP_METADATA,
3392                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3393         int num_types = 4;
3394         int alloc_size;
3395         int ret = 0;
3396         u64 slot_count = 0;
3397         int i, c;
3398
3399         if (copy_from_user(&space_args,
3400                            (struct btrfs_ioctl_space_args __user *)arg,
3401                            sizeof(space_args)))
3402                 return -EFAULT;
3403
3404         for (i = 0; i < num_types; i++) {
3405                 struct btrfs_space_info *tmp;
3406
3407                 info = NULL;
3408                 rcu_read_lock();
3409                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3410                                         list) {
3411                         if (tmp->flags == types[i]) {
3412                                 info = tmp;
3413                                 break;
3414                         }
3415                 }
3416                 rcu_read_unlock();
3417
3418                 if (!info)
3419                         continue;
3420
3421                 down_read(&info->groups_sem);
3422                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3423                         if (!list_empty(&info->block_groups[c]))
3424                                 slot_count++;
3425                 }
3426                 up_read(&info->groups_sem);
3427         }
3428
3429         /* space_slots == 0 means they are asking for a count */
3430         if (space_args.space_slots == 0) {
3431                 space_args.total_spaces = slot_count;
3432                 goto out;
3433         }
3434
3435         slot_count = min_t(u64, space_args.space_slots, slot_count);
3436
3437         alloc_size = sizeof(*dest) * slot_count;
3438
3439         /* we generally have at most 6 or so space infos, one for each raid
3440          * level.  So, a whole page should be more than enough for everyone
3441          */
3442         if (alloc_size > PAGE_CACHE_SIZE)
3443                 return -ENOMEM;
3444
3445         space_args.total_spaces = 0;
3446         dest = kmalloc(alloc_size, GFP_NOFS);
3447         if (!dest)
3448                 return -ENOMEM;
3449         dest_orig = dest;
3450
3451         /* now we have a buffer to copy into */
3452         for (i = 0; i < num_types; i++) {
3453                 struct btrfs_space_info *tmp;
3454
3455                 if (!slot_count)
3456                         break;
3457
3458                 info = NULL;
3459                 rcu_read_lock();
3460                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3461                                         list) {
3462                         if (tmp->flags == types[i]) {
3463                                 info = tmp;
3464                                 break;
3465                         }
3466                 }
3467                 rcu_read_unlock();
3468
3469                 if (!info)
3470                         continue;
3471                 down_read(&info->groups_sem);
3472                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3473                         if (!list_empty(&info->block_groups[c])) {
3474                                 btrfs_get_block_group_info(
3475                                         &info->block_groups[c], &space);
3476                                 memcpy(dest, &space, sizeof(space));
3477                                 dest++;
3478                                 space_args.total_spaces++;
3479                                 slot_count--;
3480                         }
3481                         if (!slot_count)
3482                                 break;
3483                 }
3484                 up_read(&info->groups_sem);
3485         }
3486
3487         user_dest = (struct btrfs_ioctl_space_info __user *)
3488                 (arg + sizeof(struct btrfs_ioctl_space_args));
3489
3490         if (copy_to_user(user_dest, dest_orig, alloc_size))
3491                 ret = -EFAULT;
3492
3493         kfree(dest_orig);
3494 out:
3495         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3496                 ret = -EFAULT;
3497
3498         return ret;
3499 }
3500
3501 /*
3502  * there are many ways the trans_start and trans_end ioctls can lead
3503  * to deadlocks.  They should only be used by applications that
3504  * basically own the machine, and have a very in depth understanding
3505  * of all the possible deadlocks and enospc problems.
