Pileus Git - ~andy/linux/blob - 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_dir_item *di;
1730         struct btrfs_key key;
1731         u64 dir_id;
1732         int ret;
1733
1734         path = btrfs_alloc_path();
1735         if (!path)
1736                 return -ENOMEM;
1737
1738         /* Make sure this root isn't set as the default subvol */
1739         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1740         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1741                                    dir_id, "default", 7, 0);
1742         if (di && !IS_ERR(di)) {
1743                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1744                 if (key.objectid == root->root_key.objectid) {
1745                         ret = -ENOTEMPTY;
1746                         goto out;
1747                 }
1748                 btrfs_release_path(path);
1749         }
1750
1751         key.objectid = root->root_key.objectid;
1752         key.type = BTRFS_ROOT_REF_KEY;
1753         key.offset = (u64)-1;
1754
1755         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1756                                 &key, path, 0, 0);
1757         if (ret < 0)
1758                 goto out;
1759         BUG_ON(ret == 0);
1760
1761         ret = 0;
1762         if (path->slots[0] > 0) {
1763                 path->slots[0]--;
1764                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1765                 if (key.objectid == root->root_key.objectid &&
1766                     key.type == BTRFS_ROOT_REF_KEY)
1767                         ret = -ENOTEMPTY;
1768         }
1769 out:
1770         btrfs_free_path(path);
1771         return ret;
1772 }
1773
1774 static noinline int key_in_sk(struct btrfs_key *key,
1775                               struct btrfs_ioctl_search_key *sk)
1776 {
1777         struct btrfs_key test;
1778         int ret;
1779
1780         test.objectid = sk->min_objectid;
1781         test.type = sk->min_type;
1782         test.offset = sk->min_offset;
1783
1784         ret = btrfs_comp_cpu_keys(key, &test);
1785         if (ret < 0)
1786                 return 0;
1787
1788         test.objectid = sk->max_objectid;
1789         test.type = sk->max_type;
1790         test.offset = sk->max_offset;
1791
1792         ret = btrfs_comp_cpu_keys(key, &test);
1793         if (ret > 0)
1794                 return 0;
1795         return 1;
1796 }
1797
1798 static noinline int copy_to_sk(struct btrfs_root *root,
1799                                struct btrfs_path *path,
1800                                struct btrfs_key *key,
1801                                struct btrfs_ioctl_search_key *sk,
1802                                char *buf,
1803                                unsigned long *sk_offset,
1804                                int *num_found)
1805 {
1806         u64 found_transid;
1807         struct extent_buffer *leaf;
1808         struct btrfs_ioctl_search_header sh;
1809         unsigned long item_off;
1810         unsigned long item_len;
1811         int nritems;
1812         int i;
1813         int slot;
1814         int ret = 0;
1815
1816         leaf = path->nodes[0];
1817         slot = path->slots[0];
1818         nritems = btrfs_header_nritems(leaf);
1819
1820         if (btrfs_header_generation(leaf) > sk->max_transid) {
1821                 i = nritems;
1822                 goto advance_key;
1823         }
1824         found_transid = btrfs_header_generation(leaf);
1825
1826         for (i = slot; i < nritems; i++) {
1827                 item_off = btrfs_item_ptr_offset(leaf, i);
1828                 item_len = btrfs_item_size_nr(leaf, i);
1829
1830                 btrfs_item_key_to_cpu(leaf, key, i);
1831                 if (!key_in_sk(key, sk))
1832                         continue;
1833
1834                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1835                         item_len = 0;
1836
1837                 if (sizeof(sh) + item_len + *sk_offset >
1838                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1839                         ret = 1;
1840                         goto overflow;
1841                 }
1842
1843                 sh.objectid = key->objectid;
1844                 sh.offset = key->offset;
1845                 sh.type = key->type;
1846                 sh.len = item_len;
1847                 sh.transid = found_transid;
1848
1849                 /* copy search result header */
1850                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1851                 *sk_offset += sizeof(sh);
1852
1853                 if (item_len) {
1854                         char *p = buf + *sk_offset;
1855                         /* copy the item */
1856                         read_extent_buffer(leaf, p,
1857                                            item_off, item_len);
1858                         *sk_offset += item_len;
1859                 }
1860                 (*num_found)++;
1861
1862                 if (*num_found >= sk->nr_items)
1863                         break;
1864         }
1865 advance_key:
1866         ret = 0;
1867         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1868                 key->offset++;
1869         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1870                 key->offset = 0;
1871                 key->type++;
1872         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1873                 key->offset = 0;
1874                 key->type = 0;
1875                 key->objectid++;
1876         } else
1877                 ret = 1;
1878 overflow:
1879         return ret;
1880 }
1881
1882 static noinline int search_ioctl(struct inode *inode,
1883                                  struct btrfs_ioctl_search_args *args)
1884 {
1885         struct btrfs_root *root;
1886         struct btrfs_key key;
1887         struct btrfs_key max_key;
1888         struct btrfs_path *path;
1889         struct btrfs_ioctl_search_key *sk = &args->key;
1890         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1891         int ret;
1892         int num_found = 0;
1893         unsigned long sk_offset = 0;
1894
1895         path = btrfs_alloc_path();
1896         if (!path)
1897                 return -ENOMEM;
1898
1899         if (sk->tree_id == 0) {
1900                 /* search the root of the inode that was passed */
1901                 root = BTRFS_I(inode)->root;
1902         } else {
1903                 key.objectid = sk->tree_id;
1904                 key.type = BTRFS_ROOT_ITEM_KEY;
1905                 key.offset = (u64)-1;
1906                 root = btrfs_read_fs_root_no_name(info, &key);
1907                 if (IS_ERR(root)) {
1908                         printk(KERN_ERR "could not find root %llu\n",
1909                                sk->tree_id);
1910                         btrfs_free_path(path);
1911                         return -ENOENT;
1912                 }
1913         }
1914
1915         key.objectid = sk->min_objectid;
1916         key.type = sk->min_type;
1917         key.offset = sk->min_offset;
1918
1919         max_key.objectid = sk->max_objectid;
1920         max_key.type = sk->max_type;
1921         max_key.offset = sk->max_offset;
1922
1923         path->keep_locks = 1;
1924
1925         while(1) {
1926                 ret = btrfs_search_forward(root, &key, &max_key, path,
1927                                            sk->min_transid);
1928                 if (ret != 0) {
1929                         if (ret > 0)
1930                                 ret = 0;
1931                         goto err;
1932                 }
1933                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1934                                  &sk_offset, &num_found);
1935                 btrfs_release_path(path);
1936                 if (ret || num_found >= sk->nr_items)
1937                         break;
1938
1939         }
1940         ret = 0;
1941 err:
1942         sk->nr_items = num_found;
1943         btrfs_free_path(path);
1944         return ret;
1945 }
1946
1947 static noinline int btrfs_ioctl_tree_search(struct file *file,
1948                                            void __user *argp)
1949 {
1950          struct btrfs_ioctl_search_args *args;
1951          struct inode *inode;
1952          int ret;
1953
1954         if (!capable(CAP_SYS_ADMIN))
1955                 return -EPERM;
1956
1957         args = memdup_user(argp, sizeof(*args));
1958         if (IS_ERR(args))
1959                 return PTR_ERR(args);
1960
1961         inode = file_inode(file);
1962         ret = search_ioctl(inode, args);
1963         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1964                 ret = -EFAULT;
1965         kfree(args);
1966         return ret;
1967 }
1968
1969 /*
1970  * Search INODE_REFs to identify path name of 'dirid' directory
1971  * in a 'tree_id' tree. and sets path name to 'name'.
1972  */
1973 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1974                                 u64 tree_id, u64 dirid, char *name)
1975 {
1976         struct btrfs_root *root;
1977         struct btrfs_key key;
1978         char *ptr;
1979         int ret = -1;
1980         int slot;
1981         int len;
1982         int total_len = 0;
1983         struct btrfs_inode_ref *iref;
1984         struct extent_buffer *l;
1985         struct btrfs_path *path;
1986
1987         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1988                 name[0]='\0';
1989                 return 0;
1990         }
1991
1992         path = btrfs_alloc_path();
1993         if (!path)
1994                 return -ENOMEM;
1995
1996         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1997
1998         key.objectid = tree_id;
1999         key.type = BTRFS_ROOT_ITEM_KEY;
2000         key.offset = (u64)-1;
2001         root = btrfs_read_fs_root_no_name(info, &key);
2002         if (IS_ERR(root)) {
2003                 printk(KERN_ERR "could not find root %llu\n", tree_id);
2004                 ret = -ENOENT;
2005                 goto out;
2006         }
2007
2008         key.objectid = dirid;
2009         key.type = BTRFS_INODE_REF_KEY;
2010         key.offset = (u64)-1;
2011
2012         while(1) {
2013                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2014                 if (ret < 0)
2015                         goto out;
2016
2017                 l = path->nodes[0];
2018                 slot = path->slots[0];
2019                 if (ret > 0 && slot > 0)
2020                         slot--;
2021                 btrfs_item_key_to_cpu(l, &key, slot);
2022
2023                 if (ret > 0 && (key.objectid != dirid ||
2024                                 key.type != BTRFS_INODE_REF_KEY)) {
2025                         ret = -ENOENT;
2026                         goto out;
2027                 }
2028
2029                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2030                 len = btrfs_inode_ref_name_len(l, iref);
2031                 ptr -= len + 1;
2032                 total_len += len + 1;
2033                 if (ptr < name)
2034                         goto out;
2035
2036                 *(ptr + len) = '/';
2037                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
2038
2039                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2040                         break;
2041
2042                 btrfs_release_path(path);
2043                 key.objectid = key.offset;
2044                 key.offset = (u64)-1;
2045                 dirid = key.objectid;
2046         }
2047         if (ptr < name)
2048                 goto out;
2049         memmove(name, ptr, total_len);
2050         name[total_len]='\0';
2051         ret = 0;
2052 out:
2053         btrfs_free_path(path);
2054         return ret;
2055 }
2056
2057 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2058                                            void __user *argp)
2059 {
2060          struct btrfs_ioctl_ino_lookup_args *args;
2061          struct inode *inode;
2062          int ret;
2063
2064         if (!capable(CAP_SYS_ADMIN))
2065                 return -EPERM;
2066
2067         args = memdup_user(argp, sizeof(*args));
2068         if (IS_ERR(args))
2069                 return PTR_ERR(args);
2070
2071         inode = file_inode(file);
2072
2073         if (args->treeid == 0)
2074                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2075
2076         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2077                                         args->treeid, args->objectid,
2078                                         args->name);
2079
2080         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2081                 ret = -EFAULT;
2082
2083         kfree(args);
2084         return ret;
2085 }
2086
2087 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2088                                              void __user *arg)
2089 {
2090         struct dentry *parent = fdentry(file);
2091         struct dentry *dentry;
2092         struct inode *dir = parent->d_inode;
2093         struct inode *inode;
2094         struct btrfs_root *root = BTRFS_I(dir)->root;
2095         struct btrfs_root *dest = NULL;
2096         struct btrfs_ioctl_vol_args *vol_args;
2097         struct btrfs_trans_handle *trans;
2098         struct btrfs_block_rsv block_rsv;
2099         u64 qgroup_reserved;
2100         int namelen;
2101         int ret;
2102         int err = 0;
2103
2104         vol_args = memdup_user(arg, sizeof(*vol_args));
2105         if (IS_ERR(vol_args))
2106                 return PTR_ERR(vol_args);
2107
2108         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2109         namelen = strlen(vol_args->name);
2110         if (strchr(vol_args->name, '/') ||
2111             strncmp(vol_args->name, "..", namelen) == 0) {
2112                 err = -EINVAL;
2113                 goto out;
2114         }
2115
2116         err = mnt_want_write_file(file);
2117         if (err)
2118                 goto out;
2119
2120         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2121         if (err == -EINTR)
2122                 goto out;
2123         dentry = lookup_one_len(vol_args->name, parent, namelen);
2124         if (IS_ERR(dentry)) {
2125                 err = PTR_ERR(dentry);
2126                 goto out_unlock_dir;
2127         }
2128
2129         if (!dentry->d_inode) {
2130                 err = -ENOENT;
2131                 goto out_dput;
2132         }
2133
2134         inode = dentry->d_inode;
2135         dest = BTRFS_I(inode)->root;
2136         if (!capable(CAP_SYS_ADMIN)){
2137                 /*
2138                  * Regular user.  Only allow this with a special mount
2139                  * option, when the user has write+exec access to the
2140                  * subvol root, and when rmdir(2) would have been
2141                  * allowed.
