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(file_inode(file)->i_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(leaf),
 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 btrfs_key max_key;
 846         struct extent_buffer *leaf;
 847         struct btrfs_file_extent_item *extent;
 848         int type;
 849         int ret;
 850         u64 ino = btrfs_ino(inode);
 851
 852         path = btrfs_alloc_path();
 853         if (!path)
 854                 return -ENOMEM;
 855
 856         min_key.objectid = ino;
 857         min_key.type = BTRFS_EXTENT_DATA_KEY;
 858         min_key.offset = *off;
 859
 860         max_key.objectid = ino;
 861         max_key.type = (u8)-1;
 862         max_key.offset = (u64)-1;
 863
 864         path->keep_locks = 1;
 865
 866         while(1) {
 867                 ret = btrfs_search_forward(root, &min_key, &max_key,
 868                                            path, newer_than);
 869                 if (ret != 0)
 870                         goto none;
 871                 if (min_key.objectid != ino)
 872                         goto none;
 873                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 874                         goto none;
 875
 876                 leaf = path->nodes[0];
 877                 extent = btrfs_item_ptr(leaf, path->slots[0],
 878                                         struct btrfs_file_extent_item);
 879
 880                 type = btrfs_file_extent_type(leaf, extent);
 881                 if (type == BTRFS_FILE_EXTENT_REG &&
 882                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 883                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 884                         *off = min_key.offset;
 885                         btrfs_free_path(path);
 886                         return 0;
 887                 }
 888
 889                 if (min_key.offset == (u64)-1)
 890                         goto none;
 891
 892                 min_key.offset++;
 893                 btrfs_release_path(path);
 894         }
 895 none:
 896         btrfs_free_path(path);
 897         return -ENOENT;
 898 }
 899
 900 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 901 {
 902         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 903         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 904         struct extent_map *em;
 905         u64 len = PAGE_CACHE_SIZE;
 906
 907         /*
 908          * hopefully we have this extent in the tree already, try without
 909          * the full extent lock
 910          */
 911         read_lock(&em_tree->lock);
 912         em = lookup_extent_mapping(em_tree, start, len);
 913         read_unlock(&em_tree->lock);
 914
 915         if (!em) {
 916                 /* get the big lock and read metadata off disk */
 917                 lock_extent(io_tree, start, start + len - 1);
 918                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 919                 unlock_extent(io_tree, start, start + len - 1);
 920
 921                 if (IS_ERR(em))
 922                         return NULL;
 923         }
 924
 925         return em;
 926 }
 927
 928 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 929 {
 930         struct extent_map *next;
 931         bool ret = true;
 932
 933         /* this is the last extent */
 934         if (em->start + em->len >= i_size_read(inode))
 935                 return false;
 936
 937         next = defrag_lookup_extent(inode, em->start + em->len);
 938         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 939                 ret = false;
 940
 941         free_extent_map(next);
 942         return ret;
 943 }
 944
 945 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
 946                                u64 *last_len, u64 *skip, u64 *defrag_end,
 947                                int compress)
 948 {
 949         struct extent_map *em;
 950         int ret = 1;
 951         bool next_mergeable = true;
 952
 953         /*
 954          * make sure that once we start defragging an extent, we keep on
 955          * defragging it
 956          */
 957         if (start < *defrag_end)
 958                 return 1;
 959
 960         *skip = 0;
 961
 962         em = defrag_lookup_extent(inode, start);
 963         if (!em)
 964                 return 0;
 965
 966         /* this will cover holes, and inline extents */
 967         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 968                 ret = 0;
 969                 goto out;
 970         }
 971
 972         next_mergeable = defrag_check_next_extent(inode, em);
 973
 974         /*
 975          * we hit a real extent, if it is big or the next extent is not a
 976          * real extent, don't bother defragging it
 977          */
 978         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
 979             (em->len >= thresh || !next_mergeable))
 980                 ret = 0;
 981 out:
 982         /*
 983          * last_len ends up being a counter of how many bytes we've defragged.
 984          * every time we choose not to defrag an extent, we reset *last_len
 985          * so that the next tiny extent will force a defrag.
 986          *
 987          * The end result of this is that tiny extents before a single big
 988          * extent will force at least part of that big extent to be defragged.
 989          */
 990         if (ret) {
 991                 *defrag_end = extent_map_end(em);
 992         } else {
 993                 *last_len = 0;
 994                 *skip = extent_map_end(em);
 995                 *defrag_end = 0;
 996         }
 997
 998         free_extent_map(em);
 999         return ret;
1000 }
1001
1002 /*
1003  * it doesn't do much good to defrag one or two pages
1004  * at a time.  This pulls in a nice chunk of pages
1005  * to COW and defrag.
1006  *
1007  * It also makes sure the delalloc code has enough
1008  * dirty data to avoid making new small extents as part
1009  * of the defrag
1010  *
1011  * It's a good idea to start RA on this range
1012  * before calling this.
1013  */
1014 static int cluster_pages_for_defrag(struct inode *inode,
1015                                     struct page **pages,
1016                                     unsigned long start_index,
1017                                     int num_pages)
1018 {
1019         unsigned long file_end;
1020         u64 isize = i_size_read(inode);
1021         u64 page_start;
1022         u64 page_end;
1023         u64 page_cnt;
1024         int ret;
1025         int i;
1026         int i_done;
1027         struct btrfs_ordered_extent *ordered;
1028         struct extent_state *cached_state = NULL;
1029         struct extent_io_tree *tree;
1030         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1031
1032         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1033         if (!isize || start_index > file_end)
1034                 return 0;
1035
1036         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1037
1038         ret = btrfs_delalloc_reserve_space(inode,
1039                                            page_cnt << PAGE_CACHE_SHIFT);
1040         if (ret)
1041                 return ret;
1042         i_done = 0;
1043         tree = &BTRFS_I(inode)->io_tree;
1044
1045         /* step one, lock all the pages */
1046         for (i = 0; i < page_cnt; i++) {
1047                 struct page *page;
1048 again:
1049                 page = find_or_create_page(inode->i_mapping,
1050                                            start_index + i, mask);
1051                 if (!page)
1052                         break;
1053
1054                 page_start = page_offset(page);
1055                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1056                 while (1) {
1057                         lock_extent(tree, page_start, page_end);
1058                         ordered = btrfs_lookup_ordered_extent(inode,
1059                                                               page_start);
1060                         unlock_extent(tree, page_start, page_end);
1061                         if (!ordered)
1062                                 break;
1063
1064                         unlock_page(page);
1065                         btrfs_start_ordered_extent(inode, ordered, 1);
1066                         btrfs_put_ordered_extent(ordered);
1067                         lock_page(page);
1068                         /*
1069                          * we unlocked the page above, so we need check if
1070                          * it was released or not.
1071                          */
1072                         if (page->mapping != inode->i_mapping) {
1073                                 unlock_page(page);
1074                                 page_cache_release(page);
1075                                 goto again;
1076                         }
1077                 }
1078
1079                 if (!PageUptodate(page)) {
1080                         btrfs_readpage(NULL, page);
1081                         lock_page(page);
1082                         if (!PageUptodate(page)) {
1083                                 unlock_page(page);
1084                                 page_cache_release(page);
1085                                 ret = -EIO;
1086                                 break;
1087                         }
1088                 }
1089
1090                 if (page->mapping != inode->i_mapping) {
1091                         unlock_page(page);
1092                         page_cache_release(page);
1093                         goto again;
1094                 }
1095
1096                 pages[i] = page;
1097                 i_done++;
1098         }
1099         if (!i_done || ret)
1100                 goto out;
1101
1102         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1103                 goto out;
1104
1105         /*
1106          * so now we have a nice long stream of locked
1107          * and up to date pages, lets wait on them
1108          */
1109         for (i = 0; i < i_done; i++)
1110                 wait_on_page_writeback(pages[i]);
1111
1112         page_start = page_offset(pages[0]);
1113         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1114
1115         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1116                          page_start, page_end - 1, 0, &cached_state);
1117         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1118                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1119                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1120                           &cached_state, GFP_NOFS);
1121
1122         if (i_done != page_cnt) {
1123                 spin_lock(&BTRFS_I(inode)->lock);
1124                 BTRFS_I(inode)->outstanding_extents++;
1125                 spin_unlock(&BTRFS_I(inode)->lock);
1126                 btrfs_delalloc_release_space(inode,
1127                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1128         }
1129
1130
1131         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1132                           &cached_state, GFP_NOFS);
1133
1134         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1135                              page_start, page_end - 1, &cached_state,
1136                              GFP_NOFS);
1137
1138         for (i = 0; i < i_done; i++) {
1139                 clear_page_dirty_for_io(pages[i]);
1140                 ClearPageChecked(pages[i]);
1141                 set_page_extent_mapped(pages[i]);
1142                 set_page_dirty(pages[i]);
1143                 unlock_page(pages[i]);
1144                 page_cache_release(pages[i]);
1145         }
1146         return i_done;
1147 out:
1148         for (i = 0; i < i_done; i++) {
1149                 unlock_page(pages[i]);
1150                 page_cache_release(pages[i]);
1151         }
1152         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1153         return ret;
1154
1155 }
1156
1157 int btrfs_defrag_file(struct inode *inode, struct file *file,
1158                       struct btrfs_ioctl_defrag_range_args *range,
1159                       u64 newer_than, unsigned long max_to_defrag)
1160 {
1161         struct btrfs_root *root = BTRFS_I(inode)->root;
1162         struct file_ra_state *ra = NULL;
1163         unsigned long last_index;
1164         u64 isize = i_size_read(inode);
1165         u64 last_len = 0;
1166         u64 skip = 0;
1167         u64 defrag_end = 0;
1168         u64 newer_off = range->start;
1169         unsigned long i;
1170         unsigned long ra_index = 0;
1171         int ret;
1172         int defrag_count = 0;
1173         int compress_type = BTRFS_COMPRESS_ZLIB;
1174         int extent_thresh = range->extent_thresh;
1175         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1176         int cluster = max_cluster;
1177         u64 new_align = ~((u64)128 * 1024 - 1);
1178         struct page **pages = NULL;
1179
1180         if (isize == 0)
1181                 return 0;
1182
1183         if (range->start >= isize)
1184                 return -EINVAL;
1185
1186         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1187                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1188                         return -EINVAL;
1189                 if (range->compress_type)
1190                         compress_type = range->compress_type;
1191         }
1192
1193         if (extent_thresh == 0)
1194                 extent_thresh = 256 * 1024;
1195
1196         /*
1197          * if we were not given a file, allocate a readahead
1198          * context
1199          */
1200         if (!file) {
1201                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1202                 if (!ra)
1203                         return -ENOMEM;
1204                 file_ra_state_init(ra, inode->i_mapping);
1205         } else {
1206                 ra = &file->f_ra;
1207         }
1208
1209         pages = kmalloc(sizeof(struct page *) * max_cluster,
1210                         GFP_NOFS);
1211         if (!pages) {
1212                 ret = -ENOMEM;
1213                 goto out_ra;
1214         }
1215
1216         /* find the last page to defrag */
1217         if (range->start + range->len > range->start) {
1218                 last_index = min_t(u64, isize - 1,
1219                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1220         } else {
1221                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1222         }
1223
1224         if (newer_than) {
1225                 ret = find_new_extents(root, inode, newer_than,
1226                                        &newer_off, 64 * 1024);
1227                 if (!ret) {
1228                         range->start = newer_off;
1229                         /*
1230                          * we always align our defrag to help keep
1231                          * the extents in the file evenly spaced
1232                          */
1233                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1234                 } else
1235                         goto out_ra;
1236         } else {
1237                 i = range->start >> PAGE_CACHE_SHIFT;
1238         }
1239         if (!max_to_defrag)
1240                 max_to_defrag = last_index + 1;
1241
1242         /*
1243          * make writeback starts from i, so the defrag range can be
1244          * written sequentially.
