fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "compat.h"
  48 #include "ctree.h"
  49 #include "disk-io.h"
  50 #include "transaction.h"
  51 #include "btrfs_inode.h"
  52 #include "print-tree.h"
  53 #include "volumes.h"
  54 #include "locking.h"
  55 #include "inode-map.h"
  56 #include "backref.h"
  57 #include "rcu-string.h"
  58 #include "send.h"
  59 #include "dev-replace.h"
  60
  61 static int btrfs_clone(struct inode *src, struct inode *inode,
  62                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  63
  64 /* Mask out flags that are inappropriate for the given type of inode. */
  65 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  66 {
  67         if (S_ISDIR(mode))
  68                 return flags;
  69         else if (S_ISREG(mode))
  70                 return flags & ~FS_DIRSYNC_FL;
  71         else
  72                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  73 }
  74
  75 /*
  76  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  77  */
  78 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  79 {
  80         unsigned int iflags = 0;
  81
  82         if (flags & BTRFS_INODE_SYNC)
  83                 iflags |= FS_SYNC_FL;
  84         if (flags & BTRFS_INODE_IMMUTABLE)
  85                 iflags |= FS_IMMUTABLE_FL;
  86         if (flags & BTRFS_INODE_APPEND)
  87                 iflags |= FS_APPEND_FL;
  88         if (flags & BTRFS_INODE_NODUMP)
  89                 iflags |= FS_NODUMP_FL;
  90         if (flags & BTRFS_INODE_NOATIME)
  91                 iflags |= FS_NOATIME_FL;
  92         if (flags & BTRFS_INODE_DIRSYNC)
  93                 iflags |= FS_DIRSYNC_FL;
  94         if (flags & BTRFS_INODE_NODATACOW)
  95                 iflags |= FS_NOCOW_FL;
  96
  97         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
  98                 iflags |= FS_COMPR_FL;
  99         else if (flags & BTRFS_INODE_NOCOMPRESS)
 100                 iflags |= FS_NOCOMP_FL;
 101
 102         return iflags;
 103 }
 104
 105 /*
 106  * Update inode->i_flags based on the btrfs internal flags.
 107  */
 108 void btrfs_update_iflags(struct inode *inode)
 109 {
 110         struct btrfs_inode *ip = BTRFS_I(inode);
 111
 112         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 113
 114         if (ip->flags & BTRFS_INODE_SYNC)
 115                 inode->i_flags |= S_SYNC;
 116         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 117                 inode->i_flags |= S_IMMUTABLE;
 118         if (ip->flags & BTRFS_INODE_APPEND)
 119                 inode->i_flags |= S_APPEND;
 120         if (ip->flags & BTRFS_INODE_NOATIME)
 121                 inode->i_flags |= S_NOATIME;
 122         if (ip->flags & BTRFS_INODE_DIRSYNC)
 123                 inode->i_flags |= S_DIRSYNC;
 124 }
 125
 126 /*
 127  * Inherit flags from the parent inode.
 128  *
 129  * Currently only the compression flags and the cow flags are inherited.
 130  */
 131 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 132 {
 133         unsigned int flags;
 134
 135         if (!dir)
 136                 return;
 137
 138         flags = BTRFS_I(dir)->flags;
 139
 140         if (flags & BTRFS_INODE_NOCOMPRESS) {
 141                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 142                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 143         } else if (flags & BTRFS_INODE_COMPRESS) {
 144                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 145                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 146         }
 147
 148         if (flags & BTRFS_INODE_NODATACOW) {
 149                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 150                 if (S_ISREG(inode->i_mode))
 151                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 152         }
 153
 154         btrfs_update_iflags(inode);
 155 }
 156
 157 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 158 {
 159         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 160         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 161
 162         if (copy_to_user(arg, &flags, sizeof(flags)))
 163                 return -EFAULT;
 164         return 0;
 165 }
 166
 167 static int check_flags(unsigned int flags)
 168 {
 169         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 170                       FS_NOATIME_FL | FS_NODUMP_FL | \
 171                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 172                       FS_NOCOMP_FL | FS_COMPR_FL |
 173                       FS_NOCOW_FL))
 174                 return -EOPNOTSUPP;
 175
 176         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 177                 return -EINVAL;
 178
 179         return 0;
 180 }
 181
 182 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 183 {
 184         struct inode *inode = file_inode(file);
 185         struct btrfs_inode *ip = BTRFS_I(inode);
 186         struct btrfs_root *root = ip->root;
 187         struct btrfs_trans_handle *trans;
 188         unsigned int flags, oldflags;
 189         int ret;
 190         u64 ip_oldflags;
 191         unsigned int i_oldflags;
 192         umode_t mode;
 193
 194         if (btrfs_root_readonly(root))
 195                 return -EROFS;
 196
 197         if (copy_from_user(&flags, arg, sizeof(flags)))
 198                 return -EFAULT;
 199
 200         ret = check_flags(flags);
 201         if (ret)
 202                 return ret;
 203
 204         if (!inode_owner_or_capable(inode))
 205                 return -EACCES;
 206
 207         ret = mnt_want_write_file(file);
 208         if (ret)
 209                 return ret;
 210
 211         mutex_lock(&inode->i_mutex);
 212
 213         ip_oldflags = ip->flags;
 214         i_oldflags = inode->i_flags;
 215         mode = inode->i_mode;
 216
 217         flags = btrfs_mask_flags(inode->i_mode, flags);
 218         oldflags = btrfs_flags_to_ioctl(ip->flags);
 219         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 220                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 221                         ret = -EPERM;
 222                         goto out_unlock;
 223                 }
 224         }
 225
 226         if (flags & FS_SYNC_FL)
 227                 ip->flags |= BTRFS_INODE_SYNC;
 228         else
 229                 ip->flags &= ~BTRFS_INODE_SYNC;
 230         if (flags & FS_IMMUTABLE_FL)
 231                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 232         else
 233                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 234         if (flags & FS_APPEND_FL)
 235                 ip->flags |= BTRFS_INODE_APPEND;
 236         else
 237                 ip->flags &= ~BTRFS_INODE_APPEND;
 238         if (flags & FS_NODUMP_FL)
 239                 ip->flags |= BTRFS_INODE_NODUMP;
 240         else
 241                 ip->flags &= ~BTRFS_INODE_NODUMP;
 242         if (flags & FS_NOATIME_FL)
 243                 ip->flags |= BTRFS_INODE_NOATIME;
 244         else
 245                 ip->flags &= ~BTRFS_INODE_NOATIME;
 246         if (flags & FS_DIRSYNC_FL)
 247                 ip->flags |= BTRFS_INODE_DIRSYNC;
 248         else
 249                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 250         if (flags & FS_NOCOW_FL) {
 251                 if (S_ISREG(mode)) {
 252                         /*
 253                          * It's safe to turn csums off here, no extents exist.
 254                          * Otherwise we want the flag to reflect the real COW
 255                          * status of the file and will not set it.
 256                          */
 257                         if (inode->i_size == 0)
 258                                 ip->flags |= BTRFS_INODE_NODATACOW
 259                                            | BTRFS_INODE_NODATASUM;
 260                 } else {
 261                         ip->flags |= BTRFS_INODE_NODATACOW;
 262                 }
 263         } else {
 264                 /*
 265                  * Revert back under same assuptions as above
 266                  */
 267                 if (S_ISREG(mode)) {
 268                         if (inode->i_size == 0)
 269                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 270                                              | BTRFS_INODE_NODATASUM);
 271                 } else {
 272                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 273                 }
 274         }
 275
 276         /*
 277          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 278          * flag may be changed automatically if compression code won't make
 279          * things smaller.
 280          */
 281         if (flags & FS_NOCOMP_FL) {
 282                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 283                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 284         } else if (flags & FS_COMPR_FL) {
 285                 ip->flags |= BTRFS_INODE_COMPRESS;
 286                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 287         } else {
 288                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 289         }
 290
 291         trans = btrfs_start_transaction(root, 1);
 292         if (IS_ERR(trans)) {
 293                 ret = PTR_ERR(trans);
 294                 goto out_drop;
 295         }
 296
 297         btrfs_update_iflags(inode);
 298         inode_inc_iversion(inode);
 299         inode->i_ctime = CURRENT_TIME;
 300         ret = btrfs_update_inode(trans, root, inode);
 301
 302         btrfs_end_transaction(trans, root);
 303  out_drop:
 304         if (ret) {
 305                 ip->flags = ip_oldflags;
 306                 inode->i_flags = i_oldflags;
 307         }
 308
 309  out_unlock:
 310         mutex_unlock(&inode->i_mutex);
 311         mnt_drop_write_file(file);
 312         return ret;
 313 }
 314
 315 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 316 {
 317         struct inode *inode = file_inode(file);
 318
 319         return put_user(inode->i_generation, arg);
 320 }
 321
 322 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 323 {
 324         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
 325         struct btrfs_device *device;
 326         struct request_queue *q;
 327         struct fstrim_range range;
 328         u64 minlen = ULLONG_MAX;
 329         u64 num_devices = 0;
 330         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 331         int ret;
 332
 333         if (!capable(CAP_SYS_ADMIN))
 334                 return -EPERM;
 335
 336         rcu_read_lock();
 337         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 338                                 dev_list) {
 339                 if (!device->bdev)
 340                         continue;
 341                 q = bdev_get_queue(device->bdev);
 342                 if (blk_queue_discard(q)) {
 343                         num_devices++;
 344                         minlen = min((u64)q->limits.discard_granularity,
 345                                      minlen);
 346                 }
 347         }
 348         rcu_read_unlock();
 349
 350         if (!num_devices)
 351                 return -EOPNOTSUPP;
 352         if (copy_from_user(&range, arg, sizeof(range)))
 353                 return -EFAULT;
 354         if (range.start > total_bytes ||
 355             range.len < fs_info->sb->s_blocksize)
 356                 return -EINVAL;
 357
 358         range.len = min(range.len, total_bytes - range.start);
 359         range.minlen = max(range.minlen, minlen);
 360         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 361         if (ret < 0)
 362                 return ret;
 363
 364         if (copy_to_user(arg, &range, sizeof(range)))
 365                 return -EFAULT;
 366
 367         return 0;
 368 }
 369
 370 int btrfs_is_empty_uuid(u8 *uuid)
 371 {
 372         static char empty_uuid[BTRFS_UUID_SIZE] = {0};
 373
 374         return !memcmp(uuid, empty_uuid, BTRFS_UUID_SIZE);
 375 }
 376
 377 static noinline int create_subvol(struct inode *dir,
 378                                   struct dentry *dentry,
 379                                   char *name, int namelen,
 380                                   u64 *async_transid,
 381                                   struct btrfs_qgroup_inherit *inherit)
 382 {
 383         struct btrfs_trans_handle *trans;
 384         struct btrfs_key key;
 385         struct btrfs_root_item root_item;
 386         struct btrfs_inode_item *inode_item;
 387         struct extent_buffer *leaf;
 388         struct btrfs_root *root = BTRFS_I(dir)->root;
 389         struct btrfs_root *new_root;
 390         struct btrfs_block_rsv block_rsv;
 391         struct timespec cur_time = CURRENT_TIME;
 392         int ret;
 393         int err;
 394         u64 objectid;
 395         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 396         u64 index = 0;
 397         u64 qgroup_reserved;
 398         uuid_le new_uuid;
 399
 400         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 401         if (ret)
 402                 return ret;
 403
 404         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 405         /*
 406          * The same as the snapshot creation, please see the comment
 407          * of create_snapshot().
 408          */
 409         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 410                                                8, &qgroup_reserved, false);
 411         if (ret)
 412                 return ret;
 413
 414         trans = btrfs_start_transaction(root, 0);
 415         if (IS_ERR(trans)) {
 416                 ret = PTR_ERR(trans);
 417                 goto out;
 418         }
 419         trans->block_rsv = &block_rsv;
 420         trans->bytes_reserved = block_rsv.size;
 421
 422         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 423         if (ret)
 424                 goto fail;
 425
 426         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 427                                       0, objectid, NULL, 0, 0, 0);
 428         if (IS_ERR(leaf)) {
 429                 ret = PTR_ERR(leaf);
 430                 goto fail;
 431         }
 432
 433         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 434         btrfs_set_header_bytenr(leaf, leaf->start);
 435         btrfs_set_header_generation(leaf, trans->transid);
 436         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 437         btrfs_set_header_owner(leaf, objectid);
 438
 439         write_extent_buffer(leaf, root->fs_info->fsid,
 440                             (unsigned long)btrfs_header_fsid(leaf),
 441                             BTRFS_FSID_SIZE);
 442         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 443                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 444                             BTRFS_UUID_SIZE);
 445         btrfs_mark_buffer_dirty(leaf);
 446
 447         memset(&root_item, 0, sizeof(root_item));
 448
 449         inode_item = &root_item.inode;
 450         btrfs_set_stack_inode_generation(inode_item, 1);
 451         btrfs_set_stack_inode_size(inode_item, 3);
 452         btrfs_set_stack_inode_nlink(inode_item, 1);
 453         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
 454         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 455
 456         btrfs_set_root_flags(&root_item, 0);
 457         btrfs_set_root_limit(&root_item, 0);
 458         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 459
 460         btrfs_set_root_bytenr(&root_item, leaf->start);
 461         btrfs_set_root_generation(&root_item, trans->transid);
 462         btrfs_set_root_level(&root_item, 0);
 463         btrfs_set_root_refs(&root_item, 1);
 464         btrfs_set_root_used(&root_item, leaf->len);
 465         btrfs_set_root_last_snapshot(&root_item, 0);
 466
 467         btrfs_set_root_generation_v2(&root_item,
 468                         btrfs_root_generation(&root_item));
 469         uuid_le_gen(&new_uuid);
 470         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 471         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 472         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 473         root_item.ctime = root_item.otime;
 474         btrfs_set_root_ctransid(&root_item, trans->transid);
 475         btrfs_set_root_otransid(&root_item, trans->transid);
 476
 477         btrfs_tree_unlock(leaf);
 478         free_extent_buffer(leaf);
 479         leaf = NULL;
 480
 481         btrfs_set_root_dirid(&root_item, new_dirid);
 482
 483         key.objectid = objectid;
 484         key.offset = 0;
 485         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 486         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 487                                 &root_item);
 488         if (ret)
 489                 goto fail;
 490
 491         key.offset = (u64)-1;
 492         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 493         if (IS_ERR(new_root)) {
 494                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 495                 ret = PTR_ERR(new_root);
 496                 goto fail;
 497         }
 498
 499         btrfs_record_root_in_trans(trans, new_root);
 500
 501         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
 502         if (ret) {
 503                 /* We potentially lose an unused inode item here */
 504                 btrfs_abort_transaction(trans, root, ret);
 505                 goto fail;
 506         }
 507
 508         /*
 509          * insert the directory item
 510          */
 511         ret = btrfs_set_inode_index(dir, &index);
 512         if (ret) {
 513                 btrfs_abort_transaction(trans, root, ret);
 514                 goto fail;
 515         }
 516
 517         ret = btrfs_insert_dir_item(trans, root,
 518                                     name, namelen, dir, &key,
 519                                     BTRFS_FT_DIR, index);
 520         if (ret) {
 521                 btrfs_abort_transaction(trans, root, ret);
 522                 goto fail;
 523         }
 524
 525         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 526         ret = btrfs_update_inode(trans, root, dir);
 527         BUG_ON(ret);
 528
 529         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 530                                  objectid, root->root_key.objectid,
 531                                  btrfs_ino(dir), index, name, namelen);
 532         BUG_ON(ret);
 533
 534         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 535                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 536                                   objectid);
 537         if (ret)
 538                 btrfs_abort_transaction(trans, root, ret);
 539
 540 fail:
 541         trans->block_rsv = NULL;
 542         trans->bytes_reserved = 0;
 543         if (async_transid) {
 544                 *async_transid = trans->transid;
 545                 err = btrfs_commit_transaction_async(trans, root, 1);
 546                 if (err)
 547                         err = btrfs_commit_transaction(trans, root);
 548         } else {
 549                 err = btrfs_commit_transaction(trans, root);
 550         }
 551         if (err && !ret)
 552                 ret = err;
 553
 554         if (!ret)
 555                 d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 556 out:
 557         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 558         return ret;
 559 }
 560
 561 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 562                            struct dentry *dentry, char *name, int namelen,
 563                            u64 *async_transid, bool readonly,
 564                            struct btrfs_qgroup_inherit *inherit)
 565 {
 566         struct inode *inode;
 567         struct btrfs_pending_snapshot *pending_snapshot;
 568         struct btrfs_trans_handle *trans;
 569         int ret;
 570
 571         if (!root->ref_cows)
 572                 return -EINVAL;
 573
 574         ret = btrfs_start_delalloc_inodes(root, 0);
 575         if (ret)
 576                 return ret;
 577
 578         btrfs_wait_ordered_extents(root, 0);
 579
 580         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 581         if (!pending_snapshot)
 582                 return -ENOMEM;
 583
 584         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 585                              BTRFS_BLOCK_RSV_TEMP);
 586         /*
 587          * 1 - parent dir inode
 588          * 2 - dir entries
 589          * 1 - root item
 590          * 2 - root ref/backref
 591          * 1 - root of snapshot
 592          * 1 - UUID item
 593          */
 594         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 595                                         &pending_snapshot->block_rsv, 8,
 596                                         &pending_snapshot->qgroup_reserved,
 597                                         false);
 598         if (ret)
 599                 goto out;
 600
 601         pending_snapshot->dentry = dentry;
 602         pending_snapshot->root = root;
 603         pending_snapshot->readonly = readonly;
 604         pending_snapshot->dir = dir;
 605         pending_snapshot->inherit = inherit;
 606
 607         trans = btrfs_start_transaction(root, 0);
 608         if (IS_ERR(trans)) {
 609                 ret = PTR_ERR(trans);
 610                 goto fail;
 611         }
 612
 613         spin_lock(&root->fs_info->trans_lock);
 614         list_add(&pending_snapshot->list,
 615                  &trans->transaction->pending_snapshots);
 616         spin_unlock(&root->fs_info->trans_lock);
 617         if (async_transid) {
 618                 *async_transid = trans->transid;
 619                 ret = btrfs_commit_transaction_async(trans,
 620                                      root->fs_info->extent_root, 1);
 621                 if (ret)
 622                         ret = btrfs_commit_transaction(trans, root);
 623         } else {
 624                 ret = btrfs_commit_transaction(trans,
 625                                                root->fs_info->extent_root);
 626         }
 627         if (ret)
 628                 goto fail;
 629
 630         ret = pending_snapshot->error;
 631         if (ret)
 632                 goto fail;
 633
 634         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 635         if (ret)
 636                 goto fail;
 637
 638         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 639         if (IS_ERR(inode)) {
 640                 ret = PTR_ERR(inode);
 641                 goto fail;
 642         }
 643         BUG_ON(!inode);
 644         d_instantiate(dentry, inode);
 645         ret = 0;
 646 fail:
 647         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 648                                          &pending_snapshot->block_rsv,
 649                                          pending_snapshot->qgroup_reserved);
 650 out:
 651         kfree(pending_snapshot);
 652         return ret;
 653 }
 654
 655 /*  copy of check_sticky in fs/namei.c()
 656 * It's inline, so penalty for filesystems that don't use sticky bit is
 657 * minimal.