2142                  *
2143                  * Note that this is _not_ check that the subvol is
2144                  * empty or doesn't contain data that we wouldn't
2145                  * otherwise be able to delete.
2146                  *
2147                  * Users who want to delete empty subvols should try
2148                  * rmdir(2).
2149                  */
2150                 err = -EPERM;
2151                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2152                         goto out_dput;
2153
2154                 /*
2155                  * Do not allow deletion if the parent dir is the same
2156                  * as the dir to be deleted.  That means the ioctl
2157                  * must be called on the dentry referencing the root
2158                  * of the subvol, not a random directory contained
2159                  * within it.
2160                  */
2161                 err = -EINVAL;
2162                 if (root == dest)
2163                         goto out_dput;
2164
2165                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2166                 if (err)
2167                         goto out_dput;
2168         }
2169
2170         /* check if subvolume may be deleted by a user */
2171         err = btrfs_may_delete(dir, dentry, 1);
2172         if (err)
2173                 goto out_dput;
2174
2175         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2176                 err = -EINVAL;
2177                 goto out_dput;
2178         }
2179
2180         mutex_lock(&inode->i_mutex);
2181         err = d_invalidate(dentry);
2182         if (err)
2183                 goto out_unlock;
2184
2185         down_write(&root->fs_info->subvol_sem);
2186
2187         err = may_destroy_subvol(dest);
2188         if (err)
2189                 goto out_up_write;
2190
2191         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2192         /*
2193          * One for dir inode, two for dir entries, two for root
2194          * ref/backref.
2195          */
2196         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2197                                                5, &qgroup_reserved, true);
2198         if (err)
2199                 goto out_up_write;
2200
2201         trans = btrfs_start_transaction(root, 0);
2202         if (IS_ERR(trans)) {
2203                 err = PTR_ERR(trans);
2204                 goto out_release;
2205         }
2206         trans->block_rsv = &block_rsv;
2207         trans->bytes_reserved = block_rsv.size;
2208
2209         ret = btrfs_unlink_subvol(trans, root, dir,
2210                                 dest->root_key.objectid,
2211                                 dentry->d_name.name,
2212                                 dentry->d_name.len);
2213         if (ret) {
2214                 err = ret;
2215                 btrfs_abort_transaction(trans, root, ret);
2216                 goto out_end_trans;
2217         }
2218
2219         btrfs_record_root_in_trans(trans, dest);
2220
2221         memset(&dest->root_item.drop_progress, 0,
2222                 sizeof(dest->root_item.drop_progress));
2223         dest->root_item.drop_level = 0;
2224         btrfs_set_root_refs(&dest->root_item, 0);
2225
2226         if (!xchg(&dest->orphan_item_inserted, 1)) {
2227                 ret = btrfs_insert_orphan_item(trans,
2228                                         root->fs_info->tree_root,
2229                                         dest->root_key.objectid);
2230                 if (ret) {
2231                         btrfs_abort_transaction(trans, root, ret);
2232                         err = ret;
2233                         goto out_end_trans;
2234                 }
2235         }
2236 out_end_trans:
2237         trans->block_rsv = NULL;
2238         trans->bytes_reserved = 0;
2239         ret = btrfs_end_transaction(trans, root);
2240         if (ret && !err)
2241                 err = ret;
2242         inode->i_flags |= S_DEAD;
2243 out_release:
2244         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2245 out_up_write:
2246         up_write(&root->fs_info->subvol_sem);
2247 out_unlock:
2248         mutex_unlock(&inode->i_mutex);
2249         if (!err) {
2250                 shrink_dcache_sb(root->fs_info->sb);
2251                 btrfs_invalidate_inodes(dest);
2252                 d_delete(dentry);
2253
2254                 /* the last ref */
2255                 if (dest->cache_inode) {
2256                         iput(dest->cache_inode);
2257                         dest->cache_inode = NULL;
2258                 }
2259         }
2260 out_dput:
2261         dput(dentry);
2262 out_unlock_dir:
2263         mutex_unlock(&dir->i_mutex);
2264         mnt_drop_write_file(file);
2265 out:
2266         kfree(vol_args);
2267         return err;
2268 }
2269
2270 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2271 {
2272         struct inode *inode = file_inode(file);
2273         struct btrfs_root *root = BTRFS_I(inode)->root;
2274         struct btrfs_ioctl_defrag_range_args *range;
2275         int ret;
2276
2277         ret = mnt_want_write_file(file);
2278         if (ret)
2279                 return ret;
2280
2281         if (btrfs_root_readonly(root)) {
2282                 ret = -EROFS;
2283                 goto out;
2284         }
2285
2286         switch (inode->i_mode & S_IFMT) {
2287         case S_IFDIR:
2288                 if (!capable(CAP_SYS_ADMIN)) {
2289                         ret = -EPERM;
2290                         goto out;
2291                 }
2292                 ret = btrfs_defrag_root(root);
2293                 if (ret)
2294                         goto out;
2295                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2296                 break;
2297         case S_IFREG:
2298                 if (!(file->f_mode & FMODE_WRITE)) {
2299                         ret = -EINVAL;
2300                         goto out;
2301                 }
2302
2303                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2304                 if (!range) {
2305                         ret = -ENOMEM;
2306                         goto out;
2307                 }
2308
2309                 if (argp) {
2310                         if (copy_from_user(range, argp,
2311                                            sizeof(*range))) {
2312                                 ret = -EFAULT;
2313                                 kfree(range);
2314                                 goto out;
2315                         }
2316                         /* compression requires us to start the IO */
2317                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2318                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2319                                 range->extent_thresh = (u32)-1;
2320                         }
2321                 } else {
2322                         /* the rest are all set to zero by kzalloc */
2323                         range->len = (u64)-1;
2324                 }
2325                 ret = btrfs_defrag_file(file_inode(file), file,
2326                                         range, 0, 0);
2327                 if (ret > 0)
2328                         ret = 0;
2329                 kfree(range);
2330                 break;
2331         default:
2332                 ret = -EINVAL;
2333         }
2334 out:
2335         mnt_drop_write_file(file);
2336         return ret;
2337 }
2338
2339 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2340 {
2341         struct btrfs_ioctl_vol_args *vol_args;
2342         int ret;
2343
2344         if (!capable(CAP_SYS_ADMIN))
2345                 return -EPERM;
2346
2347         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2348                         1)) {
2349                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2350                 return -EINVAL;
2351         }
2352
2353         mutex_lock(&root->fs_info->volume_mutex);
2354         vol_args = memdup_user(arg, sizeof(*vol_args));
2355         if (IS_ERR(vol_args)) {
2356                 ret = PTR_ERR(vol_args);
2357                 goto out;
2358         }
2359
2360         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2361         ret = btrfs_init_new_device(root, vol_args->name);
2362
2363         kfree(vol_args);
2364 out:
2365         mutex_unlock(&root->fs_info->volume_mutex);
2366         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2367         return ret;
2368 }
2369
2370 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2371 {
2372         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2373         struct btrfs_ioctl_vol_args *vol_args;
2374         int ret;
2375
2376         if (!capable(CAP_SYS_ADMIN))
2377                 return -EPERM;
2378
2379         ret = mnt_want_write_file(file);
2380         if (ret)
2381                 return ret;
2382
2383         vol_args = memdup_user(arg, sizeof(*vol_args));
2384         if (IS_ERR(vol_args)) {
2385                 ret = PTR_ERR(vol_args);
2386                 goto out;
2387         }
2388
2389         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2390
2391         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2392                         1)) {
2393                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2394                 goto out;
2395         }
2396
2397         mutex_lock(&root->fs_info->volume_mutex);
2398         ret = btrfs_rm_device(root, vol_args->name);
2399         mutex_unlock(&root->fs_info->volume_mutex);
2400         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2401
2402 out:
2403         kfree(vol_args);
2404         mnt_drop_write_file(file);
2405         return ret;
2406 }
2407
2408 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2409 {
2410         struct btrfs_ioctl_fs_info_args *fi_args;
2411         struct btrfs_device *device;
2412         struct btrfs_device *next;
2413         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2414         int ret = 0;
2415
2416         if (!capable(CAP_SYS_ADMIN))
2417                 return -EPERM;
2418
2419         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2420         if (!fi_args)
2421                 return -ENOMEM;
2422
2423         fi_args->num_devices = fs_devices->num_devices;
2424         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2425
2426         mutex_lock(&fs_devices->device_list_mutex);
2427         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2428                 if (device->devid > fi_args->max_id)
2429                         fi_args->max_id = device->devid;
2430         }
2431         mutex_unlock(&fs_devices->device_list_mutex);
2432
2433         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2434                 ret = -EFAULT;
2435
2436         kfree(fi_args);
2437         return ret;
2438 }
2439
2440 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2441 {
2442         struct btrfs_ioctl_dev_info_args *di_args;
2443         struct btrfs_device *dev;
2444         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2445         int ret = 0;
2446         char *s_uuid = NULL;
2447         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2448
2449         if (!capable(CAP_SYS_ADMIN))
2450                 return -EPERM;
2451
2452         di_args = memdup_user(arg, sizeof(*di_args));
2453         if (IS_ERR(di_args))
2454                 return PTR_ERR(di_args);
2455
2456         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2457                 s_uuid = di_args->uuid;
2458
2459         mutex_lock(&fs_devices->device_list_mutex);
2460         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2461
2462         if (!dev) {
2463                 ret = -ENODEV;
2464                 goto out;
2465         }
2466
2467         di_args->devid = dev->devid;
2468         di_args->bytes_used = dev->bytes_used;
2469         di_args->total_bytes = dev->total_bytes;
2470         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2471         if (dev->name) {
2472                 struct rcu_string *name;
2473
2474                 rcu_read_lock();
2475                 name = rcu_dereference(dev->name);
2476                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2477                 rcu_read_unlock();
2478                 di_args->path[sizeof(di_args->path) - 1] = 0;
2479         } else {
2480                 di_args->path[0] = '\0';