1245          */
1246         if (i < inode->i_mapping->writeback_index)
1247                 inode->i_mapping->writeback_index = i;
1248
1249         while (i <= last_index && defrag_count < max_to_defrag &&
1250                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1251                 PAGE_CACHE_SHIFT)) {
1252                 /*
1253                  * make sure we stop running if someone unmounts
1254                  * the FS
1255                  */
1256                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1257                         break;
1258
1259                 if (btrfs_defrag_cancelled(root->fs_info)) {
1260                         printk(KERN_DEBUG "btrfs: defrag_file cancelled\n");
1261                         ret = -EAGAIN;
1262                         break;
1263                 }
1264
1265                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1266                                          extent_thresh, &last_len, &skip,
1267                                          &defrag_end, range->flags &
1268                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1269                         unsigned long next;
1270                         /*
1271                          * the should_defrag function tells us how much to skip
1272                          * bump our counter by the suggested amount
1273                          */
1274                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1275                         i = max(i + 1, next);
1276                         continue;
1277                 }
1278
1279                 if (!newer_than) {
1280                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1281                                    PAGE_CACHE_SHIFT) - i;
1282                         cluster = min(cluster, max_cluster);
1283                 } else {
1284                         cluster = max_cluster;
1285                 }
1286
1287                 if (i + cluster > ra_index) {
1288                         ra_index = max(i, ra_index);
1289                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1290                                        cluster);
1291                         ra_index += max_cluster;
1292                 }
1293
1294                 mutex_lock(&inode->i_mutex);
1295                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1296                         BTRFS_I(inode)->force_compress = compress_type;
1297                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1298                 if (ret < 0) {
1299                         mutex_unlock(&inode->i_mutex);
1300                         goto out_ra;
1301                 }
1302
1303                 defrag_count += ret;
1304                 balance_dirty_pages_ratelimited(inode->i_mapping);
1305                 mutex_unlock(&inode->i_mutex);
1306
1307                 if (newer_than) {
1308                         if (newer_off == (u64)-1)
1309                                 break;
1310
1311                         if (ret > 0)
1312                                 i += ret;
1313
1314                         newer_off = max(newer_off + 1,
1315                                         (u64)i << PAGE_CACHE_SHIFT);
1316
1317                         ret = find_new_extents(root, inode,
1318                                                newer_than, &newer_off,
1319                                                64 * 1024);
1320                         if (!ret) {
1321                                 range->start = newer_off;
1322                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1323                         } else {
1324                                 break;
1325                         }
1326                 } else {
1327                         if (ret > 0) {
1328                                 i += ret;
1329                                 last_len += ret << PAGE_CACHE_SHIFT;
1330                         } else {
1331                                 i++;
1332                                 last_len = 0;
1333                         }
1334                 }
1335         }
1336
1337         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1338                 filemap_flush(inode->i_mapping);
1339
1340         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1341                 /* the filemap_flush will queue IO into the worker threads, but
1342                  * we have to make sure the IO is actually started and that
1343                  * ordered extents get created before we return
1344                  */
1345                 atomic_inc(&root->fs_info->async_submit_draining);
1346                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1347                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1348                         wait_event(root->fs_info->async_submit_wait,
1349                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1350                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1351                 }
1352                 atomic_dec(&root->fs_info->async_submit_draining);
1353         }
1354
1355         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1356                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1357         }
1358
1359         ret = defrag_count;
1360
1361 out_ra:
1362         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1363                 mutex_lock(&inode->i_mutex);
1364                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1365                 mutex_unlock(&inode->i_mutex);
1366         }
1367         if (!file)
1368                 kfree(ra);
1369         kfree(pages);
1370         return ret;
1371 }
1372
1373 static noinline int btrfs_ioctl_resize(struct file *file,
1374                                         void __user *arg)
1375 {
1376         u64 new_size;
1377         u64 old_size;
1378         u64 devid = 1;
1379         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1380         struct btrfs_ioctl_vol_args *vol_args;
1381         struct btrfs_trans_handle *trans;
1382         struct btrfs_device *device = NULL;
1383         char *sizestr;
1384         char *devstr = NULL;
1385         int ret = 0;
1386         int mod = 0;
1387
1388         if (!capable(CAP_SYS_ADMIN))
1389                 return -EPERM;
1390
1391         ret = mnt_want_write_file(file);
1392         if (ret)
1393                 return ret;
1394
1395         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1396                         1)) {
1397                 mnt_drop_write_file(file);
1398                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1399         }
1400
1401         mutex_lock(&root->fs_info->volume_mutex);
1402         vol_args = memdup_user(arg, sizeof(*vol_args));
1403         if (IS_ERR(vol_args)) {
1404                 ret = PTR_ERR(vol_args);
1405                 goto out;
1406         }
1407
1408         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1409
1410         sizestr = vol_args->name;
1411         devstr = strchr(sizestr, ':');
1412         if (devstr) {
1413                 char *end;
1414                 sizestr = devstr + 1;
1415                 *devstr = '\0';
1416                 devstr = vol_args->name;
1417                 devid = simple_strtoull(devstr, &end, 10);
1418                 if (!devid) {
1419                         ret = -EINVAL;
1420                         goto out_free;
1421                 }
1422                 printk(KERN_INFO "btrfs: resizing devid %llu\n", devid);
1423         }
1424
1425         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1426         if (!device) {
1427                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1428                        devid);
1429                 ret = -ENODEV;
1430                 goto out_free;
1431         }
1432
1433         if (!device->writeable) {
1434                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1435                        "readonly device %llu\n",
1436                        devid);
1437                 ret = -EPERM;
1438                 goto out_free;
1439         }
1440
1441         if (!strcmp(sizestr, "max"))
1442                 new_size = device->bdev->bd_inode->i_size;
1443         else {
1444                 if (sizestr[0] == '-') {
1445                         mod = -1;
1446                         sizestr++;
1447                 } else if (sizestr[0] == '+') {
1448                         mod = 1;
1449                         sizestr++;
1450                 }
1451                 new_size = memparse(sizestr, NULL);
1452                 if (new_size == 0) {
1453                         ret = -EINVAL;
1454                         goto out_free;
1455                 }
1456         }
1457
1458         if (device->is_tgtdev_for_dev_replace) {
1459                 ret = -EPERM;
1460                 goto out_free;
1461         }
1462
1463         old_size = device->total_bytes;
1464
1465         if (mod < 0) {
1466                 if (new_size > old_size) {
1467                         ret = -EINVAL;
1468                         goto out_free;
1469                 }
1470                 new_size = old_size - new_size;
1471         } else if (mod > 0) {
1472                 new_size = old_size + new_size;
1473         }
1474
1475         if (new_size < 256 * 1024 * 1024) {
1476                 ret = -EINVAL;
1477                 goto out_free;
1478         }
1479         if (new_size > device->bdev->bd_inode->i_size) {
1480                 ret = -EFBIG;
1481                 goto out_free;
1482         }
1483
1484         do_div(new_size, root->sectorsize);
1485         new_size *= root->sectorsize;
1486
1487         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1488                       rcu_str_deref(device->name), new_size);
1489
1490         if (new_size > old_size) {
1491                 trans = btrfs_start_transaction(root, 0);
1492                 if (IS_ERR(trans)) {
1493                         ret = PTR_ERR(trans);
1494                         goto out_free;
1495                 }
1496                 ret = btrfs_grow_device(trans, device, new_size);
1497                 btrfs_commit_transaction(trans, root);
1498         } else if (new_size < old_size) {
1499                 ret = btrfs_shrink_device(device, new_size);
1500         } /* equal, nothing need to do */
1501
1502 out_free:
1503         kfree(vol_args);
1504 out:
1505         mutex_unlock(&root->fs_info->volume_mutex);
1506         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1507         mnt_drop_write_file(file);
1508         return ret;
1509 }
1510
1511 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1512                                 char *name, unsigned long fd, int subvol,
1513                                 u64 *transid, bool readonly,
1514                                 struct btrfs_qgroup_inherit *inherit)
1515 {
1516         int namelen;
1517         int ret = 0;
1518
1519         ret = mnt_want_write_file(file);
1520         if (ret)
1521                 goto out;
1522
1523         namelen = strlen(name);
1524         if (strchr(name, '/')) {
1525                 ret = -EINVAL;
1526                 goto out_drop_write;
1527         }
1528
1529         if (name[0] == '.' &&
1530            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1531                 ret = -EEXIST;
1532                 goto out_drop_write;
1533         }
1534
1535         if (subvol) {
1536                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1537                                      NULL, transid, readonly, inherit);
1538         } else {
1539                 struct fd src = fdget(fd);
1540                 struct inode *src_inode;
1541                 if (!src.file) {
1542                         ret = -EINVAL;
1543                         goto out_drop_write;
1544                 }
1545
1546                 src_inode = file_inode(src.file);
1547                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1548                         printk(KERN_INFO "btrfs: Snapshot src from "
1549                                "another FS\n");
1550                         ret = -EINVAL;
1551                 } else {
1552                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1553                                              BTRFS_I(src_inode)->root,
1554                                              transid, readonly, inherit);
1555                 }
1556                 fdput(src);
1557         }
1558 out_drop_write:
1559         mnt_drop_write_file(file);
1560 out:
1561         return ret;
1562 }
1563
1564 static noinline int btrfs_ioctl_snap_create(struct file *file,
1565                                             void __user *arg, int subvol)
1566 {
1567         struct btrfs_ioctl_vol_args *vol_args;
1568         int ret;
1569
1570         vol_args = memdup_user(arg, sizeof(*vol_args));
1571         if (IS_ERR(vol_args))
1572                 return PTR_ERR(vol_args);
1573         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1574
1575         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1576                                               vol_args->fd, subvol,
1577                                               NULL, false, NULL);
1578
1579         kfree(vol_args);
1580         return ret;
1581 }
1582
1583 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1584                                                void __user *arg, int subvol)
1585 {
1586         struct btrfs_ioctl_vol_args_v2 *vol_args;
1587         int ret;
1588         u64 transid = 0;
1589         u64 *ptr = NULL;
1590         bool readonly = false;
1591         struct btrfs_qgroup_inherit *inherit = NULL;
1592
1593         vol_args = memdup_user(arg, sizeof(*vol_args));
1594         if (IS_ERR(vol_args))
1595                 return PTR_ERR(vol_args);
1596         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1597
1598         if (vol_args->flags &
1599             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1600               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1601                 ret = -EOPNOTSUPP;
1602                 goto out;
1603         }
1604
1605         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1606                 ptr = &transid;
1607         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1608                 readonly = true;
1609         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1610                 if (vol_args->size > PAGE_CACHE_SIZE) {
1611                         ret = -EINVAL;
1612                         goto out;
1613                 }
1614                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1615                 if (IS_ERR(inherit)) {
1616                         ret = PTR_ERR(inherit);
1617                         goto out;
1618                 }
1619         }
1620
1621         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1622                                               vol_args->fd, subvol, ptr,
1623                                               readonly, inherit);
1624
1625         if (ret == 0 && ptr &&
1626             copy_to_user(arg +
1627                          offsetof(struct btrfs_ioctl_vol_args_v2,
1628                                   transid), ptr, sizeof(*ptr)))
1629                 ret = -EFAULT;
1630 out:
1631         kfree(vol_args);
1632         kfree(inherit);
1633         return ret;
1634 }
1635
1636 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1637                                                 void __user *arg)
1638 {
1639         struct inode *inode = file_inode(file);
1640         struct btrfs_root *root = BTRFS_I(inode)->root;
1641         int ret = 0;
1642         u64 flags = 0;
1643
1644         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1645                 return -EINVAL;
1646
1647         down_read(&root->fs_info->subvol_sem);
1648         if (btrfs_root_readonly(root))
1649                 flags |= BTRFS_SUBVOL_RDONLY;
1650         up_read(&root->fs_info->subvol_sem);
1651
1652         if (copy_to_user(arg, &flags, sizeof(flags)))
1653                 ret = -EFAULT;
1654
1655         return ret;
1656 }
1657
1658 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1659                                               void __user *arg)
1660 {
1661         struct inode *inode = file_inode(file);
1662         struct btrfs_root *root = BTRFS_I(inode)->root;
1663         struct btrfs_trans_handle *trans;
1664         u64 root_flags;
1665         u64 flags;
1666         int ret = 0;
1667
1668         ret = mnt_want_write_file(file);
1669         if (ret)
1670                 goto out;
1671
1672         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1673                 ret = -EINVAL;
1674                 goto out_drop_write;
1675         }
1676
1677         if (copy_from_user(&flags, arg, sizeof(flags))) {
1678                 ret = -EFAULT;
1679                 goto out_drop_write;
1680         }
1681
1682         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1683                 ret = -EINVAL;
1684                 goto out_drop_write;
1685         }
1686
1687         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1688                 ret = -EOPNOTSUPP;
1689                 goto out_drop_write;
1690         }
1691
1692         if (!inode_owner_or_capable(inode)) {
1693                 ret = -EACCES;
1694                 goto out_drop_write;
1695         }
1696
1697         down_write(&root->fs_info->subvol_sem);
1698
1699         /* nothing to do */
1700         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1701                 goto out_drop_sem;
1702
1703         root_flags = btrfs_root_flags(&root->root_item);
1704         if (flags & BTRFS_SUBVOL_RDONLY)
1705                 btrfs_set_root_flags(&root->root_item,
1706                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1707         else