 658 */
 659 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 660 {
 661         kuid_t fsuid = current_fsuid();
 662
 663         if (!(dir->i_mode & S_ISVTX))
 664                 return 0;
 665         if (uid_eq(inode->i_uid, fsuid))
 666                 return 0;
 667         if (uid_eq(dir->i_uid, fsuid))
 668                 return 0;
 669         return !capable(CAP_FOWNER);
 670 }
 671
 672 /*  copy of may_delete in fs/namei.c()
 673  *      Check whether we can remove a link victim from directory dir, check
 674  *  whether the type of victim is right.
 675  *  1. We can't do it if dir is read-only (done in permission())
 676  *  2. We should have write and exec permissions on dir
 677  *  3. We can't remove anything from append-only dir
 678  *  4. We can't do anything with immutable dir (done in permission())
 679  *  5. If the sticky bit on dir is set we should either
 680  *      a. be owner of dir, or
 681  *      b. be owner of victim, or
 682  *      c. have CAP_FOWNER capability
 683  *  6. If the victim is append-only or immutable we can't do antyhing with
 684  *     links pointing to it.
 685  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 686  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 687  *  9. We can't remove a root or mountpoint.
 688  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 689  *     nfs_async_unlink().
 690  */
 691
 692 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 693 {
 694         int error;
 695
 696         if (!victim->d_inode)
 697                 return -ENOENT;
 698
 699         BUG_ON(victim->d_parent->d_inode != dir);
 700         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 701
 702         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 703         if (error)
 704                 return error;
 705         if (IS_APPEND(dir))
 706                 return -EPERM;
 707         if (btrfs_check_sticky(dir, victim->d_inode)||
 708                 IS_APPEND(victim->d_inode)||
 709             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 710                 return -EPERM;
 711         if (isdir) {
 712                 if (!S_ISDIR(victim->d_inode->i_mode))
 713                         return -ENOTDIR;
 714                 if (IS_ROOT(victim))
 715                         return -EBUSY;
 716         } else if (S_ISDIR(victim->d_inode->i_mode))
 717                 return -EISDIR;
 718         if (IS_DEADDIR(dir))
 719                 return -ENOENT;
 720         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 721                 return -EBUSY;
 722         return 0;
 723 }
 724
 725 /* copy of may_create in fs/namei.c() */
 726 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 727 {
 728         if (child->d_inode)
 729                 return -EEXIST;
 730         if (IS_DEADDIR(dir))
 731                 return -ENOENT;
 732         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 733 }
 734
 735 /*
 736  * Create a new subvolume below @parent.  This is largely modeled after
 737  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 738  * inside this filesystem so it's quite a bit simpler.
 739  */
 740 static noinline int btrfs_mksubvol(struct path *parent,
 741                                    char *name, int namelen,
 742                                    struct btrfs_root *snap_src,
 743                                    u64 *async_transid, bool readonly,
 744                                    struct btrfs_qgroup_inherit *inherit)
 745 {
 746         struct inode *dir  = parent->dentry->d_inode;
 747         struct dentry *dentry;
 748         int error;
 749
 750         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 751         if (error == -EINTR)
 752                 return error;
 753
 754         dentry = lookup_one_len(name, parent->dentry, namelen);
 755         error = PTR_ERR(dentry);
 756         if (IS_ERR(dentry))
 757                 goto out_unlock;
 758
 759         error = -EEXIST;
 760         if (dentry->d_inode)
 761                 goto out_dput;
 762
 763         error = btrfs_may_create(dir, dentry);
 764         if (error)
 765                 goto out_dput;
 766
 767         /*
 768          * even if this name doesn't exist, we may get hash collisions.
 769          * check for them now when we can safely fail
 770          */
 771         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 772                                                dir->i_ino, name,
 773                                                namelen);
 774         if (error)
 775                 goto out_dput;
 776
 777         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 778
 779         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 780                 goto out_up_read;
 781
 782         if (snap_src) {
 783                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 784                                         async_transid, readonly, inherit);
 785         } else {
 786                 error = create_subvol(dir, dentry, name, namelen,
 787                                       async_transid, inherit);
 788         }
 789         if (!error)
 790                 fsnotify_mkdir(dir, dentry);
 791 out_up_read:
 792         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 793 out_dput:
 794         dput(dentry);
 795 out_unlock:
 796         mutex_unlock(&dir->i_mutex);
 797         return error;
 798 }
 799
 800 /*
 801  * When we're defragging a range, we don't want to kick it off again
 802  * if it is really just waiting for delalloc to send it down.
 803  * If we find a nice big extent or delalloc range for the bytes in the
 804  * file you want to defrag, we return 0 to let you know to skip this
 805  * part of the file
 806  */
 807 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 808 {
 809         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 810         struct extent_map *em = NULL;
 811         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 812         u64 end;
 813
 814         read_lock(&em_tree->lock);
 815         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 816         read_unlock(&em_tree->lock);
 817
 818         if (em) {
 819                 end = extent_map_end(em);
 820                 free_extent_map(em);
 821                 if (end - offset > thresh)
 822                         return 0;
 823         }
 824         /* if we already have a nice delalloc here, just stop */
 825         thresh /= 2;
 826         end = count_range_bits(io_tree, &offset, offset + thresh,
 827                                thresh, EXTENT_DELALLOC, 1);
 828         if (end >= thresh)
 829                 return 0;
 830         return 1;
 831 }
 832
 833 /*
 834  * helper function to walk through a file and find extents
 835  * newer than a specific transid, and smaller than thresh.
 836  *
 837  * This is used by the defragging code to find new and small
 838  * extents
 839  */
 840 static int find_new_extents(struct btrfs_root *root,
 841                             struct inode *inode, u64 newer_than,
 842                             u64 *off, int thresh)
 843 {
 844         struct btrfs_path *path;
 845         struct btrfs_key min_key;
 846         struct btrfs_key max_key;
 847         struct extent_buffer *leaf;
 848         struct btrfs_file_extent_item *extent;
 849         int type;
 850         int ret;
 851         u64 ino = btrfs_ino(inode);
 852
 853         path = btrfs_alloc_path();
 854         if (!path)
 855                 return -ENOMEM;
 856
 857         min_key.objectid = ino;
 858         min_key.type = BTRFS_EXTENT_DATA_KEY;
 859         min_key.offset = *off;
 860
 861         max_key.objectid = ino;
 862         max_key.type = (u8)-1;
 863         max_key.offset = (u64)-1;
 864
 865         path->keep_locks = 1;
 866
 867         while(1) {
 868                 ret = btrfs_search_forward(root, &min_key, &max_key,
 869                                            path, newer_than);
 870                 if (ret != 0)
 871                         goto none;
 872                 if (min_key.objectid != ino)
 873                         goto none;
 874                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 875                         goto none;
 876
 877                 leaf = path->nodes[0];
 878                 extent = btrfs_item_ptr(leaf, path->slots[0],
 879                                         struct btrfs_file_extent_item);
 880
 881                 type = btrfs_file_extent_type(leaf, extent);
 882                 if (type == BTRFS_FILE_EXTENT_REG &&
 883                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 884                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 885                         *off = min_key.offset;
 886                         btrfs_free_path(path);
 887                         return 0;
 888                 }
 889
 890                 if (min_key.offset == (u64)-1)
 891                         goto none;
 892
 893                 min_key.offset++;
 894                 btrfs_release_path(path);
 895         }
 896 none:
 897         btrfs_free_path(path);
 898         return -ENOENT;
 899 }
 900
 901 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 902 {
 903         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 904         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 905         struct extent_map *em;
 906         u64 len = PAGE_CACHE_SIZE;
 907
 908         /*
 909          * hopefully we have this extent in the tree already, try without
 910          * the full extent lock
 911          */
 912         read_lock(&em_tree->lock);
 913         em = lookup_extent_mapping(em_tree, start, len);
 914         read_unlock(&em_tree->lock);
 915
 916         if (!em) {
 917                 /* get the big lock and read metadata off disk */
 918                 lock_extent(io_tree, start, start + len - 1);
 919                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 920                 unlock_extent(io_tree, start, start + len - 1);
 921
 922                 if (IS_ERR(em))
 923                         return NULL;
 924         }
 925
 926         return em;
 927 }
 928
 929 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 930 {
 931         struct extent_map *next;
 932         bool ret = true;
 933
 934         /* this is the last extent */
 935         if (em->start + em->len >= i_size_read(inode))
 936                 return false;
 937
 938         next = defrag_lookup_extent(inode, em->start + em->len);
 939         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 940                 ret = false;
 941
 942         free_extent_map(next);
 943         return ret;
 944 }
 945
 946 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
 947                                u64 *last_len, u64 *skip, u64 *defrag_end,
 948                                int compress)
 949 {
 950         struct extent_map *em;
 951         int ret = 1;
 952         bool next_mergeable = true;
 953
 954         /*
 955          * make sure that once we start defragging an extent, we keep on
 956          * defragging it
 957          */
 958         if (start < *defrag_end)
 959                 return 1;
 960
 961         *skip = 0;
 962
 963         em = defrag_lookup_extent(inode, start);
 964         if (!em)
 965                 return 0;
 966
 967         /* this will cover holes, and inline extents */
 968         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 969                 ret = 0;
 970                 goto out;
 971         }
 972
 973         next_mergeable = defrag_check_next_extent(inode, em);
 974
 975         /*
 976          * we hit a real extent, if it is big or the next extent is not a
 977          * real extent, don't bother defragging it
 978          */
 979         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
 980             (em->len >= thresh || !next_mergeable))
 981                 ret = 0;
 982 out:
 983         /*
 984          * last_len ends up being a counter of how many bytes we've defragged.
 985          * every time we choose not to defrag an extent, we reset *last_len
 986          * so that the next tiny extent will force a defrag.
 987          *
 988          * The end result of this is that tiny extents before a single big
 989          * extent will force at least part of that big extent to be defragged.
 990          */
 991         if (ret) {
 992                 *defrag_end = extent_map_end(em);
 993         } else {
 994                 *last_len = 0;
 995                 *skip = extent_map_end(em);
 996                 *defrag_end = 0;
 997         }
 998
 999         free_extent_map(em);
1000         return ret;
1001 }
1002
1003 /*
1004  * it doesn't do much good to defrag one or two pages
1005  * at a time.  This pulls in a nice chunk of pages
1006  * to COW and defrag.
1007  *
1008  * It also makes sure the delalloc code has enough
1009  * dirty data to avoid making new small extents as part
1010  * of the defrag
1011  *
1012  * It's a good idea to start RA on this range
1013  * before calling this.
1014  */
1015 static int cluster_pages_for_defrag(struct inode *inode,
1016                                     struct page **pages,
1017                                     unsigned long start_index,
1018                                     int num_pages)
1019 {
1020         unsigned long file_end;
1021         u64 isize = i_size_read(inode);
1022         u64 page_start;
1023         u64 page_end;
1024         u64 page_cnt;
1025         int ret;
1026         int i;
1027         int i_done;
1028         struct btrfs_ordered_extent *ordered;
1029         struct extent_state *cached_state = NULL;
1030         struct extent_io_tree *tree;
1031         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1032
1033         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1034         if (!isize || start_index > file_end)
1035                 return 0;
1036
1037         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1038
1039         ret = btrfs_delalloc_reserve_space(inode,
1040                                            page_cnt << PAGE_CACHE_SHIFT);
1041         if (ret)
1042                 return ret;
1043         i_done = 0;
1044         tree = &BTRFS_I(inode)->io_tree;
1045
1046         /* step one, lock all the pages */
1047         for (i = 0; i < page_cnt; i++) {
1048                 struct page *page;
1049 again:
1050                 page = find_or_create_page(inode->i_mapping,
1051                                            start_index + i, mask);
1052                 if (!page)
1053                         break;
1054
1055                 page_start = page_offset(page);
1056                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1057                 while (1) {
1058                         lock_extent(tree, page_start, page_end);
1059                         ordered = btrfs_lookup_ordered_extent(inode,
1060                                                               page_start);
1061                         unlock_extent(tree, page_start, page_end);
1062                         if (!ordered)
1063                                 break;
1064
1065                         unlock_page(page);
1066                         btrfs_start_ordered_extent(inode, ordered, 1);
1067                         btrfs_put_ordered_extent(ordered);
1068                         lock_page(page);
1069                         /*
1070                          * we unlocked the page above, so we need check if
1071                          * it was released or not.
1072                          */
1073                         if (page->mapping != inode->i_mapping) {
1074                                 unlock_page(page);
1075                                 page_cache_release(page);
1076                                 goto again;
1077                         }
1078                 }
1079
1080                 if (!PageUptodate(page)) {
1081                         btrfs_readpage(NULL, page);
1082                         lock_page(page);
1083                         if (!PageUptodate(page)) {
1084                                 unlock_page(page);
1085                                 page_cache_release(page);
1086                                 ret = -EIO;
1087                                 break;
1088                         }
1089                 }
1090
1091                 if (page->mapping != inode->i_mapping) {
1092                         unlock_page(page);
1093                         page_cache_release(page);
1094                         goto again;
1095                 }
1096
1097                 pages[i] = page;
1098                 i_done++;
1099         }
1100         if (!i_done || ret)
1101                 goto out;
1102
1103         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1104                 goto out;
1105
1106         /*
1107          * so now we have a nice long stream of locked
1108          * and up to date pages, lets wait on them
1109          */
1110         for (i = 0; i < i_done; i++)
1111                 wait_on_page_writeback(pages[i]);
1112
1113         page_start = page_offset(pages[0]);
1114         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1115
1116         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1117                          page_start, page_end - 1, 0, &cached_state);
1118         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1119                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1120                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1121                           &cached_state, GFP_NOFS);
1122
1123         if (i_done != page_cnt) {
1124                 spin_lock(&BTRFS_I(inode)->lock);
1125                 BTRFS_I(inode)->outstanding_extents++;
1126                 spin_unlock(&BTRFS_I(inode)->lock);
1127                 btrfs_delalloc_release_space(inode,
1128                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1129         }
1130
1131
1132         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1133                           &cached_state, GFP_NOFS);
1134
1135         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1136                              page_start, page_end - 1, &cached_state,
1137                              GFP_NOFS);
1138
1139         for (i = 0; i < i_done; i++) {
1140                 clear_page_dirty_for_io(pages[i]);
1141                 ClearPageChecked(pages[i]);
1142                 set_page_extent_mapped(pages[i]);
1143                 set_page_dirty(pages[i]);
1144                 unlock_page(pages[i]);
1145                 page_cache_release(pages[i]);
1146         }
1147         return i_done;
1148 out:
1149         for (i = 0; i < i_done; i++) {
1150                 unlock_page(pages[i]);
1151                 page_cache_release(pages[i]);
1152         }
1153         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1154         return ret;
1155
1156 }
1157
1158 int btrfs_defrag_file(struct inode *inode, struct file *file,
1159                       struct btrfs_ioctl_defrag_range_args *range,
1160                       u64 newer_than, unsigned long max_to_defrag)
1161 {
1162         struct btrfs_root *root = BTRFS_I(inode)->root;
1163         struct file_ra_state *ra = NULL;
1164         unsigned long last_index;
1165         u64 isize = i_size_read(inode);
1166         u64 last_len = 0;
1167         u64 skip = 0;
1168         u64 defrag_end = 0;
1169         u64 newer_off = range->start;
1170         unsigned long i;
1171         unsigned long ra_index = 0;
1172         int ret;
1173         int defrag_count = 0;
1174         int compress_type = BTRFS_COMPRESS_ZLIB;
1175         int extent_thresh = range->extent_thresh;
1176         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1177         int cluster = max_cluster;
1178         u64 new_align = ~((u64)128 * 1024 - 1);
1179         struct page **pages = NULL;
1180
1181         if (isize == 0)
1182                 return 0;
1183
1184         if (range->start >= isize)
1185                 return -EINVAL;
1186
1187         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1188                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1189                         return -EINVAL;
1190                 if (range->compress_type)
1191                         compress_type = range->compress_type;
1192         }
1193
1194         if (extent_thresh == 0)
1195                 extent_thresh = 256 * 1024;
1196
1197         /*
1198          * if we were not given a file, allocate a readahead
1199          * context
1200          */
1201         if (!file) {
1202                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1203                 if (!ra)
1204                         return -ENOMEM;
1205                 file_ra_state_init(ra, inode->i_mapping);
1206         } else {
1207                 ra = &file->f_ra;
1208         }
1209
1210         pages = kmalloc(sizeof(struct page *) * max_cluster,
1211                         GFP_NOFS);
1212         if (!pages) {
1213                 ret = -ENOMEM;
1214                 goto out_ra;
1215         }
1216
1217         /* find the last page to defrag */
1218         if (range->start + range->len > range->start) {
1219                 last_index = min_t(u64, isize - 1,
1220                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1221         } else {
1222                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1223         }
1224
1225         if (newer_than) {
1226                 ret = find_new_extents(root, inode, newer_than,
1227                                        &newer_off, 64 * 1024);
1228                 if (!ret) {
1229                         range->start = newer_off;
1230                         /*
1231                          * we always align our defrag to help keep
1232                          * the extents in the file evenly spaced
1233                          */
1234                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1235                 } else
1236                         goto out_ra;
1237         } else {
1238                 i = range->start >> PAGE_CACHE_SHIFT;
1239         }
1240         if (!max_to_defrag)
1241                 max_to_defrag = last_index + 1;
1242
1243         /*
1244          * make writeback starts from i, so the defrag range can be
1245          * written sequentially.