2481         }
2482
2483 out:
2484         mutex_unlock(&fs_devices->device_list_mutex);
2485         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2486                 ret = -EFAULT;
2487
2488         kfree(di_args);
2489         return ret;
2490 }
2491
2492 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2493 {
2494         struct page *page;
2495         pgoff_t index;
2496         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2497
2498         index = off >> PAGE_CACHE_SHIFT;
2499
2500         page = grab_cache_page(inode->i_mapping, index);
2501         if (!page)
2502                 return NULL;
2503
2504         if (!PageUptodate(page)) {
2505                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2506                                                  0))
2507                         return NULL;
2508                 lock_page(page);
2509                 if (!PageUptodate(page)) {
2510                         unlock_page(page);
2511                         page_cache_release(page);
2512                         return NULL;
2513                 }
2514         }
2515         unlock_page(page);
2516
2517         return page;
2518 }
2519
2520 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2521 {
2522         /* do any pending delalloc/csum calc on src, one way or
2523            another, and lock file content */
2524         while (1) {
2525                 struct btrfs_ordered_extent *ordered;
2526                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2527                 ordered = btrfs_lookup_first_ordered_extent(inode,
2528                                                             off + len - 1);
2529                 if (!ordered &&
2530                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2531                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2532                         break;
2533                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2534                 if (ordered)
2535                         btrfs_put_ordered_extent(ordered);
2536                 btrfs_wait_ordered_range(inode, off, len);
2537         }
2538 }
2539
2540 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2541                                 struct inode *inode2, u64 loff2, u64 len)
2542 {
2543         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2544         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2545
2546         mutex_unlock(&inode1->i_mutex);
2547         mutex_unlock(&inode2->i_mutex);
2548 }
2549
2550 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2551                               struct inode *inode2, u64 loff2, u64 len)
2552 {
2553         if (inode1 < inode2) {
2554                 swap(inode1, inode2);
2555                 swap(loff1, loff2);
2556         }
2557
2558         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2559         lock_extent_range(inode1, loff1, len);
2560         if (inode1 != inode2) {
2561                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2562                 lock_extent_range(inode2, loff2, len);
2563         }
2564 }
2565
2566 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2567                           u64 dst_loff, u64 len)
2568 {
2569         int ret = 0;
2570         struct page *src_page, *dst_page;
2571         unsigned int cmp_len = PAGE_CACHE_SIZE;
2572         void *addr, *dst_addr;
2573
2574         while (len) {
2575                 if (len < PAGE_CACHE_SIZE)
2576                         cmp_len = len;
2577
2578                 src_page = extent_same_get_page(src, loff);
2579                 if (!src_page)
2580                         return -EINVAL;
2581                 dst_page = extent_same_get_page(dst, dst_loff);
2582                 if (!dst_page) {
2583                         page_cache_release(src_page);
2584                         return -EINVAL;
2585                 }
2586                 addr = kmap_atomic(src_page);
2587                 dst_addr = kmap_atomic(dst_page);
2588
2589                 flush_dcache_page(src_page);
2590                 flush_dcache_page(dst_page);
2591
2592                 if (memcmp(addr, dst_addr, cmp_len))
2593                         ret = BTRFS_SAME_DATA_DIFFERS;
2594
2595                 kunmap_atomic(addr);
2596                 kunmap_atomic(dst_addr);
2597                 page_cache_release(src_page);
2598                 page_cache_release(dst_page);
2599
2600                 if (ret)
2601                         break;
2602
2603                 loff += cmp_len;
2604                 dst_loff += cmp_len;
2605                 len -= cmp_len;
2606         }
2607
2608         return ret;
2609 }
2610
2611 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2612 {
2613         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2614
2615         if (off + len > inode->i_size || off + len < off)
2616                 return -EINVAL;
2617         /* Check that we are block aligned - btrfs_clone() requires this */
2618         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2619                 return -EINVAL;
2620
2621         return 0;
2622 }
2623
2624 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2625                              struct inode *dst, u64 dst_loff)
2626 {
2627         int ret;
2628
2629         /*
2630          * btrfs_clone() can't handle extents in the same file
2631          * yet. Once that works, we can drop this check and replace it
2632          * with a check for the same inode, but overlapping extents.
2633          */
2634         if (src == dst)
2635                 return -EINVAL;
2636
2637         btrfs_double_lock(src, loff, dst, dst_loff, len);
2638
2639         ret = extent_same_check_offsets(src, loff, len);
2640         if (ret)
2641                 goto out_unlock;
2642
2643         ret = extent_same_check_offsets(dst, dst_loff, len);
2644         if (ret)
2645                 goto out_unlock;
2646
2647         /* don't make the dst file partly checksummed */
2648         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2649             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2650                 ret = -EINVAL;
2651                 goto out_unlock;
2652         }
2653
2654         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2655         if (ret == 0)
2656                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2657
2658 out_unlock:
2659         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2660
2661         return ret;
2662 }
2663
2664 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2665
2666 static long btrfs_ioctl_file_extent_same(struct file *file,
2667                                          void __user *argp)
2668 {
2669         struct btrfs_ioctl_same_args *args = argp;
2670         struct btrfs_ioctl_same_args same;
2671         struct btrfs_ioctl_same_extent_info info;
2672         struct inode *src = file->f_dentry->d_inode;
2673         struct file *dst_file = NULL;
2674         struct inode *dst;
2675         u64 off;
2676         u64 len;
2677         int i;
2678         int ret;
2679         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2680         bool is_admin = capable(CAP_SYS_ADMIN);
2681
2682         if (!(file->f_mode & FMODE_READ))
2683                 return -EINVAL;
2684
2685         ret = mnt_want_write_file(file);
2686         if (ret)
2687                 return ret;
2688
2689         if (copy_from_user(&same,
2690                            (struct btrfs_ioctl_same_args __user *)argp,
2691                            sizeof(same))) {
2692                 ret = -EFAULT;
2693                 goto out;
2694         }
2695
2696         off = same.logical_offset;
2697         len = same.length;
2698
2699         /*
2700          * Limit the total length we will dedupe for each operation.
2701          * This is intended to bound the total time spent in this
2702          * ioctl to something sane.
2703          */
2704         if (len > BTRFS_MAX_DEDUPE_LEN)
2705                 len = BTRFS_MAX_DEDUPE_LEN;
2706
2707         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2708                 /*
2709                  * Btrfs does not support blocksize < page_size. As a
2710                  * result, btrfs_cmp_data() won't correctly handle
2711                  * this situation without an update.
2712                  */
2713                 ret = -EINVAL;
2714                 goto out;
2715         }
2716
2717         ret = -EISDIR;
2718         if (S_ISDIR(src->i_mode))
2719                 goto out;
2720
2721         ret = -EACCES;
2722         if (!S_ISREG(src->i_mode))
2723                 goto out;
2724
2725         ret = 0;
2726         for (i = 0; i < same.dest_count; i++) {
2727                 if (copy_from_user(&info, &args->info[i], sizeof(info))) {
2728                         ret = -EFAULT;
2729                         goto out;
2730                 }
2731
2732                 info.bytes_deduped = 0;
2733
2734                 dst_file = fget(info.fd);
2735                 if (!dst_file) {
2736                         info.status = -EBADF;
2737                         goto next;
2738                 }
2739
2740                 if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2741                         info.status = -EINVAL;
2742                         goto next;
2743                 }
2744
2745                 info.status = -EXDEV;
2746                 if (file->f_path.mnt != dst_file->f_path.mnt)
2747                         goto next;
2748
2749                 dst = dst_file->f_dentry->d_inode;
2750                 if (src->i_sb != dst->i_sb)
2751                         goto next;
2752
2753                 if (S_ISDIR(dst->i_mode)) {
2754                         info.status = -EISDIR;
2755                         goto next;
2756                 }
2757
2758                 if (!S_ISREG(dst->i_mode)) {
2759                         info.status = -EACCES;
2760                         goto next;
2761                 }
2762
2763                 info.status = btrfs_extent_same(src, off, len, dst,
2764                                                 info.logical_offset);
2765                 if (info.status == 0)
2766                         info.bytes_deduped += len;
2767
2768 next:
2769                 if (dst_file)
2770                         fput(dst_file);
2771
2772                 if (__put_user_unaligned(info.status, &args->info[i].status) ||
2773                     __put_user_unaligned(info.bytes_deduped,
2774                                          &args->info[i].bytes_deduped)) {
2775                         ret = -EFAULT;
2776                         goto out;
2777                 }
2778         }
2779
2780 out:
2781         mnt_drop_write_file(file);
2782         return ret;
2783 }
2784
2785 /**
2786  * btrfs_clone() - clone a range from inode file to another
2787  *
2788  * @src: Inode to clone from
2789  * @inode: Inode to clone to
2790  * @off: Offset within source to start clone from
2791  * @olen: Original length, passed by user, of range to clone
2792  * @olen_aligned: Block-aligned value of olen, extent_same uses
2793  *               identical values here
2794  * @destoff: Offset within @inode to start clone
2795  */
2796 static int btrfs_clone(struct inode *src, struct inode *inode,
2797                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2798 {
2799         struct btrfs_root *root = BTRFS_I(inode)->root;
2800         struct btrfs_path *path = NULL;
2801         struct extent_buffer *leaf;
2802         struct btrfs_trans_handle *trans;
2803         char *buf = NULL;
2804         struct btrfs_key key;
2805         u32 nritems;
2806         int slot;
2807         int ret;
2808         u64 len = olen_aligned;
2809
2810         ret = -ENOMEM;
2811         buf = vmalloc(btrfs_level_size(root, 0));
2812         if (!buf)
2813                 return ret;
2814
2815         path = btrfs_alloc_path();
2816         if (!path) {
2817                 vfree(buf);
2818                 return ret;
2819         }
2820
2821         path->reada = 2;
2822         /* clone data */
2823         key.objectid = btrfs_ino(src);
2824         key.type = BTRFS_EXTENT_DATA_KEY;
2825         key.offset = 0;
2826
2827         while (1) {
2828                 /*
2829                  * note the key will change type as we walk through the
2830                  * tree.