1708                 btrfs_set_root_flags(&root->root_item,
1709                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1710
1711         trans = btrfs_start_transaction(root, 1);
1712         if (IS_ERR(trans)) {
1713                 ret = PTR_ERR(trans);
1714                 goto out_reset;
1715         }
1716
1717         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1718                                 &root->root_key, &root->root_item);
1719
1720         btrfs_commit_transaction(trans, root);
1721 out_reset:
1722         if (ret)
1723                 btrfs_set_root_flags(&root->root_item, root_flags);
1724 out_drop_sem:
1725         up_write(&root->fs_info->subvol_sem);
1726 out_drop_write:
1727         mnt_drop_write_file(file);
1728 out:
1729         return ret;
1730 }
1731
1732 /*
1733  * helper to check if the subvolume references other subvolumes
1734  */
1735 static noinline int may_destroy_subvol(struct btrfs_root *root)
1736 {
1737         struct btrfs_path *path;
1738         struct btrfs_dir_item *di;
1739         struct btrfs_key key;
1740         u64 dir_id;
1741         int ret;
1742
1743         path = btrfs_alloc_path();
1744         if (!path)
1745                 return -ENOMEM;
1746
1747         /* Make sure this root isn't set as the default subvol */
1748         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1749         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1750                                    dir_id, "default", 7, 0);
1751         if (di && !IS_ERR(di)) {
1752                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1753                 if (key.objectid == root->root_key.objectid) {
1754                         ret = -ENOTEMPTY;
1755                         goto out;
1756                 }
1757                 btrfs_release_path(path);
1758         }
1759
1760         key.objectid = root->root_key.objectid;
1761         key.type = BTRFS_ROOT_REF_KEY;
1762         key.offset = (u64)-1;
1763
1764         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1765                                 &key, path, 0, 0);
1766         if (ret < 0)
1767                 goto out;
1768         BUG_ON(ret == 0);
1769
1770         ret = 0;
1771         if (path->slots[0] > 0) {
1772                 path->slots[0]--;
1773                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1774                 if (key.objectid == root->root_key.objectid &&
1775                     key.type == BTRFS_ROOT_REF_KEY)
1776                         ret = -ENOTEMPTY;
1777         }
1778 out:
1779         btrfs_free_path(path);
1780         return ret;
1781 }
1782
1783 static noinline int key_in_sk(struct btrfs_key *key,
1784                               struct btrfs_ioctl_search_key *sk)
1785 {
1786         struct btrfs_key test;
1787         int ret;
1788
1789         test.objectid = sk->min_objectid;
1790         test.type = sk->min_type;
1791         test.offset = sk->min_offset;
1792
1793         ret = btrfs_comp_cpu_keys(key, &test);
1794         if (ret < 0)
1795                 return 0;
1796
1797         test.objectid = sk->max_objectid;
1798         test.type = sk->max_type;
1799         test.offset = sk->max_offset;
1800
1801         ret = btrfs_comp_cpu_keys(key, &test);
1802         if (ret > 0)
1803                 return 0;
1804         return 1;
1805 }
1806
1807 static noinline int copy_to_sk(struct btrfs_root *root,
1808                                struct btrfs_path *path,
1809                                struct btrfs_key *key,
1810                                struct btrfs_ioctl_search_key *sk,
1811                                char *buf,
1812                                unsigned long *sk_offset,
1813                                int *num_found)
1814 {
1815         u64 found_transid;
1816         struct extent_buffer *leaf;
1817         struct btrfs_ioctl_search_header sh;
1818         unsigned long item_off;
1819         unsigned long item_len;
1820         int nritems;
1821         int i;
1822         int slot;
1823         int ret = 0;
1824
1825         leaf = path->nodes[0];
1826         slot = path->slots[0];
1827         nritems = btrfs_header_nritems(leaf);
1828
1829         if (btrfs_header_generation(leaf) > sk->max_transid) {
1830                 i = nritems;
1831                 goto advance_key;
1832         }
1833         found_transid = btrfs_header_generation(leaf);
1834
1835         for (i = slot; i < nritems; i++) {
1836                 item_off = btrfs_item_ptr_offset(leaf, i);
1837                 item_len = btrfs_item_size_nr(leaf, i);
1838
1839                 btrfs_item_key_to_cpu(leaf, key, i);
1840                 if (!key_in_sk(key, sk))
1841                         continue;
1842
1843                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1844                         item_len = 0;
1845
1846                 if (sizeof(sh) + item_len + *sk_offset >
1847                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1848                         ret = 1;
1849                         goto overflow;
1850                 }
1851
1852                 sh.objectid = key->objectid;
1853                 sh.offset = key->offset;
1854                 sh.type = key->type;
1855                 sh.len = item_len;
1856                 sh.transid = found_transid;
1857
1858                 /* copy search result header */
1859                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1860                 *sk_offset += sizeof(sh);
1861
1862                 if (item_len) {
1863                         char *p = buf + *sk_offset;
1864                         /* copy the item */
1865                         read_extent_buffer(leaf, p,
1866                                            item_off, item_len);
1867                         *sk_offset += item_len;
1868                 }
1869                 (*num_found)++;
1870
1871                 if (*num_found >= sk->nr_items)
1872                         break;
1873         }
1874 advance_key:
1875         ret = 0;
1876         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1877                 key->offset++;
1878         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1879                 key->offset = 0;
1880                 key->type++;
1881         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1882                 key->offset = 0;
1883                 key->type = 0;
1884                 key->objectid++;
1885         } else
1886                 ret = 1;
1887 overflow:
1888         return ret;
1889 }
1890
1891 static noinline int search_ioctl(struct inode *inode,
1892                                  struct btrfs_ioctl_search_args *args)
1893 {
1894         struct btrfs_root *root;
1895         struct btrfs_key key;
1896         struct btrfs_key max_key;
1897         struct btrfs_path *path;
1898         struct btrfs_ioctl_search_key *sk = &args->key;
1899         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1900         int ret;
1901         int num_found = 0;
1902         unsigned long sk_offset = 0;
1903
1904         path = btrfs_alloc_path();
1905         if (!path)
1906                 return -ENOMEM;
1907
1908         if (sk->tree_id == 0) {
1909                 /* search the root of the inode that was passed */
1910                 root = BTRFS_I(inode)->root;
1911         } else {
1912                 key.objectid = sk->tree_id;
1913                 key.type = BTRFS_ROOT_ITEM_KEY;
1914                 key.offset = (u64)-1;
1915                 root = btrfs_read_fs_root_no_name(info, &key);
1916                 if (IS_ERR(root)) {
1917                         printk(KERN_ERR "could not find root %llu\n",
1918                                sk->tree_id);
1919                         btrfs_free_path(path);
1920                         return -ENOENT;
1921                 }
1922         }
1923
1924         key.objectid = sk->min_objectid;
1925         key.type = sk->min_type;
1926         key.offset = sk->min_offset;
1927
1928         max_key.objectid = sk->max_objectid;
1929         max_key.type = sk->max_type;
1930         max_key.offset = sk->max_offset;
1931
1932         path->keep_locks = 1;
1933
1934         while(1) {
1935                 ret = btrfs_search_forward(root, &key, &max_key, path,
1936                                            sk->min_transid);
1937                 if (ret != 0) {
1938                         if (ret > 0)
1939                                 ret = 0;
1940                         goto err;
1941                 }
1942                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1943                                  &sk_offset, &num_found);
1944                 btrfs_release_path(path);
1945                 if (ret || num_found >= sk->nr_items)
1946                         break;
1947
1948         }
1949         ret = 0;
1950 err:
1951         sk->nr_items = num_found;
1952         btrfs_free_path(path);
1953         return ret;
1954 }
1955
1956 static noinline int btrfs_ioctl_tree_search(struct file *file,
1957                                            void __user *argp)
1958 {
1959          struct btrfs_ioctl_search_args *args;
1960          struct inode *inode;
1961          int ret;
1962
1963         if (!capable(CAP_SYS_ADMIN))
1964                 return -EPERM;
1965
1966         args = memdup_user(argp, sizeof(*args));
1967         if (IS_ERR(args))
1968                 return PTR_ERR(args);
1969
1970         inode = file_inode(file);
1971         ret = search_ioctl(inode, args);
1972         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1973                 ret = -EFAULT;
1974         kfree(args);
1975         return ret;
1976 }
1977
1978 /*
1979  * Search INODE_REFs to identify path name of 'dirid' directory
1980  * in a 'tree_id' tree. and sets path name to 'name'.
1981  */
1982 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1983                                 u64 tree_id, u64 dirid, char *name)
1984 {
1985         struct btrfs_root *root;
1986         struct btrfs_key key;
1987         char *ptr;
1988         int ret = -1;
1989         int slot;
1990         int len;
1991         int total_len = 0;
1992         struct btrfs_inode_ref *iref;
1993         struct extent_buffer *l;
1994         struct btrfs_path *path;
1995
1996         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1997                 name[0]='\0';
1998                 return 0;
1999         }
2000
2001         path = btrfs_alloc_path();
2002         if (!path)
2003                 return -ENOMEM;
2004
2005         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2006
2007         key.objectid = tree_id;
2008         key.type = BTRFS_ROOT_ITEM_KEY;
2009         key.offset = (u64)-1;
2010         root = btrfs_read_fs_root_no_name(info, &key);
2011         if (IS_ERR(root)) {
2012                 printk(KERN_ERR "could not find root %llu\n", tree_id);
2013                 ret = -ENOENT;
2014                 goto out;
2015         }
2016
2017         key.objectid = dirid;
2018         key.type = BTRFS_INODE_REF_KEY;
2019         key.offset = (u64)-1;
2020
2021         while(1) {
2022                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2023                 if (ret < 0)
2024                         goto out;
2025                 else if (ret > 0) {
2026                         ret = btrfs_previous_item(root, path, dirid,
2027                                                   BTRFS_INODE_REF_KEY);
2028                         if (ret < 0)
2029                                 goto out;
2030                         else if (ret > 0) {
2031                                 ret = -ENOENT;
2032                                 goto out;
2033                         }
2034                 }
2035
2036                 l = path->nodes[0];
2037                 slot = path->slots[0];
2038                 btrfs_item_key_to_cpu(l, &key, slot);
2039
2040                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2041                 len = btrfs_inode_ref_name_len(l, iref);
2042                 ptr -= len + 1;
2043                 total_len += len + 1;
2044                 if (ptr < name) {
2045                         ret = -ENAMETOOLONG;
2046                         goto out;
2047                 }
2048
2049                 *(ptr + len) = '/';
2050                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
2051
2052                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2053                         break;
2054
2055                 btrfs_release_path(path);
2056                 key.objectid = key.offset;
2057                 key.offset = (u64)-1;
2058                 dirid = key.objectid;
2059         }
2060         memmove(name, ptr, total_len);
2061         name[total_len]='\0';
2062         ret = 0;
2063 out:
2064         btrfs_free_path(path);
2065         return ret;
2066 }
2067
2068 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2069                                            void __user *argp)
2070 {
2071          struct btrfs_ioctl_ino_lookup_args *args;
2072          struct inode *inode;
2073          int ret;
2074
2075         if (!capable(CAP_SYS_ADMIN))
2076                 return -EPERM;
2077
2078         args = memdup_user(argp, sizeof(*args));
2079         if (IS_ERR(args))
2080                 return PTR_ERR(args);
2081
2082         inode = file_inode(file);
2083
2084         if (args->treeid == 0)
2085                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2086
2087         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2088                                         args->treeid, args->objectid,
2089                                         args->name);
2090
2091         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2092                 ret = -EFAULT;
2093
2094         kfree(args);
2095         return ret;
2096 }
2097
2098 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2099                                              void __user *arg)
2100 {
2101         struct dentry *parent = file->f_path.dentry;
2102         struct dentry *dentry;
2103         struct inode *dir = parent->d_inode;
2104         struct inode *inode;
2105         struct btrfs_root *root = BTRFS_I(dir)->root;
2106         struct btrfs_root *dest = NULL;
2107         struct btrfs_ioctl_vol_args *vol_args;
2108         struct btrfs_trans_handle *trans;
2109         struct btrfs_block_rsv block_rsv;
2110         u64 qgroup_reserved;
2111         int namelen;
2112         int ret;
2113         int err = 0;
2114
2115         vol_args = memdup_user(arg, sizeof(*vol_args));
2116         if (IS_ERR(vol_args))
2117                 return PTR_ERR(vol_args);
2118
2119         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2120         namelen = strlen(vol_args->name);
2121         if (strchr(vol_args->name, '/') ||
2122             strncmp(vol_args->name, "..", namelen) == 0) {
2123                 err = -EINVAL;
2124                 goto out;
2125         }
2126
2127         err = mnt_want_write_file(file);
2128         if (err)
2129                 goto out;
2130
2131         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2132         if (err == -EINTR)
2133                 goto out;
2134         dentry = lookup_one_len(vol_args->name, parent, namelen);
2135         if (IS_ERR(dentry)) {
2136                 err = PTR_ERR(dentry);
2137                 goto out_unlock_dir;
2138         }
2139
2140         if (!dentry->d_inode) {
2141                 err = -ENOENT;
2142                 goto out_dput;
2143         }
2144
2145         inode = dentry->d_inode;
2146         dest = BTRFS_I(inode)->root;
2147         if (!capable(CAP_SYS_ADMIN)){
2148                 /*
2149                  * Regular user.  Only allow this with a special mount
2150                  * option, when the user has write+exec access to the
2151                  * subvol root, and when rmdir(2) would have been
2152                  * allowed.
2153                  *
2154                  * Note that this is _not_ check that the subvol is
2155                  * empty or doesn't contain data that we wouldn't
2156                  * otherwise be able to delete.
2157                  *
2158                  * Users who want to delete empty subvols should try
2159                  * rmdir(2).
2160                  */
2161                 err = -EPERM;
2162                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2163                         goto out_dput;
2164
2165                 /*
2166                  * Do not allow deletion if the parent dir is the same
2167                  * as the dir to be deleted.  That means the ioctl
2168                  * must be called on the dentry referencing the root
2169                  * of the subvol, not a random directory contained
2170                  * within it.
2171                  */
2172                 err = -EINVAL;
2173                 if (root == dest)
2174                         goto out_dput;
2175
2176                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2177                 if (err)
2178                         goto out_dput;
2179         }
2180
2181         /* check if subvolume may be deleted by a user */
2182         err = btrfs_may_delete(dir, dentry, 1);
2183         if (err)
2184                 goto out_dput;
2185
2186         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2187                 err = -EINVAL;
2188                 goto out_dput;
2189         }
2190
2191         mutex_lock(&inode->i_mutex);
2192         err = d_invalidate(dentry);
2193         if (err)
2194                 goto out_unlock;
2195
2196         down_write(&root->fs_info->subvol_sem);
2197
2198         err = may_destroy_subvol(dest);
2199         if (err)
2200                 goto out_up_write;
2201
2202         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2203         /*
2204          * One for dir inode, two for dir entries, two for root
2205          * ref/backref.
