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 "compat.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "ioctl.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58
  59 /* Mask out flags that are inappropriate for the given type of inode. */
  60 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  61 {
  62         if (S_ISDIR(mode))
  63                 return flags;
  64         else if (S_ISREG(mode))
  65                 return flags & ~FS_DIRSYNC_FL;
  66         else
  67                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  68 }
  69
  70 /*
  71  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  72  */
  73 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  74 {
  75         unsigned int iflags = 0;
  76
  77         if (flags & BTRFS_INODE_SYNC)
  78                 iflags |= FS_SYNC_FL;
  79         if (flags & BTRFS_INODE_IMMUTABLE)
  80                 iflags |= FS_IMMUTABLE_FL;
  81         if (flags & BTRFS_INODE_APPEND)
  82                 iflags |= FS_APPEND_FL;
  83         if (flags & BTRFS_INODE_NODUMP)
  84                 iflags |= FS_NODUMP_FL;
  85         if (flags & BTRFS_INODE_NOATIME)
  86                 iflags |= FS_NOATIME_FL;
  87         if (flags & BTRFS_INODE_DIRSYNC)
  88                 iflags |= FS_DIRSYNC_FL;
  89         if (flags & BTRFS_INODE_NODATACOW)
  90                 iflags |= FS_NOCOW_FL;
  91
  92         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
  93                 iflags |= FS_COMPR_FL;
  94         else if (flags & BTRFS_INODE_NOCOMPRESS)
  95                 iflags |= FS_NOCOMP_FL;
  96
  97         return iflags;
  98 }
  99
 100 /*
 101  * Update inode->i_flags based on the btrfs internal flags.
 102  */
 103 void btrfs_update_iflags(struct inode *inode)
 104 {
 105         struct btrfs_inode *ip = BTRFS_I(inode);
 106
 107         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 108
 109         if (ip->flags & BTRFS_INODE_SYNC)
 110                 inode->i_flags |= S_SYNC;
 111         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 112                 inode->i_flags |= S_IMMUTABLE;
 113         if (ip->flags & BTRFS_INODE_APPEND)
 114                 inode->i_flags |= S_APPEND;
 115         if (ip->flags & BTRFS_INODE_NOATIME)
 116                 inode->i_flags |= S_NOATIME;
 117         if (ip->flags & BTRFS_INODE_DIRSYNC)
 118                 inode->i_flags |= S_DIRSYNC;
 119 }
 120
 121 /*
 122  * Inherit flags from the parent inode.
 123  *
 124  * Currently only the compression flags and the cow flags are inherited.
 125  */
 126 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 127 {
 128         unsigned int flags;
 129
 130         if (!dir)
 131                 return;
 132
 133         flags = BTRFS_I(dir)->flags;
 134
 135         if (flags & BTRFS_INODE_NOCOMPRESS) {
 136                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 137                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 138         } else if (flags & BTRFS_INODE_COMPRESS) {
 139                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 140                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 141         }
 142
 143         if (flags & BTRFS_INODE_NODATACOW)
 144                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 145
 146         btrfs_update_iflags(inode);
 147 }
 148
 149 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 150 {
 151         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 152         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 153
 154         if (copy_to_user(arg, &flags, sizeof(flags)))
 155                 return -EFAULT;
 156         return 0;
 157 }
 158
 159 static int check_flags(unsigned int flags)
 160 {
 161         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 162                       FS_NOATIME_FL | FS_NODUMP_FL | \
 163                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 164                       FS_NOCOMP_FL | FS_COMPR_FL |
 165                       FS_NOCOW_FL))
 166                 return -EOPNOTSUPP;
 167
 168         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 169                 return -EINVAL;
 170
 171         return 0;
 172 }
 173
 174 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 175 {
 176         struct inode *inode = file->f_path.dentry->d_inode;
 177         struct btrfs_inode *ip = BTRFS_I(inode);
 178         struct btrfs_root *root = ip->root;
 179         struct btrfs_trans_handle *trans;
 180         unsigned int flags, oldflags;
 181         int ret;
 182         u64 ip_oldflags;
 183         unsigned int i_oldflags;
 184         umode_t mode;
 185
 186         if (btrfs_root_readonly(root))
 187                 return -EROFS;
 188
 189         if (copy_from_user(&flags, arg, sizeof(flags)))
 190                 return -EFAULT;
 191
 192         ret = check_flags(flags);
 193         if (ret)
 194                 return ret;
 195
 196         if (!inode_owner_or_capable(inode))
 197                 return -EACCES;
 198
 199         ret = mnt_want_write_file(file);
 200         if (ret)
 201                 return ret;
 202
 203         mutex_lock(&inode->i_mutex);
 204
 205         ip_oldflags = ip->flags;
 206         i_oldflags = inode->i_flags;
 207         mode = inode->i_mode;
 208
 209         flags = btrfs_mask_flags(inode->i_mode, flags);
 210         oldflags = btrfs_flags_to_ioctl(ip->flags);
 211         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 212                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 213                         ret = -EPERM;
 214                         goto out_unlock;
 215                 }
 216         }
 217
 218         if (flags & FS_SYNC_FL)
 219                 ip->flags |= BTRFS_INODE_SYNC;
 220         else
 221                 ip->flags &= ~BTRFS_INODE_SYNC;
 222         if (flags & FS_IMMUTABLE_FL)
 223                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 224         else
 225                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 226         if (flags & FS_APPEND_FL)
 227                 ip->flags |= BTRFS_INODE_APPEND;
 228         else
 229                 ip->flags &= ~BTRFS_INODE_APPEND;
 230         if (flags & FS_NODUMP_FL)
 231                 ip->flags |= BTRFS_INODE_NODUMP;
 232         else
 233                 ip->flags &= ~BTRFS_INODE_NODUMP;
 234         if (flags & FS_NOATIME_FL)
 235                 ip->flags |= BTRFS_INODE_NOATIME;
 236         else
 237                 ip->flags &= ~BTRFS_INODE_NOATIME;
 238         if (flags & FS_DIRSYNC_FL)
 239                 ip->flags |= BTRFS_INODE_DIRSYNC;
 240         else
 241                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 242         if (flags & FS_NOCOW_FL) {
 243                 if (S_ISREG(mode)) {
 244                         /*
 245                          * It's safe to turn csums off here, no extents exist.
 246                          * Otherwise we want the flag to reflect the real COW
 247                          * status of the file and will not set it.
 248                          */
 249                         if (inode->i_size == 0)
 250                                 ip->flags |= BTRFS_INODE_NODATACOW
 251                                            | BTRFS_INODE_NODATASUM;
 252                 } else {
 253                         ip->flags |= BTRFS_INODE_NODATACOW;
 254                 }
 255         } else {
 256                 /*
 257                  * Revert back under same assuptions as above
 258                  */
 259                 if (S_ISREG(mode)) {
 260                         if (inode->i_size == 0)
 261                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 262                                              | BTRFS_INODE_NODATASUM);
 263                 } else {
 264                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 265                 }
 266         }
 267
 268         /*
 269          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 270          * flag may be changed automatically if compression code won't make
 271          * things smaller.
 272          */
 273         if (flags & FS_NOCOMP_FL) {
 274                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 275                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 276         } else if (flags & FS_COMPR_FL) {
 277                 ip->flags |= BTRFS_INODE_COMPRESS;
 278                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 279         } else {
 280                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 281         }
 282
 283         trans = btrfs_start_transaction(root, 1);
 284         if (IS_ERR(trans)) {
 285                 ret = PTR_ERR(trans);
 286                 goto out_drop;
 287         }
 288
 289         btrfs_update_iflags(inode);
 290         inode_inc_iversion(inode);
 291         inode->i_ctime = CURRENT_TIME;
 292         ret = btrfs_update_inode(trans, root, inode);
 293
 294         btrfs_end_transaction(trans, root);
 295  out_drop:
 296         if (ret) {
 297                 ip->flags = ip_oldflags;
 298                 inode->i_flags = i_oldflags;
 299         }
 300
 301  out_unlock:
 302         mutex_unlock(&inode->i_mutex);
 303         mnt_drop_write_file(file);
 304         return ret;
 305 }
 306
 307 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 308 {
 309         struct inode *inode = file->f_path.dentry->d_inode;
 310
 311         return put_user(inode->i_generation, arg);
 312 }
 313
 314 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 315 {
 316         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
 317         struct btrfs_device *device;
 318         struct request_queue *q;
 319         struct fstrim_range range;
 320         u64 minlen = ULLONG_MAX;
 321         u64 num_devices = 0;
 322         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 323         int ret;
 324
 325         if (!capable(CAP_SYS_ADMIN))
 326                 return -EPERM;
 327
 328         rcu_read_lock();
 329         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 330                                 dev_list) {
 331                 if (!device->bdev)
 332                         continue;
 333                 q = bdev_get_queue(device->bdev);
 334                 if (blk_queue_discard(q)) {
 335                         num_devices++;
 336                         minlen = min((u64)q->limits.discard_granularity,
 337                                      minlen);
 338                 }
 339         }
 340         rcu_read_unlock();
 341
 342         if (!num_devices)
 343                 return -EOPNOTSUPP;
 344         if (copy_from_user(&range, arg, sizeof(range)))
 345                 return -EFAULT;
 346         if (range.start > total_bytes ||
 347             range.len < fs_info->sb->s_blocksize)
 348                 return -EINVAL;
 349
 350         range.len = min(range.len, total_bytes - range.start);
 351         range.minlen = max(range.minlen, minlen);
 352         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 353         if (ret < 0)
 354                 return ret;
 355
 356         if (copy_to_user(arg, &range, sizeof(range)))
 357                 return -EFAULT;
 358
 359         return 0;
 360 }
 361
 362 static noinline int create_subvol(struct btrfs_root *root,
 363                                   struct dentry *dentry,
 364                                   char *name, int namelen,
 365                                   u64 *async_transid,
 366                                   struct btrfs_qgroup_inherit **inherit)
 367 {
 368         struct btrfs_trans_handle *trans;
 369         struct btrfs_key key;
 370         struct btrfs_root_item root_item;
 371         struct btrfs_inode_item *inode_item;
 372         struct extent_buffer *leaf;
 373         struct btrfs_root *new_root;
 374         struct dentry *parent = dentry->d_parent;
 375         struct inode *dir;
 376         struct timespec cur_time = CURRENT_TIME;
 377         int ret;
 378         int err;
 379         u64 objectid;
 380         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 381         u64 index = 0;
 382         uuid_le new_uuid;
 383
 384         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 385         if (ret)
 386                 return ret;
 387
 388         dir = parent->d_inode;
 389
 390         /*
 391          * 1 - inode item
 392          * 2 - refs
 393          * 1 - root item
 394          * 2 - dir items
 395          */
 396         trans = btrfs_start_transaction(root, 6);
 397         if (IS_ERR(trans))
 398                 return PTR_ERR(trans);
 399
 400         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid,
 401                                    inherit ? *inherit : NULL);
 402         if (ret)
 403                 goto fail;
 404
 405         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 406                                       0, objectid, NULL, 0, 0, 0);
 407         if (IS_ERR(leaf)) {
 408                 ret = PTR_ERR(leaf);
 409                 goto fail;
 410         }
 411
 412         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 413         btrfs_set_header_bytenr(leaf, leaf->start);
 414         btrfs_set_header_generation(leaf, trans->transid);
 415         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 416         btrfs_set_header_owner(leaf, objectid);