2831                  */
2832                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2833                                 0, 0);
2834                 if (ret < 0)
2835                         goto out;
2836
2837                 nritems = btrfs_header_nritems(path->nodes[0]);
2838                 if (path->slots[0] >= nritems) {
2839                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2840                         if (ret < 0)
2841                                 goto out;
2842                         if (ret > 0)
2843                                 break;
2844                         nritems = btrfs_header_nritems(path->nodes[0]);
2845                 }
2846                 leaf = path->nodes[0];
2847                 slot = path->slots[0];
2848
2849                 btrfs_item_key_to_cpu(leaf, &key, slot);
2850                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2851                     key.objectid != btrfs_ino(src))
2852                         break;
2853
2854                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2855                         struct btrfs_file_extent_item *extent;
2856                         int type;
2857                         u32 size;
2858                         struct btrfs_key new_key;
2859                         u64 disko = 0, diskl = 0;
2860                         u64 datao = 0, datal = 0;
2861                         u8 comp;
2862                         u64 endoff;
2863
2864                         size = btrfs_item_size_nr(leaf, slot);
2865                         read_extent_buffer(leaf, buf,
2866                                            btrfs_item_ptr_offset(leaf, slot),
2867                                            size);
2868
2869                         extent = btrfs_item_ptr(leaf, slot,
2870                                                 struct btrfs_file_extent_item);
2871                         comp = btrfs_file_extent_compression(leaf, extent);
2872                         type = btrfs_file_extent_type(leaf, extent);
2873                         if (type == BTRFS_FILE_EXTENT_REG ||
2874                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2875                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2876                                                                       extent);
2877                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2878                                                                  extent);
2879                                 datao = btrfs_file_extent_offset(leaf, extent);
2880                                 datal = btrfs_file_extent_num_bytes(leaf,
2881                                                                     extent);
2882                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2883                                 /* take upper bound, may be compressed */
2884                                 datal = btrfs_file_extent_ram_bytes(leaf,
2885                                                                     extent);
2886                         }
2887                         btrfs_release_path(path);
2888
2889                         if (key.offset + datal <= off ||
2890                             key.offset >= off + len - 1)
2891                                 goto next;
2892
2893                         memcpy(&new_key, &key, sizeof(new_key));
2894                         new_key.objectid = btrfs_ino(inode);
2895                         if (off <= key.offset)
2896                                 new_key.offset = key.offset + destoff - off;
2897                         else
2898                                 new_key.offset = destoff;
2899
2900                         /*
2901                          * 1 - adjusting old extent (we may have to split it)
2902                          * 1 - add new extent
2903                          * 1 - inode update
2904                          */
2905                         trans = btrfs_start_transaction(root, 3);
2906                         if (IS_ERR(trans)) {
2907                                 ret = PTR_ERR(trans);
2908                                 goto out;
2909                         }
2910
2911                         if (type == BTRFS_FILE_EXTENT_REG ||
2912                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2913                                 /*
2914                                  *    a  | --- range to clone ---|  b
2915                                  * | ------------- extent ------------- |
2916                                  */
2917
2918                                 /* substract range b */
2919                                 if (key.offset + datal > off + len)
2920                                         datal = off + len - key.offset;
2921
2922                                 /* substract range a */
2923                                 if (off > key.offset) {
2924                                         datao += off - key.offset;
2925                                         datal -= off - key.offset;
2926                                 }
2927
2928                                 ret = btrfs_drop_extents(trans, root, inode,
2929                                                          new_key.offset,
2930                                                          new_key.offset + datal,
2931                                                          1);
2932                                 if (ret) {
2933                                         btrfs_abort_transaction(trans, root,
2934                                                                 ret);
2935                                         btrfs_end_transaction(trans, root);
2936                                         goto out;
2937                                 }
2938
2939                                 ret = btrfs_insert_empty_item(trans, root, path,
2940                                                               &new_key, size);
2941                                 if (ret) {
2942                                         btrfs_abort_transaction(trans, root,
2943                                                                 ret);
2944                                         btrfs_end_transaction(trans, root);
2945                                         goto out;
2946                                 }
2947
2948                                 leaf = path->nodes[0];
2949                                 slot = path->slots[0];
2950                                 write_extent_buffer(leaf, buf,
2951                                             btrfs_item_ptr_offset(leaf, slot),
2952                                             size);
2953
2954                                 extent = btrfs_item_ptr(leaf, slot,
2955                                                 struct btrfs_file_extent_item);
2956
2957                                 /* disko == 0 means it's a hole */
2958                                 if (!disko)
2959                                         datao = 0;
2960
2961                                 btrfs_set_file_extent_offset(leaf, extent,
2962                                                              datao);
2963                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2964                                                                 datal);
2965                                 if (disko) {
2966                                         inode_add_bytes(inode, datal);
2967                                         ret = btrfs_inc_extent_ref(trans, root,
2968                                                         disko, diskl, 0,
2969                                                         root->root_key.objectid,
2970                                                         btrfs_ino(inode),
2971                                                         new_key.offset - datao,
2972                                                         0);
2973                                         if (ret) {
2974                                                 btrfs_abort_transaction(trans,
2975                                                                         root,
2976                                                                         ret);
2977                                                 btrfs_end_transaction(trans,
2978                                                                       root);
2979                                                 goto out;
2980
2981                                         }
2982                                 }
2983                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2984                                 u64 skip = 0;
2985                                 u64 trim = 0;
2986                                 if (off > key.offset) {
2987                                         skip = off - key.offset;
2988                                         new_key.offset += skip;
2989                                 }
2990
2991                                 if (key.offset + datal > off + len)
2992                                         trim = key.offset + datal - (off + len);
2993
2994                                 if (comp && (skip || trim)) {
2995                                         ret = -EINVAL;
2996                                         btrfs_end_transaction(trans, root);
2997                                         goto out;
2998                                 }
2999                                 size -= skip + trim;
3000                                 datal -= skip + trim;
3001
3002                                 ret = btrfs_drop_extents(trans, root, inode,
3003                                                          new_key.offset,
3004                                                          new_key.offset + datal,
3005                                                          1);
3006                                 if (ret) {
3007                                         btrfs_abort_transaction(trans, root,
3008                                                                 ret);
3009                                         btrfs_end_transaction(trans, root);
3010                                         goto out;
3011                                 }
3012
3013                                 ret = btrfs_insert_empty_item(trans, root, path,
3014                                                               &new_key, size);
3015                                 if (ret) {
3016                                         btrfs_abort_transaction(trans, root,
3017                                                                 ret);
3018                                         btrfs_end_transaction(trans, root);
3019                                         goto out;
3020                                 }
3021
3022                                 if (skip) {
3023                                         u32 start =
3024                                           btrfs_file_extent_calc_inline_size(0);
3025                                         memmove(buf+start, buf+start+skip,
3026                                                 datal);
3027                                 }
3028
3029                                 leaf = path->nodes[0];
3030                                 slot = path->slots[0];
3031                                 write_extent_buffer(leaf, buf,
3032                                             btrfs_item_ptr_offset(leaf, slot),
3033                                             size);
3034                                 inode_add_bytes(inode, datal);
3035                         }
3036
3037                         btrfs_mark_buffer_dirty(leaf);
3038                         btrfs_release_path(path);
3039
3040                         inode_inc_iversion(inode);
3041                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3042
3043                         /*
3044                          * we round up to the block size at eof when
3045                          * determining which extents to clone above,
3046                          * but shouldn't round up the file size
3047                          */
3048                         endoff = new_key.offset + datal;
3049                         if (endoff > destoff+olen)
3050                                 endoff = destoff+olen;
3051                         if (endoff > inode->i_size)
3052                                 btrfs_i_size_write(inode, endoff);
3053
3054                         ret = btrfs_update_inode(trans, root, inode);
3055                         if (ret) {
3056                                 btrfs_abort_transaction(trans, root, ret);
3057                                 btrfs_end_transaction(trans, root);
3058                                 goto out;
3059                         }
3060                         ret = btrfs_end_transaction(trans, root);
3061                 }
3062 next:
3063                 btrfs_release_path(path);
3064                 key.offset++;
3065         }
3066         ret = 0;
3067
3068 out:
3069         btrfs_release_path(path);
3070         btrfs_free_path(path);
3071         vfree(buf);
3072         return ret;
3073 }
3074
3075 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3076                                        u64 off, u64 olen, u64 destoff)
3077 {
3078         struct inode *inode = fdentry(file)->d_inode;
3079         struct btrfs_root *root = BTRFS_I(inode)->root;
3080         struct fd src_file;
3081         struct inode *src;
3082         int ret;
3083         u64 len = olen;
3084         u64 bs = root->fs_info->sb->s_blocksize;
3085         int same_inode = 0;
3086
3087         /*
3088          * TODO:
3089          * - split compressed inline extents.  annoying: we need to
3090          *   decompress into destination's address_space (the file offset
3091          *   may change, so source mapping won't do), then recompress (or
3092          *   otherwise reinsert) a subrange.
3093          * - allow ranges within the same file to be cloned (provided
3094          *   they don't overlap)?