2206          */
2207         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2208                                                5, &qgroup_reserved, true);
2209         if (err)
2210                 goto out_up_write;
2211
2212         trans = btrfs_start_transaction(root, 0);
2213         if (IS_ERR(trans)) {
2214                 err = PTR_ERR(trans);
2215                 goto out_release;
2216         }
2217         trans->block_rsv = &block_rsv;
2218         trans->bytes_reserved = block_rsv.size;
2219
2220         ret = btrfs_unlink_subvol(trans, root, dir,
2221                                 dest->root_key.objectid,
2222                                 dentry->d_name.name,
2223                                 dentry->d_name.len);
2224         if (ret) {
2225                 err = ret;
2226                 btrfs_abort_transaction(trans, root, ret);
2227                 goto out_end_trans;
2228         }
2229
2230         btrfs_record_root_in_trans(trans, dest);
2231
2232         memset(&dest->root_item.drop_progress, 0,
2233                 sizeof(dest->root_item.drop_progress));
2234         dest->root_item.drop_level = 0;
2235         btrfs_set_root_refs(&dest->root_item, 0);
2236
2237         if (!xchg(&dest->orphan_item_inserted, 1)) {
2238                 ret = btrfs_insert_orphan_item(trans,
2239                                         root->fs_info->tree_root,
2240                                         dest->root_key.objectid);
2241                 if (ret) {
2242                         btrfs_abort_transaction(trans, root, ret);
2243                         err = ret;
2244                         goto out_end_trans;
2245                 }
2246         }
2247
2248         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2249                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2250                                   dest->root_key.objectid);
2251         if (ret && ret != -ENOENT) {
2252                 btrfs_abort_transaction(trans, root, ret);
2253                 err = ret;
2254                 goto out_end_trans;
2255         }
2256         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2257                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2258                                           dest->root_item.received_uuid,
2259                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2260                                           dest->root_key.objectid);
2261                 if (ret && ret != -ENOENT) {
2262                         btrfs_abort_transaction(trans, root, ret);
2263                         err = ret;
2264                         goto out_end_trans;
2265                 }
2266         }
2267
2268 out_end_trans:
2269         trans->block_rsv = NULL;
2270         trans->bytes_reserved = 0;
2271         ret = btrfs_end_transaction(trans, root);
2272         if (ret && !err)
2273                 err = ret;
2274         inode->i_flags |= S_DEAD;
2275 out_release:
2276         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2277 out_up_write:
2278         up_write(&root->fs_info->subvol_sem);
2279 out_unlock:
2280         mutex_unlock(&inode->i_mutex);
2281         if (!err) {
2282                 shrink_dcache_sb(root->fs_info->sb);
2283                 btrfs_invalidate_inodes(dest);
2284                 d_delete(dentry);
2285
2286                 /* the last ref */
2287                 if (dest->cache_inode) {
2288                         iput(dest->cache_inode);
2289                         dest->cache_inode = NULL;
2290                 }
2291         }
2292 out_dput:
2293         dput(dentry);
2294 out_unlock_dir:
2295         mutex_unlock(&dir->i_mutex);
2296         mnt_drop_write_file(file);
2297 out:
2298         kfree(vol_args);
2299         return err;
2300 }
2301
2302 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2303 {
2304         struct inode *inode = file_inode(file);
2305         struct btrfs_root *root = BTRFS_I(inode)->root;
2306         struct btrfs_ioctl_defrag_range_args *range;
2307         int ret;
2308
2309         ret = mnt_want_write_file(file);
2310         if (ret)
2311                 return ret;
2312
2313         if (btrfs_root_readonly(root)) {
2314                 ret = -EROFS;
2315                 goto out;
2316         }
2317
2318         switch (inode->i_mode & S_IFMT) {
2319         case S_IFDIR:
2320                 if (!capable(CAP_SYS_ADMIN)) {
2321                         ret = -EPERM;
2322                         goto out;
2323                 }
2324                 ret = btrfs_defrag_root(root);
2325                 if (ret)
2326                         goto out;
2327                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2328                 break;
2329         case S_IFREG:
2330                 if (!(file->f_mode & FMODE_WRITE)) {
2331                         ret = -EINVAL;
2332                         goto out;
2333                 }
2334
2335                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2336                 if (!range) {
2337                         ret = -ENOMEM;
2338                         goto out;
2339                 }
2340
2341                 if (argp) {
2342                         if (copy_from_user(range, argp,
2343                                            sizeof(*range))) {
2344                                 ret = -EFAULT;
2345                                 kfree(range);
2346                                 goto out;
2347                         }
2348                         /* compression requires us to start the IO */
2349                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2350                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2351                                 range->extent_thresh = (u32)-1;
2352                         }
2353                 } else {
2354                         /* the rest are all set to zero by kzalloc */
2355                         range->len = (u64)-1;
2356                 }
2357                 ret = btrfs_defrag_file(file_inode(file), file,
2358                                         range, 0, 0);
2359                 if (ret > 0)
2360                         ret = 0;
2361                 kfree(range);
2362                 break;
2363         default:
2364                 ret = -EINVAL;
2365         }
2366 out:
2367         mnt_drop_write_file(file);
2368         return ret;
2369 }
2370
2371 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2372 {
2373         struct btrfs_ioctl_vol_args *vol_args;
2374         int ret;
2375
2376         if (!capable(CAP_SYS_ADMIN))
2377                 return -EPERM;
2378
2379         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2380                         1)) {
2381                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2382         }
2383
2384         mutex_lock(&root->fs_info->volume_mutex);
2385         vol_args = memdup_user(arg, sizeof(*vol_args));
2386         if (IS_ERR(vol_args)) {
2387                 ret = PTR_ERR(vol_args);
2388                 goto out;
2389         }
2390
2391         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2392         ret = btrfs_init_new_device(root, vol_args->name);
2393
2394         kfree(vol_args);
2395 out:
2396         mutex_unlock(&root->fs_info->volume_mutex);
2397         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2398         return ret;
2399 }
2400
2401 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2402 {
2403         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2404         struct btrfs_ioctl_vol_args *vol_args;
2405         int ret;
2406
2407         if (!capable(CAP_SYS_ADMIN))
2408                 return -EPERM;
2409
2410         ret = mnt_want_write_file(file);
2411         if (ret)
2412                 return ret;
2413
2414         vol_args = memdup_user(arg, sizeof(*vol_args));
2415         if (IS_ERR(vol_args)) {
2416                 ret = PTR_ERR(vol_args);
2417                 goto out;
2418         }
2419
2420         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2421
2422         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2423                         1)) {
2424                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2425                 goto out;
2426         }
2427
2428         mutex_lock(&root->fs_info->volume_mutex);
2429         ret = btrfs_rm_device(root, vol_args->name);
2430         mutex_unlock(&root->fs_info->volume_mutex);
2431         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2432
2433 out:
2434         kfree(vol_args);
2435         mnt_drop_write_file(file);
2436         return ret;
2437 }
2438
2439 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2440 {
2441         struct btrfs_ioctl_fs_info_args *fi_args;
2442         struct btrfs_device *device;
2443         struct btrfs_device *next;
2444         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2445         int ret = 0;
2446
2447         if (!capable(CAP_SYS_ADMIN))
2448                 return -EPERM;
2449
2450         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2451         if (!fi_args)
2452                 return -ENOMEM;
2453
2454         mutex_lock(&fs_devices->device_list_mutex);
2455         fi_args->num_devices = fs_devices->num_devices;
2456         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2457
2458         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2459                 if (device->devid > fi_args->max_id)
2460                         fi_args->max_id = device->devid;
2461         }
2462         mutex_unlock(&fs_devices->device_list_mutex);
2463
2464         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2465                 ret = -EFAULT;
2466
2467         kfree(fi_args);
2468         return ret;
2469 }
2470
2471 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2472 {
2473         struct btrfs_ioctl_dev_info_args *di_args;
2474         struct btrfs_device *dev;
2475         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2476         int ret = 0;
2477         char *s_uuid = NULL;
2478
2479         if (!capable(CAP_SYS_ADMIN))
2480                 return -EPERM;
2481
2482         di_args = memdup_user(arg, sizeof(*di_args));
2483         if (IS_ERR(di_args))
2484                 return PTR_ERR(di_args);
2485
2486         if (!btrfs_is_empty_uuid(di_args->uuid))
2487                 s_uuid = di_args->uuid;
2488
2489         mutex_lock(&fs_devices->device_list_mutex);
2490         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2491
2492         if (!dev) {
2493                 ret = -ENODEV;
2494                 goto out;
2495         }
2496
2497         di_args->devid = dev->devid;
2498         di_args->bytes_used = dev->bytes_used;
2499         di_args->total_bytes = dev->total_bytes;
2500         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2501         if (dev->name) {
2502                 struct rcu_string *name;
2503
2504                 rcu_read_lock();
2505                 name = rcu_dereference(dev->name);
2506                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2507                 rcu_read_unlock();
2508                 di_args->path[sizeof(di_args->path) - 1] = 0;
2509         } else {
2510                 di_args->path[0] = '\0';
2511         }
2512
2513 out:
2514         mutex_unlock(&fs_devices->device_list_mutex);
2515         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2516                 ret = -EFAULT;
2517
2518         kfree(di_args);
2519         return ret;
2520 }
2521
2522 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2523 {
2524         struct page *page;
2525         pgoff_t index;
2526         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2527
2528         index = off >> PAGE_CACHE_SHIFT;
2529
2530         page = grab_cache_page(inode->i_mapping, index);
2531         if (!page)
2532                 return NULL;
2533
2534         if (!PageUptodate(page)) {
2535                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2536                                                  0))
2537                         return NULL;
2538                 lock_page(page);
2539                 if (!PageUptodate(page)) {
2540                         unlock_page(page);
2541                         page_cache_release(page);
2542                         return NULL;
2543                 }
2544         }
2545         unlock_page(page);
2546
2547         return page;
2548 }
2549
2550 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2551 {
2552         /* do any pending delalloc/csum calc on src, one way or
2553            another, and lock file content */
2554         while (1) {
2555                 struct btrfs_ordered_extent *ordered;
2556                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2557                 ordered = btrfs_lookup_first_ordered_extent(inode,
2558                                                             off + len - 1);
2559                 if (!ordered &&
2560                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2561                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2562                         break;
2563                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2564                 if (ordered)
2565                         btrfs_put_ordered_extent(ordered);
2566                 btrfs_wait_ordered_range(inode, off, len);
2567         }
2568 }
2569
2570 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2571                                 struct inode *inode2, u64 loff2, u64 len)
2572 {
2573         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2574         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2575
2576         mutex_unlock(&inode1->i_mutex);
2577         mutex_unlock(&inode2->i_mutex);
2578 }
2579
2580 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2581                               struct inode *inode2, u64 loff2, u64 len)
2582 {
2583         if (inode1 < inode2) {
2584                 swap(inode1, inode2);
2585                 swap(loff1, loff2);
2586         }
2587
2588         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2589         lock_extent_range(inode1, loff1, len);
2590         if (inode1 != inode2) {
2591                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2592                 lock_extent_range(inode2, loff2, len);
2593         }
2594 }
2595
2596 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2597                           u64 dst_loff, u64 len)
2598 {
2599         int ret = 0;
2600         struct page *src_page, *dst_page;
2601         unsigned int cmp_len = PAGE_CACHE_SIZE;
2602         void *addr, *dst_addr;
2603
2604         while (len) {
2605                 if (len < PAGE_CACHE_SIZE)
2606                         cmp_len = len;
2607
2608                 src_page = extent_same_get_page(src, loff);
2609                 if (!src_page)
2610                         return -EINVAL;
2611                 dst_page = extent_same_get_page(dst, dst_loff);
2612                 if (!dst_page) {
2613                         page_cache_release(src_page);
2614                         return -EINVAL;
2615                 }
2616                 addr = kmap_atomic(src_page);
2617                 dst_addr = kmap_atomic(dst_page);
2618
2619                 flush_dcache_page(src_page);
2620                 flush_dcache_page(dst_page);
2621
2622                 if (memcmp(addr, dst_addr, cmp_len))
2623                         ret = BTRFS_SAME_DATA_DIFFERS;
2624
2625                 kunmap_atomic(addr);
2626                 kunmap_atomic(dst_addr);
2627                 page_cache_release(src_page);
2628                 page_cache_release(dst_page);
2629
2630                 if (ret)
2631                         break;
2632
2633                 loff += cmp_len;
2634                 dst_loff += cmp_len;
2635                 len -= cmp_len;
2636         }
2637
2638         return ret;
2639 }
2640
2641 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2642 {
2643         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2644
2645         if (off + len > inode->i_size || off + len < off)
2646                 return -EINVAL;
2647         /* Check that we are block aligned - btrfs_clone() requires this */
2648         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2649                 return -EINVAL;
2650
2651         return 0;
2652 }
2653
2654 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2655                              struct inode *dst, u64 dst_loff)
2656 {
2657         int ret;
2658
2659         /*
2660          * btrfs_clone() can't handle extents in the same file
2661          * yet. Once that works, we can drop this check and replace it
2662          * with a check for the same inode, but overlapping extents.
2663          */
2664         if (src == dst)
2665                 return -EINVAL;
2666
2667         btrfs_double_lock(src, loff, dst, dst_loff, len);
2668
2669         ret = extent_same_check_offsets(src, loff, len);
2670         if (ret)
2671                 goto out_unlock;
2672
2673         ret = extent_same_check_offsets(dst, dst_loff, len);
2674         if (ret)
2675                 goto out_unlock;
2676
2677         /* don't make the dst file partly checksummed */
2678         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2679             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2680                 ret = -EINVAL;
2681                 goto out_unlock;
2682         }
2683
2684         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2685         if (ret == 0)
2686                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2687
2688 out_unlock:
2689         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2690
2691         return ret;
2692 }
2693
2694 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2695
2696 static long btrfs_ioctl_file_extent_same(struct file *file,
2697                                          void __user *argp)
2698 {
2699         struct btrfs_ioctl_same_args tmp;
2700         struct btrfs_ioctl_same_args *same;
2701         struct btrfs_ioctl_same_extent_info *info;
2702         struct inode *src = file->f_dentry->d_inode;
2703         struct file *dst_file = NULL;
2704         struct inode *dst;
2705         u64 off;
2706         u64 len;
2707         int i;
2708         int ret;
2709         unsigned long size;
2710         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2711         bool is_admin = capable(CAP_SYS_ADMIN);
2712
2713         if (!(file->f_mode & FMODE_READ))
2714                 return -EINVAL;
2715
2716         ret = mnt_want_write_file(file);
2717         if (ret)
2718                 return ret;
2719
2720         if (copy_from_user(&tmp,
2721                            (struct btrfs_ioctl_same_args __user *)argp,
2722                            sizeof(tmp))) {
2723                 ret = -EFAULT;
2724                 goto out;
2725         }
2726
2727         size = sizeof(tmp) +
2728                 tmp.dest_count * sizeof(struct btrfs_ioctl_same_extent_info);
2729
2730         same = kmalloc(size, GFP_NOFS);
2731         if (!same) {
2732                 ret = -EFAULT;
2733                 goto out;
2734         }
2735
2736         if (copy_from_user(same,
2737                            (struct btrfs_ioctl_same_args __user *)argp, size)) {
2738                 ret = -EFAULT;
2739                 goto out;
2740         }
2741
2742         off = same->logical_offset;
2743         len = same->length;
2744
2745         /*
2746          * Limit the total length we will dedupe for each operation.
2747          * This is intended to bound the total time spent in this
2748          * ioctl to something sane.
2749          */
2750         if (len > BTRFS_MAX_DEDUPE_LEN)
2751                 len = BTRFS_MAX_DEDUPE_LEN;
2752
2753         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2754                 /*
2755                  * Btrfs does not support blocksize < page_size. As a
2756                  * result, btrfs_cmp_data() won't correctly handle
2757                  * this situation without an update.