3506  */
3507 long btrfs_ioctl_trans_end(struct file *file)
3508 {
3509         struct inode *inode = file_inode(file);
3510         struct btrfs_root *root = BTRFS_I(inode)->root;
3511         struct btrfs_trans_handle *trans;
3512
3513         trans = file->private_data;
3514         if (!trans)
3515                 return -EINVAL;
3516         file->private_data = NULL;
3517
3518         btrfs_end_transaction(trans, root);
3519
3520         atomic_dec(&root->fs_info->open_ioctl_trans);
3521
3522         mnt_drop_write_file(file);
3523         return 0;
3524 }
3525
3526 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3527                                             void __user *argp)
3528 {
3529         struct btrfs_trans_handle *trans;
3530         u64 transid;
3531         int ret;
3532
3533         trans = btrfs_attach_transaction_barrier(root);
3534         if (IS_ERR(trans)) {
3535                 if (PTR_ERR(trans) != -ENOENT)
3536                         return PTR_ERR(trans);
3537
3538                 /* No running transaction, don't bother */
3539                 transid = root->fs_info->last_trans_committed;
3540                 goto out;
3541         }
3542         transid = trans->transid;
3543         ret = btrfs_commit_transaction_async(trans, root, 0);
3544         if (ret) {
3545                 btrfs_end_transaction(trans, root);
3546                 return ret;
3547         }
3548 out:
3549         if (argp)
3550                 if (copy_to_user(argp, &transid, sizeof(transid)))
3551                         return -EFAULT;
3552         return 0;
3553 }
3554
3555 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3556                                            void __user *argp)
3557 {
3558         u64 transid;
3559
3560         if (argp) {
3561                 if (copy_from_user(&transid, argp, sizeof(transid)))
3562                         return -EFAULT;
3563         } else {
3564                 transid = 0;  /* current trans */
3565         }
3566         return btrfs_wait_for_commit(root, transid);
3567 }
3568
3569 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3570 {
3571         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3572         struct btrfs_ioctl_scrub_args *sa;
3573         int ret;
3574
3575         if (!capable(CAP_SYS_ADMIN))
3576                 return -EPERM;
3577
3578         sa = memdup_user(arg, sizeof(*sa));
3579         if (IS_ERR(sa))
3580                 return PTR_ERR(sa);
3581
3582         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3583                 ret = mnt_want_write_file(file);
3584                 if (ret)
3585                         goto out;
3586         }
3587
3588         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3589                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3590                               0);
3591
3592         if (copy_to_user(arg, sa, sizeof(*sa)))
3593                 ret = -EFAULT;
3594
3595         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3596                 mnt_drop_write_file(file);
3597 out:
3598         kfree(sa);
3599         return ret;
3600 }
3601
3602 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3603 {
3604         if (!capable(CAP_SYS_ADMIN))
3605                 return -EPERM;
3606
3607         return btrfs_scrub_cancel(root->fs_info);
3608 }
3609
3610 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3611                                        void __user *arg)
3612 {
3613         struct btrfs_ioctl_scrub_args *sa;
3614         int ret;
3615
3616         if (!capable(CAP_SYS_ADMIN))
3617                 return -EPERM;
3618
3619         sa = memdup_user(arg, sizeof(*sa));
3620         if (IS_ERR(sa))
3621                 return PTR_ERR(sa);
3622
3623         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3624
3625         if (copy_to_user(arg, sa, sizeof(*sa)))
3626                 ret = -EFAULT;
3627
3628         kfree(sa);
3629         return ret;
3630 }
3631
3632 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3633                                       void __user *arg)
3634 {
3635         struct btrfs_ioctl_get_dev_stats *sa;
3636         int ret;
3637
3638         sa = memdup_user(arg, sizeof(*sa));
3639         if (IS_ERR(sa))
3640                 return PTR_ERR(sa);
3641
3642         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3643                 kfree(sa);
3644                 return -EPERM;
3645         }
3646
3647         ret = btrfs_get_dev_stats(root, sa);
3648
3649         if (copy_to_user(arg, sa, sizeof(*sa)))
3650                 ret = -EFAULT;
3651
3652         kfree(sa);
3653         return ret;
3654 }
3655
3656 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3657 {
3658         struct btrfs_ioctl_dev_replace_args *p;
3659         int ret;
3660
3661         if (!capable(CAP_SYS_ADMIN))
3662                 return -EPERM;
3663
3664         p = memdup_user(arg, sizeof(*p));
3665         if (IS_ERR(p))
3666                 return PTR_ERR(p);
3667
3668         switch (p->cmd) {
3669         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3670                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
3671                         ret = -EROFS;
3672                         goto out;
3673                 }
3674                 if (atomic_xchg(
3675                         &root->fs_info->mutually_exclusive_operation_running,
3676                         1)) {
3677                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3678                 } else {