1246          */
1247         if (i < inode->i_mapping->writeback_index)
1248                 inode->i_mapping->writeback_index = i;
1249
1250         while (i <= last_index && defrag_count < max_to_defrag &&
1251                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1252                 PAGE_CACHE_SHIFT)) {
1253                 /*
1254                  * make sure we stop running if someone unmounts
1255                  * the FS
1256                  */
1257                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1258                         break;
1259
1260                 if (btrfs_defrag_cancelled(root->fs_info)) {
1261                         printk(KERN_DEBUG "btrfs: defrag_file cancelled\n");
1262                         ret = -EAGAIN;
1263                         break;
1264                 }
1265
1266                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1267                                          extent_thresh, &last_len, &skip,
1268                                          &defrag_end, range->flags &
1269                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1270                         unsigned long next;
1271                         /*
1272                          * the should_defrag function tells us how much to skip
1273                          * bump our counter by the suggested amount
1274                          */
1275                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1276                         i = max(i + 1, next);
1277                         continue;
1278                 }
1279
1280                 if (!newer_than) {
1281                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1282                                    PAGE_CACHE_SHIFT) - i;
1283                         cluster = min(cluster, max_cluster);
1284                 } else {
1285                         cluster = max_cluster;
1286                 }
1287
1288                 if (i + cluster > ra_index) {
1289                         ra_index = max(i, ra_index);
1290                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1291                                        cluster);
1292                         ra_index += max_cluster;
1293                 }
1294
1295                 mutex_lock(&inode->i_mutex);
1296                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1297                         BTRFS_I(inode)->force_compress = compress_type;
1298                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1299                 if (ret < 0) {
1300                         mutex_unlock(&inode->i_mutex);
1301                         goto out_ra;
1302                 }
1303
1304                 defrag_count += ret;
1305                 balance_dirty_pages_ratelimited(inode->i_mapping);
1306                 mutex_unlock(&inode->i_mutex);
1307
1308                 if (newer_than) {
1309                         if (newer_off == (u64)-1)
1310                                 break;
1311
1312                         if (ret > 0)
1313                                 i += ret;
1314
1315                         newer_off = max(newer_off + 1,
1316                                         (u64)i << PAGE_CACHE_SHIFT);
1317
1318                         ret = find_new_extents(root, inode,
1319                                                newer_than, &newer_off,
1320                                                64 * 1024);
1321                         if (!ret) {
1322                                 range->start = newer_off;
1323                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1324                         } else {
1325                                 break;
1326                         }
1327                 } else {
1328                         if (ret > 0) {
1329                                 i += ret;
1330                                 last_len += ret << PAGE_CACHE_SHIFT;
1331                         } else {
1332                                 i++;
1333                                 last_len = 0;
1334                         }
1335                 }
1336         }
1337
1338         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1339                 filemap_flush(inode->i_mapping);
1340
1341         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1342                 /* the filemap_flush will queue IO into the worker threads, but
1343                  * we have to make sure the IO is actually started and that
1344                  * ordered extents get created before we return
1345                  */
1346                 atomic_inc(&root->fs_info->async_submit_draining);
1347                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1348                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1349                         wait_event(root->fs_info->async_submit_wait,
1350                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1351                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1352                 }
1353                 atomic_dec(&root->fs_info->async_submit_draining);
1354         }
1355
1356         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1357                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1358         }
1359
1360         ret = defrag_count;
1361
1362 out_ra:
1363         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1364                 mutex_lock(&inode->i_mutex);
1365                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1366                 mutex_unlock(&inode->i_mutex);
1367         }
1368         if (!file)
1369                 kfree(ra);
1370         kfree(pages);
1371         return ret;
1372 }
1373
1374 static noinline int btrfs_ioctl_resize(struct file *file,
1375                                         void __user *arg)
1376 {
1377         u64 new_size;
1378         u64 old_size;
1379         u64 devid = 1;
1380         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1381         struct btrfs_ioctl_vol_args *vol_args;
1382         struct btrfs_trans_handle *trans;
1383         struct btrfs_device *device = NULL;
1384         char *sizestr;
1385         char *devstr = NULL;
1386         int ret = 0;
1387         int mod = 0;
1388
1389         if (!capable(CAP_SYS_ADMIN))
1390                 return -EPERM;
1391
1392         ret = mnt_want_write_file(file);
1393         if (ret)
1394                 return ret;
1395
1396         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1397                         1)) {
1398                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
1399                 mnt_drop_write_file(file);
1400                 return -EINVAL;
1401         }
1402
1403         mutex_lock(&root->fs_info->volume_mutex);
1404         vol_args = memdup_user(arg, sizeof(*vol_args));
1405         if (IS_ERR(vol_args)) {
1406                 ret = PTR_ERR(vol_args);
1407                 goto out;
1408         }
1409
1410         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1411
1412         sizestr = vol_args->name;
1413         devstr = strchr(sizestr, ':');
1414         if (devstr) {
1415                 char *end;
1416                 sizestr = devstr + 1;
1417                 *devstr = '\0';
1418                 devstr = vol_args->name;
1419                 devid = simple_strtoull(devstr, &end, 10);
1420                 if (!devid) {
1421                         ret = -EINVAL;
1422                         goto out_free;
1423                 }
1424                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1425                        (unsigned long long)devid);
1426         }
1427
1428         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1429         if (!device) {
1430                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1431                        (unsigned long long)devid);
1432                 ret = -ENODEV;
1433                 goto out_free;
1434         }
1435
1436         if (!device->writeable) {
1437                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1438                        "readonly device %llu\n",
1439                        (unsigned long long)devid);
1440                 ret = -EPERM;
1441                 goto out_free;
1442         }
1443
1444         if (!strcmp(sizestr, "max"))
1445                 new_size = device->bdev->bd_inode->i_size;
1446         else {
1447                 if (sizestr[0] == '-') {
1448                         mod = -1;
1449                         sizestr++;
1450                 } else if (sizestr[0] == '+') {
1451                         mod = 1;
1452                         sizestr++;
1453                 }
1454                 new_size = memparse(sizestr, NULL);
1455                 if (new_size == 0) {
1456                         ret = -EINVAL;
1457                         goto out_free;
1458                 }
1459         }
1460
1461         if (device->is_tgtdev_for_dev_replace) {
1462                 ret = -EPERM;
1463                 goto out_free;
1464         }
1465
1466         old_size = device->total_bytes;
1467
1468         if (mod < 0) {
1469                 if (new_size > old_size) {
1470                         ret = -EINVAL;
1471                         goto out_free;
1472                 }
1473                 new_size = old_size - new_size;
1474         } else if (mod > 0) {
1475                 new_size = old_size + new_size;
1476         }
1477
1478         if (new_size < 256 * 1024 * 1024) {
1479                 ret = -EINVAL;
1480                 goto out_free;
1481         }
1482         if (new_size > device->bdev->bd_inode->i_size) {
1483                 ret = -EFBIG;
1484                 goto out_free;
1485         }
1486
1487         do_div(new_size, root->sectorsize);
1488         new_size *= root->sectorsize;
1489
1490         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1491                       rcu_str_deref(device->name),
1492                       (unsigned long long)new_size);
1493
1494         if (new_size > old_size) {
1495                 trans = btrfs_start_transaction(root, 0);
1496                 if (IS_ERR(trans)) {
1497                         ret = PTR_ERR(trans);
1498                         goto out_free;
1499                 }
1500                 ret = btrfs_grow_device(trans, device, new_size);
1501                 btrfs_commit_transaction(trans, root);
1502         } else if (new_size < old_size) {
1503                 ret = btrfs_shrink_device(device, new_size);
1504         } /* equal, nothing need to do */
1505
1506 out_free:
1507         kfree(vol_args);
1508 out:
1509         mutex_unlock(&root->fs_info->volume_mutex);
1510         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1511         mnt_drop_write_file(file);
1512         return ret;
1513 }
1514
1515 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1516                                 char *name, unsigned long fd, int subvol,
1517                                 u64 *transid, bool readonly,
1518                                 struct btrfs_qgroup_inherit *inherit)
1519 {
1520         int namelen;
1521         int ret = 0;
1522
1523         ret = mnt_want_write_file(file);
1524         if (ret)
1525                 goto out;
1526
1527         namelen = strlen(name);
1528         if (strchr(name, '/')) {
1529                 ret = -EINVAL;
1530                 goto out_drop_write;
1531         }
1532
1533         if (name[0] == '.' &&
1534            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1535                 ret = -EEXIST;
1536                 goto out_drop_write;
1537         }
1538
1539         if (subvol) {
1540                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1541                                      NULL, transid, readonly, inherit);
1542         } else {
1543                 struct fd src = fdget(fd);
1544                 struct inode *src_inode;
1545                 if (!src.file) {
1546                         ret = -EINVAL;
1547                         goto out_drop_write;
1548                 }
1549
1550                 src_inode = file_inode(src.file);
1551                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1552                         printk(KERN_INFO "btrfs: Snapshot src from "
1553                                "another FS\n");
1554                         ret = -EINVAL;
1555                 } else {
1556                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1557                                              BTRFS_I(src_inode)->root,
1558                                              transid, readonly, inherit);
1559                 }
1560                 fdput(src);
1561         }
1562 out_drop_write:
1563         mnt_drop_write_file(file);
1564 out:
1565         return ret;
1566 }
1567
1568 static noinline int btrfs_ioctl_snap_create(struct file *file,
1569                                             void __user *arg, int subvol)
1570 {
1571         struct btrfs_ioctl_vol_args *vol_args;
1572         int ret;
1573
1574         vol_args = memdup_user(arg, sizeof(*vol_args));
1575         if (IS_ERR(vol_args))
1576                 return PTR_ERR(vol_args);
1577         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1578
1579         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1580                                               vol_args->fd, subvol,
1581                                               NULL, false, NULL);
1582
1583         kfree(vol_args);
1584         return ret;
1585 }
1586
1587 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1588                                                void __user *arg, int subvol)
1589 {
1590         struct btrfs_ioctl_vol_args_v2 *vol_args;
1591         int ret;
1592         u64 transid = 0;
1593         u64 *ptr = NULL;
1594         bool readonly = false;
1595         struct btrfs_qgroup_inherit *inherit = NULL;
1596
1597         vol_args = memdup_user(arg, sizeof(*vol_args));
1598         if (IS_ERR(vol_args))
1599                 return PTR_ERR(vol_args);
1600         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1601
1602         if (vol_args->flags &
1603             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1604               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1605                 ret = -EOPNOTSUPP;
1606                 goto out;
1607         }
1608
1609         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1610                 ptr = &transid;
1611         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1612                 readonly = true;
1613         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1614                 if (vol_args->size > PAGE_CACHE_SIZE) {
1615                         ret = -EINVAL;
1616                         goto out;
1617                 }
1618                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1619                 if (IS_ERR(inherit)) {
1620                         ret = PTR_ERR(inherit);
1621                         goto out;
1622                 }
1623         }
1624
1625         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1626                                               vol_args->fd, subvol, ptr,
1627                                               readonly, inherit);
1628
1629         if (ret == 0 && ptr &&
1630             copy_to_user(arg +
1631                          offsetof(struct btrfs_ioctl_vol_args_v2,
1632                                   transid), ptr, sizeof(*ptr)))
1633                 ret = -EFAULT;
1634 out:
1635         kfree(vol_args);
1636         kfree(inherit);
1637         return ret;
1638 }
1639
1640 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1641                                                 void __user *arg)
1642 {
1643         struct inode *inode = file_inode(file);
1644         struct btrfs_root *root = BTRFS_I(inode)->root;
1645         int ret = 0;
1646         u64 flags = 0;
1647
1648         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1649                 return -EINVAL;
1650
1651         down_read(&root->fs_info->subvol_sem);
1652         if (btrfs_root_readonly(root))
1653                 flags |= BTRFS_SUBVOL_RDONLY;
1654         up_read(&root->fs_info->subvol_sem);
1655
1656         if (copy_to_user(arg, &flags, sizeof(flags)))
1657                 ret = -EFAULT;
1658
1659         return ret;
1660 }
1661
1662 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1663                                               void __user *arg)
1664 {
1665         struct inode *inode = file_inode(file);
1666         struct btrfs_root *root = BTRFS_I(inode)->root;
1667         struct btrfs_trans_handle *trans;
1668         u64 root_flags;
1669         u64 flags;
1670         int ret = 0;
1671
1672         ret = mnt_want_write_file(file);
1673         if (ret)
1674                 goto out;
1675
1676         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1677                 ret = -EINVAL;
1678                 goto out_drop_write;
1679         }
1680
1681         if (copy_from_user(&flags, arg, sizeof(flags))) {
1682                 ret = -EFAULT;
1683                 goto out_drop_write;
1684         }
1685
1686         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1687                 ret = -EINVAL;
1688                 goto out_drop_write;
1689         }
1690
1691         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1692                 ret = -EOPNOTSUPP;
1693                 goto out_drop_write;
1694         }
1695
1696         if (!inode_owner_or_capable(inode)) {
1697                 ret = -EACCES;
1698                 goto out_drop_write;
1699         }
1700
1701         down_write(&root->fs_info->subvol_sem);
1702
1703         /* nothing to do */
1704         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1705                 goto out_drop_sem;
1706
1707         root_flags = btrfs_root_flags(&root->root_item);
1708         if (flags & BTRFS_SUBVOL_RDONLY)
1709                 btrfs_set_root_flags(&root->root_item,