2758                  */
2759                 ret = -EINVAL;
2760                 goto out;
2761         }
2762
2763         ret = -EISDIR;
2764         if (S_ISDIR(src->i_mode))
2765                 goto out;
2766
2767         ret = -EACCES;
2768         if (!S_ISREG(src->i_mode))
2769                 goto out;
2770
2771         /* pre-format output fields to sane values */
2772         for (i = 0; i < same->dest_count; i++) {
2773                 same->info[i].bytes_deduped = 0ULL;
2774                 same->info[i].status = 0;
2775         }
2776
2777         ret = 0;
2778         for (i = 0; i < same->dest_count; i++) {
2779                 info = &same->info[i];
2780
2781                 dst_file = fget(info->fd);
2782                 if (!dst_file) {
2783                         info->status = -EBADF;
2784                         goto next;
2785                 }
2786
2787                 if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2788                         info->status = -EINVAL;
2789                         goto next;
2790                 }
2791
2792                 info->status = -EXDEV;
2793                 if (file->f_path.mnt != dst_file->f_path.mnt)
2794                         goto next;
2795
2796                 dst = dst_file->f_dentry->d_inode;
2797                 if (src->i_sb != dst->i_sb)
2798                         goto next;
2799
2800                 if (S_ISDIR(dst->i_mode)) {
2801                         info->status = -EISDIR;
2802                         goto next;
2803                 }
2804
2805                 if (!S_ISREG(dst->i_mode)) {
2806                         info->status = -EACCES;
2807                         goto next;
2808                 }
2809
2810                 info->status = btrfs_extent_same(src, off, len, dst,
2811                                                 info->logical_offset);
2812                 if (info->status == 0)
2813                         info->bytes_deduped += len;
2814
2815 next:
2816                 if (dst_file)
2817                         fput(dst_file);
2818         }
2819
2820         ret = copy_to_user(argp, same, size);
2821         if (ret)
2822                 ret = -EFAULT;
2823
2824 out:
2825         mnt_drop_write_file(file);
2826         return ret;
2827 }
2828
2829 /**
2830  * btrfs_clone() - clone a range from inode file to another
2831  *
2832  * @src: Inode to clone from
2833  * @inode: Inode to clone to
2834  * @off: Offset within source to start clone from
2835  * @olen: Original length, passed by user, of range to clone
2836  * @olen_aligned: Block-aligned value of olen, extent_same uses
2837  *               identical values here
2838  * @destoff: Offset within @inode to start clone
2839  */
2840 static int btrfs_clone(struct inode *src, struct inode *inode,
2841                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2842 {
2843         struct btrfs_root *root = BTRFS_I(inode)->root;
2844         struct btrfs_path *path = NULL;
2845         struct extent_buffer *leaf;
2846         struct btrfs_trans_handle *trans;
2847         char *buf = NULL;
2848         struct btrfs_key key;
2849         u32 nritems;
2850         int slot;
2851         int ret;
2852         u64 len = olen_aligned;
2853
2854         ret = -ENOMEM;
2855         buf = vmalloc(btrfs_level_size(root, 0));
2856         if (!buf)
2857                 return ret;
2858
2859         path = btrfs_alloc_path();
2860         if (!path) {
2861                 vfree(buf);
2862                 return ret;
2863         }
2864
2865         path->reada = 2;
2866         /* clone data */
2867         key.objectid = btrfs_ino(src);
2868         key.type = BTRFS_EXTENT_DATA_KEY;
2869         key.offset = 0;
2870
2871         while (1) {
2872                 /*
2873                  * note the key will change type as we walk through the
2874                  * tree.
2875                  */
2876                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2877                                 0, 0);
2878                 if (ret < 0)
2879                         goto out;
2880
2881                 nritems = btrfs_header_nritems(path->nodes[0]);
2882                 if (path->slots[0] >= nritems) {
2883                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2884                         if (ret < 0)
2885                                 goto out;
2886                         if (ret > 0)
2887                                 break;
2888                         nritems = btrfs_header_nritems(path->nodes[0]);
2889                 }
2890                 leaf = path->nodes[0];
2891                 slot = path->slots[0];
2892
2893                 btrfs_item_key_to_cpu(leaf, &key, slot);
2894                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2895                     key.objectid != btrfs_ino(src))
2896                         break;
2897
2898                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2899                         struct btrfs_file_extent_item *extent;
2900                         int type;
2901                         u32 size;
2902                         struct btrfs_key new_key;
2903                         u64 disko = 0, diskl = 0;
2904                         u64 datao = 0, datal = 0;
2905                         u8 comp;
2906                         u64 endoff;
2907
2908                         size = btrfs_item_size_nr(leaf, slot);
2909                         read_extent_buffer(leaf, buf,
2910                                            btrfs_item_ptr_offset(leaf, slot),
2911                                            size);
2912
2913                         extent = btrfs_item_ptr(leaf, slot,
2914                                                 struct btrfs_file_extent_item);
2915                         comp = btrfs_file_extent_compression(leaf, extent);
2916                         type = btrfs_file_extent_type(leaf, extent);
2917                         if (type == BTRFS_FILE_EXTENT_REG ||
2918                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2919                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2920                                                                       extent);
2921                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2922                                                                  extent);
2923                                 datao = btrfs_file_extent_offset(leaf, extent);
2924                                 datal = btrfs_file_extent_num_bytes(leaf,
2925                                                                     extent);
2926                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2927                                 /* take upper bound, may be compressed */
2928                                 datal = btrfs_file_extent_ram_bytes(leaf,
2929                                                                     extent);
2930                         }
2931                         btrfs_release_path(path);
2932
2933                         if (key.offset + datal <= off ||
2934                             key.offset >= off + len - 1)
2935                                 goto next;
2936
2937                         memcpy(&new_key, &key, sizeof(new_key));
2938                         new_key.objectid = btrfs_ino(inode);
2939                         if (off <= key.offset)
2940                                 new_key.offset = key.offset + destoff - off;
2941                         else
2942                                 new_key.offset = destoff;
2943
2944                         /*
2945                          * 1 - adjusting old extent (we may have to split it)
2946                          * 1 - add new extent
2947                          * 1 - inode update
2948                          */
2949                         trans = btrfs_start_transaction(root, 3);
2950                         if (IS_ERR(trans)) {
2951                                 ret = PTR_ERR(trans);
2952                                 goto out;
2953                         }
2954
2955                         if (type == BTRFS_FILE_EXTENT_REG ||
2956                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2957                                 /*
2958                                  *    a  | --- range to clone ---|  b
2959                                  * | ------------- extent ------------- |
2960                                  */
2961
2962                                 /* substract range b */
2963                                 if (key.offset + datal > off + len)
2964                                         datal = off + len - key.offset;
2965
2966                                 /* substract range a */
2967                                 if (off > key.offset) {
2968                                         datao += off - key.offset;
2969                                         datal -= off - key.offset;
2970                                 }
2971
2972                                 ret = btrfs_drop_extents(trans, root, inode,
2973                                                          new_key.offset,
2974                                                          new_key.offset + datal,
2975                                                          1);
2976                                 if (ret) {
2977                                         btrfs_abort_transaction(trans, root,
2978                                                                 ret);
2979                                         btrfs_end_transaction(trans, root);
2980                                         goto out;
2981                                 }
2982
2983                                 ret = btrfs_insert_empty_item(trans, root, path,
2984                                                               &new_key, size);
2985                                 if (ret) {
2986                                         btrfs_abort_transaction(trans, root,
2987                                                                 ret);
2988                                         btrfs_end_transaction(trans, root);
2989                                         goto out;
2990                                 }
2991
2992                                 leaf = path->nodes[0];
2993                                 slot = path->slots[0];
2994                                 write_extent_buffer(leaf, buf,
2995                                             btrfs_item_ptr_offset(leaf, slot),
2996                                             size);
2997
2998                                 extent = btrfs_item_ptr(leaf, slot,
2999                                                 struct btrfs_file_extent_item);
3000
3001                                 /* disko == 0 means it's a hole */
3002                                 if (!disko)
3003                                         datao = 0;
3004
3005                                 btrfs_set_file_extent_offset(leaf, extent,
3006                                                              datao);
3007                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3008                                                                 datal);
3009                                 if (disko) {
3010                                         inode_add_bytes(inode, datal);
3011                                         ret = btrfs_inc_extent_ref(trans, root,
3012                                                         disko, diskl, 0,
3013                                                         root->root_key.objectid,
3014                                                         btrfs_ino(inode),
3015                                                         new_key.offset - datao,
3016                                                         0);
3017                                         if (ret) {
3018                                                 btrfs_abort_transaction(trans,
3019                                                                         root,
3020                                                                         ret);
3021                                                 btrfs_end_transaction(trans,
3022                                                                       root);
3023                                                 goto out;
3024
3025                                         }
3026                                 }
3027                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3028                                 u64 skip = 0;
3029                                 u64 trim = 0;
3030                                 if (off > key.offset) {
3031                                         skip = off - key.offset;
3032                                         new_key.offset += skip;
3033                                 }
3034
3035                                 if (key.offset + datal > off + len)
3036                                         trim = key.offset + datal - (off + len);
3037
3038                                 if (comp && (skip || trim)) {
3039                                         ret = -EINVAL;
3040                                         btrfs_end_transaction(trans, root);
3041                                         goto out;
3042                                 }
3043                                 size -= skip + trim;
3044                                 datal -= skip + trim;
3045
3046                                 ret = btrfs_drop_extents(trans, root, inode,
3047                                                          new_key.offset,
3048                                                          new_key.offset + datal,
3049                                                          1);
3050                                 if (ret) {
3051                                         btrfs_abort_transaction(trans, root,
3052                                                                 ret);
3053                                         btrfs_end_transaction(trans, root);
3054                                         goto out;
3055                                 }
3056
3057                                 ret = btrfs_insert_empty_item(trans, root, path,
3058                                                               &new_key, size);
3059                                 if (ret) {
3060                                         btrfs_abort_transaction(trans, root,
3061                                                                 ret);
3062                                         btrfs_end_transaction(trans, root);
3063                                         goto out;
3064                                 }
3065
3066                                 if (skip) {
3067                                         u32 start =
3068                                           btrfs_file_extent_calc_inline_size(0);
3069                                         memmove(buf+start, buf+start+skip,
3070                                                 datal);
3071                                 }
3072
3073                                 leaf = path->nodes[0];
3074                                 slot = path->slots[0];
3075                                 write_extent_buffer(leaf, buf,
3076                                             btrfs_item_ptr_offset(leaf, slot),
3077                                             size);
3078                                 inode_add_bytes(inode, datal);
3079                         }
3080
3081                         btrfs_mark_buffer_dirty(leaf);
3082                         btrfs_release_path(path);
3083
3084                         inode_inc_iversion(inode);
3085                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3086
3087                         /*
3088                          * we round up to the block size at eof when
3089                          * determining which extents to clone above,
3090                          * but shouldn't round up the file size
3091                          */
3092                         endoff = new_key.offset + datal;
3093                         if (endoff > destoff+olen)
3094                                 endoff = destoff+olen;
3095                         if (endoff > inode->i_size)
3096                                 btrfs_i_size_write(inode, endoff);
3097
3098                         ret = btrfs_update_inode(trans, root, inode);
3099                         if (ret) {
3100                                 btrfs_abort_transaction(trans, root, ret);
3101                                 btrfs_end_transaction(trans, root);
3102                                 goto out;
3103                         }
3104                         ret = btrfs_end_transaction(trans, root);
3105                 }
3106 next:
3107                 btrfs_release_path(path);
3108                 key.offset++;
3109         }
3110         ret = 0;
3111
3112 out:
3113         btrfs_release_path(path);
3114         btrfs_free_path(path);
3115         vfree(buf);
3116         return ret;
3117 }
3118
3119 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3120                                        u64 off, u64 olen, u64 destoff)
3121 {
3122         struct inode *inode = file_inode(file);
3123         struct btrfs_root *root = BTRFS_I(inode)->root;
3124         struct fd src_file;
3125         struct inode *src;
3126         int ret;
3127         u64 len = olen;
3128         u64 bs = root->fs_info->sb->s_blocksize;
3129         int same_inode = 0;
3130
3131         /*
3132          * TODO:
3133          * - split compressed inline extents.  annoying: we need to
3134          *   decompress into destination's address_space (the file offset
3135          *   may change, so source mapping won't do), then recompress (or
3136          *   otherwise reinsert) a subrange.
3137          * - allow ranges within the same file to be cloned (provided
3138          *   they don't overlap)?
3139          */
3140
3141         /* the destination must be opened for writing */
3142         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3143                 return -EINVAL;
3144
3145         if (btrfs_root_readonly(root))
3146                 return -EROFS;
3147
3148         ret = mnt_want_write_file(file);
3149         if (ret)
3150                 return ret;
3151
3152         src_file = fdget(srcfd);
3153         if (!src_file.file) {
3154                 ret = -EBADF;
3155                 goto out_drop_write;
3156         }
3157
3158         ret = -EXDEV;
3159         if (src_file.file->f_path.mnt != file->f_path.mnt)
3160                 goto out_fput;
3161
3162         src = file_inode(src_file.file);
3163
3164         ret = -EINVAL;
3165         if (src == inode)
3166                 same_inode = 1;
3167
3168         /* the src must be open for reading */
3169         if (!(src_file.file->f_mode & FMODE_READ))
3170                 goto out_fput;
3171
3172         /* don't make the dst file partly checksummed */
3173         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3174             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3175                 goto out_fput;
3176
3177         ret = -EISDIR;
3178         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3179                 goto out_fput;
3180
3181         ret = -EXDEV;
3182         if (src->i_sb != inode->i_sb)
3183                 goto out_fput;
3184
3185         if (!same_inode) {
3186                 if (inode < src) {
3187                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3188                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3189                 } else {
3190                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3191                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3192                 }
3193         } else {
3194                 mutex_lock(&src->i_mutex);
3195         }
3196
3197         /* determine range to clone */
3198         ret = -EINVAL;
3199         if (off + len > src->i_size || off + len < off)
3200                 goto out_unlock;
3201         if (len == 0)
3202                 olen = len = src->i_size - off;
3203         /* if we extend to eof, continue to block boundary */
3204         if (off + len == src->i_size)
3205                 len = ALIGN(src->i_size, bs) - off;
3206
3207         /* verify the end result is block aligned */
3208         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3209             !IS_ALIGNED(destoff, bs))
3210                 goto out_unlock;
3211
3212         /* verify if ranges are overlapped within the same file */
3213         if (same_inode) {
3214                 if (destoff + len > off && destoff < off + len)
3215                         goto out_unlock;
3216         }
3217
3218         if (destoff > inode->i_size) {
3219                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3220                 if (ret)
3221                         goto out_unlock;
3222         }
3223
3224         /* truncate page cache pages from target inode range */
3225         truncate_inode_pages_range(&inode->i_data, destoff,
3226                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3227
3228         lock_extent_range(src, off, len);
3229
3230         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3231
3232         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3233 out_unlock:
3234         mutex_unlock(&src->i_mutex);
3235         if (!same_inode)
3236                 mutex_unlock(&inode->i_mutex);
3237 out_fput:
3238         fdput(src_file);
3239 out_drop_write:
3240         mnt_drop_write_file(file);
3241         return ret;
3242 }
3243
3244 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3245 {
3246         struct btrfs_ioctl_clone_range_args args;
3247
3248         if (copy_from_user(&args, argp, sizeof(args)))
3249                 return -EFAULT;
3250         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3251                                  args.src_length, args.dest_offset);
3252 }
3253
3254 /*
3255  * there are many ways the trans_start and trans_end ioctls can lead
3256  * to deadlocks.  They should only be used by applications that
3257  * basically own the machine, and have a very in depth understanding
3258  * of all the possible deadlocks and enospc problems.