3679                         ret = btrfs_dev_replace_start(root, p);
3680                         atomic_set(
3681                          &root->fs_info->mutually_exclusive_operation_running,
3682                          0);
3683                 }
3684                 break;
3685         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3686                 btrfs_dev_replace_status(root->fs_info, p);
3687                 ret = 0;
3688                 break;
3689         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3690                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3691                 break;
3692         default:
3693                 ret = -EINVAL;
3694                 break;
3695         }
3696
3697         if (copy_to_user(arg, p, sizeof(*p)))
3698                 ret = -EFAULT;
3699 out:
3700         kfree(p);
3701         return ret;
3702 }
3703
3704 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3705 {
3706         int ret = 0;
3707         int i;
3708         u64 rel_ptr;
3709         int size;
3710         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3711         struct inode_fs_paths *ipath = NULL;
3712         struct btrfs_path *path;
3713
3714         if (!capable(CAP_DAC_READ_SEARCH))
3715                 return -EPERM;
3716
3717         path = btrfs_alloc_path();
3718         if (!path) {
3719                 ret = -ENOMEM;
3720                 goto out;
3721         }
3722
3723         ipa = memdup_user(arg, sizeof(*ipa));
3724         if (IS_ERR(ipa)) {
3725                 ret = PTR_ERR(ipa);
3726                 ipa = NULL;
3727                 goto out;
3728         }
3729
3730         size = min_t(u32, ipa->size, 4096);
3731         ipath = init_ipath(size, root, path);
3732         if (IS_ERR(ipath)) {
3733                 ret = PTR_ERR(ipath);
3734                 ipath = NULL;
3735                 goto out;
3736         }
3737
3738         ret = paths_from_inode(ipa->inum, ipath);
3739         if (ret < 0)
3740                 goto out;
3741
3742         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3743                 rel_ptr = ipath->fspath->val[i] -
3744                           (u64)(unsigned long)ipath->fspath->val;
3745                 ipath->fspath->val[i] = rel_ptr;
3746         }
3747
3748         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3749                            (void *)(unsigned long)ipath->fspath, size);
3750         if (ret) {
3751                 ret = -EFAULT;
3752                 goto out;
3753         }
3754
3755 out:
3756         btrfs_free_path(path);
3757         free_ipath(ipath);
3758         kfree(ipa);
3759
3760         return ret;
3761 }
3762
3763 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3764 {
3765         struct btrfs_data_container *inodes = ctx;
3766         const size_t c = 3 * sizeof(u64);
3767
3768         if (inodes->bytes_left >= c) {
3769                 inodes->bytes_left -= c;
3770                 inodes->val[inodes->elem_cnt] = inum;
3771                 inodes->val[inodes->elem_cnt + 1] = offset;
3772                 inodes->val[inodes->elem_cnt + 2] = root;
3773                 inodes->elem_cnt += 3;
3774         } else {
3775                 inodes->bytes_missing += c - inodes->bytes_left;
3776                 inodes->bytes_left = 0;
3777                 inodes->elem_missed += 3;
3778         }
3779
3780         return 0;
3781 }
3782
3783 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3784                                         void __user *arg)
3785 {
3786         int ret = 0;
3787         int size;
3788         struct btrfs_ioctl_logical_ino_args *loi;
3789         struct btrfs_data_container *inodes = NULL;
3790         struct btrfs_path *path = NULL;
3791
3792         if (!capable(CAP_SYS_ADMIN))
3793                 return -EPERM;
3794
3795         loi = memdup_user(arg, sizeof(*loi));
3796         if (IS_ERR(loi)) {
3797                 ret = PTR_ERR(loi);
3798                 loi = NULL;
3799                 goto out;
3800         }
3801
3802         path = btrfs_alloc_path();
3803         if (!path) {
3804                 ret = -ENOMEM;
3805                 goto out;
3806         }
3807
3808         size = min_t(u32, loi->size, 64 * 1024);
3809         inodes = init_data_container(size);
3810         if (IS_ERR(inodes)) {
3811                 ret = PTR_ERR(inodes);
3812                 inodes = NULL;
3813                 goto out;
3814         }
3815
3816         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3817                                           build_ino_list, inodes);
3818         if (ret == -EINVAL)
3819                 ret = -ENOENT;
3820         if (ret < 0)
3821                 goto out;
3822
3823         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3824                            (void *)(unsigned long)inodes, size);
3825         if (ret)
3826                 ret = -EFAULT;
3827
3828 out:
3829         btrfs_free_path(path);
3830         vfree(inodes);
3831         kfree(loi);
3832
3833         return ret;
3834 }
3835
3836 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3837                                struct btrfs_ioctl_balance_args *bargs)
3838 {
3839         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3840
3841         bargs->flags = bctl->flags;
3842
3843         if (atomic_read(&fs_info->balance_running))
3844                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3845         if (atomic_read(&fs_info->balance_pause_req))