 417
 418         write_extent_buffer(leaf, root->fs_info->fsid,
 419                             (unsigned long)btrfs_header_fsid(leaf),
 420                             BTRFS_FSID_SIZE);
 421         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 422                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 423                             BTRFS_UUID_SIZE);
 424         btrfs_mark_buffer_dirty(leaf);
 425
 426         memset(&root_item, 0, sizeof(root_item));
 427
 428         inode_item = &root_item.inode;
 429         inode_item->generation = cpu_to_le64(1);
 430         inode_item->size = cpu_to_le64(3);
 431         inode_item->nlink = cpu_to_le32(1);
 432         inode_item->nbytes = cpu_to_le64(root->leafsize);
 433         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 434
 435         root_item.flags = 0;
 436         root_item.byte_limit = 0;
 437         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
 438
 439         btrfs_set_root_bytenr(&root_item, leaf->start);
 440         btrfs_set_root_generation(&root_item, trans->transid);
 441         btrfs_set_root_level(&root_item, 0);
 442         btrfs_set_root_refs(&root_item, 1);
 443         btrfs_set_root_used(&root_item, leaf->len);
 444         btrfs_set_root_last_snapshot(&root_item, 0);
 445
 446         btrfs_set_root_generation_v2(&root_item,
 447                         btrfs_root_generation(&root_item));
 448         uuid_le_gen(&new_uuid);
 449         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 450         root_item.otime.sec = cpu_to_le64(cur_time.tv_sec);
 451         root_item.otime.nsec = cpu_to_le32(cur_time.tv_nsec);
 452         root_item.ctime = root_item.otime;
 453         btrfs_set_root_ctransid(&root_item, trans->transid);
 454         btrfs_set_root_otransid(&root_item, trans->transid);
 455
 456         btrfs_tree_unlock(leaf);
 457         free_extent_buffer(leaf);
 458         leaf = NULL;
 459
 460         btrfs_set_root_dirid(&root_item, new_dirid);
 461
 462         key.objectid = objectid;
 463         key.offset = 0;
 464         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 465         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 466                                 &root_item);
 467         if (ret)
 468                 goto fail;
 469
 470         key.offset = (u64)-1;
 471         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 472         if (IS_ERR(new_root)) {
 473                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 474                 ret = PTR_ERR(new_root);
 475                 goto fail;
 476         }
 477
 478         btrfs_record_root_in_trans(trans, new_root);
 479
 480         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
 481         if (ret) {
 482                 /* We potentially lose an unused inode item here */
 483                 btrfs_abort_transaction(trans, root, ret);
 484                 goto fail;
 485         }
 486
 487         /*
 488          * insert the directory item
 489          */
 490         ret = btrfs_set_inode_index(dir, &index);
 491         if (ret) {
 492                 btrfs_abort_transaction(trans, root, ret);
 493                 goto fail;
 494         }
 495
 496         ret = btrfs_insert_dir_item(trans, root,
 497                                     name, namelen, dir, &key,
 498                                     BTRFS_FT_DIR, index);
 499         if (ret) {
 500                 btrfs_abort_transaction(trans, root, ret);
 501                 goto fail;
 502         }
 503
 504         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 505         ret = btrfs_update_inode(trans, root, dir);
 506         BUG_ON(ret);
 507
 508         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 509                                  objectid, root->root_key.objectid,
 510                                  btrfs_ino(dir), index, name, namelen);
 511
 512         BUG_ON(ret);
 513
 514         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 515 fail:
 516         if (async_transid) {
 517                 *async_transid = trans->transid;
 518                 err = btrfs_commit_transaction_async(trans, root, 1);
 519         } else {
 520                 err = btrfs_commit_transaction(trans, root);
 521         }
 522         if (err && !ret)
 523                 ret = err;
 524         return ret;
 525 }
 526
 527 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 528                            char *name, int namelen, u64 *async_transid,
 529                            bool readonly, struct btrfs_qgroup_inherit **inherit)
 530 {
 531         struct inode *inode;
 532         struct btrfs_pending_snapshot *pending_snapshot;
 533         struct btrfs_trans_handle *trans;
 534         int ret;
 535
 536         if (!root->ref_cows)
 537                 return -EINVAL;
 538
 539         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 540         if (!pending_snapshot)
 541                 return -ENOMEM;
 542
 543         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 544                              BTRFS_BLOCK_RSV_TEMP);
 545         pending_snapshot->dentry = dentry;
 546         pending_snapshot->root = root;
 547         pending_snapshot->readonly = readonly;
 548         if (inherit) {
 549                 pending_snapshot->inherit = *inherit;
 550                 *inherit = NULL;        /* take responsibility to free it */
 551         }
 552
 553         trans = btrfs_start_transaction(root->fs_info->extent_root, 6);
 554         if (IS_ERR(trans)) {
 555                 ret = PTR_ERR(trans);
 556                 goto fail;
 557         }
 558
 559         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 560         BUG_ON(ret);
 561
 562         spin_lock(&root->fs_info->trans_lock);
 563         list_add(&pending_snapshot->list,
 564                  &trans->transaction->pending_snapshots);
 565         spin_unlock(&root->fs_info->trans_lock);
 566         if (async_transid) {
 567                 *async_transid = trans->transid;
 568                 ret = btrfs_commit_transaction_async(trans,
 569                                      root->fs_info->extent_root, 1);
 570         } else {
 571                 ret = btrfs_commit_transaction(trans,
 572                                                root->fs_info->extent_root);
 573         }
 574         BUG_ON(ret);
 575
 576         ret = pending_snapshot->error;
 577         if (ret)
 578                 goto fail;
 579
 580         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 581         if (ret)
 582                 goto fail;
 583
 584         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 585         if (IS_ERR(inode)) {
 586                 ret = PTR_ERR(inode);
 587                 goto fail;
 588         }
 589         BUG_ON(!inode);
 590         d_instantiate(dentry, inode);
 591         ret = 0;
 592 fail:
 593         kfree(pending_snapshot);
 594         return ret;
 595 }
 596
 597 /*  copy of check_sticky in fs/namei.c()
 598 * It's inline, so penalty for filesystems that don't use sticky bit is
 599 * minimal.
 600 */
 601 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 602 {
 603         uid_t fsuid = current_fsuid();
 604
 605         if (!(dir->i_mode & S_ISVTX))
 606                 return 0;
 607         if (inode->i_uid == fsuid)
 608                 return 0;
 609         if (dir->i_uid == fsuid)
 610                 return 0;
 611         return !capable(CAP_FOWNER);
 612 }
 613
 614 /*  copy of may_delete in fs/namei.c()
 615  *      Check whether we can remove a link victim from directory dir, check
 616  *  whether the type of victim is right.
 617  *  1. We can't do it if dir is read-only (done in permission())
 618  *  2. We should have write and exec permissions on dir
 619  *  3. We can't remove anything from append-only dir
 620  *  4. We can't do anything with immutable dir (done in permission())
 621  *  5. If the sticky bit on dir is set we should either
 622  *      a. be owner of dir, or
 623  *      b. be owner of victim, or
 624  *      c. have CAP_FOWNER capability
 625  *  6. If the victim is append-only or immutable we can't do antyhing with
 626  *     links pointing to it.
 627  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 628  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 629  *  9. We can't remove a root or mountpoint.
 630  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 631  *     nfs_async_unlink().
 632  */
 633
 634 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 635 {
 636         int error;
 637
 638         if (!victim->d_inode)
 639                 return -ENOENT;
 640
 641         BUG_ON(victim->d_parent->d_inode != dir);
 642         audit_inode_child(victim, dir);
 643
 644         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 645         if (error)
 646                 return error;
 647         if (IS_APPEND(dir))
 648                 return -EPERM;
 649         if (btrfs_check_sticky(dir, victim->d_inode)||
 650                 IS_APPEND(victim->d_inode)||
 651             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 652                 return -EPERM;
 653         if (isdir) {
 654                 if (!S_ISDIR(victim->d_inode->i_mode))
 655                         return -ENOTDIR;
 656                 if (IS_ROOT(victim))
 657                         return -EBUSY;
 658         } else if (S_ISDIR(victim->d_inode->i_mode))
 659                 return -EISDIR;
 660         if (IS_DEADDIR(dir))
 661                 return -ENOENT;
 662         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 663                 return -EBUSY;
 664         return 0;
 665 }
 666
 667 /* copy of may_create in fs/namei.c() */
 668 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 669 {
 670         if (child->d_inode)
 671                 return -EEXIST;
 672         if (IS_DEADDIR(dir))
 673                 return -ENOENT;
 674         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 675 }
 676
 677 /*
 678  * Create a new subvolume below @parent.  This is largely modeled after
 679  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 680  * inside this filesystem so it's quite a bit simpler.
 681  */
 682 static noinline int btrfs_mksubvol(struct path *parent,
 683                                    char *name, int namelen,
 684                                    struct btrfs_root *snap_src,
 685                                    u64 *async_transid, bool readonly,
 686                                    struct btrfs_qgroup_inherit **inherit)
 687 {
 688         struct inode *dir  = parent->dentry->d_inode;
 689         struct dentry *dentry;
 690         int error;
 691
 692         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 693
 694         dentry = lookup_one_len(name, parent->dentry, namelen);
 695         error = PTR_ERR(dentry);
 696         if (IS_ERR(dentry))
 697                 goto out_unlock;
 698
 699         error = -EEXIST;
 700         if (dentry->d_inode)
 701                 goto out_dput;
 702
 703         error = btrfs_may_create(dir, dentry);
 704         if (error)
 705                 goto out_dput;
 706
 707         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 708
 709         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 710                 goto out_up_read;
 711
 712         if (snap_src) {
 713                 error = create_snapshot(snap_src, dentry, name, namelen,
 714                                         async_transid, readonly, inherit);
 715         } else {
 716                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 717                                       name, namelen, async_transid, inherit);
 718         }
 719         if (!error)
 720                 fsnotify_mkdir(dir, dentry);
 721 out_up_read:
 722         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 723 out_dput:
 724         dput(dentry);
 725 out_unlock:
 726         mutex_unlock(&dir->i_mutex);
 727         return error;
 728 }
 729
 730 /*
 731  * When we're defragging a range, we don't want to kick it off again
 732  * if it is really just waiting for delalloc to send it down.
 733  * If we find a nice big extent or delalloc range for the bytes in the
 734  * file you want to defrag, we return 0 to let you know to skip this
 735  * part of the file
 736  */
 737 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 738 {
 739         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 740         struct extent_map *em = NULL;
 741         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 742         u64 end;
 743
 744         read_lock(&em_tree->lock);
 745         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 746         read_unlock(&em_tree->lock);
 747
 748         if (em) {
 749                 end = extent_map_end(em);
 750                 free_extent_map(em);
 751                 if (end - offset > thresh)
 752                         return 0;
 753         }
 754         /* if we already have a nice delalloc here, just stop */
 755         thresh /= 2;
 756         end = count_range_bits(io_tree, &offset, offset + thresh,
 757                                thresh, EXTENT_DELALLOC, 1);
 758         if (end >= thresh)
 759                 return 0;
 760         return 1;
 761 }
 762
 763 /*
 764  * helper function to walk through a file and find extents
 765  * newer than a specific transid, and smaller than thresh.