1710                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1711         else
1712                 btrfs_set_root_flags(&root->root_item,
1713                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1714
1715         trans = btrfs_start_transaction(root, 1);
1716         if (IS_ERR(trans)) {
1717                 ret = PTR_ERR(trans);
1718                 goto out_reset;
1719         }
1720
1721         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1722                                 &root->root_key, &root->root_item);
1723
1724         btrfs_commit_transaction(trans, root);
1725 out_reset:
1726         if (ret)
1727                 btrfs_set_root_flags(&root->root_item, root_flags);
1728 out_drop_sem:
1729         up_write(&root->fs_info->subvol_sem);
1730 out_drop_write:
1731         mnt_drop_write_file(file);
1732 out:
1733         return ret;
1734 }
1735
1736 /*
1737  * helper to check if the subvolume references other subvolumes
1738  */
1739 static noinline int may_destroy_subvol(struct btrfs_root *root)
1740 {
1741         struct btrfs_path *path;
1742         struct btrfs_dir_item *di;
1743         struct btrfs_key key;
1744         u64 dir_id;
1745         int ret;
1746
1747         path = btrfs_alloc_path();
1748         if (!path)
1749                 return -ENOMEM;
1750
1751         /* Make sure this root isn't set as the default subvol */
1752         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1753         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1754                                    dir_id, "default", 7, 0);
1755         if (di && !IS_ERR(di)) {
1756                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1757                 if (key.objectid == root->root_key.objectid) {
1758                         ret = -ENOTEMPTY;
1759                         goto out;
1760                 }
1761                 btrfs_release_path(path);
1762         }
1763
1764         key.objectid = root->root_key.objectid;
1765         key.type = BTRFS_ROOT_REF_KEY;
1766         key.offset = (u64)-1;
1767
1768         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1769                                 &key, path, 0, 0);
1770         if (ret < 0)
1771                 goto out;
1772         BUG_ON(ret == 0);
1773
1774         ret = 0;
1775         if (path->slots[0] > 0) {
1776                 path->slots[0]--;
1777                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1778                 if (key.objectid == root->root_key.objectid &&
1779                     key.type == BTRFS_ROOT_REF_KEY)
1780                         ret = -ENOTEMPTY;
1781         }
1782 out:
1783         btrfs_free_path(path);
1784         return ret;
1785 }
1786
1787 static noinline int key_in_sk(struct btrfs_key *key,
1788                               struct btrfs_ioctl_search_key *sk)
1789 {
1790         struct btrfs_key test;
1791         int ret;
1792
1793         test.objectid = sk->min_objectid;
1794         test.type = sk->min_type;
1795         test.offset = sk->min_offset;
1796
1797         ret = btrfs_comp_cpu_keys(key, &test);
1798         if (ret < 0)
1799                 return 0;
1800
1801         test.objectid = sk->max_objectid;
1802         test.type = sk->max_type;
1803         test.offset = sk->max_offset;
1804
1805         ret = btrfs_comp_cpu_keys(key, &test);
1806         if (ret > 0)
1807                 return 0;
1808         return 1;
1809 }
1810
1811 static noinline int copy_to_sk(struct btrfs_root *root,
1812                                struct btrfs_path *path,
1813                                struct btrfs_key *key,
1814                                struct btrfs_ioctl_search_key *sk,
1815                                char *buf,
1816                                unsigned long *sk_offset,
1817                                int *num_found)
1818 {
1819         u64 found_transid;
1820         struct extent_buffer *leaf;
1821         struct btrfs_ioctl_search_header sh;
1822         unsigned long item_off;
1823         unsigned long item_len;
1824         int nritems;
1825         int i;
1826         int slot;
1827         int ret = 0;
1828
1829         leaf = path->nodes[0];
1830         slot = path->slots[0];
1831         nritems = btrfs_header_nritems(leaf);
1832
1833         if (btrfs_header_generation(leaf) > sk->max_transid) {
1834                 i = nritems;
1835                 goto advance_key;
1836         }
1837         found_transid = btrfs_header_generation(leaf);
1838
1839         for (i = slot; i < nritems; i++) {
1840                 item_off = btrfs_item_ptr_offset(leaf, i);
1841                 item_len = btrfs_item_size_nr(leaf, i);
1842
1843                 btrfs_item_key_to_cpu(leaf, key, i);
1844                 if (!key_in_sk(key, sk))
1845                         continue;
1846
1847                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1848                         item_len = 0;
1849
1850                 if (sizeof(sh) + item_len + *sk_offset >
1851                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1852                         ret = 1;
1853                         goto overflow;
1854                 }
1855
1856                 sh.objectid = key->objectid;
1857                 sh.offset = key->offset;
1858                 sh.type = key->type;
1859                 sh.len = item_len;
1860                 sh.transid = found_transid;
1861
1862                 /* copy search result header */
1863                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1864                 *sk_offset += sizeof(sh);
1865
1866                 if (item_len) {
1867                         char *p = buf + *sk_offset;
1868                         /* copy the item */
1869                         read_extent_buffer(leaf, p,
1870                                            item_off, item_len);
1871                         *sk_offset += item_len;
1872                 }
1873                 (*num_found)++;
1874
1875                 if (*num_found >= sk->nr_items)
1876                         break;
1877         }
1878 advance_key:
1879         ret = 0;
1880         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1881                 key->offset++;
1882         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1883                 key->offset = 0;
1884                 key->type++;
1885         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1886                 key->offset = 0;
1887                 key->type = 0;
1888                 key->objectid++;
1889         } else
1890                 ret = 1;
1891 overflow:
1892         return ret;
1893 }
1894
1895 static noinline int search_ioctl(struct inode *inode,
1896                                  struct btrfs_ioctl_search_args *args)
1897 {
1898         struct btrfs_root *root;
1899         struct btrfs_key key;
1900         struct btrfs_key max_key;
1901         struct btrfs_path *path;
1902         struct btrfs_ioctl_search_key *sk = &args->key;
1903         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1904         int ret;
1905         int num_found = 0;
1906         unsigned long sk_offset = 0;
1907
1908         path = btrfs_alloc_path();
1909         if (!path)
1910                 return -ENOMEM;
1911
1912         if (sk->tree_id == 0) {
1913                 /* search the root of the inode that was passed */
1914                 root = BTRFS_I(inode)->root;
1915         } else {
1916                 key.objectid = sk->tree_id;
1917                 key.type = BTRFS_ROOT_ITEM_KEY;
1918                 key.offset = (u64)-1;
1919                 root = btrfs_read_fs_root_no_name(info, &key);
1920                 if (IS_ERR(root)) {
1921                         printk(KERN_ERR "could not find root %llu\n",
1922                                sk->tree_id);
1923                         btrfs_free_path(path);
1924                         return -ENOENT;
1925                 }
1926         }
1927
1928         key.objectid = sk->min_objectid;
1929         key.type = sk->min_type;
1930         key.offset = sk->min_offset;
1931
1932         max_key.objectid = sk->max_objectid;
1933         max_key.type = sk->max_type;
1934         max_key.offset = sk->max_offset;
1935
1936         path->keep_locks = 1;
1937
1938         while(1) {
1939                 ret = btrfs_search_forward(root, &key, &max_key, path,
1940                                            sk->min_transid);
1941                 if (ret != 0) {
1942                         if (ret > 0)
1943                                 ret = 0;
1944                         goto err;
1945                 }
1946                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1947                                  &sk_offset, &num_found);
1948                 btrfs_release_path(path);
1949                 if (ret || num_found >= sk->nr_items)
1950                         break;
1951
1952         }
1953         ret = 0;
1954 err:
1955         sk->nr_items = num_found;
1956         btrfs_free_path(path);
1957         return ret;
1958 }
1959
1960 static noinline int btrfs_ioctl_tree_search(struct file *file,
1961                                            void __user *argp)
1962 {
1963          struct btrfs_ioctl_search_args *args;
1964          struct inode *inode;
1965          int ret;
1966
1967         if (!capable(CAP_SYS_ADMIN))
1968                 return -EPERM;
1969
1970         args = memdup_user(argp, sizeof(*args));
1971         if (IS_ERR(args))
1972                 return PTR_ERR(args);
1973
1974         inode = file_inode(file);
1975         ret = search_ioctl(inode, args);
1976         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1977                 ret = -EFAULT;
1978         kfree(args);
1979         return ret;
1980 }
1981
1982 /*
1983  * Search INODE_REFs to identify path name of 'dirid' directory
1984  * in a 'tree_id' tree. and sets path name to 'name'.
1985  */
1986 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1987                                 u64 tree_id, u64 dirid, char *name)
1988 {
1989         struct btrfs_root *root;
1990         struct btrfs_key key;
1991         char *ptr;
1992         int ret = -1;
1993         int slot;
1994         int len;
1995         int total_len = 0;
1996         struct btrfs_inode_ref *iref;
1997         struct extent_buffer *l;
1998         struct btrfs_path *path;
1999
2000         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2001                 name[0]='\0';
2002                 return 0;
2003         }
2004
2005         path = btrfs_alloc_path();
2006         if (!path)
2007                 return -ENOMEM;
2008
2009         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2010
2011         key.objectid = tree_id;
2012         key.type = BTRFS_ROOT_ITEM_KEY;
2013         key.offset = (u64)-1;
2014         root = btrfs_read_fs_root_no_name(info, &key);
2015         if (IS_ERR(root)) {
2016                 printk(KERN_ERR "could not find root %llu\n", tree_id);
2017                 ret = -ENOENT;
2018                 goto out;
2019         }
2020
2021         key.objectid = dirid;
2022         key.type = BTRFS_INODE_REF_KEY;
2023         key.offset = (u64)-1;
2024
2025         while(1) {
2026                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2027                 if (ret < 0)
2028                         goto out;
2029                 else if (ret > 0) {
2030                         ret = btrfs_previous_item(root, path, dirid,
2031                                                   BTRFS_INODE_REF_KEY);
2032                         if (ret < 0)
2033                                 goto out;
2034                         else if (ret > 0) {
2035                                 ret = -ENOENT;
2036                                 goto out;
2037                         }
2038                 }
2039
2040                 l = path->nodes[0];
2041                 slot = path->slots[0];
2042                 btrfs_item_key_to_cpu(l, &key, slot);
2043
2044                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2045                 len = btrfs_inode_ref_name_len(l, iref);
2046                 ptr -= len + 1;
2047                 total_len += len + 1;
2048                 if (ptr < name) {
2049                         ret = -ENAMETOOLONG;
2050                         goto out;
2051                 }
2052
2053                 *(ptr + len) = '/';
2054                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
2055
2056                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2057                         break;
2058
2059                 btrfs_release_path(path);
2060                 key.objectid = key.offset;
2061                 key.offset = (u64)-1;
2062                 dirid = key.objectid;
2063         }
2064         memmove(name, ptr, total_len);
2065         name[total_len]='\0';
2066         ret = 0;
2067 out:
2068         btrfs_free_path(path);
2069         return ret;
2070 }
2071
2072 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2073                                            void __user *argp)
2074 {
2075          struct btrfs_ioctl_ino_lookup_args *args;
2076          struct inode *inode;
2077          int ret;
2078
2079         if (!capable(CAP_SYS_ADMIN))
2080                 return -EPERM;
2081
2082         args = memdup_user(argp, sizeof(*args));
2083         if (IS_ERR(args))
2084                 return PTR_ERR(args);
2085
2086         inode = file_inode(file);
2087
2088         if (args->treeid == 0)
2089                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2090
2091         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2092                                         args->treeid, args->objectid,
2093                                         args->name);
2094
2095         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2096                 ret = -EFAULT;
2097
2098         kfree(args);
2099         return ret;
2100 }
2101
2102 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2103                                              void __user *arg)
2104 {
2105         struct dentry *parent = fdentry(file);
2106         struct dentry *dentry;
2107         struct inode *dir = parent->d_inode;
2108         struct inode *inode;
2109         struct btrfs_root *root = BTRFS_I(dir)->root;
2110         struct btrfs_root *dest = NULL;
2111         struct btrfs_ioctl_vol_args *vol_args;
2112         struct btrfs_trans_handle *trans;
2113         struct btrfs_block_rsv block_rsv;
2114         u64 qgroup_reserved;
2115         int namelen;
2116         int ret;
2117         int err = 0;
2118
2119         vol_args = memdup_user(arg, sizeof(*vol_args));
2120         if (IS_ERR(vol_args))
2121                 return PTR_ERR(vol_args);
2122
2123         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2124         namelen = strlen(vol_args->name);
2125         if (strchr(vol_args->name, '/') ||
2126             strncmp(vol_args->name, "..", namelen) == 0) {
2127                 err = -EINVAL;
2128                 goto out;
2129         }
2130
2131         err = mnt_want_write_file(file);
2132         if (err)
2133                 goto out;
2134
2135         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2136         if (err == -EINTR)
2137                 goto out;
2138         dentry = lookup_one_len(vol_args->name, parent, namelen);
2139         if (IS_ERR(dentry)) {
2140                 err = PTR_ERR(dentry);
2141                 goto out_unlock_dir;
2142         }
2143
2144         if (!dentry->d_inode) {
2145                 err = -ENOENT;
2146                 goto out_dput;
2147         }
2148
2149         inode = dentry->d_inode;
2150         dest = BTRFS_I(inode)->root;
2151         if (!capable(CAP_SYS_ADMIN)){
2152                 /*
2153                  * Regular user.  Only allow this with a special mount
2154                  * option, when the user has write+exec access to the
2155                  * subvol root, and when rmdir(2) would have been
2156                  * allowed.
2157                  *
2158                  * Note that this is _not_ check that the subvol is
2159                  * empty or doesn't contain data that we wouldn't
2160                  * otherwise be able to delete.
2161                  *
2162                  * Users who want to delete empty subvols should try
2163                  * rmdir(2).
2164                  */
2165                 err = -EPERM;
2166                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2167                         goto out_dput;
2168
2169                 /*
2170                  * Do not allow deletion if the parent dir is the same
2171                  * as the dir to be deleted.  That means the ioctl
2172                  * must be called on the dentry referencing the root
2173                  * of the subvol, not a random directory contained
2174                  * within it.
2175                  */
2176                 err = -EINVAL;
2177                 if (root == dest)
2178                         goto out_dput;
2179
2180                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2181                 if (err)
2182                         goto out_dput;
2183         }
2184
2185         /* check if subvolume may be deleted by a user */
2186         err = btrfs_may_delete(dir, dentry, 1);
2187         if (err)
2188                 goto out_dput;
2189
2190         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2191                 err = -EINVAL;
2192                 goto out_dput;
2193         }
2194
2195         mutex_lock(&inode->i_mutex);
2196         err = d_invalidate(dentry);
2197         if (err)
2198                 goto out_unlock;
2199
2200         down_write(&root->fs_info->subvol_sem);
2201
2202         err = may_destroy_subvol(dest);
2203         if (err)
2204                 goto out_up_write;
2205
2206         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2207         /*
2208          * One for dir inode, two for dir entries, two for root
2209          * ref/backref.