3846                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3847         if (atomic_read(&fs_info->balance_cancel_req))
3848                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3849
3850         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3851         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3852         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3853
3854         if (lock) {
3855                 spin_lock(&fs_info->balance_lock);
3856                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3857                 spin_unlock(&fs_info->balance_lock);
3858         } else {
3859                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3860         }
3861 }
3862
3863 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3864 {
3865         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3866         struct btrfs_fs_info *fs_info = root->fs_info;
3867         struct btrfs_ioctl_balance_args *bargs;
3868         struct btrfs_balance_control *bctl;
3869         bool need_unlock; /* for mut. excl. ops lock */
3870         int ret;
3871
3872         if (!capable(CAP_SYS_ADMIN))
3873                 return -EPERM;
3874
3875         ret = mnt_want_write_file(file);
3876         if (ret)
3877                 return ret;
3878
3879 again:
3880         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
3881                 mutex_lock(&fs_info->volume_mutex);
3882                 mutex_lock(&fs_info->balance_mutex);
3883                 need_unlock = true;
3884                 goto locked;
3885         }
3886
3887         /*
3888          * mut. excl. ops lock is locked.  Three possibilites:
3889          *   (1) some other op is running
3890          *   (2) balance is running
3891          *   (3) balance is paused -- special case (think resume)
3892          */
3893         mutex_lock(&fs_info->balance_mutex);
3894         if (fs_info->balance_ctl) {
3895                 /* this is either (2) or (3) */
3896                 if (!atomic_read(&fs_info->balance_running)) {
3897                         mutex_unlock(&fs_info->balance_mutex);
3898                         if (!mutex_trylock(&fs_info->volume_mutex))
3899                                 goto again;
3900                         mutex_lock(&fs_info->balance_mutex);
3901
3902                         if (fs_info->balance_ctl &&
3903                             !atomic_read(&fs_info->balance_running)) {
3904                                 /* this is (3) */
3905                                 need_unlock = false;
3906                                 goto locked;
3907                         }
3908
3909                         mutex_unlock(&fs_info->balance_mutex);
3910                         mutex_unlock(&fs_info->volume_mutex);
3911                         goto again;
3912                 } else {
3913                         /* this is (2) */
3914                         mutex_unlock(&fs_info->balance_mutex);
3915                         ret = -EINPROGRESS;
3916                         goto out;
3917                 }
3918         } else {
3919                 /* this is (1) */
3920                 mutex_unlock(&fs_info->balance_mutex);
3921                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3922                 goto out;
3923         }
3924
3925 locked:
3926         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
3927
3928         if (arg) {
3929                 bargs = memdup_user(arg, sizeof(*bargs));
3930                 if (IS_ERR(bargs)) {
3931                         ret = PTR_ERR(bargs);
3932                         goto out_unlock;
3933                 }
3934
3935                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3936                         if (!fs_info->balance_ctl) {
3937                                 ret = -ENOTCONN;
3938                                 goto out_bargs;
3939                         }
3940
3941                         bctl = fs_info->balance_ctl;
3942                         spin_lock(&fs_info->balance_lock);
3943                         bctl->flags |= BTRFS_BALANCE_RESUME;
3944                         spin_unlock(&fs_info->balance_lock);
3945
3946                         goto do_balance;
3947                 }
3948         } else {
3949                 bargs = NULL;
3950         }
3951
3952         if (fs_info->balance_ctl) {
3953                 ret = -EINPROGRESS;
3954                 goto out_bargs;
3955         }
3956
3957         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3958         if (!bctl) {
3959                 ret = -ENOMEM;
3960                 goto out_bargs;
3961         }
3962
3963         bctl->fs_info = fs_info;
3964         if (arg) {
3965                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3966                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3967                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3968
3969                 bctl->flags = bargs->flags;
3970         } else {
3971                 /* balance everything - no filters */
3972                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3973         }
3974
3975 do_balance:
3976         /*
3977          * Ownership of bctl and mutually_exclusive_operation_running
3978          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
3979          * or, if restriper was paused all the way until unmount, in
3980          * free_fs_info.  mutually_exclusive_operation_running is
3981          * cleared in __cancel_balance.
3982          */
3983         need_unlock = false;
3984
3985         ret = btrfs_balance(bctl, bargs);