3259  */
3260 static long btrfs_ioctl_trans_start(struct file *file)
3261 {
3262         struct inode *inode = file_inode(file);
3263         struct btrfs_root *root = BTRFS_I(inode)->root;
3264         struct btrfs_trans_handle *trans;
3265         int ret;
3266
3267         ret = -EPERM;
3268         if (!capable(CAP_SYS_ADMIN))
3269                 goto out;
3270
3271         ret = -EINPROGRESS;
3272         if (file->private_data)
3273                 goto out;
3274
3275         ret = -EROFS;
3276         if (btrfs_root_readonly(root))
3277                 goto out;
3278
3279         ret = mnt_want_write_file(file);
3280         if (ret)
3281                 goto out;
3282
3283         atomic_inc(&root->fs_info->open_ioctl_trans);
3284
3285         ret = -ENOMEM;
3286         trans = btrfs_start_ioctl_transaction(root);
3287         if (IS_ERR(trans))
3288                 goto out_drop;
3289
3290         file->private_data = trans;
3291         return 0;
3292
3293 out_drop:
3294         atomic_dec(&root->fs_info->open_ioctl_trans);
3295         mnt_drop_write_file(file);
3296 out:
3297         return ret;
3298 }
3299
3300 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3301 {
3302         struct inode *inode = file_inode(file);
3303         struct btrfs_root *root = BTRFS_I(inode)->root;
3304         struct btrfs_root *new_root;
3305         struct btrfs_dir_item *di;
3306         struct btrfs_trans_handle *trans;
3307         struct btrfs_path *path;
3308         struct btrfs_key location;
3309         struct btrfs_disk_key disk_key;
3310         u64 objectid = 0;
3311         u64 dir_id;
3312         int ret;
3313
3314         if (!capable(CAP_SYS_ADMIN))
3315                 return -EPERM;
3316
3317         ret = mnt_want_write_file(file);
3318         if (ret)
3319                 return ret;
3320
3321         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3322                 ret = -EFAULT;
3323                 goto out;
3324         }
3325
3326         if (!objectid)
3327                 objectid = BTRFS_FS_TREE_OBJECTID;
3328
3329         location.objectid = objectid;
3330         location.type = BTRFS_ROOT_ITEM_KEY;
3331         location.offset = (u64)-1;
3332
3333         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3334         if (IS_ERR(new_root)) {
3335                 ret = PTR_ERR(new_root);
3336                 goto out;
3337         }
3338
3339         path = btrfs_alloc_path();
3340         if (!path) {
3341                 ret = -ENOMEM;
3342                 goto out;
3343         }
3344         path->leave_spinning = 1;
3345
3346         trans = btrfs_start_transaction(root, 1);
3347         if (IS_ERR(trans)) {
3348                 btrfs_free_path(path);
3349                 ret = PTR_ERR(trans);
3350                 goto out;
3351         }
3352
3353         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3354         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3355                                    dir_id, "default", 7, 1);
3356         if (IS_ERR_OR_NULL(di)) {
3357                 btrfs_free_path(path);
3358                 btrfs_end_transaction(trans, root);
3359                 printk(KERN_ERR "Umm, you don't have the default dir item, "
3360                        "this isn't going to work\n");
3361                 ret = -ENOENT;
3362                 goto out;
3363         }
3364
3365         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3366         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3367         btrfs_mark_buffer_dirty(path->nodes[0]);
3368         btrfs_free_path(path);
3369
3370         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3371         btrfs_end_transaction(trans, root);
3372 out:
3373         mnt_drop_write_file(file);
3374         return ret;
3375 }
3376
3377 void btrfs_get_block_group_info(struct list_head *groups_list,
3378                                 struct btrfs_ioctl_space_info *space)
3379 {
3380         struct btrfs_block_group_cache *block_group;
3381
3382         space->total_bytes = 0;
3383         space->used_bytes = 0;
3384         space->flags = 0;
3385         list_for_each_entry(block_group, groups_list, list) {
3386                 space->flags = block_group->flags;
3387                 space->total_bytes += block_group->key.offset;
3388                 space->used_bytes +=
3389                         btrfs_block_group_used(&block_group->item);
3390         }
3391 }
3392
3393 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3394 {
3395         struct btrfs_ioctl_space_args space_args;
3396         struct btrfs_ioctl_space_info space;
3397         struct btrfs_ioctl_space_info *dest;
3398         struct btrfs_ioctl_space_info *dest_orig;
3399         struct btrfs_ioctl_space_info __user *user_dest;
3400         struct btrfs_space_info *info;
3401         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3402                        BTRFS_BLOCK_GROUP_SYSTEM,
3403                        BTRFS_BLOCK_GROUP_METADATA,
3404                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3405         int num_types = 4;
3406         int alloc_size;
3407         int ret = 0;
3408         u64 slot_count = 0;
3409         int i, c;
3410
3411         if (copy_from_user(&space_args,
3412                            (struct btrfs_ioctl_space_args __user *)arg,
3413                            sizeof(space_args)))
3414                 return -EFAULT;
3415
3416         for (i = 0; i < num_types; i++) {
3417                 struct btrfs_space_info *tmp;
3418
3419                 info = NULL;
3420                 rcu_read_lock();
3421                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3422                                         list) {
3423                         if (tmp->flags == types[i]) {
3424                                 info = tmp;
3425                                 break;
3426                         }
3427                 }
3428                 rcu_read_unlock();
3429
3430                 if (!info)
3431                         continue;
3432
3433                 down_read(&info->groups_sem);
3434                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3435                         if (!list_empty(&info->block_groups[c]))
3436                                 slot_count++;
3437                 }
3438                 up_read(&info->groups_sem);
3439         }
3440
3441         /* space_slots == 0 means they are asking for a count */
3442         if (space_args.space_slots == 0) {
3443                 space_args.total_spaces = slot_count;
3444                 goto out;
3445         }
3446
3447         slot_count = min_t(u64, space_args.space_slots, slot_count);
3448
3449         alloc_size = sizeof(*dest) * slot_count;
3450
3451         /* we generally have at most 6 or so space infos, one for each raid
3452          * level.  So, a whole page should be more than enough for everyone
3453          */
3454         if (alloc_size > PAGE_CACHE_SIZE)
3455                 return -ENOMEM;
3456
3457         space_args.total_spaces = 0;
3458         dest = kmalloc(alloc_size, GFP_NOFS);
3459         if (!dest)
3460                 return -ENOMEM;
3461         dest_orig = dest;
3462
3463         /* now we have a buffer to copy into */
3464         for (i = 0; i < num_types; i++) {
3465                 struct btrfs_space_info *tmp;
3466
3467                 if (!slot_count)
3468                         break;
3469
3470                 info = NULL;
3471                 rcu_read_lock();
3472                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3473                                         list) {
3474                         if (tmp->flags == types[i]) {
3475                                 info = tmp;
3476                                 break;
3477                         }
3478                 }
3479                 rcu_read_unlock();
3480
3481                 if (!info)
3482                         continue;
3483                 down_read(&info->groups_sem);
3484                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3485                         if (!list_empty(&info->block_groups[c])) {
3486                                 btrfs_get_block_group_info(
3487                                         &info->block_groups[c], &space);
3488                                 memcpy(dest, &space, sizeof(space));
3489                                 dest++;
3490                                 space_args.total_spaces++;
3491                                 slot_count--;
3492                         }
3493                         if (!slot_count)
3494                                 break;
3495                 }
3496                 up_read(&info->groups_sem);
3497         }
3498
3499         user_dest = (struct btrfs_ioctl_space_info __user *)
3500                 (arg + sizeof(struct btrfs_ioctl_space_args));
3501
3502         if (copy_to_user(user_dest, dest_orig, alloc_size))
3503                 ret = -EFAULT;
3504
3505         kfree(dest_orig);
3506 out:
3507         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3508                 ret = -EFAULT;
3509
3510         return ret;
3511 }
3512
3513 /*
3514  * there are many ways the trans_start and trans_end ioctls can lead
3515  * to deadlocks.  They should only be used by applications that
3516  * basically own the machine, and have a very in depth understanding
3517  * of all the possible deadlocks and enospc problems.
3518  */
3519 long btrfs_ioctl_trans_end(struct file *file)
3520 {
3521         struct inode *inode = file_inode(file);
3522         struct btrfs_root *root = BTRFS_I(inode)->root;
3523         struct btrfs_trans_handle *trans;
3524
3525         trans = file->private_data;
3526         if (!trans)
3527                 return -EINVAL;
3528         file->private_data = NULL;
3529
3530         btrfs_end_transaction(trans, root);
3531
3532         atomic_dec(&root->fs_info->open_ioctl_trans);
3533
3534         mnt_drop_write_file(file);
3535         return 0;
3536 }
3537
3538 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3539                                             void __user *argp)
3540 {
3541         struct btrfs_trans_handle *trans;
3542         u64 transid;
3543         int ret;
3544
3545         trans = btrfs_attach_transaction_barrier(root);
3546         if (IS_ERR(trans)) {
3547                 if (PTR_ERR(trans) != -ENOENT)
3548                         return PTR_ERR(trans);
3549
3550                 /* No running transaction, don't bother */
3551                 transid = root->fs_info->last_trans_committed;
3552                 goto out;
3553         }
3554         transid = trans->transid;
3555         ret = btrfs_commit_transaction_async(trans, root, 0);
3556         if (ret) {
3557                 btrfs_end_transaction(trans, root);
3558                 return ret;
3559         }
3560 out:
3561         if (argp)
3562                 if (copy_to_user(argp, &transid, sizeof(transid)))
3563                         return -EFAULT;
3564         return 0;
3565 }
3566
3567 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3568                                            void __user *argp)
3569 {
3570         u64 transid;
3571
3572         if (argp) {
3573                 if (copy_from_user(&transid, argp, sizeof(transid)))
3574                         return -EFAULT;
3575         } else {
3576                 transid = 0;  /* current trans */
3577         }
3578         return btrfs_wait_for_commit(root, transid);
3579 }
3580
3581 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3582 {
3583         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3584         struct btrfs_ioctl_scrub_args *sa;
3585         int ret;
3586
3587         if (!capable(CAP_SYS_ADMIN))
3588                 return -EPERM;
3589
3590         sa = memdup_user(arg, sizeof(*sa));
3591         if (IS_ERR(sa))
3592                 return PTR_ERR(sa);
3593
3594         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3595                 ret = mnt_want_write_file(file);
3596                 if (ret)
3597                         goto out;
3598         }
3599
3600         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3601                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3602                               0);
3603
3604         if (copy_to_user(arg, sa, sizeof(*sa)))
3605                 ret = -EFAULT;
3606
3607         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3608                 mnt_drop_write_file(file);
3609 out:
3610         kfree(sa);
3611         return ret;
3612 }
3613
3614 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3615 {
3616         if (!capable(CAP_SYS_ADMIN))
3617                 return -EPERM;
3618
3619         return btrfs_scrub_cancel(root->fs_info);
3620 }
3621
3622 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3623                                        void __user *arg)
3624 {
3625         struct btrfs_ioctl_scrub_args *sa;
3626         int ret;
3627
3628         if (!capable(CAP_SYS_ADMIN))
3629                 return -EPERM;
3630
3631         sa = memdup_user(arg, sizeof(*sa));
3632         if (IS_ERR(sa))
3633                 return PTR_ERR(sa);
3634
3635         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3636
3637         if (copy_to_user(arg, sa, sizeof(*sa)))
3638                 ret = -EFAULT;
3639
3640         kfree(sa);
3641         return ret;
3642 }
3643
3644 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3645                                       void __user *arg)
3646 {
3647         struct btrfs_ioctl_get_dev_stats *sa;
3648         int ret;
3649
3650         sa = memdup_user(arg, sizeof(*sa));
3651         if (IS_ERR(sa))
3652                 return PTR_ERR(sa);
3653
3654         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3655                 kfree(sa);
3656                 return -EPERM;
3657         }
3658
3659         ret = btrfs_get_dev_stats(root, sa);
3660
3661         if (copy_to_user(arg, sa, sizeof(*sa)))
3662                 ret = -EFAULT;
3663
3664         kfree(sa);
3665         return ret;
3666 }
3667
3668 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3669 {
3670         struct btrfs_ioctl_dev_replace_args *p;
3671         int ret;
3672
3673         if (!capable(CAP_SYS_ADMIN))
3674                 return -EPERM;
3675
3676         p = memdup_user(arg, sizeof(*p));
3677         if (IS_ERR(p))
3678                 return PTR_ERR(p);
3679
3680         switch (p->cmd) {
3681         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3682                 if (root->fs_info->sb->s_flags & MS_RDONLY)
3683                         return -EROFS;
3684
3685                 if (atomic_xchg(
3686                         &root->fs_info->mutually_exclusive_operation_running,
3687                         1)) {
3688                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3689                 } else {
3690                         ret = btrfs_dev_replace_start(root, p);
3691                         atomic_set(
3692                          &root->fs_info->mutually_exclusive_operation_running,
3693                          0);
3694                 }
3695                 break;
3696         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3697                 btrfs_dev_replace_status(root->fs_info, p);
3698                 ret = 0;
3699                 break;
3700         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3701                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3702                 break;
3703         default:
3704                 ret = -EINVAL;
3705                 break;
3706         }
3707
3708         if (copy_to_user(arg, p, sizeof(*p)))
3709                 ret = -EFAULT;
3710
3711         kfree(p);
3712         return ret;
3713 }
3714
3715 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3716 {
3717         int ret = 0;
3718         int i;
3719         u64 rel_ptr;
3720         int size;
