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