2210          */
2211         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2212                                                5, &qgroup_reserved, true);
2213         if (err)
2214                 goto out_up_write;
2215
2216         trans = btrfs_start_transaction(root, 0);
2217         if (IS_ERR(trans)) {
2218                 err = PTR_ERR(trans);
2219                 goto out_release;
2220         }
2221         trans->block_rsv = &block_rsv;
2222         trans->bytes_reserved = block_rsv.size;
2223
2224         ret = btrfs_unlink_subvol(trans, root, dir,
2225                                 dest->root_key.objectid,
2226                                 dentry->d_name.name,
2227                                 dentry->d_name.len);
2228         if (ret) {
2229                 err = ret;
2230                 btrfs_abort_transaction(trans, root, ret);
2231                 goto out_end_trans;
2232         }
2233
2234         btrfs_record_root_in_trans(trans, dest);
2235
2236         memset(&dest->root_item.drop_progress, 0,
2237                 sizeof(dest->root_item.drop_progress));
2238         dest->root_item.drop_level = 0;
2239         btrfs_set_root_refs(&dest->root_item, 0);
2240
2241         if (!xchg(&dest->orphan_item_inserted, 1)) {
2242                 ret = btrfs_insert_orphan_item(trans,
2243                                         root->fs_info->tree_root,
2244                                         dest->root_key.objectid);
2245                 if (ret) {
2246                         btrfs_abort_transaction(trans, root, ret);
2247                         err = ret;
2248                         goto out_end_trans;
2249                 }
2250         }
2251
2252         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2253                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2254                                   dest->root_key.objectid);
2255         if (ret && ret != -ENOENT) {
2256                 btrfs_abort_transaction(trans, root, ret);
2257                 err = ret;
2258                 goto out_end_trans;
2259         }
2260         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2261                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2262                                           dest->root_item.received_uuid,
2263                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2264                                           dest->root_key.objectid);
2265                 if (ret && ret != -ENOENT) {
2266                         btrfs_abort_transaction(trans, root, ret);
2267                         err = ret;
2268                         goto out_end_trans;
2269                 }
2270         }
2271
2272 out_end_trans:
2273         trans->block_rsv = NULL;
2274         trans->bytes_reserved = 0;
2275         ret = btrfs_end_transaction(trans, root);
2276         if (ret && !err)
2277                 err = ret;
2278         inode->i_flags |= S_DEAD;
2279 out_release:
2280         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2281 out_up_write:
2282         up_write(&root->fs_info->subvol_sem);
2283 out_unlock:
2284         mutex_unlock(&inode->i_mutex);
2285         if (!err) {
2286                 shrink_dcache_sb(root->fs_info->sb);
2287                 btrfs_invalidate_inodes(dest);
2288                 d_delete(dentry);
2289
2290                 /* the last ref */
2291                 if (dest->cache_inode) {
2292                         iput(dest->cache_inode);
2293                         dest->cache_inode = NULL;
2294                 }
2295         }
2296 out_dput:
2297         dput(dentry);
2298 out_unlock_dir:
2299         mutex_unlock(&dir->i_mutex);
2300         mnt_drop_write_file(file);
2301 out:
2302         kfree(vol_args);
2303         return err;
2304 }
2305
2306 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2307 {
2308         struct inode *inode = file_inode(file);
2309         struct btrfs_root *root = BTRFS_I(inode)->root;
2310         struct btrfs_ioctl_defrag_range_args *range;
2311         int ret;
2312
2313         ret = mnt_want_write_file(file);
2314         if (ret)
2315                 return ret;
2316
2317         if (btrfs_root_readonly(root)) {
2318                 ret = -EROFS;
2319                 goto out;
2320         }
2321
2322         switch (inode->i_mode & S_IFMT) {
2323         case S_IFDIR:
2324                 if (!capable(CAP_SYS_ADMIN)) {
2325                         ret = -EPERM;
2326                         goto out;
2327                 }
2328                 ret = btrfs_defrag_root(root);
2329                 if (ret)
2330                         goto out;
2331                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2332                 break;
2333         case S_IFREG:
2334                 if (!(file->f_mode & FMODE_WRITE)) {
2335                         ret = -EINVAL;
2336                         goto out;
2337                 }
2338
2339                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2340                 if (!range) {
2341                         ret = -ENOMEM;
2342                         goto out;
2343                 }
2344
2345                 if (argp) {
2346                         if (copy_from_user(range, argp,
2347                                            sizeof(*range))) {
2348                                 ret = -EFAULT;
2349                                 kfree(range);
2350                                 goto out;
2351                         }
2352                         /* compression requires us to start the IO */
2353                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2354                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2355                                 range->extent_thresh = (u32)-1;
2356                         }
2357                 } else {
2358                         /* the rest are all set to zero by kzalloc */
2359                         range->len = (u64)-1;
2360                 }
2361                 ret = btrfs_defrag_file(file_inode(file), file,
2362                                         range, 0, 0);
2363                 if (ret > 0)
2364                         ret = 0;
2365                 kfree(range);
2366                 break;
2367         default:
2368                 ret = -EINVAL;
2369         }
2370 out:
2371         mnt_drop_write_file(file);
2372         return ret;
2373 }
2374
2375 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2376 {
2377         struct btrfs_ioctl_vol_args *vol_args;
2378         int ret;
2379
2380         if (!capable(CAP_SYS_ADMIN))
2381                 return -EPERM;
2382
2383         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2384                         1)) {
2385                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2386                 return -EINVAL;
2387         }
2388
2389         mutex_lock(&root->fs_info->volume_mutex);
2390         vol_args = memdup_user(arg, sizeof(*vol_args));
2391         if (IS_ERR(vol_args)) {
2392                 ret = PTR_ERR(vol_args);
2393                 goto out;
2394         }
2395
2396         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2397         ret = btrfs_init_new_device(root, vol_args->name);
2398
2399         kfree(vol_args);
2400 out:
2401         mutex_unlock(&root->fs_info->volume_mutex);
2402         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2403         return ret;
2404 }
2405
2406 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2407 {
2408         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2409         struct btrfs_ioctl_vol_args *vol_args;
2410         int ret;
2411
2412         if (!capable(CAP_SYS_ADMIN))
2413                 return -EPERM;
2414
2415         ret = mnt_want_write_file(file);
2416         if (ret)
2417                 return ret;
2418
2419         vol_args = memdup_user(arg, sizeof(*vol_args));
2420         if (IS_ERR(vol_args)) {
2421                 ret = PTR_ERR(vol_args);
2422                 goto out;
2423         }
2424
2425         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2426
2427         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2428                         1)) {
2429                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2430                 goto out;
2431         }
2432
2433         mutex_lock(&root->fs_info->volume_mutex);
2434         ret = btrfs_rm_device(root, vol_args->name);
2435         mutex_unlock(&root->fs_info->volume_mutex);
2436         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2437
2438 out:
2439         kfree(vol_args);
2440         mnt_drop_write_file(file);
2441         return ret;
2442 }
2443
2444 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2445 {
2446         struct btrfs_ioctl_fs_info_args *fi_args;
2447         struct btrfs_device *device;
2448         struct btrfs_device *next;
2449         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2450         int ret = 0;
2451
2452         if (!capable(CAP_SYS_ADMIN))
2453                 return -EPERM;
2454
2455         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2456         if (!fi_args)
2457                 return -ENOMEM;
2458
2459         fi_args->num_devices = fs_devices->num_devices;
2460         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2461
2462         mutex_lock(&fs_devices->device_list_mutex);
2463         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2464                 if (device->devid > fi_args->max_id)
2465                         fi_args->max_id = device->devid;
2466         }
2467         mutex_unlock(&fs_devices->device_list_mutex);
2468
2469         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2470                 ret = -EFAULT;
2471
2472         kfree(fi_args);
2473         return ret;
2474 }
2475
2476 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2477 {
2478         struct btrfs_ioctl_dev_info_args *di_args;
2479         struct btrfs_device *dev;
2480         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2481         int ret = 0;
2482         char *s_uuid = NULL;
2483
2484         if (!capable(CAP_SYS_ADMIN))
2485                 return -EPERM;
2486
2487         di_args = memdup_user(arg, sizeof(*di_args));
2488         if (IS_ERR(di_args))
2489                 return PTR_ERR(di_args);
2490
2491         if (!btrfs_is_empty_uuid(di_args->uuid))
2492                 s_uuid = di_args->uuid;
2493
2494         mutex_lock(&fs_devices->device_list_mutex);
2495         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2496
2497         if (!dev) {
2498                 ret = -ENODEV;
2499                 goto out;
2500         }
2501
2502         di_args->devid = dev->devid;
2503         di_args->bytes_used = dev->bytes_used;
2504         di_args->total_bytes = dev->total_bytes;
2505         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2506         if (dev->name) {
2507                 struct rcu_string *name;
2508
2509                 rcu_read_lock();
2510                 name = rcu_dereference(dev->name);
2511                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2512                 rcu_read_unlock();
2513                 di_args->path[sizeof(di_args->path) - 1] = 0;
2514         } else {
2515                 di_args->path[0] = '\0';
2516         }
2517
2518 out:
2519         mutex_unlock(&fs_devices->device_list_mutex);
2520         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2521                 ret = -EFAULT;
2522
2523         kfree(di_args);
2524         return ret;
2525 }
2526
2527 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2528 {
2529         struct page *page;
2530         pgoff_t index;
2531         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2532
2533         index = off >> PAGE_CACHE_SHIFT;
2534
2535         page = grab_cache_page(inode->i_mapping, index);
2536         if (!page)
2537                 return NULL;
2538
2539         if (!PageUptodate(page)) {
2540                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2541                                                  0))
2542                         return NULL;
2543                 lock_page(page);
2544                 if (!PageUptodate(page)) {
2545                         unlock_page(page);
2546                         page_cache_release(page);
2547                         return NULL;
2548                 }
2549         }
2550         unlock_page(page);
2551
2552         return page;
2553 }
2554
2555 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2556 {
2557         /* do any pending delalloc/csum calc on src, one way or
2558            another, and lock file content */
2559         while (1) {
2560                 struct btrfs_ordered_extent *ordered;
2561                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2562                 ordered = btrfs_lookup_first_ordered_extent(inode,
2563                                                             off + len - 1);
2564                 if (!ordered &&
2565                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2566                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2567                         break;
2568                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2569                 if (ordered)
2570                         btrfs_put_ordered_extent(ordered);
2571                 btrfs_wait_ordered_range(inode, off, len);
2572         }
2573 }
2574
2575 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2576                                 struct inode *inode2, u64 loff2, u64 len)
2577 {
2578         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2579         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2580
2581         mutex_unlock(&inode1->i_mutex);
2582         mutex_unlock(&inode2->i_mutex);
2583 }
2584
2585 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2586                               struct inode *inode2, u64 loff2, u64 len)
2587 {
2588         if (inode1 < inode2) {
2589                 swap(inode1, inode2);
2590                 swap(loff1, loff2);
2591         }
2592
2593         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2594         lock_extent_range(inode1, loff1, len);
2595         if (inode1 != inode2) {
2596                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2597                 lock_extent_range(inode2, loff2, len);
2598         }
2599 }
2600
2601 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2602                           u64 dst_loff, u64 len)
2603 {
2604         int ret = 0;
2605         struct page *src_page, *dst_page;
2606         unsigned int cmp_len = PAGE_CACHE_SIZE;
2607         void *addr, *dst_addr;
2608
2609         while (len) {
2610                 if (len < PAGE_CACHE_SIZE)
2611                         cmp_len = len;
2612
2613                 src_page = extent_same_get_page(src, loff);
2614                 if (!src_page)
2615                         return -EINVAL;
2616                 dst_page = extent_same_get_page(dst, dst_loff);
2617                 if (!dst_page) {
2618                         page_cache_release(src_page);
2619                         return -EINVAL;
2620                 }
2621                 addr = kmap_atomic(src_page);
2622                 dst_addr = kmap_atomic(dst_page);
2623
2624                 flush_dcache_page(src_page);
2625                 flush_dcache_page(dst_page);
2626
2627                 if (memcmp(addr, dst_addr, cmp_len))
2628                         ret = BTRFS_SAME_DATA_DIFFERS;
2629
2630                 kunmap_atomic(addr);
2631                 kunmap_atomic(dst_addr);
2632                 page_cache_release(src_page);
2633                 page_cache_release(dst_page);
2634
2635                 if (ret)
2636                         break;
2637
2638                 loff += cmp_len;
2639                 dst_loff += cmp_len;
2640                 len -= cmp_len;
2641         }
2642
2643         return ret;
2644 }
2645
2646 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2647 {
2648         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2649
2650         if (off + len > inode->i_size || off + len < off)
2651                 return -EINVAL;
2652         /* Check that we are block aligned - btrfs_clone() requires this */
2653         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2654                 return -EINVAL;
2655
2656         return 0;
2657 }
2658
2659 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2660                              struct inode *dst, u64 dst_loff)
2661 {
2662         int ret;
2663
2664         /*
2665          * btrfs_clone() can't handle extents in the same file
2666          * yet. Once that works, we can drop this check and replace it
2667          * with a check for the same inode, but overlapping extents.
2668          */
2669         if (src == dst)
2670                 return -EINVAL;
2671
2672         btrfs_double_lock(src, loff, dst, dst_loff, len);
2673
2674         ret = extent_same_check_offsets(src, loff, len);
2675         if (ret)
2676                 goto out_unlock;
2677
2678         ret = extent_same_check_offsets(dst, dst_loff, len);
2679         if (ret)
2680                 goto out_unlock;
2681
2682         /* don't make the dst file partly checksummed */
2683         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2684             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2685                 ret = -EINVAL;
2686                 goto out_unlock;
2687         }
2688
2689         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2690         if (ret == 0)
2691                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2692
2693 out_unlock:
2694         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2695
2696         return ret;
2697 }
2698
2699 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2700
2701 static long btrfs_ioctl_file_extent_same(struct file *file,
2702                                          void __user *argp)
2703 {
2704         struct btrfs_ioctl_same_args *args = argp;
2705         struct btrfs_ioctl_same_args same;
2706         struct btrfs_ioctl_same_extent_info info;
2707         struct inode *src = file->f_dentry->d_inode;
2708         struct file *dst_file = NULL;
2709         struct inode *dst;
2710         u64 off;
2711         u64 len;
2712         int i;
2713         int ret;
2714         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2715         bool is_admin = capable(CAP_SYS_ADMIN);
2716
2717         if (!(file->f_mode & FMODE_READ))
2718                 return -EINVAL;
2719
2720         ret = mnt_want_write_file(file);
2721         if (ret)
2722                 return ret;
2723
2724         if (copy_from_user(&same,
2725                            (struct btrfs_ioctl_same_args __user *)argp,
2726                            sizeof(same))) {
2727                 ret = -EFAULT;
2728                 goto out;
2729         }
2730
2731         off = same.logical_offset;
2732         len = same.length;
2733
2734         /*
2735          * Limit the total length we will dedupe for each operation.
2736          * This is intended to bound the total time spent in this
2737          * ioctl to something sane.
2738          */
2739         if (len > BTRFS_MAX_DEDUPE_LEN)
2740                 len = BTRFS_MAX_DEDUPE_LEN;
2741
2742         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2743                 /*
2744                  * Btrfs does not support blocksize < page_size. As a
2745                  * result, btrfs_cmp_data() won't correctly handle
2746                  * this situation without an update.
2747                  */
2748                 ret = -EINVAL;
2749                 goto out;
2750         }
2751
2752         ret = -EISDIR;
2753         if (S_ISDIR(src->i_mode))
2754                 goto out;
2755
2756         ret = -EACCES;
2757         if (!S_ISREG(src->i_mode))
2758                 goto out;
2759
2760         ret = 0;
2761         for (i = 0; i < same.dest_count; i++) {
2762                 if (copy_from_user(&info, &args->info[i], sizeof(info))) {
2763                         ret = -EFAULT;
2764                         goto out;
2765                 }
2766
2767                 info.bytes_deduped = 0;
2768
2769                 dst_file = fget(info.fd);
2770                 if (!dst_file) {
2771                         info.status = -EBADF;
2772                         goto next;
2773                 }
2774
2775                 if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2776                         info.status = -EINVAL;
2777                         goto next;
2778                 }
2779
2780                 info.status = -EXDEV;
2781                 if (file->f_path.mnt != dst_file->f_path.mnt)
2782                         goto next;
2783
2784                 dst = dst_file->f_dentry->d_inode;
2785                 if (src->i_sb != dst->i_sb)
2786                         goto next;
2787
2788                 if (S_ISDIR(dst->i_mode)) {
2789                         info.status = -EISDIR;
2790                         goto next;
2791                 }
2792
2793                 if (!S_ISREG(dst->i_mode)) {
2794                         info.status = -EACCES;
2795                         goto next;
2796                 }
2797
2798                 info.status = btrfs_extent_same(src, off, len, dst,
2799                                                 info.logical_offset);
2800                 if (info.status == 0)
2801                         info.bytes_deduped += len;
2802
2803 next:
2804                 if (dst_file)
2805                         fput(dst_file);
2806
2807                 if (__put_user_unaligned(info.status, &args->info[i].status) ||
2808                     __put_user_unaligned(info.bytes_deduped,
2809                                          &args->info[i].bytes_deduped)) {
2810                         ret = -EFAULT;
2811                         goto out;
2812                 }
2813         }
2814
2815 out:
2816         mnt_drop_write_file(file);
2817         return ret;
2818 }
2819
2820 /**
2821  * btrfs_clone() - clone a range from inode file to another
2822  *
2823  * @src: Inode to clone from
2824  * @inode: Inode to clone to
2825  * @off: Offset within source to start clone from
2826  * @olen: Original length, passed by user, of range to clone
2827  * @olen_aligned: Block-aligned value of olen, extent_same uses
2828  *               identical values here
2829  * @destoff: Offset within @inode to start clone
2830  */
2831 static int btrfs_clone(struct inode *src, struct inode *inode,
2832                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2833 {
2834         struct btrfs_root *root = BTRFS_I(inode)->root;
2835         struct btrfs_path *path = NULL;
2836         struct extent_buffer *leaf;
2837         struct btrfs_trans_handle *trans;
2838         char *buf = NULL;
2839         struct btrfs_key key;
2840         u32 nritems;
2841         int slot;
2842         int ret;
2843         u64 len = olen_aligned;
2844
2845         ret = -ENOMEM;
2846         buf = vmalloc(btrfs_level_size(root, 0));
2847         if (!buf)
2848                 return ret;
2849
2850         path = btrfs_alloc_path();
2851         if (!path) {
2852                 vfree(buf);
2853                 return ret;
2854         }
2855
2856         path->reada = 2;
2857         /* clone data */
2858         key.objectid = btrfs_ino(src);
2859         key.type = BTRFS_EXTENT_DATA_KEY;
2860         key.offset = 0;
2861
2862         while (1) {
2863                 /*
2864                  * note the key will change type as we walk through the
2865                  * tree.