3095          */
3096
3097         /* the destination must be opened for writing */
3098         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3099                 return -EINVAL;
3100
3101         if (btrfs_root_readonly(root))
3102                 return -EROFS;
3103
3104         ret = mnt_want_write_file(file);
3105         if (ret)
3106                 return ret;
3107
3108         src_file = fdget(srcfd);
3109         if (!src_file.file) {
3110                 ret = -EBADF;
3111                 goto out_drop_write;
3112         }
3113
3114         ret = -EXDEV;
3115         if (src_file.file->f_path.mnt != file->f_path.mnt)
3116                 goto out_fput;
3117
3118         src = file_inode(src_file.file);
3119
3120         ret = -EINVAL;
3121         if (src == inode)
3122                 same_inode = 1;
3123
3124         /* the src must be open for reading */
3125         if (!(src_file.file->f_mode & FMODE_READ))
3126                 goto out_fput;
3127
3128         /* don't make the dst file partly checksummed */
3129         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3130             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3131                 goto out_fput;
3132
3133         ret = -EISDIR;
3134         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3135                 goto out_fput;
3136
3137         ret = -EXDEV;
3138         if (src->i_sb != inode->i_sb)
3139                 goto out_fput;
3140
3141         if (!same_inode) {
3142                 if (inode < src) {
3143                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3144                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3145                 } else {
3146                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3147                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3148                 }
3149         } else {
3150                 mutex_lock(&src->i_mutex);
3151         }
3152
3153         /* determine range to clone */
3154         ret = -EINVAL;
3155         if (off + len > src->i_size || off + len < off)
3156                 goto out_unlock;
3157         if (len == 0)
3158                 olen = len = src->i_size - off;
3159         /* if we extend to eof, continue to block boundary */
3160         if (off + len == src->i_size)
3161                 len = ALIGN(src->i_size, bs) - off;
3162
3163         /* verify the end result is block aligned */
3164         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3165             !IS_ALIGNED(destoff, bs))
3166                 goto out_unlock;
3167
3168         /* verify if ranges are overlapped within the same file */
3169         if (same_inode) {
3170                 if (destoff + len > off && destoff < off + len)
3171                         goto out_unlock;
3172         }
3173
3174         if (destoff > inode->i_size) {
3175                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3176                 if (ret)
3177                         goto out_unlock;
3178         }
3179
3180         /* truncate page cache pages from target inode range */
3181         truncate_inode_pages_range(&inode->i_data, destoff,
3182                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3183
3184         lock_extent_range(src, off, len);
3185
3186         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3187
3188         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3189 out_unlock:
3190         mutex_unlock(&src->i_mutex);
3191         if (!same_inode)
3192                 mutex_unlock(&inode->i_mutex);
3193 out_fput:
3194         fdput(src_file);
3195 out_drop_write:
3196         mnt_drop_write_file(file);
3197         return ret;
3198 }
3199
3200 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3201 {
3202         struct btrfs_ioctl_clone_range_args args;
3203
3204         if (copy_from_user(&args, argp, sizeof(args)))
3205                 return -EFAULT;
3206         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3207                                  args.src_length, args.dest_offset);
3208 }
3209
3210 /*
3211  * there are many ways the trans_start and trans_end ioctls can lead
3212  * to deadlocks.  They should only be used by applications that
3213  * basically own the machine, and have a very in depth understanding
3214  * of all the possible deadlocks and enospc problems.
3215  */
3216 static long btrfs_ioctl_trans_start(struct file *file)
3217 {
3218         struct inode *inode = file_inode(file);
3219         struct btrfs_root *root = BTRFS_I(inode)->root;
3220         struct btrfs_trans_handle *trans;
3221         int ret;
3222
3223         ret = -EPERM;
3224         if (!capable(CAP_SYS_ADMIN))
3225                 goto out;
3226
3227         ret = -EINPROGRESS;
3228         if (file->private_data)
3229                 goto out;
3230
3231         ret = -EROFS;
3232         if (btrfs_root_readonly(root))
3233                 goto out;
3234
3235         ret = mnt_want_write_file(file);
3236         if (ret)
3237                 goto out;
3238
3239         atomic_inc(&root->fs_info->open_ioctl_trans);
3240
3241         ret = -ENOMEM;
3242         trans = btrfs_start_ioctl_transaction(root);
3243         if (IS_ERR(trans))
3244                 goto out_drop;
3245
3246         file->private_data = trans;
3247         return 0;
3248
3249 out_drop:
3250         atomic_dec(&root->fs_info->open_ioctl_trans);
3251         mnt_drop_write_file(file);
3252 out:
3253         return ret;
3254 }
3255
3256 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3257 {
3258         struct inode *inode = file_inode(file);
3259         struct btrfs_root *root = BTRFS_I(inode)->root;
3260         struct btrfs_root *new_root;
3261         struct btrfs_dir_item *di;
3262         struct btrfs_trans_handle *trans;
3263         struct btrfs_path *path;
3264         struct btrfs_key location;
3265         struct btrfs_disk_key disk_key;
3266         u64 objectid = 0;
3267         u64 dir_id;
3268         int ret;
3269
3270         if (!capable(CAP_SYS_ADMIN))
3271                 return -EPERM;
3272
3273         ret = mnt_want_write_file(file);
3274         if (ret)
3275                 return ret;
3276
3277         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3278                 ret = -EFAULT;
3279                 goto out;
3280         }
3281
3282         if (!objectid)
3283                 objectid = root->root_key.objectid;
3284
3285         location.objectid = objectid;
3286         location.type = BTRFS_ROOT_ITEM_KEY;
3287         location.offset = (u64)-1;
3288
3289         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3290         if (IS_ERR(new_root)) {
3291                 ret = PTR_ERR(new_root);
3292                 goto out;
3293         }
3294
3295         path = btrfs_alloc_path();
3296         if (!path) {
3297                 ret = -ENOMEM;
3298                 goto out;
3299         }
3300         path->leave_spinning = 1;
3301
3302         trans = btrfs_start_transaction(root, 1);
3303         if (IS_ERR(trans)) {
3304                 btrfs_free_path(path);
3305                 ret = PTR_ERR(trans);
3306                 goto out;
3307         }
3308
3309         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3310         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3311                                    dir_id, "default", 7, 1);
3312         if (IS_ERR_OR_NULL(di)) {
3313                 btrfs_free_path(path);
3314                 btrfs_end_transaction(trans, root);
3315                 printk(KERN_ERR "Umm, you don't have the default dir item, "
3316                        "this isn't going to work\n");
3317                 ret = -ENOENT;
3318                 goto out;
3319         }
3320
3321         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3322         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3323         btrfs_mark_buffer_dirty(path->nodes[0]);
3324         btrfs_free_path(path);
3325
3326         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3327         btrfs_end_transaction(trans, root);
3328 out:
3329         mnt_drop_write_file(file);
3330         return ret;
3331 }
3332
3333 void btrfs_get_block_group_info(struct list_head *groups_list,
3334                                 struct btrfs_ioctl_space_info *space)
3335 {
3336         struct btrfs_block_group_cache *block_group;
3337
3338         space->total_bytes = 0;
3339         space->used_bytes = 0;
3340         space->flags = 0;
3341         list_for_each_entry(block_group, groups_list, list) {
3342                 space->flags = block_group->flags;
3343                 space->total_bytes += block_group->key.offset;
3344                 space->used_bytes +=
3345                         btrfs_block_group_used(&block_group->item);
3346         }
3347 }
3348
3349 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3350 {
3351         struct btrfs_ioctl_space_args space_args;
3352         struct btrfs_ioctl_space_info space;
3353         struct btrfs_ioctl_space_info *dest;
3354         struct btrfs_ioctl_space_info *dest_orig;
3355         struct btrfs_ioctl_space_info __user *user_dest;
3356         struct btrfs_space_info *info;
3357         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3358                        BTRFS_BLOCK_GROUP_SYSTEM,
3359                        BTRFS_BLOCK_GROUP_METADATA,
3360                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3361         int num_types = 4;
3362         int alloc_size;
3363         int ret = 0;
3364         u64 slot_count = 0;
3365         int i, c;
3366
3367         if (copy_from_user(&space_args,
3368                            (struct btrfs_ioctl_space_args __user *)arg,
3369                            sizeof(space_args)))
3370                 return -EFAULT;
3371
3372         for (i = 0; i < num_types; i++) {
3373                 struct btrfs_space_info *tmp;
3374
3375                 info = NULL;
3376                 rcu_read_lock();
3377                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3378                                         list) {
3379                         if (tmp->flags == types[i]) {
3380                                 info = tmp;
3381                                 break;
3382                         }
3383                 }
3384                 rcu_read_unlock();
3385
3386                 if (!info)
3387                         continue;
3388
3389                 down_read(&info->groups_sem);
3390                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3391                         if (!list_empty(&info->block_groups[c]))
3392                                 slot_count++;
3393                 }
3394                 up_read(&info->groups_sem);
3395         }
3396
3397         /* space_slots == 0 means they are asking for a count */
3398         if (space_args.space_slots == 0) {
3399                 space_args.total_spaces = slot_count;
3400                 goto out;
3401         }
3402
3403         slot_count = min_t(u64, space_args.space_slots, slot_count);
3404
3405         alloc_size = sizeof(*dest) * slot_count;
3406
3407         /* we generally have at most 6 or so space infos, one for each raid
3408          * level.  So, a whole page should be more than enough for everyone
3409          */
3410         if (alloc_size > PAGE_CACHE_SIZE)
3411                 return -ENOMEM;
3412
3413         space_args.total_spaces = 0;
3414         dest = kmalloc(alloc_size, GFP_NOFS);
3415         if (!dest)
3416                 return -ENOMEM;
3417         dest_orig = dest;
3418
3419         /* now we have a buffer to copy into */
3420         for (i = 0; i < num_types; i++) {
3421                 struct btrfs_space_info *tmp;
3422
3423                 if (!slot_count)
3424                         break;
3425
3426                 info = NULL;
3427                 rcu_read_lock();
3428                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3429                                         list) {
3430                         if (tmp->flags == types[i]) {
3431                                 info = tmp;
3432                                 break;
3433                         }
3434                 }
3435                 rcu_read_unlock();
3436
3437                 if (!info)
3438                         continue;
3439                 down_read(&info->groups_sem);
3440                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3441                         if (!list_empty(&info->block_groups[c])) {
3442                                 btrfs_get_block_group_info(
3443                                         &info->block_groups[c], &space);
3444                                 memcpy(dest, &space, sizeof(space));
3445                                 dest++;
3446                                 space_args.total_spaces++;
3447                                 slot_count--;
3448                         }
3449                         if (!slot_count)
3450                                 break;
3451                 }
3452                 up_read(&info->groups_sem);
3453         }
3454
3455         user_dest = (struct btrfs_ioctl_space_info __user *)
3456                 (arg + sizeof(struct btrfs_ioctl_space_args));
3457
3458         if (copy_to_user(user_dest, dest_orig, alloc_size))
3459                 ret = -EFAULT;
3460
3461         kfree(dest_orig);
3462 out:
3463         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3464                 ret = -EFAULT;
3465
3466         return ret;
3467 }
3468
3469 /*
3470  * there are many ways the trans_start and trans_end ioctls can lead
3471  * to deadlocks.  They should only be used by applications that
3472  * basically own the machine, and have a very in depth understanding
3473  * of all the possible deadlocks and enospc problems.