 766  *
 767  * This is used by the defragging code to find new and small
 768  * extents
 769  */
 770 static int find_new_extents(struct btrfs_root *root,
 771                             struct inode *inode, u64 newer_than,
 772                             u64 *off, int thresh)
 773 {
 774         struct btrfs_path *path;
 775         struct btrfs_key min_key;
 776         struct btrfs_key max_key;
 777         struct extent_buffer *leaf;
 778         struct btrfs_file_extent_item *extent;
 779         int type;
 780         int ret;
 781         u64 ino = btrfs_ino(inode);
 782
 783         path = btrfs_alloc_path();
 784         if (!path)
 785                 return -ENOMEM;
 786
 787         min_key.objectid = ino;
 788         min_key.type = BTRFS_EXTENT_DATA_KEY;
 789         min_key.offset = *off;
 790
 791         max_key.objectid = ino;
 792         max_key.type = (u8)-1;
 793         max_key.offset = (u64)-1;
 794
 795         path->keep_locks = 1;
 796
 797         while(1) {
 798                 ret = btrfs_search_forward(root, &min_key, &max_key,
 799                                            path, 0, newer_than);
 800                 if (ret != 0)
 801                         goto none;
 802                 if (min_key.objectid != ino)
 803                         goto none;
 804                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 805                         goto none;
 806
 807                 leaf = path->nodes[0];
 808                 extent = btrfs_item_ptr(leaf, path->slots[0],
 809                                         struct btrfs_file_extent_item);
 810
 811                 type = btrfs_file_extent_type(leaf, extent);
 812                 if (type == BTRFS_FILE_EXTENT_REG &&
 813                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 814                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 815                         *off = min_key.offset;
 816                         btrfs_free_path(path);
 817                         return 0;
 818                 }
 819
 820                 if (min_key.offset == (u64)-1)
 821                         goto none;
 822
 823                 min_key.offset++;
 824                 btrfs_release_path(path);
 825         }
 826 none:
 827         btrfs_free_path(path);
 828         return -ENOENT;
 829 }
 830
 831 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 832 {
 833         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 834         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 835         struct extent_map *em;
 836         u64 len = PAGE_CACHE_SIZE;
 837
 838         /*
 839          * hopefully we have this extent in the tree already, try without
 840          * the full extent lock
 841          */
 842         read_lock(&em_tree->lock);
 843         em = lookup_extent_mapping(em_tree, start, len);
 844         read_unlock(&em_tree->lock);
 845
 846         if (!em) {
 847                 /* get the big lock and read metadata off disk */
 848                 lock_extent(io_tree, start, start + len - 1);
 849                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 850                 unlock_extent(io_tree, start, start + len - 1);
 851
 852                 if (IS_ERR(em))
 853                         return NULL;
 854         }
 855
 856         return em;
 857 }
 858
 859 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 860 {
 861         struct extent_map *next;
 862         bool ret = true;
 863
 864         /* this is the last extent */
 865         if (em->start + em->len >= i_size_read(inode))
 866                 return false;
 867
 868         next = defrag_lookup_extent(inode, em->start + em->len);
 869         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 870                 ret = false;
 871
 872         free_extent_map(next);
 873         return ret;
 874 }
 875
 876 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
 877                                u64 *last_len, u64 *skip, u64 *defrag_end,
 878                                int compress)
 879 {
 880         struct extent_map *em;
 881         int ret = 1;
 882         bool next_mergeable = true;
 883
 884         /*
 885          * make sure that once we start defragging an extent, we keep on
 886          * defragging it
 887          */
 888         if (start < *defrag_end)
 889                 return 1;
 890
 891         *skip = 0;
 892
 893         em = defrag_lookup_extent(inode, start);
 894         if (!em)
 895                 return 0;
 896
 897         /* this will cover holes, and inline extents */
 898         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 899                 ret = 0;
 900                 goto out;
 901         }
 902
 903         next_mergeable = defrag_check_next_extent(inode, em);
 904
 905         /*
 906          * we hit a real extent, if it is big or the next extent is not a
 907          * real extent, don't bother defragging it
 908          */
 909         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
 910             (em->len >= thresh || !next_mergeable))
 911                 ret = 0;
 912 out:
 913         /*
 914          * last_len ends up being a counter of how many bytes we've defragged.
 915          * every time we choose not to defrag an extent, we reset *last_len
 916          * so that the next tiny extent will force a defrag.
 917          *
 918          * The end result of this is that tiny extents before a single big
 919          * extent will force at least part of that big extent to be defragged.
 920          */
 921         if (ret) {
 922                 *defrag_end = extent_map_end(em);
 923         } else {
 924                 *last_len = 0;
 925                 *skip = extent_map_end(em);
 926                 *defrag_end = 0;
 927         }
 928
 929         free_extent_map(em);
 930         return ret;
 931 }
 932
 933 /*
 934  * it doesn't do much good to defrag one or two pages
 935  * at a time.  This pulls in a nice chunk of pages
 936  * to COW and defrag.
 937  *
 938  * It also makes sure the delalloc code has enough
 939  * dirty data to avoid making new small extents as part
 940  * of the defrag
 941  *
 942  * It's a good idea to start RA on this range
 943  * before calling this.
 944  */
 945 static int cluster_pages_for_defrag(struct inode *inode,
 946                                     struct page **pages,
 947                                     unsigned long start_index,
 948                                     int num_pages)
 949 {
 950         unsigned long file_end;
 951         u64 isize = i_size_read(inode);
 952         u64 page_start;
 953         u64 page_end;
 954         u64 page_cnt;
 955         int ret;
 956         int i;
 957         int i_done;
 958         struct btrfs_ordered_extent *ordered;
 959         struct extent_state *cached_state = NULL;
 960         struct extent_io_tree *tree;
 961         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 962
 963         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 964         if (!isize || start_index > file_end)
 965                 return 0;
 966
 967         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
 968
 969         ret = btrfs_delalloc_reserve_space(inode,
 970                                            page_cnt << PAGE_CACHE_SHIFT);
 971         if (ret)
 972                 return ret;
 973         i_done = 0;
 974         tree = &BTRFS_I(inode)->io_tree;
 975
 976         /* step one, lock all the pages */
 977         for (i = 0; i < page_cnt; i++) {
 978                 struct page *page;
 979 again:
 980                 page = find_or_create_page(inode->i_mapping,
 981                                            start_index + i, mask);
 982                 if (!page)
 983                         break;
 984
 985                 page_start = page_offset(page);
 986                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 987                 while (1) {
 988                         lock_extent(tree, page_start, page_end);
 989                         ordered = btrfs_lookup_ordered_extent(inode,
 990                                                               page_start);
 991                         unlock_extent(tree, page_start, page_end);
 992                         if (!ordered)
 993                                 break;
 994
 995                         unlock_page(page);
 996                         btrfs_start_ordered_extent(inode, ordered, 1);
 997                         btrfs_put_ordered_extent(ordered);
 998                         lock_page(page);
 999                         /*
1000                          * we unlocked the page above, so we need check if
1001                          * it was released or not.
1002                          */
1003                         if (page->mapping != inode->i_mapping) {
1004                                 unlock_page(page);
1005                                 page_cache_release(page);
1006                                 goto again;
1007                         }
1008                 }
1009
1010                 if (!PageUptodate(page)) {
1011                         btrfs_readpage(NULL, page);
1012                         lock_page(page);
1013                         if (!PageUptodate(page)) {
1014                                 unlock_page(page);
1015                                 page_cache_release(page);
1016                                 ret = -EIO;
1017                                 break;
1018                         }
1019                 }
1020
1021                 if (page->mapping != inode->i_mapping) {
1022                         unlock_page(page);
1023                         page_cache_release(page);
1024                         goto again;
1025                 }
1026
1027                 pages[i] = page;
1028                 i_done++;
1029         }
1030         if (!i_done || ret)
1031                 goto out;
1032
1033         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1034                 goto out;
1035
1036         /*
1037          * so now we have a nice long stream of locked
1038          * and up to date pages, lets wait on them
1039          */
1040         for (i = 0; i < i_done; i++)
1041                 wait_on_page_writeback(pages[i]);
1042
1043         page_start = page_offset(pages[0]);
1044         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1045
1046         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1047                          page_start, page_end - 1, 0, &cached_state);
1048         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1049                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1050                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1051                           &cached_state, GFP_NOFS);
1052
1053         if (i_done != page_cnt) {
1054                 spin_lock(&BTRFS_I(inode)->lock);
1055                 BTRFS_I(inode)->outstanding_extents++;
1056                 spin_unlock(&BTRFS_I(inode)->lock);
1057                 btrfs_delalloc_release_space(inode,
1058                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1059         }
1060
1061
1062         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1063                           &cached_state, GFP_NOFS);
1064
1065         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1066                              page_start, page_end - 1, &cached_state,
1067                              GFP_NOFS);
1068
1069         for (i = 0; i < i_done; i++) {
1070                 clear_page_dirty_for_io(pages[i]);
1071                 ClearPageChecked(pages[i]);
1072                 set_page_extent_mapped(pages[i]);
1073                 set_page_dirty(pages[i]);
1074                 unlock_page(pages[i]);
1075                 page_cache_release(pages[i]);
1076         }
1077         return i_done;
1078 out:
1079         for (i = 0; i < i_done; i++) {
1080                 unlock_page(pages[i]);
1081                 page_cache_release(pages[i]);
1082         }
1083         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1084         return ret;
1085
1086 }
1087
1088 int btrfs_defrag_file(struct inode *inode, struct file *file,
1089                       struct btrfs_ioctl_defrag_range_args *range,
1090                       u64 newer_than, unsigned long max_to_defrag)
1091 {
1092         struct btrfs_root *root = BTRFS_I(inode)->root;
1093         struct file_ra_state *ra = NULL;
1094         unsigned long last_index;
1095         u64 isize = i_size_read(inode);
1096         u64 last_len = 0;
1097         u64 skip = 0;
1098         u64 defrag_end = 0;
1099         u64 newer_off = range->start;
1100         unsigned long i;
1101         unsigned long ra_index = 0;
1102         int ret;
1103         int defrag_count = 0;
1104         int compress_type = BTRFS_COMPRESS_ZLIB;
1105         int extent_thresh = range->extent_thresh;
1106         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1107         int cluster = max_cluster;
1108         u64 new_align = ~((u64)128 * 1024 - 1);
1109         struct page **pages = NULL;
1110
1111         if (extent_thresh == 0)
1112                 extent_thresh = 256 * 1024;
1113
1114         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1115                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1116                         return -EINVAL;
1117                 if (range->compress_type)
1118                         compress_type = range->compress_type;
1119         }
1120
1121         if (isize == 0)
1122                 return 0;
1123
1124         /*
1125          * if we were not given a file, allocate a readahead
1126          * context
1127          */
1128         if (!file) {
1129                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1130                 if (!ra)
1131                         return -ENOMEM;
1132                 file_ra_state_init(ra, inode->i_mapping);
1133         } else {
1134                 ra = &file->f_ra;
1135         }
1136
1137         pages = kmalloc(sizeof(struct page *) * max_cluster,
1138                         GFP_NOFS);
1139         if (!pages) {
1140                 ret = -ENOMEM;
1141                 goto out_ra;
1142         }
1143
1144         /* find the last page to defrag */
1145         if (range->start + range->len > range->start) {
1146                 last_index = min_t(u64, isize - 1,
1147                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1148         } else {
1149                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1150         }
1151
1152         if (newer_than) {
1153                 ret = find_new_extents(root, inode, newer_than,
1154                                        &newer_off, 64 * 1024);
1155                 if (!ret) {
1156                         range->start = newer_off;
1157                         /*
1158                          * we always align our defrag to help keep
1159                          * the extents in the file evenly spaced
1160                          */
1161                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1162                 } else
1163                         goto out_ra;
1164         } else {
1165                 i = range->start >> PAGE_CACHE_SHIFT;
1166         }
1167         if (!max_to_defrag)
1168                 max_to_defrag = last_index + 1;
1169
1170         /*
1171          * make writeback starts from i, so the defrag range can be
1172          * written sequentially.