2866                  */
2867                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2868                                 0, 0);
2869                 if (ret < 0)
2870                         goto out;
2871
2872                 nritems = btrfs_header_nritems(path->nodes[0]);
2873                 if (path->slots[0] >= nritems) {
2874                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2875                         if (ret < 0)
2876                                 goto out;
2877                         if (ret > 0)
2878                                 break;
2879                         nritems = btrfs_header_nritems(path->nodes[0]);
2880                 }
2881                 leaf = path->nodes[0];
2882                 slot = path->slots[0];
2883
2884                 btrfs_item_key_to_cpu(leaf, &key, slot);
2885                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2886                     key.objectid != btrfs_ino(src))
2887                         break;
2888
2889                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2890                         struct btrfs_file_extent_item *extent;
2891                         int type;
2892                         u32 size;
2893                         struct btrfs_key new_key;
2894                         u64 disko = 0, diskl = 0;
2895                         u64 datao = 0, datal = 0;
2896                         u8 comp;
2897                         u64 endoff;
2898
2899                         size = btrfs_item_size_nr(leaf, slot);
2900                         read_extent_buffer(leaf, buf,
2901                                            btrfs_item_ptr_offset(leaf, slot),
2902                                            size);
2903
2904                         extent = btrfs_item_ptr(leaf, slot,
2905                                                 struct btrfs_file_extent_item);
2906                         comp = btrfs_file_extent_compression(leaf, extent);
2907                         type = btrfs_file_extent_type(leaf, extent);
2908                         if (type == BTRFS_FILE_EXTENT_REG ||
2909                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2910                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2911                                                                       extent);
2912                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2913                                                                  extent);
2914                                 datao = btrfs_file_extent_offset(leaf, extent);
2915                                 datal = btrfs_file_extent_num_bytes(leaf,
2916                                                                     extent);
2917                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2918                                 /* take upper bound, may be compressed */
2919                                 datal = btrfs_file_extent_ram_bytes(leaf,
2920                                                                     extent);
2921                         }
2922                         btrfs_release_path(path);
2923
2924                         if (key.offset + datal <= off ||
2925                             key.offset >= off + len - 1)
2926                                 goto next;
2927
2928                         memcpy(&new_key, &key, sizeof(new_key));
2929                         new_key.objectid = btrfs_ino(inode);
2930                         if (off <= key.offset)
2931                                 new_key.offset = key.offset + destoff - off;
2932                         else
2933                                 new_key.offset = destoff;
2934
2935                         /*
2936                          * 1 - adjusting old extent (we may have to split it)
2937                          * 1 - add new extent
2938                          * 1 - inode update
2939                          */
2940                         trans = btrfs_start_transaction(root, 3);
2941                         if (IS_ERR(trans)) {
2942                                 ret = PTR_ERR(trans);
2943                                 goto out;
2944                         }
2945
2946                         if (type == BTRFS_FILE_EXTENT_REG ||
2947                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2948                                 /*
2949                                  *    a  | --- range to clone ---|  b
2950                                  * | ------------- extent ------------- |
2951                                  */
2952
2953                                 /* substract range b */
2954                                 if (key.offset + datal > off + len)
2955                                         datal = off + len - key.offset;
2956
2957                                 /* substract range a */
2958                                 if (off > key.offset) {
2959                                         datao += off - key.offset;
2960                                         datal -= off - key.offset;
2961                                 }
2962
2963                                 ret = btrfs_drop_extents(trans, root, inode,
2964                                                          new_key.offset,
2965                                                          new_key.offset + datal,
2966                                                          1);
2967                                 if (ret) {
2968                                         btrfs_abort_transaction(trans, root,
2969                                                                 ret);
2970                                         btrfs_end_transaction(trans, root);
2971                                         goto out;
2972                                 }
2973
2974                                 ret = btrfs_insert_empty_item(trans, root, path,
2975                                                               &new_key, size);
2976                                 if (ret) {
2977                                         btrfs_abort_transaction(trans, root,
2978                                                                 ret);
2979                                         btrfs_end_transaction(trans, root);
2980                                         goto out;
2981                                 }
2982
2983                                 leaf = path->nodes[0];
2984                                 slot = path->slots[0];
2985                                 write_extent_buffer(leaf, buf,
2986                                             btrfs_item_ptr_offset(leaf, slot),
2987                                             size);
2988
2989                                 extent = btrfs_item_ptr(leaf, slot,
2990                                                 struct btrfs_file_extent_item);
2991
2992                                 /* disko == 0 means it's a hole */
2993                                 if (!disko)
2994                                         datao = 0;
2995
2996                                 btrfs_set_file_extent_offset(leaf, extent,
2997                                                              datao);
2998                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2999                                                                 datal);
3000                                 if (disko) {
3001                                         inode_add_bytes(inode, datal);
3002                                         ret = btrfs_inc_extent_ref(trans, root,
3003                                                         disko, diskl, 0,
3004                                                         root->root_key.objectid,
3005                                                         btrfs_ino(inode),
3006                                                         new_key.offset - datao,
3007                                                         0);
3008                                         if (ret) {
3009                                                 btrfs_abort_transaction(trans,
3010                                                                         root,
3011                                                                         ret);
3012                                                 btrfs_end_transaction(trans,
3013                                                                       root);
3014                                                 goto out;
3015
3016                                         }
3017                                 }
3018                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3019                                 u64 skip = 0;
3020                                 u64 trim = 0;
3021                                 if (off > key.offset) {
3022                                         skip = off - key.offset;
3023                                         new_key.offset += skip;
3024                                 }
3025
3026                                 if (key.offset + datal > off + len)
3027                                         trim = key.offset + datal - (off + len);
3028
3029                                 if (comp && (skip || trim)) {
3030                                         ret = -EINVAL;
3031                                         btrfs_end_transaction(trans, root);
3032                                         goto out;
3033                                 }
3034                                 size -= skip + trim;
3035                                 datal -= skip + trim;
3036
3037                                 ret = btrfs_drop_extents(trans, root, inode,
3038                                                          new_key.offset,
3039                                                          new_key.offset + datal,
3040                                                          1);
3041                                 if (ret) {
3042                                         btrfs_abort_transaction(trans, root,
3043                                                                 ret);
3044                                         btrfs_end_transaction(trans, root);
3045                                         goto out;
3046                                 }
3047
3048                                 ret = btrfs_insert_empty_item(trans, root, path,
3049                                                               &new_key, size);
3050                                 if (ret) {
3051                                         btrfs_abort_transaction(trans, root,
3052                                                                 ret);
3053                                         btrfs_end_transaction(trans, root);
3054                                         goto out;
3055                                 }
3056
3057                                 if (skip) {
3058                                         u32 start =
3059                                           btrfs_file_extent_calc_inline_size(0);
3060                                         memmove(buf+start, buf+start+skip,
3061                                                 datal);
3062                                 }
3063
3064                                 leaf = path->nodes[0];
3065                                 slot = path->slots[0];
3066                                 write_extent_buffer(leaf, buf,
3067                                             btrfs_item_ptr_offset(leaf, slot),
3068                                             size);
3069                                 inode_add_bytes(inode, datal);
3070                         }
3071
3072                         btrfs_mark_buffer_dirty(leaf);
3073                         btrfs_release_path(path);
3074
3075                         inode_inc_iversion(inode);
3076                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3077
3078                         /*
3079                          * we round up to the block size at eof when
3080                          * determining which extents to clone above,
3081                          * but shouldn't round up the file size
3082                          */
3083                         endoff = new_key.offset + datal;
3084                         if (endoff > destoff+olen)
3085                                 endoff = destoff+olen;
3086                         if (endoff > inode->i_size)
3087                                 btrfs_i_size_write(inode, endoff);
3088
3089                         ret = btrfs_update_inode(trans, root, inode);
3090                         if (ret) {
3091                                 btrfs_abort_transaction(trans, root, ret);
3092                                 btrfs_end_transaction(trans, root);
3093                                 goto out;
3094                         }
3095                         ret = btrfs_end_transaction(trans, root);
3096                 }
3097 next:
3098                 btrfs_release_path(path);
3099                 key.offset++;
3100         }
3101         ret = 0;
3102
3103 out:
3104         btrfs_release_path(path);
3105         btrfs_free_path(path);
3106         vfree(buf);
3107         return ret;
3108 }
3109
3110 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3111                                        u64 off, u64 olen, u64 destoff)
3112 {
3113         struct inode *inode = fdentry(file)->d_inode;
3114         struct btrfs_root *root = BTRFS_I(inode)->root;
3115         struct fd src_file;
3116         struct inode *src;
3117         int ret;
3118         u64 len = olen;
3119         u64 bs = root->fs_info->sb->s_blocksize;
3120         int same_inode = 0;
3121
3122         /*
3123          * TODO:
3124          * - split compressed inline extents.  annoying: we need to
3125          *   decompress into destination's address_space (the file offset
3126          *   may change, so source mapping won't do), then recompress (or
3127          *   otherwise reinsert) a subrange.
3128          * - allow ranges within the same file to be cloned (provided
3129          *   they don't overlap)?
3130          */
3131
3132         /* the destination must be opened for writing */
3133         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3134                 return -EINVAL;
3135
3136         if (btrfs_root_readonly(root))
3137                 return -EROFS;
3138
3139         ret = mnt_want_write_file(file);
3140         if (ret)
3141                 return ret;
3142
3143         src_file = fdget(srcfd);
3144         if (!src_file.file) {
3145                 ret = -EBADF;
3146                 goto out_drop_write;
3147         }
3148
3149         ret = -EXDEV;
3150         if (src_file.file->f_path.mnt != file->f_path.mnt)
3151                 goto out_fput;
3152
3153         src = file_inode(src_file.file);
3154
3155         ret = -EINVAL;
3156         if (src == inode)
3157                 same_inode = 1;
3158
3159         /* the src must be open for reading */
3160         if (!(src_file.file->f_mode & FMODE_READ))
3161                 goto out_fput;
3162
3163         /* don't make the dst file partly checksummed */
3164         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3165             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3166                 goto out_fput;
3167
3168         ret = -EISDIR;
3169         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3170                 goto out_fput;
3171
3172         ret = -EXDEV;
3173         if (src->i_sb != inode->i_sb)
3174                 goto out_fput;
3175
3176         if (!same_inode) {
3177                 if (inode < src) {
3178                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3179                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3180                 } else {
3181                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3182                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3183                 }
3184         } else {
3185                 mutex_lock(&src->i_mutex);
3186         }
3187
3188         /* determine range to clone */
3189         ret = -EINVAL;
3190         if (off + len > src->i_size || off + len < off)
3191                 goto out_unlock;
3192         if (len == 0)
3193                 olen = len = src->i_size - off;
3194         /* if we extend to eof, continue to block boundary */
3195         if (off + len == src->i_size)
3196                 len = ALIGN(src->i_size, bs) - off;
3197
3198         /* verify the end result is block aligned */
3199         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3200             !IS_ALIGNED(destoff, bs))
3201                 goto out_unlock;
3202
3203         /* verify if ranges are overlapped within the same file */
3204         if (same_inode) {
3205                 if (destoff + len > off && destoff < off + len)
3206                         goto out_unlock;
3207         }
3208
3209         if (destoff > inode->i_size) {
3210                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3211                 if (ret)
3212                         goto out_unlock;
3213         }
3214
3215         /* truncate page cache pages from target inode range */
3216         truncate_inode_pages_range(&inode->i_data, destoff,
3217                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3218
3219         lock_extent_range(src, off, len);
3220
3221         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3222
3223         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3224 out_unlock:
3225         mutex_unlock(&src->i_mutex);
3226         if (!same_inode)
3227                 mutex_unlock(&inode->i_mutex);
3228 out_fput:
3229         fdput(src_file);
3230 out_drop_write:
3231         mnt_drop_write_file(file);
3232         return ret;
3233 }
3234
3235 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3236 {
3237         struct btrfs_ioctl_clone_range_args args;
3238
3239         if (copy_from_user(&args, argp, sizeof(args)))
3240                 return -EFAULT;
3241         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3242                                  args.src_length, args.dest_offset);
3243 }
3244
3245 /*
3246  * there are many ways the trans_start and trans_end ioctls can lead
3247  * to deadlocks.  They should only be used by applications that
3248  * basically own the machine, and have a very in depth understanding
3249  * of all the possible deadlocks and enospc problems.
3250  */
3251 static long btrfs_ioctl_trans_start(struct file *file)
3252 {
3253         struct inode *inode = file_inode(file);
3254         struct btrfs_root *root = BTRFS_I(inode)->root;
3255         struct btrfs_trans_handle *trans;
3256         int ret;
3257
3258         ret = -EPERM;
3259         if (!capable(CAP_SYS_ADMIN))
3260                 goto out;
3261
3262         ret = -EINPROGRESS;
3263         if (file->private_data)
3264                 goto out;
3265
3266         ret = -EROFS;
3267         if (btrfs_root_readonly(root))
3268                 goto out;
3269
3270         ret = mnt_want_write_file(file);
3271         if (ret)
3272                 goto out;
3273
3274         atomic_inc(&root->fs_info->open_ioctl_trans);
3275
3276         ret = -ENOMEM;
3277         trans = btrfs_start_ioctl_transaction(root);
3278         if (IS_ERR(trans))
3279                 goto out_drop;
3280
3281         file->private_data = trans;
3282         return 0;
3283
3284 out_drop:
3285         atomic_dec(&root->fs_info->open_ioctl_trans);
3286         mnt_drop_write_file(file);
3287 out:
3288         return ret;
3289 }
3290
3291 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3292 {
3293         struct inode *inode = file_inode(file);
3294         struct btrfs_root *root = BTRFS_I(inode)->root;
3295         struct btrfs_root *new_root;
3296         struct btrfs_dir_item *di;
3297         struct btrfs_trans_handle *trans;
3298         struct btrfs_path *path;
3299         struct btrfs_key location;
3300         struct btrfs_disk_key disk_key;
3301         u64 objectid = 0;
3302         u64 dir_id;
3303         int ret;
3304
3305         if (!capable(CAP_SYS_ADMIN))
3306                 return -EPERM;
3307
3308         ret = mnt_want_write_file(file);
3309         if (ret)
3310                 return ret;
3311
3312         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3313                 ret = -EFAULT;
3314                 goto out;
3315         }
3316
3317         if (!objectid)
3318                 objectid = root->root_key.objectid;
3319
3320         location.objectid = objectid;
3321         location.type = BTRFS_ROOT_ITEM_KEY;
3322         location.offset = (u64)-1;
3323
3324         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3325         if (IS_ERR(new_root)) {
3326                 ret = PTR_ERR(new_root);
3327                 goto out;
3328         }
3329
3330         path = btrfs_alloc_path();
3331         if (!path) {
3332                 ret = -ENOMEM;
3333                 goto out;
3334         }
3335         path->leave_spinning = 1;
3336
3337         trans = btrfs_start_transaction(root, 1);
3338         if (IS_ERR(trans)) {
3339                 btrfs_free_path(path);
3340                 ret = PTR_ERR(trans);
3341                 goto out;
3342         }
3343
3344         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3345         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3346                                    dir_id, "default", 7, 1);
3347         if (IS_ERR_OR_NULL(di)) {
3348                 btrfs_free_path(path);
3349                 btrfs_end_transaction(trans, root);
3350                 printk(KERN_ERR "Umm, you don't have the default dir item, "
3351                        "this isn't going to work\n");
3352                 ret = -ENOENT;
3353                 goto out;
3354         }
3355
3356         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3357         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3358         btrfs_mark_buffer_dirty(path->nodes[0]);
3359         btrfs_free_path(path);
3360
3361         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3362         btrfs_end_transaction(trans, root);
3363 out:
3364         mnt_drop_write_file(file);
3365         return ret;
3366 }
3367
3368 void btrfs_get_block_group_info(struct list_head *groups_list,
3369                                 struct btrfs_ioctl_space_info *space)
3370 {
3371         struct btrfs_block_group_cache *block_group;
3372
3373         space->total_bytes = 0;
3374         space->used_bytes = 0;
3375         space->flags = 0;
3376         list_for_each_entry(block_group, groups_list, list) {
3377                 space->flags = block_group->flags;
3378                 space->total_bytes += block_group->key.offset;
3379                 space->used_bytes +=
3380                         btrfs_block_group_used(&block_group->item);
3381         }
3382 }
3383
3384 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3385 {
3386         struct btrfs_ioctl_space_args space_args;
3387         struct btrfs_ioctl_space_info space;
3388         struct btrfs_ioctl_space_info *dest;
3389         struct btrfs_ioctl_space_info *dest_orig;
3390         struct btrfs_ioctl_space_info __user *user_dest;
3391         struct btrfs_space_info *info;
3392         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3393                        BTRFS_BLOCK_GROUP_SYSTEM,
3394                        BTRFS_BLOCK_GROUP_METADATA,
3395                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3396         int num_types = 4;
3397         int alloc_size;
3398         int ret = 0;
3399         u64 slot_count = 0;
3400         int i, c;
3401
3402         if (copy_from_user(&space_args,
3403                            (struct btrfs_ioctl_space_args __user *)arg,
3404                            sizeof(space_args)))
3405                 return -EFAULT;
3406
3407         for (i = 0; i < num_types; i++) {
3408                 struct btrfs_space_info *tmp;
3409
3410                 info = NULL;
3411                 rcu_read_lock();
3412                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3413                                         list) {
3414                         if (tmp->flags == types[i]) {
3415                                 info = tmp;
3416                                 break;
3417                         }
3418                 }
3419                 rcu_read_unlock();
3420
3421                 if (!info)
3422                         continue;
3423
3424                 down_read(&info->groups_sem);
3425                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3426                         if (!list_empty(&info->block_groups[c]))
3427                                 slot_count++;
3428                 }
3429                 up_read(&info->groups_sem);
3430         }
3431
3432         /* space_slots == 0 means they are asking for a count */
3433         if (space_args.space_slots == 0) {
3434                 space_args.total_spaces = slot_count;
3435                 goto out;
3436         }
3437
3438         slot_count = min_t(u64, space_args.space_slots, slot_count);
3439
3440         alloc_size = sizeof(*dest) * slot_count;
3441
3442         /* we generally have at most 6 or so space infos, one for each raid
3443          * level.  So, a whole page should be more than enough for everyone
3444          */
3445         if (alloc_size > PAGE_CACHE_SIZE)
3446                 return -ENOMEM;
3447
3448         space_args.total_spaces = 0;
3449         dest = kmalloc(alloc_size, GFP_NOFS);
3450         if (!dest)
3451                 return -ENOMEM;
3452         dest_orig = dest;
3453
3454         /* now we have a buffer to copy into */
3455         for (i = 0; i < num_types; i++) {
3456                 struct btrfs_space_info *tmp;
3457
3458                 if (!slot_count)
3459                         break;
3460
3461                 info = NULL;
3462                 rcu_read_lock();
3463                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3464                                         list) {
3465                         if (tmp->flags == types[i]) {
3466                                 info = tmp;
3467                                 break;
3468                         }
3469                 }
3470                 rcu_read_unlock();
3471
3472                 if (!info)
3473                         continue;
3474                 down_read(&info->groups_sem);
3475                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3476                         if (!list_empty(&info->block_groups[c])) {
3477                                 btrfs_get_block_group_info(
3478                                         &info->block_groups[c], &space);
3479                                 memcpy(dest, &space, sizeof(space));
3480                                 dest++;
3481                                 space_args.total_spaces++;
3482                                 slot_count--;
3483                         }
3484                         if (!slot_count)
3485                                 break;
3486                 }
3487                 up_read(&info->groups_sem);
3488         }
3489
3490         user_dest = (struct btrfs_ioctl_space_info __user *)
3491                 (arg + sizeof(struct btrfs_ioctl_space_args));
3492
3493         if (copy_to_user(user_dest, dest_orig, alloc_size))
3494                 ret = -EFAULT;
3495
3496         kfree(dest_orig);
3497 out:
3498         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3499                 ret = -EFAULT;
3500
3501         return ret;
3502 }
3503
3504 /*
3505  * there are many ways the trans_start and trans_end ioctls can lead
3506  * to deadlocks.  They should only be used by applications that
3507  * basically own the machine, and have a very in depth understanding
3508  * of all the possible deadlocks and enospc problems.