3474  */
3475 long btrfs_ioctl_trans_end(struct file *file)
3476 {
3477         struct inode *inode = file_inode(file);
3478         struct btrfs_root *root = BTRFS_I(inode)->root;
3479         struct btrfs_trans_handle *trans;
3480
3481         trans = file->private_data;
3482         if (!trans)
3483                 return -EINVAL;
3484         file->private_data = NULL;
3485
3486         btrfs_end_transaction(trans, root);
3487
3488         atomic_dec(&root->fs_info->open_ioctl_trans);
3489
3490         mnt_drop_write_file(file);
3491         return 0;
3492 }
3493
3494 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3495                                             void __user *argp)
3496 {
3497         struct btrfs_trans_handle *trans;
3498         u64 transid;
3499         int ret;
3500
3501         trans = btrfs_attach_transaction_barrier(root);
3502         if (IS_ERR(trans)) {
3503                 if (PTR_ERR(trans) != -ENOENT)
3504                         return PTR_ERR(trans);
3505
3506                 /* No running transaction, don't bother */
3507                 transid = root->fs_info->last_trans_committed;
3508                 goto out;
3509         }
3510         transid = trans->transid;
3511         ret = btrfs_commit_transaction_async(trans, root, 0);
3512         if (ret) {
3513                 btrfs_end_transaction(trans, root);
3514                 return ret;
3515         }
3516 out:
3517         if (argp)
3518                 if (copy_to_user(argp, &transid, sizeof(transid)))
3519                         return -EFAULT;
3520         return 0;
3521 }
3522
3523 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3524                                            void __user *argp)
3525 {
3526         u64 transid;
3527
3528         if (argp) {
3529                 if (copy_from_user(&transid, argp, sizeof(transid)))
3530                         return -EFAULT;
3531         } else {
3532                 transid = 0;  /* current trans */
3533         }
3534         return btrfs_wait_for_commit(root, transid);
3535 }
3536
3537 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3538 {
3539         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3540         struct btrfs_ioctl_scrub_args *sa;
3541         int ret;
3542
3543         if (!capable(CAP_SYS_ADMIN))
3544                 return -EPERM;
3545
3546         sa = memdup_user(arg, sizeof(*sa));
3547         if (IS_ERR(sa))
3548                 return PTR_ERR(sa);
3549
3550         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3551                 ret = mnt_want_write_file(file);
3552                 if (ret)
3553                         goto out;
3554         }
3555
3556         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3557                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3558                               0);
3559
3560         if (copy_to_user(arg, sa, sizeof(*sa)))
3561                 ret = -EFAULT;
3562
3563         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3564                 mnt_drop_write_file(file);
3565 out:
3566         kfree(sa);
3567         return ret;
3568 }
3569
3570 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3571 {
3572         if (!capable(CAP_SYS_ADMIN))
3573                 return -EPERM;
3574
3575         return btrfs_scrub_cancel(root->fs_info);
3576 }
3577
3578 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3579                                        void __user *arg)
3580 {
3581         struct btrfs_ioctl_scrub_args *sa;
3582         int ret;
3583
3584         if (!capable(CAP_SYS_ADMIN))
3585                 return -EPERM;
3586
3587         sa = memdup_user(arg, sizeof(*sa));
3588         if (IS_ERR(sa))
3589                 return PTR_ERR(sa);
3590
3591         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3592
3593         if (copy_to_user(arg, sa, sizeof(*sa)))
3594                 ret = -EFAULT;
3595
3596         kfree(sa);
3597         return ret;
3598 }
3599
3600 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3601                                       void __user *arg)
3602 {
3603         struct btrfs_ioctl_get_dev_stats *sa;
3604         int ret;
3605
3606         sa = memdup_user(arg, sizeof(*sa));
3607         if (IS_ERR(sa))
3608                 return PTR_ERR(sa);
3609
3610         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3611                 kfree(sa);
3612                 return -EPERM;
3613         }
3614
3615         ret = btrfs_get_dev_stats(root, sa);
3616
3617         if (copy_to_user(arg, sa, sizeof(*sa)))
3618                 ret = -EFAULT;
3619
3620         kfree(sa);
3621         return ret;
3622 }
3623
3624 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3625 {
3626         struct btrfs_ioctl_dev_replace_args *p;
3627         int ret;
3628
3629         if (!capable(CAP_SYS_ADMIN))
3630                 return -EPERM;
3631
3632         p = memdup_user(arg, sizeof(*p));
3633         if (IS_ERR(p))
3634                 return PTR_ERR(p);
3635
3636         switch (p->cmd) {
3637         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3638                 if (atomic_xchg(
3639                         &root->fs_info->mutually_exclusive_operation_running,
3640                         1)) {
3641                         pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3642                         ret = -EINPROGRESS;
3643                 } else {
3644                         ret = btrfs_dev_replace_start(root, p);
3645                         atomic_set(
3646                          &root->fs_info->mutually_exclusive_operation_running,
3647                          0);
3648                 }
3649                 break;
3650         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3651                 btrfs_dev_replace_status(root->fs_info, p);
3652                 ret = 0;
3653                 break;
3654         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3655                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3656                 break;
3657         default:
3658                 ret = -EINVAL;
3659                 break;
3660         }
3661
3662         if (copy_to_user(arg, p, sizeof(*p)))
3663                 ret = -EFAULT;
3664
3665         kfree(p);
3666         return ret;
3667 }
3668
3669 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3670 {
3671         int ret = 0;
3672         int i;
3673         u64 rel_ptr;
3674         int size;
3675         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3676         struct inode_fs_paths *ipath = NULL;
3677         struct btrfs_path *path;
3678
3679         if (!capable(CAP_DAC_READ_SEARCH))
3680                 return -EPERM;
3681
3682         path = btrfs_alloc_path();
3683         if (!path) {
3684                 ret = -ENOMEM;
3685                 goto out;
3686         }
3687
3688         ipa = memdup_user(arg, sizeof(*ipa));
3689         if (IS_ERR(ipa)) {
3690                 ret = PTR_ERR(ipa);
3691                 ipa = NULL;
3692                 goto out;
3693         }
3694
3695         size = min_t(u32, ipa->size, 4096);
3696         ipath = init_ipath(size, root, path);
3697         if (IS_ERR(ipath)) {
3698                 ret = PTR_ERR(ipath);
3699                 ipath = NULL;
3700                 goto out;
3701         }
3702
3703         ret = paths_from_inode(ipa->inum, ipath);
3704         if (ret < 0)
3705                 goto out;
3706
3707         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3708                 rel_ptr = ipath->fspath->val[i] -
3709                           (u64)(unsigned long)ipath->fspath->val;
3710                 ipath->fspath->val[i] = rel_ptr;
3711         }
3712
3713         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3714                            (void *)(unsigned long)ipath->fspath, size);
3715         if (ret) {
3716                 ret = -EFAULT;
3717                 goto out;
3718         }
3719
3720 out:
3721         btrfs_free_path(path);
3722         free_ipath(ipath);
3723         kfree(ipa);
3724
3725         return ret;
3726 }
3727
3728 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3729 {
3730         struct btrfs_data_container *inodes = ctx;
3731         const size_t c = 3 * sizeof(u64);
3732
3733         if (inodes->bytes_left >= c) {
3734                 inodes->bytes_left -= c;
3735                 inodes->val[inodes->elem_cnt] = inum;
3736                 inodes->val[inodes->elem_cnt + 1] = offset;
3737                 inodes->val[inodes->elem_cnt + 2] = root;
3738                 inodes->elem_cnt += 3;
3739         } else {
3740                 inodes->bytes_missing += c - inodes->bytes_left;
3741                 inodes->bytes_left = 0;
3742                 inodes->elem_missed += 3;
3743         }
3744
3745         return 0;
3746 }
3747
3748 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3749                                         void __user *arg)
3750 {
3751         int ret = 0;
3752         int size;
3753         struct btrfs_ioctl_logical_ino_args *loi;
3754         struct btrfs_data_container *inodes = NULL;
3755         struct btrfs_path *path = NULL;
3756
3757         if (!capable(CAP_SYS_ADMIN))
3758                 return -EPERM;
3759
3760         loi = memdup_user(arg, sizeof(*loi));
3761         if (IS_ERR(loi)) {
3762                 ret = PTR_ERR(loi);
3763                 loi = NULL;
3764                 goto out;
3765         }
3766
3767         path = btrfs_alloc_path();
3768         if (!path) {
3769                 ret = -ENOMEM;
3770                 goto out;
3771         }
3772
3773         size = min_t(u32, loi->size, 64 * 1024);
3774         inodes = init_data_container(size);
3775         if (IS_ERR(inodes)) {
3776                 ret = PTR_ERR(inodes);
3777                 inodes = NULL;
3778                 goto out;
3779         }
3780
3781         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3782                                           build_ino_list, inodes);
3783         if (ret == -EINVAL)
3784                 ret = -ENOENT;
3785         if (ret < 0)
3786                 goto out;
3787
3788         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3789                            (void *)(unsigned long)inodes, size);
3790         if (ret)
3791                 ret = -EFAULT;
3792
3793 out:
3794         btrfs_free_path(path);
3795         vfree(inodes);
3796         kfree(loi);
3797
3798         return ret;
3799 }
3800
3801 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3802                                struct btrfs_ioctl_balance_args *bargs)
3803 {
3804         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3805
3806         bargs->flags = bctl->flags;
3807
3808         if (atomic_read(&fs_info->balance_running))
3809                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3810         if (atomic_read(&fs_info->balance_pause_req))
3811                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3812         if (atomic_read(&fs_info->balance_cancel_req))
3813                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3814
3815         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3816         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3817         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3818
3819         if (lock) {
3820                 spin_lock(&fs_info->balance_lock);
3821                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3822                 spin_unlock(&fs_info->balance_lock);
3823         } else {
3824                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3825         }
3826 }
3827
3828 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3829 {
3830         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3831         struct btrfs_fs_info *fs_info = root->fs_info;
3832         struct btrfs_ioctl_balance_args *bargs;
3833         struct btrfs_balance_control *bctl;
3834         bool need_unlock; /* for mut. excl. ops lock */
3835         int ret;
3836
3837         if (!capable(CAP_SYS_ADMIN))
3838                 return -EPERM;
3839
3840         ret = mnt_want_write_file(file);
3841         if (ret)
3842                 return ret;
3843
3844 again:
3845         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
3846                 mutex_lock(&fs_info->volume_mutex);
3847                 mutex_lock(&fs_info->balance_mutex);
3848                 need_unlock = true;
3849                 goto locked;
3850         }
3851
3852         /*
3853          * mut. excl. ops lock is locked.  Three possibilites:
3854          *   (1) some other op is running
3855          *   (2) balance is running
3856          *   (3) balance is paused -- special case (think resume)
3857          */
3858         mutex_lock(&fs_info->balance_mutex);
3859         if (fs_info->balance_ctl) {
3860                 /* this is either (2) or (3) */
3861                 if (!atomic_read(&fs_info->balance_running)) {
3862                         mutex_unlock(&fs_info->balance_mutex);
3863                         if (!mutex_trylock(&fs_info->volume_mutex))
3864                                 goto again;
3865                         mutex_lock(&fs_info->balance_mutex);
3866
3867                         if (fs_info->balance_ctl &&
3868                             !atomic_read(&fs_info->balance_running)) {
3869                                 /* this is (3) */
3870                                 need_unlock = false;
3871                                 goto locked;
3872                         }
3873
3874                         mutex_unlock(&fs_info->balance_mutex);
3875                         mutex_unlock(&fs_info->volume_mutex);
3876                         goto again;
3877                 } else {
3878                         /* this is (2) */
3879                         mutex_unlock(&fs_info->balance_mutex);
3880                         ret = -EINPROGRESS;
3881                         goto out;
3882                 }
3883         } else {
3884                 /* this is (1) */
3885                 mutex_unlock(&fs_info->balance_mutex);
3886                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3887                 ret = -EINVAL;
3888                 goto out;
3889         }
3890
3891 locked:
3892         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
3893
3894         if (arg) {
3895                 bargs = memdup_user(arg, sizeof(*bargs));
3896                 if (IS_ERR(bargs)) {
3897                         ret = PTR_ERR(bargs);
3898                         goto out_unlock;
3899                 }
3900
3901                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3902                         if (!fs_info->balance_ctl) {
3903                                 ret = -ENOTCONN;
3904                                 goto out_bargs;
3905                         }
3906
3907                         bctl = fs_info->balance_ctl;
3908                         spin_lock(&fs_info->balance_lock);
3909                         bctl->flags |= BTRFS_BALANCE_RESUME;
3910                         spin_unlock(&fs_info->balance_lock);
3911
3912                         goto do_balance;
3913                 }
3914         } else {
3915                 bargs = NULL;
3916         }
3917
3918         if (fs_info->balance_ctl) {
3919                 ret = -EINPROGRESS;
3920                 goto out_bargs;
3921         }
3922
3923         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3924         if (!bctl) {
3925                 ret = -ENOMEM;
3926                 goto out_bargs;
3927         }
3928
3929         bctl->fs_info = fs_info;
3930         if (arg) {
3931                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3932                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3933                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3934
3935                 bctl->flags = bargs->flags;
3936         } else {
3937                 /* balance everything - no filters */
3938                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3939         }
3940
3941 do_balance:
3942         /*
3943          * Ownership of bctl and mutually_exclusive_operation_running
3944          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
3945          * or, if restriper was paused all the way until unmount, in
3946          * free_fs_info.  mutually_exclusive_operation_running is
3947          * cleared in __cancel_balance.