1173          */
1174         if (i < inode->i_mapping->writeback_index)
1175                 inode->i_mapping->writeback_index = i;
1176
1177         while (i <= last_index && defrag_count < max_to_defrag &&
1178                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1179                 PAGE_CACHE_SHIFT)) {
1180                 /*
1181                  * make sure we stop running if someone unmounts
1182                  * the FS
1183                  */
1184                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1185                         break;
1186
1187                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1188                                          extent_thresh, &last_len, &skip,
1189                                          &defrag_end, range->flags &
1190                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1191                         unsigned long next;
1192                         /*
1193                          * the should_defrag function tells us how much to skip
1194                          * bump our counter by the suggested amount
1195                          */
1196                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1197                         i = max(i + 1, next);
1198                         continue;
1199                 }
1200
1201                 if (!newer_than) {
1202                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1203                                    PAGE_CACHE_SHIFT) - i;
1204                         cluster = min(cluster, max_cluster);
1205                 } else {
1206                         cluster = max_cluster;
1207                 }
1208
1209                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1210                         BTRFS_I(inode)->force_compress = compress_type;
1211
1212                 if (i + cluster > ra_index) {
1213                         ra_index = max(i, ra_index);
1214                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1215                                        cluster);
1216                         ra_index += max_cluster;
1217                 }
1218
1219                 mutex_lock(&inode->i_mutex);
1220                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1221                 if (ret < 0) {
1222                         mutex_unlock(&inode->i_mutex);
1223                         goto out_ra;
1224                 }
1225
1226                 defrag_count += ret;
1227                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1228                 mutex_unlock(&inode->i_mutex);
1229
1230                 if (newer_than) {
1231                         if (newer_off == (u64)-1)
1232                                 break;
1233
1234                         if (ret > 0)
1235                                 i += ret;
1236
1237                         newer_off = max(newer_off + 1,
1238                                         (u64)i << PAGE_CACHE_SHIFT);
1239
1240                         ret = find_new_extents(root, inode,
1241                                                newer_than, &newer_off,
1242                                                64 * 1024);
1243                         if (!ret) {
1244                                 range->start = newer_off;
1245                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1246                         } else {
1247                                 break;
1248                         }
1249                 } else {
1250                         if (ret > 0) {
1251                                 i += ret;
1252                                 last_len += ret << PAGE_CACHE_SHIFT;
1253                         } else {
1254                                 i++;
1255                                 last_len = 0;
1256                         }
1257                 }
1258         }
1259
1260         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1261                 filemap_flush(inode->i_mapping);
1262
1263         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1264                 /* the filemap_flush will queue IO into the worker threads, but
1265                  * we have to make sure the IO is actually started and that
1266                  * ordered extents get created before we return
1267                  */
1268                 atomic_inc(&root->fs_info->async_submit_draining);
1269                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1270                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1271                         wait_event(root->fs_info->async_submit_wait,
1272                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1273                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1274                 }
1275                 atomic_dec(&root->fs_info->async_submit_draining);
1276
1277                 mutex_lock(&inode->i_mutex);
1278                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1279                 mutex_unlock(&inode->i_mutex);
1280         }
1281
1282         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1283                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1284         }
1285
1286         ret = defrag_count;
1287
1288 out_ra:
1289         if (!file)
1290                 kfree(ra);
1291         kfree(pages);
1292         return ret;
1293 }
1294
1295 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1296                                         void __user *arg)
1297 {
1298         u64 new_size;
1299         u64 old_size;
1300         u64 devid = 1;
1301         struct btrfs_ioctl_vol_args *vol_args;
1302         struct btrfs_trans_handle *trans;
1303         struct btrfs_device *device = NULL;
1304         char *sizestr;
1305         char *devstr = NULL;
1306         int ret = 0;
1307         int mod = 0;
1308
1309         if (root->fs_info->sb->s_flags & MS_RDONLY)
1310                 return -EROFS;
1311
1312         if (!capable(CAP_SYS_ADMIN))
1313                 return -EPERM;
1314
1315         mutex_lock(&root->fs_info->volume_mutex);
1316         if (root->fs_info->balance_ctl) {
1317                 printk(KERN_INFO "btrfs: balance in progress\n");
1318                 ret = -EINVAL;
1319                 goto out;
1320         }
1321
1322         vol_args = memdup_user(arg, sizeof(*vol_args));
1323         if (IS_ERR(vol_args)) {
1324                 ret = PTR_ERR(vol_args);
1325                 goto out;
1326         }
1327
1328         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1329
1330         sizestr = vol_args->name;
1331         devstr = strchr(sizestr, ':');
1332         if (devstr) {
1333                 char *end;
1334                 sizestr = devstr + 1;
1335                 *devstr = '\0';
1336                 devstr = vol_args->name;
1337                 devid = simple_strtoull(devstr, &end, 10);
1338                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1339                        (unsigned long long)devid);
1340         }
1341         device = btrfs_find_device(root, devid, NULL, NULL);
1342         if (!device) {
1343                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1344                        (unsigned long long)devid);
1345                 ret = -EINVAL;
1346                 goto out_free;
1347         }
1348         if (device->fs_devices && device->fs_devices->seeding) {
1349                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1350                        "seeding device %llu\n",
1351                        (unsigned long long)devid);
1352                 ret = -EINVAL;
1353                 goto out_free;
1354         }
1355
1356         if (!strcmp(sizestr, "max"))
1357                 new_size = device->bdev->bd_inode->i_size;
1358         else {
1359                 if (sizestr[0] == '-') {
1360                         mod = -1;
1361                         sizestr++;
1362                 } else if (sizestr[0] == '+') {
1363                         mod = 1;
1364                         sizestr++;
1365                 }
1366                 new_size = memparse(sizestr, NULL);
1367                 if (new_size == 0) {
1368                         ret = -EINVAL;
1369                         goto out_free;
1370                 }
1371         }
1372
1373         old_size = device->total_bytes;
1374
1375         if (mod < 0) {
1376                 if (new_size > old_size) {
1377                         ret = -EINVAL;
1378                         goto out_free;
1379                 }
1380                 new_size = old_size - new_size;
1381         } else if (mod > 0) {
1382                 new_size = old_size + new_size;
1383         }
1384
1385         if (new_size < 256 * 1024 * 1024) {
1386                 ret = -EINVAL;
1387                 goto out_free;
1388         }
1389         if (new_size > device->bdev->bd_inode->i_size) {
1390                 ret = -EFBIG;
1391                 goto out_free;
1392         }
1393
1394         do_div(new_size, root->sectorsize);
1395         new_size *= root->sectorsize;
1396
1397         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1398                       rcu_str_deref(device->name),
1399                       (unsigned long long)new_size);
1400
1401         if (new_size > old_size) {
1402                 trans = btrfs_start_transaction(root, 0);
1403                 if (IS_ERR(trans)) {
1404                         ret = PTR_ERR(trans);
1405                         goto out_free;
1406                 }
1407                 ret = btrfs_grow_device(trans, device, new_size);
1408                 btrfs_commit_transaction(trans, root);
1409         } else if (new_size < old_size) {
1410                 ret = btrfs_shrink_device(device, new_size);
1411         }
1412
1413 out_free:
1414         kfree(vol_args);
1415 out:
1416         mutex_unlock(&root->fs_info->volume_mutex);
1417         return ret;
1418 }
1419
1420 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1421                                 char *name, unsigned long fd, int subvol,
1422                                 u64 *transid, bool readonly,
1423                                 struct btrfs_qgroup_inherit **inherit)
1424 {
1425         struct file *src_file;
1426         int namelen;
1427         int ret = 0;
1428
1429         ret = mnt_want_write_file(file);
1430         if (ret)
1431                 goto out;
1432
1433         namelen = strlen(name);
1434         if (strchr(name, '/')) {
1435                 ret = -EINVAL;
1436                 goto out_drop_write;
1437         }
1438
1439         if (name[0] == '.' &&
1440            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1441                 ret = -EEXIST;
1442                 goto out_drop_write;
1443         }
1444
1445         if (subvol) {
1446                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1447                                      NULL, transid, readonly, inherit);
1448         } else {
1449                 struct inode *src_inode;
1450                 src_file = fget(fd);
1451                 if (!src_file) {
1452                         ret = -EINVAL;
1453                         goto out_drop_write;
1454                 }
1455
1456                 src_inode = src_file->f_path.dentry->d_inode;
1457                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1458                         printk(KERN_INFO "btrfs: Snapshot src from "
1459                                "another FS\n");
1460                         ret = -EINVAL;
1461                         fput(src_file);
1462                         goto out_drop_write;
1463                 }
1464                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1465                                      BTRFS_I(src_inode)->root,
1466                                      transid, readonly, inherit);
1467                 fput(src_file);
1468         }
1469 out_drop_write:
1470         mnt_drop_write_file(file);
1471 out:
1472         return ret;
1473 }
1474
1475 static noinline int btrfs_ioctl_snap_create(struct file *file,
1476                                             void __user *arg, int subvol)
1477 {
1478         struct btrfs_ioctl_vol_args *vol_args;
1479         int ret;
1480
1481         vol_args = memdup_user(arg, sizeof(*vol_args));
1482         if (IS_ERR(vol_args))
1483                 return PTR_ERR(vol_args);
1484         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1485
1486         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1487                                               vol_args->fd, subvol,
1488                                               NULL, false, NULL);
1489
1490         kfree(vol_args);
1491         return ret;
1492 }
1493
1494 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1495                                                void __user *arg, int subvol)
1496 {
1497         struct btrfs_ioctl_vol_args_v2 *vol_args;
1498         int ret;
1499         u64 transid = 0;
1500         u64 *ptr = NULL;
1501         bool readonly = false;
1502         struct btrfs_qgroup_inherit *inherit = NULL;
1503
1504         vol_args = memdup_user(arg, sizeof(*vol_args));
1505         if (IS_ERR(vol_args))
1506                 return PTR_ERR(vol_args);
1507         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1508
1509         if (vol_args->flags &
1510             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1511               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1512                 ret = -EOPNOTSUPP;
1513                 goto out;
1514         }
1515
1516         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1517                 ptr = &transid;
1518         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1519                 readonly = true;
1520         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1521                 if (vol_args->size > PAGE_CACHE_SIZE) {
1522                         ret = -EINVAL;
1523                         goto out;
1524                 }
1525                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1526                 if (IS_ERR(inherit)) {
1527                         ret = PTR_ERR(inherit);
1528                         goto out;
1529                 }
1530         }
1531
1532         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1533                                               vol_args->fd, subvol, ptr,
1534                                               readonly, &inherit);
1535
1536         if (ret == 0 && ptr &&
1537             copy_to_user(arg +
1538                          offsetof(struct btrfs_ioctl_vol_args_v2,
1539                                   transid), ptr, sizeof(*ptr)))
1540                 ret = -EFAULT;
1541 out:
1542         kfree(vol_args);
1543         kfree(inherit);
1544         return ret;
1545 }
1546
1547 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1548                                                 void __user *arg)
1549 {
1550         struct inode *inode = fdentry(file)->d_inode;
1551         struct btrfs_root *root = BTRFS_I(inode)->root;
1552         int ret = 0;
1553         u64 flags = 0;
1554
1555         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1556                 return -EINVAL;
1557
1558         down_read(&root->fs_info->subvol_sem);
1559         if (btrfs_root_readonly(root))
1560                 flags |= BTRFS_SUBVOL_RDONLY;
1561         up_read(&root->fs_info->subvol_sem);
1562
1563         if (copy_to_user(arg, &flags, sizeof(flags)))
1564                 ret = -EFAULT;
1565
1566         return ret;
1567 }
1568
1569 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1570                                               void __user *arg)
1571 {
1572         struct inode *inode = fdentry(file)->d_inode;
1573         struct btrfs_root *root = BTRFS_I(inode)->root;
1574         struct btrfs_trans_handle *trans;
1575         u64 root_flags;
1576         u64 flags;
1577         int ret = 0;
1578
1579         ret = mnt_want_write_file(file);
1580         if (ret)