3721         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3722         struct inode_fs_paths *ipath = NULL;
3723         struct btrfs_path *path;
3724
3725         if (!capable(CAP_DAC_READ_SEARCH))
3726                 return -EPERM;
3727
3728         path = btrfs_alloc_path();
3729         if (!path) {
3730                 ret = -ENOMEM;
3731                 goto out;
3732         }
3733
3734         ipa = memdup_user(arg, sizeof(*ipa));
3735         if (IS_ERR(ipa)) {
3736                 ret = PTR_ERR(ipa);
3737                 ipa = NULL;
3738                 goto out;
3739         }
3740
3741         size = min_t(u32, ipa->size, 4096);
3742         ipath = init_ipath(size, root, path);
3743         if (IS_ERR(ipath)) {
3744                 ret = PTR_ERR(ipath);
3745                 ipath = NULL;
3746                 goto out;
3747         }
3748
3749         ret = paths_from_inode(ipa->inum, ipath);
3750         if (ret < 0)
3751                 goto out;
3752
3753         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3754                 rel_ptr = ipath->fspath->val[i] -
3755                           (u64)(unsigned long)ipath->fspath->val;
3756                 ipath->fspath->val[i] = rel_ptr;
3757         }
3758
3759         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3760                            (void *)(unsigned long)ipath->fspath, size);
3761         if (ret) {
3762                 ret = -EFAULT;
3763                 goto out;
3764         }
3765
3766 out:
3767         btrfs_free_path(path);
3768         free_ipath(ipath);
3769         kfree(ipa);
3770
3771         return ret;
3772 }
3773
3774 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3775 {
3776         struct btrfs_data_container *inodes = ctx;
3777         const size_t c = 3 * sizeof(u64);
3778
3779         if (inodes->bytes_left >= c) {
3780                 inodes->bytes_left -= c;
3781                 inodes->val[inodes->elem_cnt] = inum;
3782                 inodes->val[inodes->elem_cnt + 1] = offset;
3783                 inodes->val[inodes->elem_cnt + 2] = root;
3784                 inodes->elem_cnt += 3;
3785         } else {
3786                 inodes->bytes_missing += c - inodes->bytes_left;
3787                 inodes->bytes_left = 0;
3788                 inodes->elem_missed += 3;
3789         }
3790
3791         return 0;
3792 }
3793
3794 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3795                                         void __user *arg)
3796 {
3797         int ret = 0;
3798         int size;
3799         struct btrfs_ioctl_logical_ino_args *loi;
3800         struct btrfs_data_container *inodes = NULL;
3801         struct btrfs_path *path = NULL;
3802
3803         if (!capable(CAP_SYS_ADMIN))
3804                 return -EPERM;
3805
3806         loi = memdup_user(arg, sizeof(*loi));
3807         if (IS_ERR(loi)) {
3808                 ret = PTR_ERR(loi);
3809                 loi = NULL;
3810                 goto out;
3811         }
3812
3813         path = btrfs_alloc_path();
3814         if (!path) {
3815                 ret = -ENOMEM;
3816                 goto out;
3817         }
3818
3819         size = min_t(u32, loi->size, 64 * 1024);
3820         inodes = init_data_container(size);
3821         if (IS_ERR(inodes)) {
3822                 ret = PTR_ERR(inodes);
3823                 inodes = NULL;
3824                 goto out;
3825         }
3826
3827         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3828                                           build_ino_list, inodes);
3829         if (ret == -EINVAL)
3830                 ret = -ENOENT;
3831         if (ret < 0)
3832                 goto out;
3833
3834         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3835                            (void *)(unsigned long)inodes, size);
3836         if (ret)
3837                 ret = -EFAULT;
3838
3839 out:
3840         btrfs_free_path(path);
3841         vfree(inodes);
3842         kfree(loi);
3843
3844         return ret;
3845 }
3846
3847 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3848                                struct btrfs_ioctl_balance_args *bargs)
3849 {
3850         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3851
3852         bargs->flags = bctl->flags;
3853
3854         if (atomic_read(&fs_info->balance_running))
3855                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3856         if (atomic_read(&fs_info->balance_pause_req))
3857                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3858         if (atomic_read(&fs_info->balance_cancel_req))
3859                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3860
3861         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3862         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3863         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3864
3865         if (lock) {
3866                 spin_lock(&fs_info->balance_lock);
3867                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3868                 spin_unlock(&fs_info->balance_lock);
3869         } else {
3870                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3871         }
3872 }
3873
3874 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3875 {
3876         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3877         struct btrfs_fs_info *fs_info = root->fs_info;
3878         struct btrfs_ioctl_balance_args *bargs;
3879         struct btrfs_balance_control *bctl;
3880         bool need_unlock; /* for mut. excl. ops lock */
3881         int ret;
3882
3883         if (!capable(CAP_SYS_ADMIN))
3884                 return -EPERM;
3885
3886         ret = mnt_want_write_file(file);
3887         if (ret)
3888                 return ret;
3889
3890 again:
3891         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
3892                 mutex_lock(&fs_info->volume_mutex);
3893                 mutex_lock(&fs_info->balance_mutex);
3894                 need_unlock = true;
3895                 goto locked;
3896         }
3897
3898         /*
3899          * mut. excl. ops lock is locked.  Three possibilites:
3900          *   (1) some other op is running
3901          *   (2) balance is running
3902          *   (3) balance is paused -- special case (think resume)
3903          */
3904         mutex_lock(&fs_info->balance_mutex);
3905         if (fs_info->balance_ctl) {
3906                 /* this is either (2) or (3) */
3907                 if (!atomic_read(&fs_info->balance_running)) {
3908                         mutex_unlock(&fs_info->balance_mutex);
3909                         if (!mutex_trylock(&fs_info->volume_mutex))
3910                                 goto again;
3911                         mutex_lock(&fs_info->balance_mutex);
3912
3913                         if (fs_info->balance_ctl &&
3914                             !atomic_read(&fs_info->balance_running)) {
3915                                 /* this is (3) */
3916                                 need_unlock = false;
3917                                 goto locked;
3918                         }
3919
3920                         mutex_unlock(&fs_info->balance_mutex);
3921                         mutex_unlock(&fs_info->volume_mutex);
3922                         goto again;
3923                 } else {
3924                         /* this is (2) */
3925                         mutex_unlock(&fs_info->balance_mutex);
3926                         ret = -EINPROGRESS;
3927                         goto out;
3928                 }
3929         } else {
3930                 /* this is (1) */
3931                 mutex_unlock(&fs_info->balance_mutex);
3932                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3933                 goto out;
3934         }
3935
3936 locked:
3937         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
3938
3939         if (arg) {
3940                 bargs = memdup_user(arg, sizeof(*bargs));
3941                 if (IS_ERR(bargs)) {
3942                         ret = PTR_ERR(bargs);
3943                         goto out_unlock;
3944                 }
3945
3946                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3947                         if (!fs_info->balance_ctl) {
3948                                 ret = -ENOTCONN;
3949                                 goto out_bargs;
3950                         }
3951
3952                         bctl = fs_info->balance_ctl;
3953                         spin_lock(&fs_info->balance_lock);
3954                         bctl->flags |= BTRFS_BALANCE_RESUME;
3955                         spin_unlock(&fs_info->balance_lock);
3956
3957                         goto do_balance;
3958                 }
3959         } else {
3960                 bargs = NULL;
3961         }
3962
3963         if (fs_info->balance_ctl) {
3964                 ret = -EINPROGRESS;
3965                 goto out_bargs;
3966         }
3967
3968         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3969         if (!bctl) {
3970                 ret = -ENOMEM;
3971                 goto out_bargs;
3972         }
3973
3974         bctl->fs_info = fs_info;
3975         if (arg) {
3976                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3977                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3978                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3979
3980                 bctl->flags = bargs->flags;
3981         } else {
3982                 /* balance everything - no filters */
3983                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3984         }
3985
3986 do_balance:
3987         /*
3988          * Ownership of bctl and mutually_exclusive_operation_running
3989          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
3990          * or, if restriper was paused all the way until unmount, in
3991          * free_fs_info.  mutually_exclusive_operation_running is
3992          * cleared in __cancel_balance.
3993          */
3994         need_unlock = false;
3995
3996         ret = btrfs_balance(bctl, bargs);
3997
3998         if (arg) {
3999                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4000                         ret = -EFAULT;
4001         }
4002
4003 out_bargs:
4004         kfree(bargs);
4005 out_unlock:
4006         mutex_unlock(&fs_info->balance_mutex);
4007         mutex_unlock(&fs_info->volume_mutex);
4008         if (need_unlock)
4009                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4010 out:
4011         mnt_drop_write_file(file);
4012         return ret;
4013 }
4014
4015 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4016 {
4017         if (!capable(CAP_SYS_ADMIN))
4018                 return -EPERM;
4019
4020         switch (cmd) {
4021         case BTRFS_BALANCE_CTL_PAUSE:
4022                 return btrfs_pause_balance(root->fs_info);
4023         case BTRFS_BALANCE_CTL_CANCEL:
4024                 return btrfs_cancel_balance(root->fs_info);
4025         }
4026
4027         return -EINVAL;
4028 }
4029
4030 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4031                                          void __user *arg)
4032 {
4033         struct btrfs_fs_info *fs_info = root->fs_info;
4034         struct btrfs_ioctl_balance_args *bargs;
4035         int ret = 0;
4036
4037         if (!capable(CAP_SYS_ADMIN))
4038                 return -EPERM;
4039
4040         mutex_lock(&fs_info->balance_mutex);
4041         if (!fs_info->balance_ctl) {
4042                 ret = -ENOTCONN;
4043                 goto out;
4044         }
4045
4046         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4047         if (!bargs) {
4048                 ret = -ENOMEM;
4049                 goto out;
4050         }
4051
4052         update_ioctl_balance_args(fs_info, 1, bargs);
4053
4054         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4055                 ret = -EFAULT;
4056
4057         kfree(bargs);
4058 out:
4059         mutex_unlock(&fs_info->balance_mutex);
4060         return ret;
4061 }
4062
4063 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4064 {
4065         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4066         struct btrfs_ioctl_quota_ctl_args *sa;
4067         struct btrfs_trans_handle *trans = NULL;
4068         int ret;
4069         int err;
4070
4071         if (!capable(CAP_SYS_ADMIN))
4072                 return -EPERM;
4073
4074         ret = mnt_want_write_file(file);
4075         if (ret)
4076                 return ret;
4077
4078         sa = memdup_user(arg, sizeof(*sa));
4079         if (IS_ERR(sa)) {
4080                 ret = PTR_ERR(sa);
4081                 goto drop_write;
4082         }
4083
4084         down_write(&root->fs_info->subvol_sem);
4085         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4086         if (IS_ERR(trans)) {
4087                 ret = PTR_ERR(trans);
4088                 goto out;
4089         }
4090
4091         switch (sa->cmd) {
4092         case BTRFS_QUOTA_CTL_ENABLE:
4093                 ret = btrfs_quota_enable(trans, root->fs_info);
4094                 break;
4095         case BTRFS_QUOTA_CTL_DISABLE:
4096                 ret = btrfs_quota_disable(trans, root->fs_info);
4097                 break;
4098         default:
4099                 ret = -EINVAL;
4100                 break;
4101         }
4102
4103         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4104         if (err && !ret)
4105                 ret = err;
4106 out:
4107         kfree(sa);
4108         up_write(&root->fs_info->subvol_sem);
4109 drop_write:
4110         mnt_drop_write_file(file);
4111         return ret;
4112 }
4113
4114 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4115 {
4116         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4117         struct btrfs_ioctl_qgroup_assign_args *sa;
4118         struct btrfs_trans_handle *trans;
4119         int ret;
4120         int err;
4121
4122         if (!capable(CAP_SYS_ADMIN))
4123                 return -EPERM;
4124
4125         ret = mnt_want_write_file(file);
4126         if (ret)
4127                 return ret;
4128
4129         sa = memdup_user(arg, sizeof(*sa));
4130         if (IS_ERR(sa)) {
4131                 ret = PTR_ERR(sa);
4132                 goto drop_write;
4133         }
4134
4135         trans = btrfs_join_transaction(root);
4136         if (IS_ERR(trans)) {
4137                 ret = PTR_ERR(trans);
4138                 goto out;
4139         }
4140
4141         /* FIXME: check if the IDs really exist */
4142         if (sa->assign) {
4143                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4144                                                 sa->src, sa->dst);
4145         } else {
4146                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4147                                                 sa->src, sa->dst);
4148         }
4149
4150         err = btrfs_end_transaction(trans, root);
4151         if (err && !ret)
4152                 ret = err;
4153
4154 out:
4155         kfree(sa);
4156 drop_write:
4157         mnt_drop_write_file(file);
4158         return ret;
4159 }
4160
4161 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4162 {
4163         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4164         struct btrfs_ioctl_qgroup_create_args *sa;
4165         struct btrfs_trans_handle *trans;
4166         int ret;
4167         int err;
4168
4169         if (!capable(CAP_SYS_ADMIN))
4170                 return -EPERM;
4171
4172         ret = mnt_want_write_file(file);
4173         if (ret)
4174                 return ret;
4175
4176         sa = memdup_user(arg, sizeof(*sa));
4177         if (IS_ERR(sa)) {
4178                 ret = PTR_ERR(sa);
4179                 goto drop_write;
4180         }
4181
4182         if (!sa->qgroupid) {
4183                 ret = -EINVAL;
4184                 goto out;
4185         }
4186
4187         trans = btrfs_join_transaction(root);
4188         if (IS_ERR(trans)) {
4189                 ret = PTR_ERR(trans);