3948          */
3949         need_unlock = false;
3950
3951         ret = btrfs_balance(bctl, bargs);
3952
3953         if (arg) {
3954                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3955                         ret = -EFAULT;
3956         }
3957
3958 out_bargs:
3959         kfree(bargs);
3960 out_unlock:
3961         mutex_unlock(&fs_info->balance_mutex);
3962         mutex_unlock(&fs_info->volume_mutex);
3963         if (need_unlock)
3964                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3965 out:
3966         mnt_drop_write_file(file);
3967         return ret;
3968 }
3969
3970 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3971 {
3972         if (!capable(CAP_SYS_ADMIN))
3973                 return -EPERM;
3974
3975         switch (cmd) {
3976         case BTRFS_BALANCE_CTL_PAUSE:
3977                 return btrfs_pause_balance(root->fs_info);
3978         case BTRFS_BALANCE_CTL_CANCEL:
3979                 return btrfs_cancel_balance(root->fs_info);
3980         }
3981
3982         return -EINVAL;
3983 }
3984
3985 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3986                                          void __user *arg)
3987 {
3988         struct btrfs_fs_info *fs_info = root->fs_info;
3989         struct btrfs_ioctl_balance_args *bargs;
3990         int ret = 0;
3991
3992         if (!capable(CAP_SYS_ADMIN))
3993                 return -EPERM;
3994
3995         mutex_lock(&fs_info->balance_mutex);
3996         if (!fs_info->balance_ctl) {
3997                 ret = -ENOTCONN;
3998                 goto out;
3999         }
4000
4001         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4002         if (!bargs) {
4003                 ret = -ENOMEM;
4004                 goto out;
4005         }
4006
4007         update_ioctl_balance_args(fs_info, 1, bargs);
4008
4009         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4010                 ret = -EFAULT;
4011
4012         kfree(bargs);
4013 out:
4014         mutex_unlock(&fs_info->balance_mutex);
4015         return ret;
4016 }
4017
4018 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4019 {
4020         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4021         struct btrfs_ioctl_quota_ctl_args *sa;
4022         struct btrfs_trans_handle *trans = NULL;
4023         int ret;
4024         int err;
4025
4026         if (!capable(CAP_SYS_ADMIN))
4027                 return -EPERM;
4028
4029         ret = mnt_want_write_file(file);
4030         if (ret)
4031                 return ret;
4032
4033         sa = memdup_user(arg, sizeof(*sa));
4034         if (IS_ERR(sa)) {
4035                 ret = PTR_ERR(sa);
4036                 goto drop_write;
4037         }
4038
4039         down_write(&root->fs_info->subvol_sem);
4040         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4041         if (IS_ERR(trans)) {
4042                 ret = PTR_ERR(trans);
4043                 goto out;
4044         }
4045
4046         switch (sa->cmd) {
4047         case BTRFS_QUOTA_CTL_ENABLE:
4048                 ret = btrfs_quota_enable(trans, root->fs_info);
4049                 break;
4050         case BTRFS_QUOTA_CTL_DISABLE:
4051                 ret = btrfs_quota_disable(trans, root->fs_info);
4052                 break;
4053         default:
4054                 ret = -EINVAL;
4055                 break;
4056         }
4057
4058         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4059         if (err && !ret)
4060                 ret = err;
4061 out:
4062         kfree(sa);
4063         up_write(&root->fs_info->subvol_sem);
4064 drop_write:
4065         mnt_drop_write_file(file);
4066         return ret;
4067 }
4068
4069 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4070 {
4071         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4072         struct btrfs_ioctl_qgroup_assign_args *sa;
4073         struct btrfs_trans_handle *trans;
4074         int ret;
4075         int err;
4076
4077         if (!capable(CAP_SYS_ADMIN))
4078                 return -EPERM;
4079
4080         ret = mnt_want_write_file(file);
4081         if (ret)
4082                 return ret;
4083
4084         sa = memdup_user(arg, sizeof(*sa));
4085         if (IS_ERR(sa)) {
4086                 ret = PTR_ERR(sa);
4087                 goto drop_write;
4088         }
4089
4090         trans = btrfs_join_transaction(root);
4091         if (IS_ERR(trans)) {
4092                 ret = PTR_ERR(trans);
4093                 goto out;
4094         }
4095
4096         /* FIXME: check if the IDs really exist */
4097         if (sa->assign) {
4098                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4099                                                 sa->src, sa->dst);
4100         } else {
4101                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4102                                                 sa->src, sa->dst);
4103         }
4104
4105         err = btrfs_end_transaction(trans, root);
4106         if (err && !ret)
4107                 ret = err;
4108
4109 out:
4110         kfree(sa);
4111 drop_write:
4112         mnt_drop_write_file(file);
4113         return ret;
4114 }
4115
4116 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4117 {
4118         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4119         struct btrfs_ioctl_qgroup_create_args *sa;
4120         struct btrfs_trans_handle *trans;
4121         int ret;
4122         int err;
4123
4124         if (!capable(CAP_SYS_ADMIN))
4125                 return -EPERM;
4126
4127         ret = mnt_want_write_file(file);
4128         if (ret)
4129                 return ret;
4130
4131         sa = memdup_user(arg, sizeof(*sa));
4132         if (IS_ERR(sa)) {
4133                 ret = PTR_ERR(sa);
4134                 goto drop_write;
4135         }
4136
4137         if (!sa->qgroupid) {
4138                 ret = -EINVAL;
4139                 goto out;
4140         }
4141
4142         trans = btrfs_join_transaction(root);
4143         if (IS_ERR(trans)) {
4144                 ret = PTR_ERR(trans);
4145                 goto out;
4146         }
4147
4148         /* FIXME: check if the IDs really exist */
4149         if (sa->create) {
4150                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4151                                           NULL);
4152         } else {
4153                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4154         }
4155
4156         err = btrfs_end_transaction(trans, root);
4157         if (err && !ret)
4158                 ret = err;
4159
4160 out:
4161         kfree(sa);
4162 drop_write:
4163         mnt_drop_write_file(file);
4164         return ret;
4165 }
4166
4167 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4168 {
4169         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4170         struct btrfs_ioctl_qgroup_limit_args *sa;
4171         struct btrfs_trans_handle *trans;
4172         int ret;
4173         int err;
4174         u64 qgroupid;
4175
4176         if (!capable(CAP_SYS_ADMIN))
4177                 return -EPERM;
4178
4179         ret = mnt_want_write_file(file);
4180         if (ret)
4181                 return ret;
4182
4183         sa = memdup_user(arg, sizeof(*sa));
4184         if (IS_ERR(sa)) {
4185                 ret = PTR_ERR(sa);
4186                 goto drop_write;
4187         }
4188
4189         trans = btrfs_join_transaction(root);
4190         if (IS_ERR(trans)) {
4191                 ret = PTR_ERR(trans);
4192                 goto out;
4193         }
4194
4195         qgroupid = sa->qgroupid;
4196         if (!qgroupid) {
4197                 /* take the current subvol as qgroup */
4198                 qgroupid = root->root_key.objectid;
4199         }
4200
4201         /* FIXME: check if the IDs really exist */
4202         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4203
4204         err = btrfs_end_transaction(trans, root);
4205         if (err && !ret)
4206                 ret = err;
4207
4208 out:
4209         kfree(sa);
4210 drop_write:
4211         mnt_drop_write_file(file);
4212         return ret;
4213 }
4214
4215 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4216 {
4217         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4218         struct btrfs_ioctl_quota_rescan_args *qsa;
4219         int ret;
4220
4221         if (!capable(CAP_SYS_ADMIN))
4222                 return -EPERM;
4223
4224         ret = mnt_want_write_file(file);
4225         if (ret)
4226                 return ret;
4227
4228         qsa = memdup_user(arg, sizeof(*qsa));
4229         if (IS_ERR(qsa)) {
4230                 ret = PTR_ERR(qsa);
4231                 goto drop_write;
4232         }
4233
4234         if (qsa->flags) {
4235                 ret = -EINVAL;
4236                 goto out;
4237         }
4238
4239         ret = btrfs_qgroup_rescan(root->fs_info);
4240
4241 out:
4242         kfree(qsa);
4243 drop_write:
4244         mnt_drop_write_file(file);
4245         return ret;
4246 }
4247
4248 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4249 {
4250         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4251         struct btrfs_ioctl_quota_rescan_args *qsa;
4252         int ret = 0;
4253
4254         if (!capable(CAP_SYS_ADMIN))
4255                 return -EPERM;
4256
4257         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4258         if (!qsa)
4259                 return -ENOMEM;
4260
4261         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4262                 qsa->flags = 1;
4263                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4264         }
4265
4266         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4267                 ret = -EFAULT;
4268
4269         kfree(qsa);
4270         return ret;
4271 }
4272
4273 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4274 {
4275         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
4276
4277         if (!capable(CAP_SYS_ADMIN))
4278                 return -EPERM;
4279
4280         return btrfs_qgroup_wait_for_completion(root->fs_info);
4281 }
4282
4283 static long btrfs_ioctl_set_received_subvol(struct file *file,
4284                                             void __user *arg)
4285 {
4286         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4287         struct inode *inode = file_inode(file);
4288         struct btrfs_root *root = BTRFS_I(inode)->root;