1581                 goto out;
1582
1583         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1584                 ret = -EINVAL;
1585                 goto out_drop_write;
1586         }
1587
1588         if (copy_from_user(&flags, arg, sizeof(flags))) {
1589                 ret = -EFAULT;
1590                 goto out_drop_write;
1591         }
1592
1593         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1594                 ret = -EINVAL;
1595                 goto out_drop_write;
1596         }
1597
1598         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1599                 ret = -EOPNOTSUPP;
1600                 goto out_drop_write;
1601         }
1602
1603         if (!inode_owner_or_capable(inode)) {
1604                 ret = -EACCES;
1605                 goto out_drop_write;
1606         }
1607
1608         down_write(&root->fs_info->subvol_sem);
1609
1610         /* nothing to do */
1611         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1612                 goto out_drop_sem;
1613
1614         root_flags = btrfs_root_flags(&root->root_item);
1615         if (flags & BTRFS_SUBVOL_RDONLY)
1616                 btrfs_set_root_flags(&root->root_item,
1617                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1618         else
1619                 btrfs_set_root_flags(&root->root_item,
1620                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1621
1622         trans = btrfs_start_transaction(root, 1);
1623         if (IS_ERR(trans)) {
1624                 ret = PTR_ERR(trans);
1625                 goto out_reset;
1626         }
1627
1628         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1629                                 &root->root_key, &root->root_item);
1630
1631         btrfs_commit_transaction(trans, root);
1632 out_reset:
1633         if (ret)
1634                 btrfs_set_root_flags(&root->root_item, root_flags);
1635 out_drop_sem:
1636         up_write(&root->fs_info->subvol_sem);
1637 out_drop_write:
1638         mnt_drop_write_file(file);
1639 out:
1640         return ret;
1641 }
1642
1643 /*
1644  * helper to check if the subvolume references other subvolumes
1645  */
1646 static noinline int may_destroy_subvol(struct btrfs_root *root)
1647 {
1648         struct btrfs_path *path;
1649         struct btrfs_key key;
1650         int ret;
1651
1652         path = btrfs_alloc_path();
1653         if (!path)
1654                 return -ENOMEM;
1655
1656         key.objectid = root->root_key.objectid;
1657         key.type = BTRFS_ROOT_REF_KEY;
1658         key.offset = (u64)-1;
1659
1660         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1661                                 &key, path, 0, 0);
1662         if (ret < 0)
1663                 goto out;
1664         BUG_ON(ret == 0);
1665
1666         ret = 0;
1667         if (path->slots[0] > 0) {
1668                 path->slots[0]--;
1669                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1670                 if (key.objectid == root->root_key.objectid &&
1671                     key.type == BTRFS_ROOT_REF_KEY)
1672                         ret = -ENOTEMPTY;
1673         }
1674 out:
1675         btrfs_free_path(path);
1676         return ret;
1677 }
1678
1679 static noinline int key_in_sk(struct btrfs_key *key,
1680                               struct btrfs_ioctl_search_key *sk)
1681 {
1682         struct btrfs_key test;
1683         int ret;
1684
1685         test.objectid = sk->min_objectid;
1686         test.type = sk->min_type;
1687         test.offset = sk->min_offset;
1688
1689         ret = btrfs_comp_cpu_keys(key, &test);
1690         if (ret < 0)
1691                 return 0;
1692
1693         test.objectid = sk->max_objectid;
1694         test.type = sk->max_type;
1695         test.offset = sk->max_offset;
1696
1697         ret = btrfs_comp_cpu_keys(key, &test);
1698         if (ret > 0)
1699                 return 0;
1700         return 1;
1701 }
1702
1703 static noinline int copy_to_sk(struct btrfs_root *root,
1704                                struct btrfs_path *path,
1705                                struct btrfs_key *key,
1706                                struct btrfs_ioctl_search_key *sk,
1707                                char *buf,
1708                                unsigned long *sk_offset,
1709                                int *num_found)
1710 {
1711         u64 found_transid;
1712         struct extent_buffer *leaf;
1713         struct btrfs_ioctl_search_header sh;
1714         unsigned long item_off;
1715         unsigned long item_len;
1716         int nritems;
1717         int i;
1718         int slot;
1719         int ret = 0;
1720
1721         leaf = path->nodes[0];
1722         slot = path->slots[0];
1723         nritems = btrfs_header_nritems(leaf);
1724
1725         if (btrfs_header_generation(leaf) > sk->max_transid) {
1726                 i = nritems;
1727                 goto advance_key;
1728         }
1729         found_transid = btrfs_header_generation(leaf);
1730
1731         for (i = slot; i < nritems; i++) {
1732                 item_off = btrfs_item_ptr_offset(leaf, i);
1733                 item_len = btrfs_item_size_nr(leaf, i);
1734
1735                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1736                         item_len = 0;
1737
1738                 if (sizeof(sh) + item_len + *sk_offset >
1739                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1740                         ret = 1;
1741                         goto overflow;
1742                 }
1743
1744                 btrfs_item_key_to_cpu(leaf, key, i);
1745                 if (!key_in_sk(key, sk))
1746                         continue;
1747
1748                 sh.objectid = key->objectid;
1749                 sh.offset = key->offset;
1750                 sh.type = key->type;
1751                 sh.len = item_len;
1752                 sh.transid = found_transid;
1753
1754                 /* copy search result header */
1755                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1756                 *sk_offset += sizeof(sh);
1757
1758                 if (item_len) {
1759                         char *p = buf + *sk_offset;
1760                         /* copy the item */
1761                         read_extent_buffer(leaf, p,
1762                                            item_off, item_len);
1763                         *sk_offset += item_len;
1764                 }
1765                 (*num_found)++;
1766
1767                 if (*num_found >= sk->nr_items)
1768                         break;
1769         }
1770 advance_key:
1771         ret = 0;
1772         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1773                 key->offset++;
1774         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1775                 key->offset = 0;
1776                 key->type++;
1777         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1778                 key->offset = 0;
1779                 key->type = 0;
1780                 key->objectid++;
1781         } else
1782                 ret = 1;
1783 overflow:
1784         return ret;
1785 }
1786
1787 static noinline int search_ioctl(struct inode *inode,
1788                                  struct btrfs_ioctl_search_args *args)
1789 {
1790         struct btrfs_root *root;
1791         struct btrfs_key key;
1792         struct btrfs_key max_key;
1793         struct btrfs_path *path;
1794         struct btrfs_ioctl_search_key *sk = &args->key;
1795         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1796         int ret;
1797         int num_found = 0;
1798         unsigned long sk_offset = 0;
1799
1800         path = btrfs_alloc_path();
1801         if (!path)
1802                 return -ENOMEM;
1803
1804         if (sk->tree_id == 0) {
1805                 /* search the root of the inode that was passed */
1806                 root = BTRFS_I(inode)->root;
1807         } else {
1808                 key.objectid = sk->tree_id;
1809                 key.type = BTRFS_ROOT_ITEM_KEY;
1810                 key.offset = (u64)-1;
1811                 root = btrfs_read_fs_root_no_name(info, &key);
1812                 if (IS_ERR(root)) {
1813                         printk(KERN_ERR "could not find root %llu\n",
1814                                sk->tree_id);
1815                         btrfs_free_path(path);
1816                         return -ENOENT;
1817                 }
1818         }
1819
1820         key.objectid = sk->min_objectid;
1821         key.type = sk->min_type;
1822         key.offset = sk->min_offset;
1823
1824         max_key.objectid = sk->max_objectid;
1825         max_key.type = sk->max_type;
1826         max_key.offset = sk->max_offset;
1827
1828         path->keep_locks = 1;
1829
1830         while(1) {
1831                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1832                                            sk->min_transid);
1833                 if (ret != 0) {
1834                         if (ret > 0)
1835                                 ret = 0;
1836                         goto err;
1837                 }
1838                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1839                                  &sk_offset, &num_found);
1840                 btrfs_release_path(path);
1841                 if (ret || num_found >= sk->nr_items)
1842                         break;
1843
1844         }
1845         ret = 0;
1846 err:
1847         sk->nr_items = num_found;
1848         btrfs_free_path(path);
1849         return ret;
1850 }
1851
1852 static noinline int btrfs_ioctl_tree_search(struct file *file,
1853                                            void __user *argp)
1854 {
1855          struct btrfs_ioctl_search_args *args;
1856          struct inode *inode;
1857          int ret;
1858
1859         if (!capable(CAP_SYS_ADMIN))
1860                 return -EPERM;
1861
1862         args = memdup_user(argp, sizeof(*args));
1863         if (IS_ERR(args))
1864                 return PTR_ERR(args);
1865
1866         inode = fdentry(file)->d_inode;
1867         ret = search_ioctl(inode, args);
1868         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1869                 ret = -EFAULT;
1870         kfree(args);
1871         return ret;
1872 }
1873
1874 /*
1875  * Search INODE_REFs to identify path name of 'dirid' directory
1876  * in a 'tree_id' tree. and sets path name to 'name'.
1877  */
1878 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1879                                 u64 tree_id, u64 dirid, char *name)
1880 {
1881         struct btrfs_root *root;
1882         struct btrfs_key key;
1883         char *ptr;
1884         int ret = -1;
1885         int slot;
1886         int len;
1887         int total_len = 0;
1888         struct btrfs_inode_ref *iref;
1889         struct extent_buffer *l;
1890         struct btrfs_path *path;
1891
1892         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1893                 name[0]='\0';
1894                 return 0;
1895         }
1896
1897         path = btrfs_alloc_path();
1898         if (!path)
1899                 return -ENOMEM;
1900
1901         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1902
1903         key.objectid = 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", tree_id);
1909                 ret = -ENOENT;
1910                 goto out;
1911         }
1912
1913         key.objectid = dirid;
1914         key.type = BTRFS_INODE_REF_KEY;
1915         key.offset = (u64)-1;
1916
1917         while(1) {
1918                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1919                 if (ret < 0)
1920                         goto out;
1921
1922                 l = path->nodes[0];
1923                 slot = path->slots[0];
1924                 if (ret > 0 && slot > 0)
1925                         slot--;
1926                 btrfs_item_key_to_cpu(l, &key, slot);
1927
1928                 if (ret > 0 && (key.objectid != dirid ||
1929                                 key.type != BTRFS_INODE_REF_KEY)) {
1930                         ret = -ENOENT;
1931                         goto out;
1932                 }
1933
1934                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1935                 len = btrfs_inode_ref_name_len(l, iref);
1936                 ptr -= len + 1;
1937                 total_len += len + 1;
1938                 if (ptr < name)
1939                         goto out;
1940
1941                 *(ptr + len) = '/';
1942                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1943
1944                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1945                         break;
1946
1947                 btrfs_release_path(path);
1948                 key.objectid = key.offset;
1949                 key.offset = (u64)-1;
1950                 dirid = key.objectid;
1951         }
1952         if (ptr < name)
1953                 goto out;
1954         memmove(name, ptr, total_len);
1955         name[total_len]='\0';
1956         ret = 0;
1957 out:
1958         btrfs_free_path(path);
1959         return ret;
1960 }
1961
1962 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1963                                            void __user *argp)
1964 {
1965          struct btrfs_ioctl_ino_lookup_args *args;
1966          struct inode *inode;
1967          int ret;
1968
1969         if (!capable(CAP_SYS_ADMIN))
1970                 return -EPERM;
1971
1972         args = memdup_user(argp, sizeof(*args));
1973         if (IS_ERR(args))
1974                 return PTR_ERR(args);
1975
1976         inode = fdentry(file)->d_inode;
1977
1978         if (args->treeid == 0)
1979                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1980
1981         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1982                                         args->treeid, args->objectid,
1983                                         args->name);
1984
1985         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1986                 ret = -EFAULT;
1987
1988         kfree(args);
1989         return ret;
1990 }
1991
1992 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1993                                              void __user *arg)
1994 {
1995         struct dentry *parent = fdentry(file);
1996         struct dentry *dentry;
1997         struct inode *dir = parent->d_inode;
1998         struct inode *inode;
1999         struct btrfs_root *root = BTRFS_I(dir)->root;
2000         struct btrfs_root *dest = NULL;
2001         struct btrfs_ioctl_vol_args *vol_args;
2002         struct btrfs_trans_handle *trans;
2003         int namelen;
2004         int ret;
2005         int err = 0;
2006
2007         vol_args = memdup_user(arg, sizeof(*vol_args));
2008         if (IS_ERR(vol_args))
2009                 return PTR_ERR(vol_args);
2010
2011         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2012         namelen = strlen(vol_args->name);
2013         if (strchr(vol_args->name, '/') ||
2014             strncmp(vol_args->name, "..", namelen) == 0) {
2015                 err = -EINVAL;
2016                 goto out;
2017         }
2018
2019         err = mnt_want_write_file(file);
2020         if (err)
2021                 goto out;
2022
2023         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
2024         dentry = lookup_one_len(vol_args->name, parent, namelen);
2025         if (IS_ERR(dentry)) {
2026                 err = PTR_ERR(dentry);
2027                 goto out_unlock_dir;
2028         }
2029
2030         if (!dentry->d_inode) {
2031                 err = -ENOENT;
2032                 goto out_dput;
2033         }
2034
2035         inode = dentry->d_inode;
2036         dest = BTRFS_I(inode)->root;
2037         if (!capable(CAP_SYS_ADMIN)){
2038                 /*
2039                  * Regular user.  Only allow this with a special mount
2040                  * option, when the user has write+exec access to the
2041                  * subvol root, and when rmdir(2) would have been
2042                  * allowed.