3509  */
3510 long btrfs_ioctl_trans_end(struct file *file)
3511 {
3512         struct inode *inode = file_inode(file);
3513         struct btrfs_root *root = BTRFS_I(inode)->root;
3514         struct btrfs_trans_handle *trans;
3515
3516         trans = file->private_data;
3517         if (!trans)
3518                 return -EINVAL;
3519         file->private_data = NULL;
3520
3521         btrfs_end_transaction(trans, root);
3522
3523         atomic_dec(&root->fs_info->open_ioctl_trans);
3524
3525         mnt_drop_write_file(file);
3526         return 0;
3527 }
3528
3529 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3530                                             void __user *argp)
3531 {
3532         struct btrfs_trans_handle *trans;
3533         u64 transid;
3534         int ret;
3535
3536         trans = btrfs_attach_transaction_barrier(root);
3537         if (IS_ERR(trans)) {
3538                 if (PTR_ERR(trans) != -ENOENT)
3539                         return PTR_ERR(trans);
3540
3541                 /* No running transaction, don't bother */
3542                 transid = root->fs_info->last_trans_committed;
3543                 goto out;
3544         }
3545         transid = trans->transid;
3546         ret = btrfs_commit_transaction_async(trans, root, 0);
3547         if (ret) {
3548                 btrfs_end_transaction(trans, root);
3549                 return ret;
3550         }
3551 out:
3552         if (argp)
3553                 if (copy_to_user(argp, &transid, sizeof(transid)))
3554                         return -EFAULT;
3555         return 0;
3556 }
3557
3558 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3559                                            void __user *argp)
3560 {
3561         u64 transid;
3562
3563         if (argp) {
3564                 if (copy_from_user(&transid, argp, sizeof(transid)))
3565                         return -EFAULT;
3566         } else {
3567                 transid = 0;  /* current trans */
3568         }
3569         return btrfs_wait_for_commit(root, transid);
3570 }
3571
3572 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3573 {
3574         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3575         struct btrfs_ioctl_scrub_args *sa;
3576         int ret;
3577
3578         if (!capable(CAP_SYS_ADMIN))
3579                 return -EPERM;
3580
3581         sa = memdup_user(arg, sizeof(*sa));
3582         if (IS_ERR(sa))
3583                 return PTR_ERR(sa);
3584
3585         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3586                 ret = mnt_want_write_file(file);
3587                 if (ret)
3588                         goto out;
3589         }
3590
3591         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3592                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3593                               0);
3594
3595         if (copy_to_user(arg, sa, sizeof(*sa)))
3596                 ret = -EFAULT;
3597
3598         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3599                 mnt_drop_write_file(file);
3600 out:
3601         kfree(sa);
3602         return ret;
3603 }
3604
3605 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3606 {
3607         if (!capable(CAP_SYS_ADMIN))
3608                 return -EPERM;
3609
3610         return btrfs_scrub_cancel(root->fs_info);
3611 }
3612
3613 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3614                                        void __user *arg)
3615 {
3616         struct btrfs_ioctl_scrub_args *sa;
3617         int ret;
3618
3619         if (!capable(CAP_SYS_ADMIN))
3620                 return -EPERM;
3621
3622         sa = memdup_user(arg, sizeof(*sa));
3623         if (IS_ERR(sa))
3624                 return PTR_ERR(sa);
3625
3626         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3627
3628         if (copy_to_user(arg, sa, sizeof(*sa)))
3629                 ret = -EFAULT;
3630
3631         kfree(sa);
3632         return ret;
3633 }
3634
3635 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3636                                       void __user *arg)
3637 {
3638         struct btrfs_ioctl_get_dev_stats *sa;
3639         int ret;
3640
3641         sa = memdup_user(arg, sizeof(*sa));
3642         if (IS_ERR(sa))
3643                 return PTR_ERR(sa);
3644
3645         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3646                 kfree(sa);
3647                 return -EPERM;
3648         }
3649
3650         ret = btrfs_get_dev_stats(root, sa);
3651
3652         if (copy_to_user(arg, sa, sizeof(*sa)))
3653                 ret = -EFAULT;
3654
3655         kfree(sa);
3656         return ret;
3657 }
3658
3659 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3660 {
3661         struct btrfs_ioctl_dev_replace_args *p;
3662         int ret;
3663
3664         if (!capable(CAP_SYS_ADMIN))
3665                 return -EPERM;
3666
3667         p = memdup_user(arg, sizeof(*p));
3668         if (IS_ERR(p))
3669                 return PTR_ERR(p);
3670
3671         switch (p->cmd) {
3672         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3673                 if (root->fs_info->sb->s_flags & MS_RDONLY)
3674                         return -EROFS;
3675
3676                 if (atomic_xchg(
3677                         &root->fs_info->mutually_exclusive_operation_running,
3678                         1)) {
3679                         pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3680                         ret = -EINPROGRESS;
3681                 } else {
3682                         ret = btrfs_dev_replace_start(root, p);
3683                         atomic_set(
3684                          &root->fs_info->mutually_exclusive_operation_running,
3685                          0);
3686                 }
3687                 break;
3688         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3689                 btrfs_dev_replace_status(root->fs_info, p);
3690                 ret = 0;
3691                 break;
3692         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3693                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3694                 break;
3695         default:
3696                 ret = -EINVAL;
3697                 break;
3698         }
3699
3700         if (copy_to_user(arg, p, sizeof(*p)))
3701                 ret = -EFAULT;
3702
3703         kfree(p);
3704         return ret;
3705 }
3706
3707 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3708 {
3709         int ret = 0;
3710         int i;
3711         u64 rel_ptr;
3712         int size;
3713         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3714         struct inode_fs_paths *ipath = NULL;
3715         struct btrfs_path *path;
3716
3717         if (!capable(CAP_DAC_READ_SEARCH))
3718                 return -EPERM;
3719
3720         path = btrfs_alloc_path();
3721         if (!path) {
3722                 ret = -ENOMEM;
3723                 goto out;
3724         }
3725
3726         ipa = memdup_user(arg, sizeof(*ipa));
3727         if (IS_ERR(ipa)) {
3728                 ret = PTR_ERR(ipa);
3729                 ipa = NULL;
3730                 goto out;
3731         }
3732
3733         size = min_t(u32, ipa->size, 4096);
3734         ipath = init_ipath(size, root, path);
3735         if (IS_ERR(ipath)) {
3736                 ret = PTR_ERR(ipath);
3737                 ipath = NULL;
3738                 goto out;
3739         }
3740
3741         ret = paths_from_inode(ipa->inum, ipath);
3742         if (ret < 0)
3743                 goto out;
3744
3745         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3746                 rel_ptr = ipath->fspath->val[i] -
3747                           (u64)(unsigned long)ipath->fspath->val;
3748                 ipath->fspath->val[i] = rel_ptr;
3749         }
3750
3751         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3752                            (void *)(unsigned long)ipath->fspath, size);
3753         if (ret) {
3754                 ret = -EFAULT;
3755                 goto out;
3756         }
3757
3758 out:
3759         btrfs_free_path(path);
3760         free_ipath(ipath);
3761         kfree(ipa);
3762
3763         return ret;
3764 }
3765
3766 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3767 {
3768         struct btrfs_data_container *inodes = ctx;
3769         const size_t c = 3 * sizeof(u64);
3770
3771         if (inodes->bytes_left >= c) {
3772                 inodes->bytes_left -= c;
3773                 inodes->val[inodes->elem_cnt] = inum;
3774                 inodes->val[inodes->elem_cnt + 1] = offset;
3775                 inodes->val[inodes->elem_cnt + 2] = root;
3776                 inodes->elem_cnt += 3;
3777         } else {
3778                 inodes->bytes_missing += c - inodes->bytes_left;
3779                 inodes->bytes_left = 0;
3780                 inodes->elem_missed += 3;
3781         }
3782
3783         return 0;
3784 }
3785
3786 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3787                                         void __user *arg)
3788 {
3789         int ret = 0;
3790         int size;
3791         struct btrfs_ioctl_logical_ino_args *loi;
3792         struct btrfs_data_container *inodes = NULL;
3793         struct btrfs_path *path = NULL;
3794
3795         if (!capable(CAP_SYS_ADMIN))
3796                 return -EPERM;
3797
3798         loi = memdup_user(arg, sizeof(*loi));
3799         if (IS_ERR(loi)) {
3800                 ret = PTR_ERR(loi);
3801                 loi = NULL;
3802                 goto out;
3803         }
3804
3805         path = btrfs_alloc_path();
3806         if (!path) {
3807                 ret = -ENOMEM;
3808                 goto out;
3809         }
3810
3811         size = min_t(u32, loi->size, 64 * 1024);
3812         inodes = init_data_container(size);
3813         if (IS_ERR(inodes)) {
3814                 ret = PTR_ERR(inodes);
3815                 inodes = NULL;
3816                 goto out;
3817         }
3818
3819         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3820                                           build_ino_list, inodes);
3821         if (ret == -EINVAL)
3822                 ret = -ENOENT;
3823         if (ret < 0)
3824                 goto out;
3825
3826         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3827                            (void *)(unsigned long)inodes, size);
3828         if (ret)
3829                 ret = -EFAULT;
3830
3831 out:
3832         btrfs_free_path(path);
3833         vfree(inodes);
3834         kfree(loi);
3835
3836         return ret;
3837 }
3838
3839 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3840                                struct btrfs_ioctl_balance_args *bargs)
3841 {
3842         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3843
3844         bargs->flags = bctl->flags;
3845
3846         if (atomic_read(&fs_info->balance_running))
3847                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3848         if (atomic_read(&fs_info->balance_pause_req))
3849                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3850         if (atomic_read(&fs_info->balance_cancel_req))
3851                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3852
3853         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3854         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3855         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3856
3857         if (lock) {
3858                 spin_lock(&fs_info->balance_lock);
3859                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3860                 spin_unlock(&fs_info->balance_lock);
3861         } else {
3862                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3863         }
3864 }
3865
3866 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3867 {
3868         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3869         struct btrfs_fs_info *fs_info = root->fs_info;
3870         struct btrfs_ioctl_balance_args *bargs;
3871         struct btrfs_balance_control *bctl;
3872         bool need_unlock; /* for mut. excl. ops lock */
3873         int ret;
3874
3875         if (!capable(CAP_SYS_ADMIN))
3876                 return -EPERM;
3877
3878         ret = mnt_want_write_file(file);
3879         if (ret)
3880                 return ret;
3881
3882 again:
3883         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
3884                 mutex_lock(&fs_info->volume_mutex);
3885                 mutex_lock(&fs_info->balance_mutex);
3886                 need_unlock = true;
3887                 goto locked;
3888         }
3889
3890         /*
3891          * mut. excl. ops lock is locked.  Three possibilites:
3892          *   (1) some other op is running
3893          *   (2) balance is running
3894          *   (3) balance is paused -- special case (think resume)
3895          */
3896         mutex_lock(&fs_info->balance_mutex);
3897         if (fs_info->balance_ctl) {
3898                 /* this is either (2) or (3) */
3899                 if (!atomic_read(&fs_info->balance_running)) {
3900                         mutex_unlock(&fs_info->balance_mutex);
3901                         if (!mutex_trylock(&fs_info->volume_mutex))
3902                                 goto again;
3903                         mutex_lock(&fs_info->balance_mutex);
3904
3905                         if (fs_info->balance_ctl &&
3906                             !atomic_read(&fs_info->balance_running)) {
3907                                 /* this is (3) */
3908                                 need_unlock = false;
3909                                 goto locked;
3910                         }
3911
3912                         mutex_unlock(&fs_info->balance_mutex);
3913                         mutex_unlock(&fs_info->volume_mutex);
3914                         goto again;
3915                 } else {
3916                         /* this is (2) */
3917                         mutex_unlock(&fs_info->balance_mutex);
3918                         ret = -EINPROGRESS;
3919                         goto out;
3920                 }
3921         } else {
3922                 /* this is (1) */
3923                 mutex_unlock(&fs_info->balance_mutex);
3924                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3925                 ret = -EINVAL;
3926                 goto out;
3927         }
3928
3929 locked:
3930         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
3931
3932         if (arg) {
3933                 bargs = memdup_user(arg, sizeof(*bargs));
3934                 if (IS_ERR(bargs)) {
3935                         ret = PTR_ERR(bargs);
3936                         goto out_unlock;
3937                 }
3938
3939                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3940                         if (!fs_info->balance_ctl) {
3941                                 ret = -ENOTCONN;
3942                                 goto out_bargs;
3943                         }
3944
3945                         bctl = fs_info->balance_ctl;
3946                         spin_lock(&fs_info->balance_lock);
3947                         bctl->flags |= BTRFS_BALANCE_RESUME;
3948                         spin_unlock(&fs_info->balance_lock);
3949
3950                         goto do_balance;
3951                 }
3952         } else {
3953                 bargs = NULL;
3954         }
3955
3956         if (fs_info->balance_ctl) {
3957                 ret = -EINPROGRESS;
3958                 goto out_bargs;
3959         }
3960
3961         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3962         if (!bctl) {
3963                 ret = -ENOMEM;
3964                 goto out_bargs;
3965         }
3966
3967         bctl->fs_info = fs_info;
3968         if (arg) {
3969                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3970                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3971                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3972
3973                 bctl->flags = bargs->flags;
3974         } else {
3975                 /* balance everything - no filters */
3976                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3977         }
3978
3979 do_balance:
3980         /*
3981          * Ownership of bctl and mutually_exclusive_operation_running
3982          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
3983          * or, if restriper was paused all the way until unmount, in
3984          * free_fs_info.  mutually_exclusive_operation_running is
3985          * cleared in __cancel_balance.