3986
3987         if (arg) {
3988                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3989                         ret = -EFAULT;
3990         }
3991
3992 out_bargs:
3993         kfree(bargs);
3994 out_unlock:
3995         mutex_unlock(&fs_info->balance_mutex);
3996         mutex_unlock(&fs_info->volume_mutex);
3997         if (need_unlock)
3998                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3999 out:
4000         mnt_drop_write_file(file);
4001         return ret;
4002 }
4003
4004 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4005 {
4006         if (!capable(CAP_SYS_ADMIN))
4007                 return -EPERM;
4008
4009         switch (cmd) {
4010         case BTRFS_BALANCE_CTL_PAUSE:
4011                 return btrfs_pause_balance(root->fs_info);
4012         case BTRFS_BALANCE_CTL_CANCEL:
4013                 return btrfs_cancel_balance(root->fs_info);
4014         }
4015
4016         return -EINVAL;
4017 }
4018
4019 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4020                                          void __user *arg)
4021 {
4022         struct btrfs_fs_info *fs_info = root->fs_info;
4023         struct btrfs_ioctl_balance_args *bargs;
4024         int ret = 0;
4025
4026         if (!capable(CAP_SYS_ADMIN))
4027                 return -EPERM;
4028
4029         mutex_lock(&fs_info->balance_mutex);
4030         if (!fs_info->balance_ctl) {
4031                 ret = -ENOTCONN;
4032                 goto out;
4033         }
4034
4035         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4036         if (!bargs) {
4037                 ret = -ENOMEM;
4038                 goto out;
4039         }
4040
4041         update_ioctl_balance_args(fs_info, 1, bargs);
4042
4043         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4044                 ret = -EFAULT;
4045
4046         kfree(bargs);
4047 out:
4048         mutex_unlock(&fs_info->balance_mutex);
4049         return ret;
4050 }
4051
4052 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4053 {
4054         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4055         struct btrfs_ioctl_quota_ctl_args *sa;
4056         struct btrfs_trans_handle *trans = NULL;
4057         int ret;
4058         int err;
4059
4060         if (!capable(CAP_SYS_ADMIN))
4061                 return -EPERM;
4062
4063         ret = mnt_want_write_file(file);
4064         if (ret)
4065                 return ret;
4066
4067         sa = memdup_user(arg, sizeof(*sa));
4068         if (IS_ERR(sa)) {
4069                 ret = PTR_ERR(sa);
4070                 goto drop_write;
4071         }
4072
4073         down_write(&root->fs_info->subvol_sem);
4074         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4075         if (IS_ERR(trans)) {
4076                 ret = PTR_ERR(trans);
4077                 goto out;
4078         }
4079
4080         switch (sa->cmd) {
4081         case BTRFS_QUOTA_CTL_ENABLE:
4082                 ret = btrfs_quota_enable(trans, root->fs_info);
4083                 break;
4084         case BTRFS_QUOTA_CTL_DISABLE:
4085                 ret = btrfs_quota_disable(trans, root->fs_info);
4086                 break;
4087         default:
4088                 ret = -EINVAL;
4089                 break;
4090         }
4091
4092         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4093         if (err && !ret)
4094                 ret = err;
4095 out:
4096         kfree(sa);
4097         up_write(&root->fs_info->subvol_sem);
4098 drop_write:
4099         mnt_drop_write_file(file);
4100         return ret;
4101 }
4102
4103 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4104 {
4105         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4106         struct btrfs_ioctl_qgroup_assign_args *sa;
4107         struct btrfs_trans_handle *trans;
4108         int ret;
4109         int err;
4110
4111         if (!capable(CAP_SYS_ADMIN))
4112                 return -EPERM;
4113
4114         ret = mnt_want_write_file(file);
4115         if (ret)
4116                 return ret;
4117
4118         sa = memdup_user(arg, sizeof(*sa));
4119         if (IS_ERR(sa)) {
4120                 ret = PTR_ERR(sa);
4121                 goto drop_write;
4122         }
4123
4124         trans = btrfs_join_transaction(root);
4125         if (IS_ERR(trans)) {
4126                 ret = PTR_ERR(trans);
4127                 goto out;
4128         }
4129
4130         /* FIXME: check if the IDs really exist */
4131         if (sa->assign) {
4132                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4133                                                 sa->src, sa->dst);
4134         } else {
4135                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4136                                                 sa->src, sa->dst);
4137         }
4138
4139         err = btrfs_end_transaction(trans, root);
4140         if (err && !ret)
4141                 ret = err;
4142
4143 out:
4144         kfree(sa);
4145 drop_write:
4146         mnt_drop_write_file(file);
4147         return ret;
4148 }
4149
4150 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4151 {
4152         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4153         struct btrfs_ioctl_qgroup_create_args *sa;
4154         struct btrfs_trans_handle *trans;
4155         int ret;
4156         int err;
4157
4158         if (!capable(CAP_SYS_ADMIN))
4159                 return -EPERM;
4160
4161         ret = mnt_want_write_file(file);
4162         if (ret)
4163                 return ret;
4164