2043                  *
2044                  * Note that this is _not_ check that the subvol is
2045                  * empty or doesn't contain data that we wouldn't
2046                  * otherwise be able to delete.
2047                  *
2048                  * Users who want to delete empty subvols should try
2049                  * rmdir(2).
2050                  */
2051                 err = -EPERM;
2052                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2053                         goto out_dput;
2054
2055                 /*
2056                  * Do not allow deletion if the parent dir is the same
2057                  * as the dir to be deleted.  That means the ioctl
2058                  * must be called on the dentry referencing the root
2059                  * of the subvol, not a random directory contained
2060                  * within it.
2061                  */
2062                 err = -EINVAL;
2063                 if (root == dest)
2064                         goto out_dput;
2065
2066                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2067                 if (err)
2068                         goto out_dput;
2069
2070                 /* check if subvolume may be deleted by a non-root user */
2071                 err = btrfs_may_delete(dir, dentry, 1);
2072                 if (err)
2073                         goto out_dput;
2074         }
2075
2076         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2077                 err = -EINVAL;
2078                 goto out_dput;
2079         }
2080
2081         mutex_lock(&inode->i_mutex);
2082         err = d_invalidate(dentry);
2083         if (err)
2084                 goto out_unlock;
2085
2086         down_write(&root->fs_info->subvol_sem);
2087
2088         err = may_destroy_subvol(dest);
2089         if (err)
2090                 goto out_up_write;
2091
2092         trans = btrfs_start_transaction(root, 0);
2093         if (IS_ERR(trans)) {
2094                 err = PTR_ERR(trans);
2095                 goto out_up_write;
2096         }
2097         trans->block_rsv = &root->fs_info->global_block_rsv;
2098
2099         ret = btrfs_unlink_subvol(trans, root, dir,
2100                                 dest->root_key.objectid,
2101                                 dentry->d_name.name,
2102                                 dentry->d_name.len);
2103         if (ret) {
2104                 err = ret;
2105                 btrfs_abort_transaction(trans, root, ret);
2106                 goto out_end_trans;
2107         }
2108
2109         btrfs_record_root_in_trans(trans, dest);
2110
2111         memset(&dest->root_item.drop_progress, 0,
2112                 sizeof(dest->root_item.drop_progress));
2113         dest->root_item.drop_level = 0;
2114         btrfs_set_root_refs(&dest->root_item, 0);
2115
2116         if (!xchg(&dest->orphan_item_inserted, 1)) {
2117                 ret = btrfs_insert_orphan_item(trans,
2118                                         root->fs_info->tree_root,
2119                                         dest->root_key.objectid);
2120                 if (ret) {
2121                         btrfs_abort_transaction(trans, root, ret);
2122                         err = ret;
2123                         goto out_end_trans;
2124                 }
2125         }
2126 out_end_trans:
2127         ret = btrfs_end_transaction(trans, root);
2128         if (ret && !err)
2129                 err = ret;
2130         inode->i_flags |= S_DEAD;
2131 out_up_write:
2132         up_write(&root->fs_info->subvol_sem);
2133 out_unlock:
2134         mutex_unlock(&inode->i_mutex);
2135         if (!err) {
2136                 shrink_dcache_sb(root->fs_info->sb);
2137                 btrfs_invalidate_inodes(dest);
2138                 d_delete(dentry);
2139         }
2140 out_dput:
2141         dput(dentry);
2142 out_unlock_dir:
2143         mutex_unlock(&dir->i_mutex);
2144         mnt_drop_write_file(file);
2145 out:
2146         kfree(vol_args);
2147         return err;
2148 }
2149
2150 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2151 {
2152         struct inode *inode = fdentry(file)->d_inode;
2153         struct btrfs_root *root = BTRFS_I(inode)->root;
2154         struct btrfs_ioctl_defrag_range_args *range;
2155         int ret;
2156
2157         if (btrfs_root_readonly(root))
2158                 return -EROFS;
2159
2160         ret = mnt_want_write_file(file);
2161         if (ret)
2162                 return ret;
2163
2164         switch (inode->i_mode & S_IFMT) {
2165         case S_IFDIR:
2166                 if (!capable(CAP_SYS_ADMIN)) {
2167                         ret = -EPERM;
2168                         goto out;
2169                 }
2170                 ret = btrfs_defrag_root(root, 0);
2171                 if (ret)
2172                         goto out;
2173                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2174                 break;
2175         case S_IFREG:
2176                 if (!(file->f_mode & FMODE_WRITE)) {
2177                         ret = -EINVAL;
2178                         goto out;
2179                 }
2180
2181                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2182                 if (!range) {
2183                         ret = -ENOMEM;
2184                         goto out;
2185                 }
2186
2187                 if (argp) {
2188                         if (copy_from_user(range, argp,
2189                                            sizeof(*range))) {
2190                                 ret = -EFAULT;
2191                                 kfree(range);
2192                                 goto out;
2193                         }
2194                         /* compression requires us to start the IO */
2195                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2196                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2197                                 range->extent_thresh = (u32)-1;
2198                         }
2199                 } else {
2200                         /* the rest are all set to zero by kzalloc */
2201                         range->len = (u64)-1;
2202                 }
2203                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2204                                         range, 0, 0);
2205                 if (ret > 0)
2206                         ret = 0;
2207                 kfree(range);
2208                 break;
2209         default:
2210                 ret = -EINVAL;
2211         }
2212 out:
2213         mnt_drop_write_file(file);
2214         return ret;
2215 }
2216
2217 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2218 {
2219         struct btrfs_ioctl_vol_args *vol_args;
2220         int ret;
2221
2222         if (!capable(CAP_SYS_ADMIN))
2223                 return -EPERM;
2224
2225         mutex_lock(&root->fs_info->volume_mutex);
2226         if (root->fs_info->balance_ctl) {
2227                 printk(KERN_INFO "btrfs: balance in progress\n");
2228                 ret = -EINVAL;
2229                 goto out;
2230         }
2231
2232         vol_args = memdup_user(arg, sizeof(*vol_args));
2233         if (IS_ERR(vol_args)) {
2234                 ret = PTR_ERR(vol_args);
2235                 goto out;
2236         }
2237
2238         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2239         ret = btrfs_init_new_device(root, vol_args->name);
2240
2241         kfree(vol_args);
2242 out:
2243         mutex_unlock(&root->fs_info->volume_mutex);
2244         return ret;
2245 }
2246
2247 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2248 {
2249         struct btrfs_ioctl_vol_args *vol_args;
2250         int ret;
2251
2252         if (!capable(CAP_SYS_ADMIN))
2253                 return -EPERM;
2254
2255         if (root->fs_info->sb->s_flags & MS_RDONLY)
2256                 return -EROFS;
2257
2258         mutex_lock(&root->fs_info->volume_mutex);
2259         if (root->fs_info->balance_ctl) {
2260                 printk(KERN_INFO "btrfs: balance in progress\n");
2261                 ret = -EINVAL;
2262                 goto out;
2263         }
2264
2265         vol_args = memdup_user(arg, sizeof(*vol_args));
2266         if (IS_ERR(vol_args)) {
2267                 ret = PTR_ERR(vol_args);
2268                 goto out;
2269         }
2270
2271         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2272         ret = btrfs_rm_device(root, vol_args->name);
2273
2274         kfree(vol_args);
2275 out:
2276         mutex_unlock(&root->fs_info->volume_mutex);
2277         return ret;
2278 }
2279
2280 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2281 {
2282         struct btrfs_ioctl_fs_info_args *fi_args;
2283         struct btrfs_device *device;
2284         struct btrfs_device *next;
2285         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2286         int ret = 0;
2287
2288         if (!capable(CAP_SYS_ADMIN))
2289                 return -EPERM;
2290
2291         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2292         if (!fi_args)
2293                 return -ENOMEM;
2294
2295         fi_args->num_devices = fs_devices->num_devices;
2296         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2297
2298         mutex_lock(&fs_devices->device_list_mutex);
2299         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2300                 if (device->devid > fi_args->max_id)
2301                         fi_args->max_id = device->devid;
2302         }
2303         mutex_unlock(&fs_devices->device_list_mutex);
2304
2305         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2306                 ret = -EFAULT;
2307
2308         kfree(fi_args);
2309         return ret;
2310 }
2311
2312 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2313 {
2314         struct btrfs_ioctl_dev_info_args *di_args;
2315         struct btrfs_device *dev;
2316         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2317         int ret = 0;
2318         char *s_uuid = NULL;
2319         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2320
2321         if (!capable(CAP_SYS_ADMIN))
2322                 return -EPERM;
2323
2324         di_args = memdup_user(arg, sizeof(*di_args));
2325         if (IS_ERR(di_args))
2326                 return PTR_ERR(di_args);
2327
2328         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2329                 s_uuid = di_args->uuid;
2330
2331         mutex_lock(&fs_devices->device_list_mutex);
2332         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2333         mutex_unlock(&fs_devices->device_list_mutex);
2334
2335         if (!dev) {
2336                 ret = -ENODEV;
2337                 goto out;
2338         }
2339
2340         di_args->devid = dev->devid;
2341         di_args->bytes_used = dev->bytes_used;
2342         di_args->total_bytes = dev->total_bytes;
2343         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2344         if (dev->name) {
2345                 struct rcu_string *name;
2346
2347                 rcu_read_lock();
2348                 name = rcu_dereference(dev->name);
2349                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2350                 rcu_read_unlock();
2351                 di_args->path[sizeof(di_args->path) - 1] = 0;
2352         } else {
2353                 di_args->path[0] = '\0';
2354         }
2355
2356 out:
2357         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2358                 ret = -EFAULT;
2359
2360         kfree(di_args);
2361         return ret;
2362 }
2363
2364 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2365                                        u64 off, u64 olen, u64 destoff)
2366 {
2367         struct inode *inode = fdentry(file)->d_inode;
2368         struct btrfs_root *root = BTRFS_I(inode)->root;
2369         struct file *src_file;
2370         struct inode *src;
2371         struct btrfs_trans_handle *trans;
2372         struct btrfs_path *path;
2373         struct extent_buffer *leaf;
2374         char *buf;
2375         struct btrfs_key key;
2376         u32 nritems;
2377         int slot;
2378         int ret;
2379         u64 len = olen;
2380         u64 bs = root->fs_info->sb->s_blocksize;
2381
2382         /*
2383          * TODO:
2384          * - split compressed inline extents.  annoying: we need to
2385          *   decompress into destination's address_space (the file offset
2386          *   may change, so source mapping won't do), then recompress (or
2387          *   otherwise reinsert) a subrange.
2388          * - allow ranges within the same file to be cloned (provided
2389          *   they don't overlap)?