4289         struct btrfs_root_item *root_item = &root->root_item;
4290         struct btrfs_trans_handle *trans;
4291         struct timespec ct = CURRENT_TIME;
4292         int ret = 0;
4293
4294         ret = mnt_want_write_file(file);
4295         if (ret < 0)
4296                 return ret;
4297
4298         down_write(&root->fs_info->subvol_sem);
4299
4300         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4301                 ret = -EINVAL;
4302                 goto out;
4303         }
4304
4305         if (btrfs_root_readonly(root)) {
4306                 ret = -EROFS;
4307                 goto out;
4308         }
4309
4310         if (!inode_owner_or_capable(inode)) {
4311                 ret = -EACCES;
4312                 goto out;
4313         }
4314
4315         sa = memdup_user(arg, sizeof(*sa));
4316         if (IS_ERR(sa)) {
4317                 ret = PTR_ERR(sa);
4318                 sa = NULL;
4319                 goto out;
4320         }
4321
4322         trans = btrfs_start_transaction(root, 1);
4323         if (IS_ERR(trans)) {
4324                 ret = PTR_ERR(trans);
4325                 trans = NULL;
4326                 goto out;
4327         }
4328
4329         sa->rtransid = trans->transid;
4330         sa->rtime.sec = ct.tv_sec;
4331         sa->rtime.nsec = ct.tv_nsec;
4332
4333         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4334         btrfs_set_root_stransid(root_item, sa->stransid);
4335         btrfs_set_root_rtransid(root_item, sa->rtransid);
4336         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4337         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4338         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4339         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4340
4341         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4342                                 &root->root_key, &root->root_item);
4343         if (ret < 0) {
4344                 btrfs_end_transaction(trans, root);
4345                 trans = NULL;
4346                 goto out;
4347         } else {
4348                 ret = btrfs_commit_transaction(trans, root);
4349                 if (ret < 0)
4350                         goto out;
4351         }
4352
4353         ret = copy_to_user(arg, sa, sizeof(*sa));
4354         if (ret)
4355                 ret = -EFAULT;
4356
4357 out:
4358         kfree(sa);
4359         up_write(&root->fs_info->subvol_sem);
4360         mnt_drop_write_file(file);
4361         return ret;
4362 }
4363
4364 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4365 {
4366         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4367         size_t len;
4368         int ret;
4369         char label[BTRFS_LABEL_SIZE];
4370
4371         spin_lock(&root->fs_info->super_lock);
4372         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4373         spin_unlock(&root->fs_info->super_lock);
4374
4375         len = strnlen(label, BTRFS_LABEL_SIZE);
4376
4377         if (len == BTRFS_LABEL_SIZE) {
4378                 pr_warn("btrfs: label is too long, return the first %zu bytes\n",
4379                         --len);
4380         }
4381
4382         ret = copy_to_user(arg, label, len);
4383
4384         return ret ? -EFAULT : 0;
4385 }
4386
4387 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4388 {
4389         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4390         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4391         struct btrfs_trans_handle *trans;
4392         char label[BTRFS_LABEL_SIZE];
4393         int ret;
4394
4395         if (!capable(CAP_SYS_ADMIN))
4396                 return -EPERM;
4397
4398         if (copy_from_user(label, arg, sizeof(label)))
4399                 return -EFAULT;
4400
4401         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4402                 pr_err("btrfs: unable to set label with more than %d bytes\n",
4403                        BTRFS_LABEL_SIZE - 1);
4404                 return -EINVAL;
4405         }
4406
4407         ret = mnt_want_write_file(file);
4408         if (ret)
4409                 return ret;
4410
4411         trans = btrfs_start_transaction(root, 0);
4412         if (IS_ERR(trans)) {
4413                 ret = PTR_ERR(trans);
4414                 goto out_unlock;
4415         }
4416
4417         spin_lock(&root->fs_info->super_lock);
4418         strcpy(super_block->label, label);
4419         spin_unlock(&root->fs_info->super_lock);
4420         ret = btrfs_end_transaction(trans, root);
4421
4422 out_unlock:
4423         mnt_drop_write_file(file);
4424         return ret;
4425 }
4426
4427 long btrfs_ioctl(struct file *file, unsigned int
4428                 cmd, unsigned long arg)
4429 {
4430         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4431         void __user *argp = (void __user *)arg;
4432
4433         switch (cmd) {
4434         case FS_IOC_GETFLAGS:
4435                 return btrfs_ioctl_getflags(file, argp);
4436         case FS_IOC_SETFLAGS:
4437                 return btrfs_ioctl_setflags(file, argp);
4438         case FS_IOC_GETVERSION:
4439                 return btrfs_ioctl_getversion(file, argp);
4440         case FITRIM:
4441                 return btrfs_ioctl_fitrim(file, argp);
4442         case BTRFS_IOC_SNAP_CREATE:
4443                 return btrfs_ioctl_snap_create(file, argp, 0);
4444         case BTRFS_IOC_SNAP_CREATE_V2:
4445                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4446         case BTRFS_IOC_SUBVOL_CREATE:
4447                 return btrfs_ioctl_snap_create(file, argp, 1);
4448         case BTRFS_IOC_SUBVOL_CREATE_V2:
4449                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4450         case BTRFS_IOC_SNAP_DESTROY:
4451                 return btrfs_ioctl_snap_destroy(file, argp);
4452         case BTRFS_IOC_SUBVOL_GETFLAGS:
4453                 return btrfs_ioctl_subvol_getflags(file, argp);
4454         case BTRFS_IOC_SUBVOL_SETFLAGS:
4455                 return btrfs_ioctl_subvol_setflags(file, argp);
4456         case BTRFS_IOC_DEFAULT_SUBVOL:
4457                 return btrfs_ioctl_default_subvol(file, argp);
4458         case BTRFS_IOC_DEFRAG:
4459                 return btrfs_ioctl_defrag(file, NULL);
4460         case BTRFS_IOC_DEFRAG_RANGE:
4461                 return btrfs_ioctl_defrag(file, argp);
4462         case BTRFS_IOC_RESIZE:
4463                 return btrfs_ioctl_resize(file, argp);
4464         case BTRFS_IOC_ADD_DEV:
4465                 return btrfs_ioctl_add_dev(root, argp);
4466         case BTRFS_IOC_RM_DEV:
4467                 return btrfs_ioctl_rm_dev(file, argp);
4468         case BTRFS_IOC_FS_INFO:
4469                 return btrfs_ioctl_fs_info(root, argp);
4470         case BTRFS_IOC_DEV_INFO:
4471                 return btrfs_ioctl_dev_info(root, argp);
4472         case BTRFS_IOC_BALANCE:
4473                 return btrfs_ioctl_balance(file, NULL);
4474         case BTRFS_IOC_CLONE:
4475                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4476         case BTRFS_IOC_CLONE_RANGE:
4477                 return btrfs_ioctl_clone_range(file, argp);
4478         case BTRFS_IOC_TRANS_START:
4479                 return btrfs_ioctl_trans_start(file);
4480         case BTRFS_IOC_TRANS_END:
4481                 return btrfs_ioctl_trans_end(file);
4482         case BTRFS_IOC_TREE_SEARCH:
4483                 return btrfs_ioctl_tree_search(file, argp);
4484         case BTRFS_IOC_INO_LOOKUP:
4485                 return btrfs_ioctl_ino_lookup(file, argp);
4486         case BTRFS_IOC_INO_PATHS:
4487                 return btrfs_ioctl_ino_to_path(root, argp);
4488         case BTRFS_IOC_LOGICAL_INO:
4489                 return btrfs_ioctl_logical_to_ino(root, argp);
4490         case BTRFS_IOC_SPACE_INFO:
4491                 return btrfs_ioctl_space_info(root, argp);
4492         case BTRFS_IOC_SYNC:
4493                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
4494                 return 0;
4495         case BTRFS_IOC_START_SYNC:
4496                 return btrfs_ioctl_start_sync(root, argp);
4497         case BTRFS_IOC_WAIT_SYNC:
4498                 return btrfs_ioctl_wait_sync(root, argp);
4499         case BTRFS_IOC_SCRUB:
4500                 return btrfs_ioctl_scrub(file, argp);
4501         case BTRFS_IOC_SCRUB_CANCEL:
4502                 return btrfs_ioctl_scrub_cancel(root, argp);
4503         case BTRFS_IOC_SCRUB_PROGRESS:
4504                 return btrfs_ioctl_scrub_progress(root, argp);
4505         case BTRFS_IOC_BALANCE_V2:
4506                 return btrfs_ioctl_balance(file, argp);
4507         case BTRFS_IOC_BALANCE_CTL:
4508                 return btrfs_ioctl_balance_ctl(root, arg);
4509         case BTRFS_IOC_BALANCE_PROGRESS:
4510                 return btrfs_ioctl_balance_progress(root, argp);
4511         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4512                 return btrfs_ioctl_set_received_subvol(file, argp);
4513         case BTRFS_IOC_SEND:
4514                 return btrfs_ioctl_send(file, argp);
4515         case BTRFS_IOC_GET_DEV_STATS:
4516                 return btrfs_ioctl_get_dev_stats(root, argp);
4517         case BTRFS_IOC_QUOTA_CTL:
4518                 return btrfs_ioctl_quota_ctl(file, argp);
4519         case BTRFS_IOC_QGROUP_ASSIGN:
4520                 return btrfs_ioctl_qgroup_assign(file, argp);
4521         case BTRFS_IOC_QGROUP_CREATE:
4522                 return btrfs_ioctl_qgroup_create(file, argp);
4523         case BTRFS_IOC_QGROUP_LIMIT:
4524                 return btrfs_ioctl_qgroup_limit(file, argp);
4525         case BTRFS_IOC_QUOTA_RESCAN:
4526                 return btrfs_ioctl_quota_rescan(file, argp);
4527         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4528                 return btrfs_ioctl_quota_rescan_status(file, argp);
4529         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4530                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4531         case BTRFS_IOC_DEV_REPLACE:
4532                 return btrfs_ioctl_dev_replace(root, argp);
4533         case BTRFS_IOC_GET_FSLABEL:
4534                 return btrfs_ioctl_get_fslabel(file, argp);
4535         case BTRFS_IOC_SET_FSLABEL:
4536                 return btrfs_ioctl_set_fslabel(file, argp);
4537         case BTRFS_IOC_FILE_EXTENT_SAME:
4538                 return btrfs_ioctl_file_extent_same(file, argp);
4539         }
4540
4541         return -ENOTTY;
4542 }