4190                 goto out;
4191         }
4192
4193         /* FIXME: check if the IDs really exist */
4194         if (sa->create) {
4195                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4196                                           NULL);
4197         } else {
4198                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4199         }
4200
4201         err = btrfs_end_transaction(trans, root);
4202         if (err && !ret)
4203                 ret = err;
4204
4205 out:
4206         kfree(sa);
4207 drop_write:
4208         mnt_drop_write_file(file);
4209         return ret;
4210 }
4211
4212 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4213 {
4214         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4215         struct btrfs_ioctl_qgroup_limit_args *sa;
4216         struct btrfs_trans_handle *trans;
4217         int ret;
4218         int err;
4219         u64 qgroupid;
4220
4221         if (!capable(CAP_SYS_ADMIN))
4222                 return -EPERM;
4223
4224         ret = mnt_want_write_file(file);
4225         if (ret)
4226                 return ret;
4227
4228         sa = memdup_user(arg, sizeof(*sa));
4229         if (IS_ERR(sa)) {
4230                 ret = PTR_ERR(sa);
4231                 goto drop_write;
4232         }
4233
4234         trans = btrfs_join_transaction(root);
4235         if (IS_ERR(trans)) {
4236                 ret = PTR_ERR(trans);
4237                 goto out;
4238         }
4239
4240         qgroupid = sa->qgroupid;
4241         if (!qgroupid) {
4242                 /* take the current subvol as qgroup */
4243                 qgroupid = root->root_key.objectid;
4244         }
4245
4246         /* FIXME: check if the IDs really exist */
4247         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4248
4249         err = btrfs_end_transaction(trans, root);
4250         if (err && !ret)
4251                 ret = err;
4252
4253 out:
4254         kfree(sa);
4255 drop_write:
4256         mnt_drop_write_file(file);
4257         return ret;
4258 }
4259
4260 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4261 {
4262         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4263         struct btrfs_ioctl_quota_rescan_args *qsa;
4264         int ret;
4265
4266         if (!capable(CAP_SYS_ADMIN))
4267                 return -EPERM;
4268
4269         ret = mnt_want_write_file(file);
4270         if (ret)
4271                 return ret;
4272
4273         qsa = memdup_user(arg, sizeof(*qsa));
4274         if (IS_ERR(qsa)) {
4275                 ret = PTR_ERR(qsa);
4276                 goto drop_write;
4277         }
4278
4279         if (qsa->flags) {
4280                 ret = -EINVAL;
4281                 goto out;
4282         }
4283
4284         ret = btrfs_qgroup_rescan(root->fs_info);
4285
4286 out:
4287         kfree(qsa);
4288 drop_write:
4289         mnt_drop_write_file(file);
4290         return ret;
4291 }
4292
4293 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4294 {
4295         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4296         struct btrfs_ioctl_quota_rescan_args *qsa;
4297         int ret = 0;
4298
4299         if (!capable(CAP_SYS_ADMIN))
4300                 return -EPERM;
4301
4302         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4303         if (!qsa)
4304                 return -ENOMEM;
4305
4306         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4307                 qsa->flags = 1;
4308                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4309         }
4310
4311         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4312                 ret = -EFAULT;
4313
4314         kfree(qsa);
4315         return ret;
4316 }
4317
4318 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4319 {
4320         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4321
4322         if (!capable(CAP_SYS_ADMIN))
4323                 return -EPERM;
4324
4325         return btrfs_qgroup_wait_for_completion(root->fs_info);
4326 }
4327
4328 static long btrfs_ioctl_set_received_subvol(struct file *file,
4329                                             void __user *arg)
4330 {
4331         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4332         struct inode *inode = file_inode(file);
4333         struct btrfs_root *root = BTRFS_I(inode)->root;
4334         struct btrfs_root_item *root_item = &root->root_item;
4335         struct btrfs_trans_handle *trans;
4336         struct timespec ct = CURRENT_TIME;
4337         int ret = 0;
4338         int received_uuid_changed;
4339
4340         ret = mnt_want_write_file(file);
4341         if (ret < 0)
4342                 return ret;
4343
4344         down_write(&root->fs_info->subvol_sem);
4345
4346         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4347                 ret = -EINVAL;
4348                 goto out;
4349         }
4350
4351         if (btrfs_root_readonly(root)) {
4352                 ret = -EROFS;
4353                 goto out;
4354         }
4355
4356         if (!inode_owner_or_capable(inode)) {
4357                 ret = -EACCES;
4358                 goto out;
4359         }
4360
4361         sa = memdup_user(arg, sizeof(*sa));
4362         if (IS_ERR(sa)) {
4363                 ret = PTR_ERR(sa);
4364                 sa = NULL;
4365                 goto out;
4366         }
4367
4368         /*
4369          * 1 - root item
4370          * 2 - uuid items (received uuid + subvol uuid)
4371          */
4372         trans = btrfs_start_transaction(root, 3);
4373         if (IS_ERR(trans)) {
4374                 ret = PTR_ERR(trans);
4375                 trans = NULL;
4376                 goto out;
4377         }
4378
4379         sa->rtransid = trans->transid;
4380         sa->rtime.sec = ct.tv_sec;
4381         sa->rtime.nsec = ct.tv_nsec;
4382
4383         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4384                                        BTRFS_UUID_SIZE);
4385         if (received_uuid_changed &&
4386             !btrfs_is_empty_uuid(root_item->received_uuid))
4387                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4388                                     root_item->received_uuid,
4389                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4390                                     root->root_key.objectid);
4391         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4392         btrfs_set_root_stransid(root_item, sa->stransid);
4393         btrfs_set_root_rtransid(root_item, sa->rtransid);
4394         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4395         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4396         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4397         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4398
4399         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4400                                 &root->root_key, &root->root_item);
4401         if (ret < 0) {
4402                 btrfs_end_transaction(trans, root);
4403                 goto out;
4404         }
4405         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4406                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4407                                           sa->uuid,
4408                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4409                                           root->root_key.objectid);
4410                 if (ret < 0 && ret != -EEXIST) {
4411                         btrfs_abort_transaction(trans, root, ret);
4412                         goto out;
4413                 }
4414         }
4415         ret = btrfs_commit_transaction(trans, root);
4416         if (ret < 0) {
4417                 btrfs_abort_transaction(trans, root, ret);
4418                 goto out;
4419         }
4420
4421         ret = copy_to_user(arg, sa, sizeof(*sa));
4422         if (ret)
4423                 ret = -EFAULT;
4424
4425 out:
4426         kfree(sa);
4427         up_write(&root->fs_info->subvol_sem);
4428         mnt_drop_write_file(file);
4429         return ret;
4430 }
4431
4432 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4433 {
4434         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4435         size_t len;
4436         int ret;
4437         char label[BTRFS_LABEL_SIZE];
4438
4439         spin_lock(&root->fs_info->super_lock);
4440         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4441         spin_unlock(&root->fs_info->super_lock);
4442
4443         len = strnlen(label, BTRFS_LABEL_SIZE);
4444
4445         if (len == BTRFS_LABEL_SIZE) {
4446                 pr_warn("btrfs: label is too long, return the first %zu bytes\n",
4447                         --len);
4448         }
4449
4450         ret = copy_to_user(arg, label, len);
4451
4452         return ret ? -EFAULT : 0;
4453 }
4454
4455 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4456 {
4457         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4458         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4459         struct btrfs_trans_handle *trans;
4460         char label[BTRFS_LABEL_SIZE];
4461         int ret;
4462
4463         if (!capable(CAP_SYS_ADMIN))
4464                 return -EPERM;
4465
4466         if (copy_from_user(label, arg, sizeof(label)))
4467                 return -EFAULT;
4468
4469         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4470                 pr_err("btrfs: unable to set label with more than %d bytes\n",
4471                        BTRFS_LABEL_SIZE - 1);
4472                 return -EINVAL;
4473         }
4474
4475         ret = mnt_want_write_file(file);
4476         if (ret)
4477                 return ret;
4478
4479         trans = btrfs_start_transaction(root, 0);
4480         if (IS_ERR(trans)) {
4481                 ret = PTR_ERR(trans);
4482                 goto out_unlock;
4483         }
4484
4485         spin_lock(&root->fs_info->super_lock);
4486         strcpy(super_block->label, label);
4487         spin_unlock(&root->fs_info->super_lock);
4488         ret = btrfs_end_transaction(trans, root);
4489
4490 out_unlock:
4491         mnt_drop_write_file(file);
4492         return ret;
4493 }
4494
4495 long btrfs_ioctl(struct file *file, unsigned int
4496                 cmd, unsigned long arg)
4497 {
4498         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4499         void __user *argp = (void __user *)arg;
4500
4501         switch (cmd) {
4502         case FS_IOC_GETFLAGS:
4503                 return btrfs_ioctl_getflags(file, argp);
4504         case FS_IOC_SETFLAGS:
4505                 return btrfs_ioctl_setflags(file, argp);
4506         case FS_IOC_GETVERSION:
4507                 return btrfs_ioctl_getversion(file, argp);
4508         case FITRIM:
4509                 return btrfs_ioctl_fitrim(file, argp);
4510         case BTRFS_IOC_SNAP_CREATE:
4511                 return btrfs_ioctl_snap_create(file, argp, 0);
4512         case BTRFS_IOC_SNAP_CREATE_V2:
4513                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4514         case BTRFS_IOC_SUBVOL_CREATE:
4515                 return btrfs_ioctl_snap_create(file, argp, 1);
4516         case BTRFS_IOC_SUBVOL_CREATE_V2:
4517                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4518         case BTRFS_IOC_SNAP_DESTROY:
4519                 return btrfs_ioctl_snap_destroy(file, argp);
4520         case BTRFS_IOC_SUBVOL_GETFLAGS:
4521                 return btrfs_ioctl_subvol_getflags(file, argp);
4522         case BTRFS_IOC_SUBVOL_SETFLAGS:
4523                 return btrfs_ioctl_subvol_setflags(file, argp);
4524         case BTRFS_IOC_DEFAULT_SUBVOL:
4525                 return btrfs_ioctl_default_subvol(file, argp);
4526         case BTRFS_IOC_DEFRAG:
4527                 return btrfs_ioctl_defrag(file, NULL);
4528         case BTRFS_IOC_DEFRAG_RANGE:
4529                 return btrfs_ioctl_defrag(file, argp);
4530         case BTRFS_IOC_RESIZE:
4531                 return btrfs_ioctl_resize(file, argp);
4532         case BTRFS_IOC_ADD_DEV:
4533                 return btrfs_ioctl_add_dev(root, argp);
4534         case BTRFS_IOC_RM_DEV:
4535                 return btrfs_ioctl_rm_dev(file, argp);
4536         case BTRFS_IOC_FS_INFO:
4537                 return btrfs_ioctl_fs_info(root, argp);
4538         case BTRFS_IOC_DEV_INFO:
4539                 return btrfs_ioctl_dev_info(root, argp);
4540         case BTRFS_IOC_BALANCE:
4541                 return btrfs_ioctl_balance(file, NULL);
4542         case BTRFS_IOC_CLONE:
4543                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4544         case BTRFS_IOC_CLONE_RANGE:
4545                 return btrfs_ioctl_clone_range(file, argp);
4546         case BTRFS_IOC_TRANS_START:
4547                 return btrfs_ioctl_trans_start(file);
4548         case BTRFS_IOC_TRANS_END:
4549                 return btrfs_ioctl_trans_end(file);
4550         case BTRFS_IOC_TREE_SEARCH:
4551                 return btrfs_ioctl_tree_search(file, argp);
4552         case BTRFS_IOC_INO_LOOKUP:
4553                 return btrfs_ioctl_ino_lookup(file, argp);
4554         case BTRFS_IOC_INO_PATHS:
4555                 return btrfs_ioctl_ino_to_path(root, argp);
4556         case BTRFS_IOC_LOGICAL_INO:
4557                 return btrfs_ioctl_logical_to_ino(root, argp);
4558         case BTRFS_IOC_SPACE_INFO:
4559                 return btrfs_ioctl_space_info(root, argp);
4560         case BTRFS_IOC_SYNC:
4561                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
4562                 return 0;
4563         case BTRFS_IOC_START_SYNC:
4564                 return btrfs_ioctl_start_sync(root, argp);
4565         case BTRFS_IOC_WAIT_SYNC:
4566                 return btrfs_ioctl_wait_sync(root, argp);
4567         case BTRFS_IOC_SCRUB:
4568                 return btrfs_ioctl_scrub(file, argp);
4569         case BTRFS_IOC_SCRUB_CANCEL:
4570                 return btrfs_ioctl_scrub_cancel(root, argp);
4571         case BTRFS_IOC_SCRUB_PROGRESS:
4572                 return btrfs_ioctl_scrub_progress(root, argp);
4573         case BTRFS_IOC_BALANCE_V2:
4574                 return btrfs_ioctl_balance(file, argp);
4575         case BTRFS_IOC_BALANCE_CTL:
4576                 return btrfs_ioctl_balance_ctl(root, arg);
4577         case BTRFS_IOC_BALANCE_PROGRESS:
4578                 return btrfs_ioctl_balance_progress(root, argp);
4579         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4580                 return btrfs_ioctl_set_received_subvol(file, argp);
4581         case BTRFS_IOC_SEND:
4582                 return btrfs_ioctl_send(file, argp);
4583         case BTRFS_IOC_GET_DEV_STATS:
4584                 return btrfs_ioctl_get_dev_stats(root, argp);
4585         case BTRFS_IOC_QUOTA_CTL:
4586                 return btrfs_ioctl_quota_ctl(file, argp);
4587         case BTRFS_IOC_QGROUP_ASSIGN:
4588                 return btrfs_ioctl_qgroup_assign(file, argp);
4589         case BTRFS_IOC_QGROUP_CREATE:
4590                 return btrfs_ioctl_qgroup_create(file, argp);
4591         case BTRFS_IOC_QGROUP_LIMIT:
4592                 return btrfs_ioctl_qgroup_limit(file, argp);
4593         case BTRFS_IOC_QUOTA_RESCAN:
4594                 return btrfs_ioctl_quota_rescan(file, argp);
4595         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4596                 return btrfs_ioctl_quota_rescan_status(file, argp);
4597         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4598                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4599         case BTRFS_IOC_DEV_REPLACE:
4600                 return btrfs_ioctl_dev_replace(root, argp);
4601         case BTRFS_IOC_GET_FSLABEL:
4602                 return btrfs_ioctl_get_fslabel(file, argp);
4603         case BTRFS_IOC_SET_FSLABEL:
4604                 return btrfs_ioctl_set_fslabel(file, argp);
4605         case BTRFS_IOC_FILE_EXTENT_SAME:
4606                 return btrfs_ioctl_file_extent_same(file, argp);
4607         }
4608
4609         return -ENOTTY;
4610 }