2390          */
2391
2392         /* the destination must be opened for writing */
2393         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2394                 return -EINVAL;
2395
2396         if (btrfs_root_readonly(root))
2397                 return -EROFS;
2398
2399         ret = mnt_want_write_file(file);
2400         if (ret)
2401                 return ret;
2402
2403         src_file = fget(srcfd);
2404         if (!src_file) {
2405                 ret = -EBADF;
2406                 goto out_drop_write;
2407         }
2408
2409         ret = -EXDEV;
2410         if (src_file->f_path.mnt != file->f_path.mnt)
2411                 goto out_fput;
2412
2413         src = src_file->f_dentry->d_inode;
2414
2415         ret = -EINVAL;
2416         if (src == inode)
2417                 goto out_fput;
2418
2419         /* the src must be open for reading */
2420         if (!(src_file->f_mode & FMODE_READ))
2421                 goto out_fput;
2422
2423         /* don't make the dst file partly checksummed */
2424         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2425             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2426                 goto out_fput;
2427
2428         ret = -EISDIR;
2429         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2430                 goto out_fput;
2431
2432         ret = -EXDEV;
2433         if (src->i_sb != inode->i_sb)
2434                 goto out_fput;
2435
2436         ret = -ENOMEM;
2437         buf = vmalloc(btrfs_level_size(root, 0));
2438         if (!buf)
2439                 goto out_fput;
2440
2441         path = btrfs_alloc_path();
2442         if (!path) {
2443                 vfree(buf);
2444                 goto out_fput;
2445         }
2446         path->reada = 2;
2447
2448         if (inode < src) {
2449                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2450                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2451         } else {
2452                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2453                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2454         }
2455
2456         /* determine range to clone */
2457         ret = -EINVAL;
2458         if (off + len > src->i_size || off + len < off)
2459                 goto out_unlock;
2460         if (len == 0)
2461                 olen = len = src->i_size - off;
2462         /* if we extend to eof, continue to block boundary */
2463         if (off + len == src->i_size)
2464                 len = ALIGN(src->i_size, bs) - off;
2465
2466         /* verify the end result is block aligned */
2467         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2468             !IS_ALIGNED(destoff, bs))
2469                 goto out_unlock;
2470
2471         if (destoff > inode->i_size) {
2472                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2473                 if (ret)
2474                         goto out_unlock;
2475         }
2476
2477         /* truncate page cache pages from target inode range */
2478         truncate_inode_pages_range(&inode->i_data, destoff,
2479                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2480
2481         /* do any pending delalloc/csum calc on src, one way or
2482            another, and lock file content */
2483         while (1) {
2484                 struct btrfs_ordered_extent *ordered;
2485                 lock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2486                 ordered = btrfs_lookup_first_ordered_extent(src, off + len - 1);
2487                 if (!ordered &&
2488                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off + len - 1,
2489                                     EXTENT_DELALLOC, 0, NULL))
2490                         break;
2491                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2492                 if (ordered)
2493                         btrfs_put_ordered_extent(ordered);
2494                 btrfs_wait_ordered_range(src, off, len);
2495         }
2496
2497         /* clone data */
2498         key.objectid = btrfs_ino(src);
2499         key.type = BTRFS_EXTENT_DATA_KEY;
2500         key.offset = 0;
2501
2502         while (1) {
2503                 /*
2504                  * note the key will change type as we walk through the
2505                  * tree.
2506                  */
2507                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2508                                 0, 0);
2509                 if (ret < 0)
2510                         goto out;
2511
2512                 nritems = btrfs_header_nritems(path->nodes[0]);
2513                 if (path->slots[0] >= nritems) {
2514                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2515                         if (ret < 0)
2516                                 goto out;
2517                         if (ret > 0)
2518                                 break;
2519                         nritems = btrfs_header_nritems(path->nodes[0]);
2520                 }
2521                 leaf = path->nodes[0];
2522                 slot = path->slots[0];
2523
2524                 btrfs_item_key_to_cpu(leaf, &key, slot);
2525                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2526                     key.objectid != btrfs_ino(src))
2527                         break;
2528
2529                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2530                         struct btrfs_file_extent_item *extent;
2531                         int type;
2532                         u32 size;
2533                         struct btrfs_key new_key;
2534                         u64 disko = 0, diskl = 0;
2535                         u64 datao = 0, datal = 0;
2536                         u8 comp;
2537                         u64 endoff;
2538
2539                         size = btrfs_item_size_nr(leaf, slot);
2540                         read_extent_buffer(leaf, buf,
2541                                            btrfs_item_ptr_offset(leaf, slot),
2542                                            size);
2543
2544                         extent = btrfs_item_ptr(leaf, slot,
2545                                                 struct btrfs_file_extent_item);
2546                         comp = btrfs_file_extent_compression(leaf, extent);
2547                         type = btrfs_file_extent_type(leaf, extent);
2548                         if (type == BTRFS_FILE_EXTENT_REG ||
2549                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2550                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2551                                                                       extent);
2552                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2553                                                                  extent);
2554                                 datao = btrfs_file_extent_offset(leaf, extent);
2555                                 datal = btrfs_file_extent_num_bytes(leaf,
2556                                                                     extent);
2557                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2558                                 /* take upper bound, may be compressed */
2559                                 datal = btrfs_file_extent_ram_bytes(leaf,
2560                                                                     extent);
2561                         }
2562                         btrfs_release_path(path);
2563
2564                         if (key.offset + datal <= off ||
2565                             key.offset >= off + len - 1)
2566                                 goto next;
2567
2568                         memcpy(&new_key, &key, sizeof(new_key));
2569                         new_key.objectid = btrfs_ino(inode);
2570                         if (off <= key.offset)
2571                                 new_key.offset = key.offset + destoff - off;
2572                         else
2573                                 new_key.offset = destoff;
2574
2575                         /*
2576                          * 1 - adjusting old extent (we may have to split it)
2577                          * 1 - add new extent
2578                          * 1 - inode update
2579                          */
2580                         trans = btrfs_start_transaction(root, 3);
2581                         if (IS_ERR(trans)) {
2582                                 ret = PTR_ERR(trans);
2583                                 goto out;
2584                         }
2585
2586                         if (type == BTRFS_FILE_EXTENT_REG ||
2587                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2588                                 /*
2589                                  *    a  | --- range to clone ---|  b
2590                                  * | ------------- extent ------------- |
2591                                  */
2592
2593                                 /* substract range b */
2594                                 if (key.offset + datal > off + len)
2595                                         datal = off + len - key.offset;
2596
2597                                 /* substract range a */
2598                                 if (off > key.offset) {
2599                                         datao += off - key.offset;
2600                                         datal -= off - key.offset;
2601                                 }
2602
2603                                 ret = btrfs_drop_extents(trans, root, inode,
2604                                                          new_key.offset,
2605                                                          new_key.offset + datal,
2606                                                          1);
2607                                 if (ret) {
2608                                         btrfs_abort_transaction(trans, root,
2609                                                                 ret);
2610                                         btrfs_end_transaction(trans, root);
2611                                         goto out;
2612                                 }
2613
2614                                 ret = btrfs_insert_empty_item(trans, root, path,
2615                                                               &new_key, size);
2616                                 if (ret) {
2617                                         btrfs_abort_transaction(trans, root,
2618                                                                 ret);
2619                                         btrfs_end_transaction(trans, root);
2620                                         goto out;
2621                                 }
2622
2623                                 leaf = path->nodes[0];
2624                                 slot = path->slots[0];
2625                                 write_extent_buffer(leaf, buf,
2626                                             btrfs_item_ptr_offset(leaf, slot),
2627                                             size);
2628
2629                                 extent = btrfs_item_ptr(leaf, slot,
2630                                                 struct btrfs_file_extent_item);
2631
2632                                 /* disko == 0 means it's a hole */
2633                                 if (!disko)
2634                                         datao = 0;
2635
2636                                 btrfs_set_file_extent_offset(leaf, extent,
2637                                                              datao);
2638                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2639                                                                 datal);
2640                                 if (disko) {
2641                                         inode_add_bytes(inode, datal);
2642                                         ret = btrfs_inc_extent_ref(trans, root,
2643                                                         disko, diskl, 0,
2644                                                         root->root_key.objectid,
2645                                                         btrfs_ino(inode),
2646                                                         new_key.offset - datao,
2647                                                         0);
2648                                         if (ret) {
2649                                                 btrfs_abort_transaction(trans,
2650                                                                         root,
2651                                                                         ret);
2652                                                 btrfs_end_transaction(trans,
2653                                                                       root);
2654                                                 goto out;
2655
2656                                         }
2657                                 }
2658                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2659                                 u64 skip = 0;
2660                                 u64 trim = 0;
2661                                 if (off > key.offset) {
2662                                         skip = off - key.offset;
2663                                         new_key.offset += skip;
2664                                 }
2665
2666                                 if (key.offset + datal > off + len)
2667                                         trim = key.offset + datal - (off + len);
2668
2669                                 if (comp && (skip || trim)) {
2670                                         ret = -EINVAL;
2671                                         btrfs_end_transaction(trans, root);
2672                                         goto out;
2673                                 }
2674                                 size -= skip + trim;
2675                                 datal -= skip + trim;
2676
2677                                 ret = btrfs_drop_extents(trans, root, inode,
2678                                                          new_key.offset,
2679                                                          new_key.offset + datal,
2680                                                          1);
2681                                 if (ret) {
2682                                         btrfs_abort_transaction(trans, root,
2683                                                                 ret);
2684                                         btrfs_end_transaction(trans, root);
2685                                         goto out;
2686                                 }
2687
2688                                 ret = btrfs_insert_empty_item(trans, root, path,
2689                                                               &new_key, size);
2690                                 if (ret) {
2691                                         btrfs_abort_transaction(trans, root,
2692                                                                 ret);
2693                                         btrfs_end_transaction(trans, root);
2694                                         goto out;
2695                                 }
2696
2697                                 if (skip) {
2698                                         u32 start =
2699                                           btrfs_file_extent_calc_inline_size(0);
2700                                         memmove(buf+start, buf+start+skip,
2701                                                 datal);
2702                                 }
2703
2704                                 leaf = path->nodes[0];
2705                                 slot = path->slots[0];
2706                                 write_extent_buffer(leaf, buf,
2707                                             btrfs_item_ptr_offset(leaf, slot),
2708                                             size);
2709                                 inode_add_bytes(inode, datal);
2710                         }
2711
2712                         btrfs_mark_buffer_dirty(leaf);
2713                         btrfs_release_path(path);
2714
2715                         inode_inc_iversion(inode);
2716                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2717
2718                         /*
2719                          * we round up to the block size at eof when
2720                          * determining which extents to clone above,
2721                          * but shouldn't round up the file size
2722                          */
2723                         endoff = new_key.offset + datal;
2724                         if (endoff > destoff+olen)
2725                                 endoff = destoff+olen;
2726                         if (endoff > inode->i_size)
2727                                 btrfs_i_size_write(inode, endoff);
2728
2729                         ret = btrfs_update_inode(trans, root, inode);
2730                         if (ret) {
2731                                 btrfs_abort_transaction(trans, root, ret);
2732                                 btrfs_end_transaction(trans, root);
2733                                 goto out;
2734                         }
2735                         ret = btrfs_end_transaction(trans, root);
2736                 }
2737 next:
2738                 btrfs_release_path(path);
2739                 key.offset++;
2740         }
2741         ret = 0;
2742 out:
2743         btrfs_release_path(path);
2744         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2745 out_unlock:
2746         mutex_unlock(&src->i_mutex);
2747         mutex_unlock(&inode->i_mutex);
2748         vfree(buf);
2749         btrfs_free_path(path);
2750 out_fput:
2751         fput(src_file);
2752 out_drop_write:
2753         mnt_drop_write_file(file);
2754         return ret;
2755 }
2756
2757 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2758 {
2759         struct btrfs_ioctl_clone_range_args args;
2760
2761         if (copy_from_user(&args, argp, sizeof(args)))
2762                 return -EFAULT;
2763         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2764                                  args.src_length, args.dest_offset);
2765 }
2766
2767 /*
2768  * there are many ways the trans_start and trans_end ioctls can lead
2769  * to deadlocks.  They should only be used by applications that
2770  * basically own the machine, and have a very in depth understanding
2771  * of all the possible deadlocks and enospc problems.