3986          */
3987         need_unlock = false;
3988
3989         ret = btrfs_balance(bctl, bargs);
3990
3991         if (arg) {
3992                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3993                         ret = -EFAULT;
3994         }
3995
3996 out_bargs:
3997         kfree(bargs);
3998 out_unlock:
3999         mutex_unlock(&fs_info->balance_mutex);
4000         mutex_unlock(&fs_info->volume_mutex);
4001         if (need_unlock)
4002                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4003 out:
4004         mnt_drop_write_file(file);
4005         return ret;
4006 }
4007
4008 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4009 {
4010         if (!capable(CAP_SYS_ADMIN))
4011                 return -EPERM;
4012
4013         switch (cmd) {
4014         case BTRFS_BALANCE_CTL_PAUSE:
4015                 return btrfs_pause_balance(root->fs_info);
4016         case BTRFS_BALANCE_CTL_CANCEL:
4017                 return btrfs_cancel_balance(root->fs_info);
4018         }
4019
4020         return -EINVAL;
4021 }
4022
4023 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4024                                          void __user *arg)
4025 {
4026         struct btrfs_fs_info *fs_info = root->fs_info;
4027         struct btrfs_ioctl_balance_args *bargs;
4028         int ret = 0;
4029
4030         if (!capable(CAP_SYS_ADMIN))
4031                 return -EPERM;
4032
4033         mutex_lock(&fs_info->balance_mutex);
4034         if (!fs_info->balance_ctl) {
4035                 ret = -ENOTCONN;
4036                 goto out;
4037         }
4038
4039         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4040         if (!bargs) {
4041                 ret = -ENOMEM;
4042                 goto out;
4043         }
4044
4045         update_ioctl_balance_args(fs_info, 1, bargs);
4046
4047         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4048                 ret = -EFAULT;
4049
4050         kfree(bargs);
4051 out:
4052         mutex_unlock(&fs_info->balance_mutex);
4053         return ret;
4054 }
4055
4056 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4057 {
4058         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4059         struct btrfs_ioctl_quota_ctl_args *sa;
4060         struct btrfs_trans_handle *trans = NULL;
4061         int ret;
4062         int err;
4063
4064         if (!capable(CAP_SYS_ADMIN))
4065                 return -EPERM;
4066
4067         ret = mnt_want_write_file(file);
4068         if (ret)
4069                 return ret;
4070
4071         sa = memdup_user(arg, sizeof(*sa));
4072         if (IS_ERR(sa)) {
4073                 ret = PTR_ERR(sa);
4074                 goto drop_write;
4075         }
4076
4077         down_write(&root->fs_info->subvol_sem);
4078         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4079         if (IS_ERR(trans)) {
4080                 ret = PTR_ERR(trans);
4081                 goto out;
4082         }
4083
4084         switch (sa->cmd) {
4085         case BTRFS_QUOTA_CTL_ENABLE:
4086                 ret = btrfs_quota_enable(trans, root->fs_info);
4087                 break;
4088         case BTRFS_QUOTA_CTL_DISABLE:
4089                 ret = btrfs_quota_disable(trans, root->fs_info);
4090                 break;
4091         default:
4092                 ret = -EINVAL;
4093                 break;
4094         }
4095
4096         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4097         if (err && !ret)
4098                 ret = err;
4099 out:
4100         kfree(sa);
4101         up_write(&root->fs_info->subvol_sem);
4102 drop_write:
4103         mnt_drop_write_file(file);
4104         return ret;
4105 }
4106
4107 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4108 {
4109         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4110         struct btrfs_ioctl_qgroup_assign_args *sa;
4111         struct btrfs_trans_handle *trans;
4112         int ret;
4113         int err;
4114
4115         if (!capable(CAP_SYS_ADMIN))
4116                 return -EPERM;
4117
4118         ret = mnt_want_write_file(file);
4119         if (ret)
4120                 return ret;
4121
4122         sa = memdup_user(arg, sizeof(*sa));
4123         if (IS_ERR(sa)) {
4124                 ret = PTR_ERR(sa);
4125                 goto drop_write;
4126         }
4127
4128         trans = btrfs_join_transaction(root);
4129         if (IS_ERR(trans)) {
4130                 ret = PTR_ERR(trans);
4131                 goto out;
4132         }
4133
4134         /* FIXME: check if the IDs really exist */
4135         if (sa->assign) {
4136                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4137                                                 sa->src, sa->dst);
4138         } else {
4139                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4140                                                 sa->src, sa->dst);
4141         }
4142
4143         err = btrfs_end_transaction(trans, root);
4144         if (err && !ret)
4145                 ret = err;
4146
4147 out:
4148         kfree(sa);
4149 drop_write:
4150         mnt_drop_write_file(file);
4151         return ret;
4152 }
4153
4154 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4155 {
4156         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4157         struct btrfs_ioctl_qgroup_create_args *sa;
4158         struct btrfs_trans_handle *trans;
4159         int ret;
4160         int err;
4161
4162         if (!capable(CAP_SYS_ADMIN))
4163                 return -EPERM;
4164
4165         ret = mnt_want_write_file(file);
4166         if (ret)
4167                 return ret;
4168
4169         sa = memdup_user(arg, sizeof(*sa));
4170         if (IS_ERR(sa)) {
4171                 ret = PTR_ERR(sa);
4172                 goto drop_write;
4173         }
4174
4175         if (!sa->qgroupid) {
4176                 ret = -EINVAL;
4177                 goto out;
4178         }
4179
4180         trans = btrfs_join_transaction(root);
4181         if (IS_ERR(trans)) {
4182                 ret = PTR_ERR(trans);
4183                 goto out;
4184         }
4185
4186         /* FIXME: check if the IDs really exist */
4187         if (sa->create) {
4188                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4189                                           NULL);
4190         } else {
4191                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4192         }
4193
4194         err = btrfs_end_transaction(trans, root);
4195         if (err && !ret)
4196                 ret = err;
4197
4198 out:
4199         kfree(sa);
4200 drop_write:
4201         mnt_drop_write_file(file);
4202         return ret;
4203 }
4204
4205 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4206 {
4207         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4208         struct btrfs_ioctl_qgroup_limit_args *sa;
4209         struct btrfs_trans_handle *trans;
4210         int ret;
4211         int err;
4212         u64 qgroupid;
4213
4214         if (!capable(CAP_SYS_ADMIN))
4215                 return -EPERM;
4216
4217         ret = mnt_want_write_file(file);
4218         if (ret)
4219                 return ret;
4220
4221         sa = memdup_user(arg, sizeof(*sa));
4222         if (IS_ERR(sa)) {
4223                 ret = PTR_ERR(sa);
4224                 goto drop_write;
4225         }
4226
4227         trans = btrfs_join_transaction(root);
4228         if (IS_ERR(trans)) {
4229                 ret = PTR_ERR(trans);
4230                 goto out;
4231         }
4232
4233         qgroupid = sa->qgroupid;
4234         if (!qgroupid) {
4235                 /* take the current subvol as qgroup */
4236                 qgroupid = root->root_key.objectid;
4237         }
4238
4239         /* FIXME: check if the IDs really exist */
4240         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4241
4242         err = btrfs_end_transaction(trans, root);
4243         if (err && !ret)
4244                 ret = err;
4245
4246 out:
4247         kfree(sa);
4248 drop_write:
4249         mnt_drop_write_file(file);
4250         return ret;
4251 }
4252
4253 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4254 {
4255         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4256         struct btrfs_ioctl_quota_rescan_args *qsa;
4257         int ret;
4258
4259         if (!capable(CAP_SYS_ADMIN))
4260                 return -EPERM;
4261
4262         ret = mnt_want_write_file(file);
4263         if (ret)
4264                 return ret;
4265
4266         qsa = memdup_user(arg, sizeof(*qsa));
4267         if (IS_ERR(qsa)) {
4268                 ret = PTR_ERR(qsa);
4269                 goto drop_write;
4270         }
4271
4272         if (qsa->flags) {
4273                 ret = -EINVAL;
4274                 goto out;
4275         }
4276
4277         ret = btrfs_qgroup_rescan(root->fs_info);
4278
4279 out:
4280         kfree(qsa);
4281 drop_write:
4282         mnt_drop_write_file(file);
4283         return ret;
4284 }
4285
4286 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4287 {
4288         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4289         struct btrfs_ioctl_quota_rescan_args *qsa;
4290         int ret = 0;
4291
4292         if (!capable(CAP_SYS_ADMIN))
4293                 return -EPERM;
4294
4295         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4296         if (!qsa)
4297                 return -ENOMEM;
4298
4299         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4300                 qsa->flags = 1;
4301                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4302         }
4303
4304         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4305                 ret = -EFAULT;
4306
4307         kfree(qsa);
4308         return ret;
4309 }
4310
4311 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4312 {
4313         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
4314
4315         if (!capable(CAP_SYS_ADMIN))
4316                 return -EPERM;
4317
4318         return btrfs_qgroup_wait_for_completion(root->fs_info);
4319 }
4320
4321 static long btrfs_ioctl_set_received_subvol(struct file *file,
4322                                             void __user *arg)
4323 {
4324         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4325         struct inode *inode = file_inode(file);
4326         struct btrfs_root *root = BTRFS_I(inode)->root;
4327         struct btrfs_root_item *root_item = &root->root_item;
4328         struct btrfs_trans_handle *trans;
4329         struct timespec ct = CURRENT_TIME;
4330         int ret = 0;
4331         int received_uuid_changed;
4332
4333         ret = mnt_want_write_file(file);
4334         if (ret < 0)
4335                 return ret;
4336
4337         down_write(&root->fs_info->subvol_sem);
4338
4339         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4340                 ret = -EINVAL;
4341                 goto out;
4342         }
4343
4344         if (btrfs_root_readonly(root)) {
4345                 ret = -EROFS;
4346                 goto out;
4347         }
4348
4349         if (!inode_owner_or_capable(inode)) {
4350                 ret = -EACCES;
4351                 goto out;
4352         }
4353
4354         sa = memdup_user(arg, sizeof(*sa));
4355         if (IS_ERR(sa)) {
4356                 ret = PTR_ERR(sa);
4357                 sa = NULL;
4358                 goto out;
4359         }
4360
4361         /*
4362          * 1 - root item
4363          * 2 - uuid items (received uuid + subvol uuid)
4364          */
4365         trans = btrfs_start_transaction(root, 3);
4366         if (IS_ERR(trans)) {
4367                 ret = PTR_ERR(trans);
4368                 trans = NULL;
4369                 goto out;
4370         }
4371
4372         sa->rtransid = trans->transid;
4373         sa->rtime.sec = ct.tv_sec;
4374         sa->rtime.nsec = ct.tv_nsec;
4375
4376         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4377                                        BTRFS_UUID_SIZE);
4378         if (received_uuid_changed &&
4379             !btrfs_is_empty_uuid(root_item->received_uuid))
4380                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4381                                     root_item->received_uuid,
4382                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4383                                     root->root_key.objectid);
4384         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4385         btrfs_set_root_stransid(root_item, sa->stransid);
4386         btrfs_set_root_rtransid(root_item, sa->rtransid);
4387         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4388         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4389         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4390         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4391
4392         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4393                                 &root->root_key, &root->root_item);
4394         if (ret < 0) {
4395                 btrfs_end_transaction(trans, root);
4396                 goto out;
4397         }
4398         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4399                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4400                                           sa->uuid,
4401                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4402                                           root->root_key.objectid);
4403                 if (ret < 0 && ret != -EEXIST) {
4404                         btrfs_abort_transaction(trans, root, ret);
4405                         goto out;
4406                 }
4407         }
4408         ret = btrfs_commit_transaction(trans, root);
4409         if (ret < 0) {
4410                 btrfs_abort_transaction(trans, root, ret);
4411                 goto out;
4412         }
4413
4414         ret = copy_to_user(arg, sa, sizeof(*sa));
4415         if (ret)
4416                 ret = -EFAULT;
4417
4418 out:
4419         kfree(sa);
4420         up_write(&root->fs_info->subvol_sem);
4421         mnt_drop_write_file(file);
4422         return ret;
4423 }
4424
4425 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4426 {
4427         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4428         size_t len;
4429         int ret;
4430         char label[BTRFS_LABEL_SIZE];
4431
4432         spin_lock(&root->fs_info->super_lock);
4433         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4434         spin_unlock(&root->fs_info->super_lock);
4435
4436         len = strnlen(label, BTRFS_LABEL_SIZE);
4437
4438         if (len == BTRFS_LABEL_SIZE) {
4439                 pr_warn("btrfs: label is too long, return the first %zu bytes\n",
4440                         --len);
4441         }
4442
4443         ret = copy_to_user(arg, label, len);
4444
4445         return ret ? -EFAULT : 0;
4446 }
4447
4448 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4449 {
4450         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4451         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4452         struct btrfs_trans_handle *trans;
4453         char label[BTRFS_LABEL_SIZE];
4454         int ret;
4455
4456         if (!capable(CAP_SYS_ADMIN))
4457                 return -EPERM;
4458
4459         if (copy_from_user(label, arg, sizeof(label)))
4460                 return -EFAULT;
4461
4462         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4463                 pr_err("btrfs: unable to set label with more than %d bytes\n",
4464                        BTRFS_LABEL_SIZE - 1);
4465                 return -EINVAL;
4466         }
4467
4468         ret = mnt_want_write_file(file);
4469         if (ret)
4470                 return ret;
4471
4472         trans = btrfs_start_transaction(root, 0);
4473         if (IS_ERR(trans)) {
4474                 ret = PTR_ERR(trans);
4475                 goto out_unlock;
4476         }
4477
4478         spin_lock(&root->fs_info->super_lock);
4479         strcpy(super_block->label, label);
4480         spin_unlock(&root->fs_info->super_lock);
4481         ret = btrfs_end_transaction(trans, root);
4482
4483 out_unlock:
4484         mnt_drop_write_file(file);
4485         return ret;
4486 }
4487
4488 long btrfs_ioctl(struct file *file, unsigned int
4489                 cmd, unsigned long arg)
4490 {
4491         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4492         void __user *argp = (void __user *)arg;
4493
4494         switch (cmd) {
4495         case FS_IOC_GETFLAGS:
4496                 return btrfs_ioctl_getflags(file, argp);
4497         case FS_IOC_SETFLAGS:
4498                 return btrfs_ioctl_setflags(file, argp);
4499         case FS_IOC_GETVERSION:
4500                 return btrfs_ioctl_getversion(file, argp);
4501         case FITRIM:
4502                 return btrfs_ioctl_fitrim(file, argp);
4503         case BTRFS_IOC_SNAP_CREATE:
4504                 return btrfs_ioctl_snap_create(file, argp, 0);
4505         case BTRFS_IOC_SNAP_CREATE_V2:
4506                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4507         case BTRFS_IOC_SUBVOL_CREATE:
4508                 return btrfs_ioctl_snap_create(file, argp, 1);
4509         case BTRFS_IOC_SUBVOL_CREATE_V2:
4510                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4511         case BTRFS_IOC_SNAP_DESTROY:
4512                 return btrfs_ioctl_snap_destroy(file, argp);
4513         case BTRFS_IOC_SUBVOL_GETFLAGS:
4514                 return btrfs_ioctl_subvol_getflags(file, argp);
4515         case BTRFS_IOC_SUBVOL_SETFLAGS:
4516                 return btrfs_ioctl_subvol_setflags(file, argp);
4517         case BTRFS_IOC_DEFAULT_SUBVOL:
4518                 return btrfs_ioctl_default_subvol(file, argp);
4519         case BTRFS_IOC_DEFRAG:
4520                 return btrfs_ioctl_defrag(file, NULL);
4521         case BTRFS_IOC_DEFRAG_RANGE:
4522                 return btrfs_ioctl_defrag(file, argp);
4523         case BTRFS_IOC_RESIZE:
4524                 return btrfs_ioctl_resize(file, argp);
4525         case BTRFS_IOC_ADD_DEV:
4526                 return btrfs_ioctl_add_dev(root, argp);
4527         case BTRFS_IOC_RM_DEV:
4528                 return btrfs_ioctl_rm_dev(file, argp);
4529         case BTRFS_IOC_FS_INFO:
4530                 return btrfs_ioctl_fs_info(root, argp);
4531         case BTRFS_IOC_DEV_INFO:
4532                 return btrfs_ioctl_dev_info(root, argp);
4533         case BTRFS_IOC_BALANCE:
4534                 return btrfs_ioctl_balance(file, NULL);
4535         case BTRFS_IOC_CLONE:
4536                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4537         case BTRFS_IOC_CLONE_RANGE:
4538                 return btrfs_ioctl_clone_range(file, argp);
4539         case BTRFS_IOC_TRANS_START:
4540                 return btrfs_ioctl_trans_start(file);
4541         case BTRFS_IOC_TRANS_END:
4542                 return btrfs_ioctl_trans_end(file);
4543         case BTRFS_IOC_TREE_SEARCH:
4544                 return btrfs_ioctl_tree_search(file, argp);
4545         case BTRFS_IOC_INO_LOOKUP:
4546                 return btrfs_ioctl_ino_lookup(file, argp);
4547         case BTRFS_IOC_INO_PATHS:
4548                 return btrfs_ioctl_ino_to_path(root, argp);
4549         case BTRFS_IOC_LOGICAL_INO:
4550                 return btrfs_ioctl_logical_to_ino(root, argp);
4551         case BTRFS_IOC_SPACE_INFO:
4552                 return btrfs_ioctl_space_info(root, argp);
4553         case BTRFS_IOC_SYNC:
4554                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
4555                 return 0;
4556         case BTRFS_IOC_START_SYNC:
4557                 return btrfs_ioctl_start_sync(root, argp);
4558         case BTRFS_IOC_WAIT_SYNC:
4559                 return btrfs_ioctl_wait_sync(root, argp);
4560         case BTRFS_IOC_SCRUB:
4561                 return btrfs_ioctl_scrub(file, argp);
4562         case BTRFS_IOC_SCRUB_CANCEL:
4563                 return btrfs_ioctl_scrub_cancel(root, argp);
4564         case BTRFS_IOC_SCRUB_PROGRESS:
4565                 return btrfs_ioctl_scrub_progress(root, argp);
4566         case BTRFS_IOC_BALANCE_V2:
4567                 return btrfs_ioctl_balance(file, argp);
4568         case BTRFS_IOC_BALANCE_CTL:
4569                 return btrfs_ioctl_balance_ctl(root, arg);
4570         case BTRFS_IOC_BALANCE_PROGRESS:
4571                 return btrfs_ioctl_balance_progress(root, argp);
4572         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4573                 return btrfs_ioctl_set_received_subvol(file, argp);
4574         case BTRFS_IOC_SEND:
4575                 return btrfs_ioctl_send(file, argp);
4576         case BTRFS_IOC_GET_DEV_STATS:
4577                 return btrfs_ioctl_get_dev_stats(root, argp);
4578         case BTRFS_IOC_QUOTA_CTL:
4579                 return btrfs_ioctl_quota_ctl(file, argp);
4580         case BTRFS_IOC_QGROUP_ASSIGN:
4581                 return btrfs_ioctl_qgroup_assign(file, argp);
4582         case BTRFS_IOC_QGROUP_CREATE:
4583                 return btrfs_ioctl_qgroup_create(file, argp);
4584         case BTRFS_IOC_QGROUP_LIMIT:
4585                 return btrfs_ioctl_qgroup_limit(file, argp);
4586         case BTRFS_IOC_QUOTA_RESCAN:
4587                 return btrfs_ioctl_quota_rescan(file, argp);
4588         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4589                 return btrfs_ioctl_quota_rescan_status(file, argp);
4590         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4591                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4592         case BTRFS_IOC_DEV_REPLACE:
4593                 return btrfs_ioctl_dev_replace(root, argp);
4594         case BTRFS_IOC_GET_FSLABEL:
4595                 return btrfs_ioctl_get_fslabel(file, argp);
4596         case BTRFS_IOC_SET_FSLABEL:
4597                 return btrfs_ioctl_set_fslabel(file, argp);
4598         case BTRFS_IOC_FILE_EXTENT_SAME:
4599                 return btrfs_ioctl_file_extent_same(file, argp);
4600         }
4601
4602         return -ENOTTY;
4603 }