4165         sa = memdup_user(arg, sizeof(*sa));
4166         if (IS_ERR(sa)) {
4167                 ret = PTR_ERR(sa);
4168                 goto drop_write;
4169         }
4170
4171         if (!sa->qgroupid) {
4172                 ret = -EINVAL;
4173                 goto out;
4174         }
4175
4176         trans = btrfs_join_transaction(root);
4177         if (IS_ERR(trans)) {
4178                 ret = PTR_ERR(trans);
4179                 goto out;
4180         }
4181
4182         /* FIXME: check if the IDs really exist */
4183         if (sa->create) {
4184                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4185                                           NULL);
4186         } else {
4187                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4188         }
4189
4190         err = btrfs_end_transaction(trans, root);
4191         if (err && !ret)
4192                 ret = err;
4193
4194 out:
4195         kfree(sa);
4196 drop_write:
4197         mnt_drop_write_file(file);
4198         return ret;
4199 }
4200
4201 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4202 {
4203         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4204         struct btrfs_ioctl_qgroup_limit_args *sa;
4205         struct btrfs_trans_handle *trans;
4206         int ret;
4207         int err;
4208         u64 qgroupid;
4209
4210         if (!capable(CAP_SYS_ADMIN))
4211                 return -EPERM;
4212
4213         ret = mnt_want_write_file(file);
4214         if (ret)
4215                 return ret;
4216
4217         sa = memdup_user(arg, sizeof(*sa));
4218         if (IS_ERR(sa)) {
4219                 ret = PTR_ERR(sa);
4220                 goto drop_write;
4221         }
4222
4223         trans = btrfs_join_transaction(root);
4224         if (IS_ERR(trans)) {
4225                 ret = PTR_ERR(trans);
4226                 goto out;
4227         }
4228
4229         qgroupid = sa->qgroupid;
4230         if (!qgroupid) {
4231                 /* take the current subvol as qgroup */
4232                 qgroupid = root->root_key.objectid;
4233         }
4234
4235         /* FIXME: check if the IDs really exist */
4236         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4237
4238         err = btrfs_end_transaction(trans, root);
4239         if (err && !ret)
4240                 ret = err;
4241
4242 out:
4243         kfree(sa);
4244 drop_write:
4245         mnt_drop_write_file(file);
4246         return ret;
4247 }
4248
4249 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4250 {
4251         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4252         struct btrfs_ioctl_quota_rescan_args *qsa;
4253         int ret;
4254
4255         if (!capable(CAP_SYS_ADMIN))
4256                 return -EPERM;
4257
4258         ret = mnt_want_write_file(file);
4259         if (ret)
4260                 return ret;
4261
4262         qsa = memdup_user(arg, sizeof(*qsa));
4263         if (IS_ERR(qsa)) {
4264                 ret = PTR_ERR(qsa);
4265                 goto drop_write;
4266         }
4267
4268         if (qsa->flags) {
4269                 ret = -EINVAL;
4270                 goto out;
4271         }
4272
4273         ret = btrfs_qgroup_rescan(root->fs_info);
4274
4275 out:
4276         kfree(qsa);
4277 drop_write:
4278         mnt_drop_write_file(file);
4279         return ret;
4280 }
4281
4282 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4283 {
4284         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4285         struct btrfs_ioctl_quota_rescan_args *qsa;
4286         int ret = 0;
4287
4288         if (!capable(CAP_SYS_ADMIN))
4289                 return -EPERM;
4290
4291         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4292         if (!qsa)
4293                 return -ENOMEM;
4294
4295         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4296                 qsa->flags = 1;
4297                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4298         }
4299
4300         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4301                 ret = -EFAULT;
4302
4303         kfree(qsa);
4304         return ret;
4305 }
4306
4307 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4308 {
4309         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
4310
4311         if (!capable(CAP_SYS_ADMIN))
4312                 return -EPERM;
4313
4314         return btrfs_qgroup_wait_for_completion(root->fs_info);
4315 }
4316
4317 static long btrfs_ioctl_set_received_subvol(struct file *file,
4318                                             void __user *arg)
4319 {
4320         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4321         struct inode *inode = file_inode(file);
4322         struct btrfs_root *root = BTRFS_I(inode)->root;
4323         struct btrfs_root_item *root_item = &root->root_item;
4324         struct btrfs_trans_handle *trans;
4325         struct timespec ct = CURRENT_TIME;
4326         int ret = 0;
4327         int received_uuid_changed;
4328
4329         ret = mnt_want_write_file(file);
4330         if (ret < 0)
4331                 return ret;
4332
4333         down_write(&root->fs_info->subvol_sem);
4334
4335         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4336                 ret = -EINVAL;
4337                 goto out;
4338         }
4339
4340         if (btrfs_root_readonly(root)) {
4341                 ret = -EROFS;
4342                 goto out;
4343         }
4344
4345         if (!inode_owner_or_capable(inode)) {
4346                 ret = -EACCES;
