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