4347                 goto out;
4348         }
4349
4350         sa = memdup_user(arg, sizeof(*sa));
4351         if (IS_ERR(sa)) {
4352                 ret = PTR_ERR(sa);
4353                 sa = NULL;
4354                 goto out;
4355         }
4356
4357         /*
4358          * 1 - root item
4359          * 2 - uuid items (received uuid + subvol uuid)
4360          */
4361         trans = btrfs_start_transaction(root, 3);
4362         if (IS_ERR(trans)) {
4363                 ret = PTR_ERR(trans);
4364                 trans = NULL;
4365                 goto out;
4366         }
4367
4368         sa->rtransid = trans->transid;
4369         sa->rtime.sec = ct.tv_sec;
4370         sa->rtime.nsec = ct.tv_nsec;
4371
4372         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4373                                        BTRFS_UUID_SIZE);
4374         if (received_uuid_changed &&
4375             !btrfs_is_empty_uuid(root_item->received_uuid))
4376                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4377                                     root_item->received_uuid,
4378                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4379                                     root->root_key.objectid);
4380         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4381         btrfs_set_root_stransid(root_item, sa->stransid);
4382         btrfs_set_root_rtransid(root_item, sa->rtransid);
4383         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4384         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4385         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4386         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4387
4388         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4389                                 &root->root_key, &root->root_item);
4390         if (ret < 0) {
4391                 btrfs_end_transaction(trans, root);
4392                 goto out;
4393         }
4394         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4395                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4396                                           sa->uuid,
4397                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4398                                           root->root_key.objectid);
4399                 if (ret < 0 && ret != -EEXIST) {
4400                         btrfs_abort_transaction(trans, root, ret);
4401                         goto out;
4402                 }
4403         }
4404         ret = btrfs_commit_transaction(trans, root);
4405         if (ret < 0) {
4406                 btrfs_abort_transaction(trans, root, ret);
4407                 goto out;
4408         }
4409
4410         ret = copy_to_user(arg, sa, sizeof(*sa));
4411         if (ret)
4412                 ret = -EFAULT;
4413
4414 out:
4415         kfree(sa);
4416         up_write(&root->fs_info->subvol_sem);
4417         mnt_drop_write_file(file);
4418         return ret;
4419 }
4420
4421 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4422 {
4423         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4424         size_t len;
4425         int ret;
4426         char label[BTRFS_LABEL_SIZE];
4427
4428         spin_lock(&root->fs_info->super_lock);
4429         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4430         spin_unlock(&root->fs_info->super_lock);
4431
4432         len = strnlen(label, BTRFS_LABEL_SIZE);
4433
4434         if (len == BTRFS_LABEL_SIZE) {
4435                 pr_warn("btrfs: label is too long, return the first %zu bytes\n",
4436                         --len);
4437         }
4438
4439         ret = copy_to_user(arg, label, len);
4440
4441         return ret ? -EFAULT : 0;
4442 }
4443
4444 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4445 {
4446         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4447         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4448         struct btrfs_trans_handle *trans;
4449         char label[BTRFS_LABEL_SIZE];
4450         int ret;
4451
4452         if (!capable(CAP_SYS_ADMIN))
4453                 return -EPERM;
4454
4455         if (copy_from_user(label, arg, sizeof(label)))
4456                 return -EFAULT;
4457
4458         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4459                 pr_err("btrfs: unable to set label with more than %d bytes\n",
4460                        BTRFS_LABEL_SIZE - 1);
4461                 return -EINVAL;
4462         }
4463
4464         ret = mnt_want_write_file(file);
4465         if (ret)
4466                 return ret;
4467
4468         trans = btrfs_start_transaction(root, 0);
4469         if (IS_ERR(trans)) {
4470                 ret = PTR_ERR(trans);
4471                 goto out_unlock;
4472         }
4473
4474         spin_lock(&root->fs_info->super_lock);
4475         strcpy(super_block->label, label);
4476         spin_unlock(&root->fs_info->super_lock);
4477         ret = btrfs_end_transaction(trans, root);
4478
4479 out_unlock:
4480         mnt_drop_write_file(file);
4481         return ret;
4482 }
4483
4484 long btrfs_ioctl(struct file *file, unsigned int
4485                 cmd, unsigned long arg)
4486 {
4487         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4488         void __user *argp = (void __user *)arg;
4489
4490         switch (cmd) {
4491         case FS_IOC_GETFLAGS:
4492                 return btrfs_ioctl_getflags(file, argp);
4493         case FS_IOC_SETFLAGS:
4494                 return btrfs_ioctl_setflags(file, argp);
4495         case FS_IOC_GETVERSION:
4496                 return btrfs_ioctl_getversion(file, argp);
4497         case FITRIM:
4498                 return btrfs_ioctl_fitrim(file, argp);
4499         case BTRFS_IOC_SNAP_CREATE:
4500                 return btrfs_ioctl_snap_create(file, argp, 0);
4501         case BTRFS_IOC_SNAP_CREATE_V2:
4502                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4503         case BTRFS_IOC_SUBVOL_CREATE:
4504                 return btrfs_ioctl_snap_create(file, argp, 1);
4505         case BTRFS_IOC_SUBVOL_CREATE_V2:
4506                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4507         case BTRFS_IOC_SNAP_DESTROY:
4508                 return btrfs_ioctl_snap_destroy(file, argp);
4509         case BTRFS_IOC_SUBVOL_GETFLAGS:
4510                 return btrfs_ioctl_subvol_getflags(file, argp);
4511         case BTRFS_IOC_SUBVOL_SETFLAGS:
4512                 return btrfs_ioctl_subvol_setflags(file, argp);
4513         case BTRFS_IOC_DEFAULT_SUBVOL:
4514                 return btrfs_ioctl_default_subvol(file, argp);
4515         case BTRFS_IOC_DEFRAG:
4516                 return btrfs_ioctl_defrag(file, NULL);
4517         case BTRFS_IOC_DEFRAG_RANGE:
4518                 return btrfs_ioctl_defrag(file, argp);
4519         case BTRFS_IOC_RESIZE:
4520                 return btrfs_ioctl_resize(file, argp);
4521         case BTRFS_IOC_ADD_DEV:
4522                 return btrfs_ioctl_add_dev(root, argp);
4523         case BTRFS_IOC_RM_DEV:
4524                 return btrfs_ioctl_rm_dev(file, argp);
4525         case BTRFS_IOC_FS_INFO:
4526                 return btrfs_ioctl_fs_info(root, argp);
4527         case BTRFS_IOC_DEV_INFO:
4528                 return btrfs_ioctl_dev_info(root, argp);
4529         case BTRFS_IOC_BALANCE:
4530                 return btrfs_ioctl_balance(file, NULL);
4531         case BTRFS_IOC_CLONE:
4532                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4533         case BTRFS_IOC_CLONE_RANGE:
4534                 return btrfs_ioctl_clone_range(file, argp);
4535         case BTRFS_IOC_TRANS_START:
4536                 return btrfs_ioctl_trans_start(file);
4537         case BTRFS_IOC_TRANS_END:
4538                 return btrfs_ioctl_trans_end(file);
4539         case BTRFS_IOC_TREE_SEARCH:
4540                 return btrfs_ioctl_tree_search(file, argp);
4541         case BTRFS_IOC_INO_LOOKUP:
4542                 return btrfs_ioctl_ino_lookup(file, argp);
4543         case BTRFS_IOC_INO_PATHS:
4544                 return btrfs_ioctl_ino_to_path(root, argp);
4545         case BTRFS_IOC_LOGICAL_INO:
4546                 return btrfs_ioctl_logical_to_ino(root, argp);
4547         case BTRFS_IOC_SPACE_INFO:
4548                 return btrfs_ioctl_space_info(root, argp);
4549         case BTRFS_IOC_SYNC: {
4550                 int ret;
4551
4552                 ret = btrfs_start_all_delalloc_inodes(root->fs_info, 0);
4553                 if (ret)
4554                         return ret;
4555                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
4556                 return ret;
4557         }
4558         case BTRFS_IOC_START_SYNC:
4559                 return btrfs_ioctl_start_sync(root, argp);
4560         case BTRFS_IOC_WAIT_SYNC:
4561                 return btrfs_ioctl_wait_sync(root, argp);
4562         case BTRFS_IOC_SCRUB:
4563                 return btrfs_ioctl_scrub(file, argp);
4564         case BTRFS_IOC_SCRUB_CANCEL:
4565                 return btrfs_ioctl_scrub_cancel(root, argp);
4566         case BTRFS_IOC_SCRUB_PROGRESS:
4567                 return btrfs_ioctl_scrub_progress(root, argp);
4568         case BTRFS_IOC_BALANCE_V2:
4569                 return btrfs_ioctl_balance(file, argp);
4570         case BTRFS_IOC_BALANCE_CTL:
4571                 return btrfs_ioctl_balance_ctl(root, arg);
4572         case BTRFS_IOC_BALANCE_PROGRESS:
4573                 return btrfs_ioctl_balance_progress(root, argp);
4574         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4575                 return btrfs_ioctl_set_received_subvol(file, argp);
4576         case BTRFS_IOC_SEND:
4577                 return btrfs_ioctl_send(file, argp);
4578         case BTRFS_IOC_GET_DEV_STATS:
4579                 return btrfs_ioctl_get_dev_stats(root, argp);
4580         case BTRFS_IOC_QUOTA_CTL:
4581                 return btrfs_ioctl_quota_ctl(file, argp);
4582         case BTRFS_IOC_QGROUP_ASSIGN:
4583                 return btrfs_ioctl_qgroup_assign(file, argp);
4584         case BTRFS_IOC_QGROUP_CREATE:
4585                 return btrfs_ioctl_qgroup_create(file, argp);
4586         case BTRFS_IOC_QGROUP_LIMIT:
4587                 return btrfs_ioctl_qgroup_limit(file, argp);
4588         case BTRFS_IOC_QUOTA_RESCAN:
4589                 return btrfs_ioctl_quota_rescan(file, argp);
4590         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4591                 return btrfs_ioctl_quota_rescan_status(file, argp);
4592         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4593                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4594         case BTRFS_IOC_DEV_REPLACE:
4595                 return btrfs_ioctl_dev_replace(root, argp);
4596         case BTRFS_IOC_GET_FSLABEL:
4597                 return btrfs_ioctl_get_fslabel(file, argp);
4598         case BTRFS_IOC_SET_FSLABEL:
4599                 return btrfs_ioctl_set_fslabel(file, argp);
4600         case BTRFS_IOC_FILE_EXTENT_SAME:
4601                 return btrfs_ioctl_file_extent_same(file, argp);
4602         }
4603
4604         return -ENOTTY;
4605 }