2772  */
2773 static long btrfs_ioctl_trans_start(struct file *file)
2774 {
2775         struct inode *inode = fdentry(file)->d_inode;
2776         struct btrfs_root *root = BTRFS_I(inode)->root;
2777         struct btrfs_trans_handle *trans;
2778         int ret;
2779
2780         ret = -EPERM;
2781         if (!capable(CAP_SYS_ADMIN))
2782                 goto out;
2783
2784         ret = -EINPROGRESS;
2785         if (file->private_data)
2786                 goto out;
2787
2788         ret = -EROFS;
2789         if (btrfs_root_readonly(root))
2790                 goto out;
2791
2792         ret = mnt_want_write_file(file);
2793         if (ret)
2794                 goto out;
2795
2796         atomic_inc(&root->fs_info->open_ioctl_trans);
2797
2798         ret = -ENOMEM;
2799         trans = btrfs_start_ioctl_transaction(root);
2800         if (IS_ERR(trans))
2801                 goto out_drop;
2802
2803         file->private_data = trans;
2804         return 0;
2805
2806 out_drop:
2807         atomic_dec(&root->fs_info->open_ioctl_trans);
2808         mnt_drop_write_file(file);
2809 out:
2810         return ret;
2811 }
2812
2813 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2814 {
2815         struct inode *inode = fdentry(file)->d_inode;
2816         struct btrfs_root *root = BTRFS_I(inode)->root;
2817         struct btrfs_root *new_root;
2818         struct btrfs_dir_item *di;
2819         struct btrfs_trans_handle *trans;
2820         struct btrfs_path *path;
2821         struct btrfs_key location;
2822         struct btrfs_disk_key disk_key;
2823         u64 objectid = 0;
2824         u64 dir_id;
2825
2826         if (!capable(CAP_SYS_ADMIN))
2827                 return -EPERM;
2828
2829         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2830                 return -EFAULT;
2831
2832         if (!objectid)
2833                 objectid = root->root_key.objectid;
2834
2835         location.objectid = objectid;
2836         location.type = BTRFS_ROOT_ITEM_KEY;
2837         location.offset = (u64)-1;
2838
2839         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2840         if (IS_ERR(new_root))
2841                 return PTR_ERR(new_root);
2842
2843         if (btrfs_root_refs(&new_root->root_item) == 0)
2844                 return -ENOENT;
2845
2846         path = btrfs_alloc_path();
2847         if (!path)
2848                 return -ENOMEM;
2849         path->leave_spinning = 1;
2850
2851         trans = btrfs_start_transaction(root, 1);
2852         if (IS_ERR(trans)) {
2853                 btrfs_free_path(path);
2854                 return PTR_ERR(trans);
2855         }
2856
2857         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2858         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2859                                    dir_id, "default", 7, 1);
2860         if (IS_ERR_OR_NULL(di)) {
2861                 btrfs_free_path(path);
2862                 btrfs_end_transaction(trans, root);
2863                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2864                        "this isn't going to work\n");
2865                 return -ENOENT;
2866         }
2867
2868         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2869         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2870         btrfs_mark_buffer_dirty(path->nodes[0]);
2871         btrfs_free_path(path);
2872
2873         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
2874         btrfs_end_transaction(trans, root);
2875
2876         return 0;
2877 }
2878
2879 void btrfs_get_block_group_info(struct list_head *groups_list,
2880                                 struct btrfs_ioctl_space_info *space)
2881 {
2882         struct btrfs_block_group_cache *block_group;
2883
2884         space->total_bytes = 0;
2885         space->used_bytes = 0;
2886         space->flags = 0;
2887         list_for_each_entry(block_group, groups_list, list) {
2888                 space->flags = block_group->flags;
2889                 space->total_bytes += block_group->key.offset;
2890                 space->used_bytes +=
2891                         btrfs_block_group_used(&block_group->item);
2892         }
2893 }
2894
2895 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2896 {
2897         struct btrfs_ioctl_space_args space_args;
2898         struct btrfs_ioctl_space_info space;
2899         struct btrfs_ioctl_space_info *dest;
2900         struct btrfs_ioctl_space_info *dest_orig;
2901         struct btrfs_ioctl_space_info __user *user_dest;
2902         struct btrfs_space_info *info;
2903         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2904                        BTRFS_BLOCK_GROUP_SYSTEM,
2905                        BTRFS_BLOCK_GROUP_METADATA,
2906                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2907         int num_types = 4;
2908         int alloc_size;
2909         int ret = 0;
2910         u64 slot_count = 0;
2911         int i, c;
2912
2913         if (copy_from_user(&space_args,
2914                            (struct btrfs_ioctl_space_args __user *)arg,
2915                            sizeof(space_args)))
2916                 return -EFAULT;
2917
2918         for (i = 0; i < num_types; i++) {
2919                 struct btrfs_space_info *tmp;
2920
2921                 info = NULL;
2922                 rcu_read_lock();
2923                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2924                                         list) {
2925                         if (tmp->flags == types[i]) {
2926                                 info = tmp;
2927                                 break;
2928                         }
2929                 }
2930                 rcu_read_unlock();
2931
2932                 if (!info)
2933                         continue;
2934
2935                 down_read(&info->groups_sem);
2936                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2937                         if (!list_empty(&info->block_groups[c]))
2938                                 slot_count++;
2939                 }
2940                 up_read(&info->groups_sem);
2941         }
2942
2943         /* space_slots == 0 means they are asking for a count */
2944         if (space_args.space_slots == 0) {
2945                 space_args.total_spaces = slot_count;
2946                 goto out;
2947         }
2948
2949         slot_count = min_t(u64, space_args.space_slots, slot_count);
2950
2951         alloc_size = sizeof(*dest) * slot_count;
2952
2953         /* we generally have at most 6 or so space infos, one for each raid
2954          * level.  So, a whole page should be more than enough for everyone
2955          */
2956         if (alloc_size > PAGE_CACHE_SIZE)
2957                 return -ENOMEM;
2958
2959         space_args.total_spaces = 0;
2960         dest = kmalloc(alloc_size, GFP_NOFS);
2961         if (!dest)
2962                 return -ENOMEM;
2963         dest_orig = dest;
2964
2965         /* now we have a buffer to copy into */
2966         for (i = 0; i < num_types; i++) {
2967                 struct btrfs_space_info *tmp;
2968
2969                 if (!slot_count)
2970                         break;
2971
2972                 info = NULL;
2973                 rcu_read_lock();
2974                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2975                                         list) {
2976                         if (tmp->flags == types[i]) {
2977                                 info = tmp;
2978                                 break;
2979                         }
2980                 }
2981                 rcu_read_unlock();
2982
2983                 if (!info)
2984                         continue;
2985                 down_read(&info->groups_sem);
2986                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2987                         if (!list_empty(&info->block_groups[c])) {
2988                                 btrfs_get_block_group_info(
2989                                         &info->block_groups[c], &space);
2990                                 memcpy(dest, &space, sizeof(space));
2991                                 dest++;
2992                                 space_args.total_spaces++;
2993                                 slot_count--;
2994                         }
2995                         if (!slot_count)
2996                                 break;
2997                 }
2998                 up_read(&info->groups_sem);
2999         }
3000
3001         user_dest = (struct btrfs_ioctl_space_info __user *)
3002                 (arg + sizeof(struct btrfs_ioctl_space_args));
3003
3004         if (copy_to_user(user_dest, dest_orig, alloc_size))
3005                 ret = -EFAULT;
3006
3007         kfree(dest_orig);
3008 out:
3009         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3010                 ret = -EFAULT;
3011
3012         return ret;
3013 }
3014
3015 /*
3016  * there are many ways the trans_start and trans_end ioctls can lead
3017  * to deadlocks.  They should only be used by applications that
3018  * basically own the machine, and have a very in depth understanding
3019  * of all the possible deadlocks and enospc problems.
3020  */
3021 long btrfs_ioctl_trans_end(struct file *file)
3022 {
3023         struct inode *inode = fdentry(file)->d_inode;
3024         struct btrfs_root *root = BTRFS_I(inode)->root;
3025         struct btrfs_trans_handle *trans;
3026
3027         trans = file->private_data;
3028         if (!trans)
3029                 return -EINVAL;
3030         file->private_data = NULL;
3031
3032         btrfs_end_transaction(trans, root);
3033
3034         atomic_dec(&root->fs_info->open_ioctl_trans);
3035
3036         mnt_drop_write_file(file);
3037         return 0;
3038 }
3039
3040 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
3041 {
3042         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
3043         struct btrfs_trans_handle *trans;
3044         u64 transid;
3045         int ret;
3046
3047         trans = btrfs_start_transaction(root, 0);
3048         if (IS_ERR(trans))
3049                 return PTR_ERR(trans);
3050         transid = trans->transid;
3051         ret = btrfs_commit_transaction_async(trans, root, 0);
3052         if (ret) {
3053                 btrfs_end_transaction(trans, root);
3054                 return ret;
3055         }
3056
3057         if (argp)
3058                 if (copy_to_user(argp, &transid, sizeof(transid)))
3059                         return -EFAULT;
3060         return 0;
3061 }
3062
3063 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
3064 {
3065         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
3066         u64 transid;
3067
3068         if (argp) {
3069                 if (copy_from_user(&transid, argp, sizeof(transid)))
3070                         return -EFAULT;
3071         } else {
3072                 transid = 0;  /* current trans */
3073         }
3074         return btrfs_wait_for_commit(root, transid);
3075 }
3076
3077 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
3078 {
3079         int ret;
3080         struct btrfs_ioctl_scrub_args *sa;
3081
3082         if (!capable(CAP_SYS_ADMIN))
3083                 return -EPERM;
3084
3085         sa = memdup_user(arg, sizeof(*sa));
3086         if (IS_ERR(sa))
3087                 return PTR_ERR(sa);
3088
3089         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
3090                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
3091
3092         if (copy_to_user(arg, sa, sizeof(*sa)))
3093                 ret = -EFAULT;
3094
3095         kfree(sa);
3096         return ret;
3097 }
3098
3099 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3100 {
3101         if (!capable(CAP_SYS_ADMIN))
3102                 return -EPERM;
3103
3104         return btrfs_scrub_cancel(root);
3105 }
3106
3107 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3108                                        void __user *arg)
3109 {
3110         struct btrfs_ioctl_scrub_args *sa;
3111         int ret;
3112
3113         if (!capable(CAP_SYS_ADMIN))
3114                 return -EPERM;
3115
3116         sa = memdup_user(arg, sizeof(*sa));
3117         if (IS_ERR(sa))
3118                 return PTR_ERR(sa);
3119
3120         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3121
3122         if (copy_to_user(arg, sa, sizeof(*sa)))
3123                 ret = -EFAULT;
3124
3125         kfree(sa);
3126         return ret;
3127 }
3128
3129 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3130                                       void __user *arg)
3131 {
3132         struct btrfs_ioctl_get_dev_stats *sa;
3133         int ret;
3134
3135         sa = memdup_user(arg, sizeof(*sa));
3136         if (IS_ERR(sa))
3137                 return PTR_ERR(sa);
3138
3139         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3140                 kfree(sa);
3141                 return -EPERM;
3142         }
3143
3144         ret = btrfs_get_dev_stats(root, sa);
3145
3146         if (copy_to_user(arg, sa, sizeof(*sa)))
3147                 ret = -EFAULT;
3148
3149         kfree(sa);
3150         return ret;
3151 }
3152
3153 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3154 {
3155         int ret = 0;
3156         int i;
3157         u64 rel_ptr;
3158         int size;
3159         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3160         struct inode_fs_paths *ipath = NULL;
3161         struct btrfs_path *path;
3162
3163         if (!capable(CAP_SYS_ADMIN))
3164                 return -EPERM;
3165
3166         path = btrfs_alloc_path();
3167         if (!path) {
3168                 ret = -ENOMEM;
3169                 goto out;
3170         }
3171
3172         ipa = memdup_user(arg, sizeof(*ipa));
3173         if (IS_ERR(ipa)) {
3174                 ret = PTR_ERR(ipa);
3175                 ipa = NULL;
3176                 goto out;
3177         }
3178
3179         size = min_t(u32, ipa->size, 4096);
3180         ipath = init_ipath(size, root, path);
3181         if (IS_ERR(ipath)) {
3182                 ret = PTR_ERR(ipath);
3183                 ipath = NULL;
3184                 goto out;
3185         }
3186
3187         ret = paths_from_inode(ipa->inum, ipath);
3188         if (ret < 0)
3189                 goto out;
3190
3191         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3192                 rel_ptr = ipath->fspath->val[i] -
3193                           (u64)(unsigned long)ipath->fspath->val;
3194                 ipath->fspath->val[i] = rel_ptr;
3195         }
3196
3197         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3198                            (void *)(unsigned long)ipath->fspath, size);
3199         if (ret) {
3200                 ret = -EFAULT;
3201                 goto out;
3202         }
3203
3204 out:
3205         btrfs_free_path(path);
3206         free_ipath(ipath);
3207         kfree(ipa);
3208
3209         return ret;
3210 }
3211
3212 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3213 {
3214         struct btrfs_data_container *inodes = ctx;
3215         const size_t c = 3 * sizeof(u64);
3216
3217         if (inodes->bytes_left >= c) {
3218                 inodes->bytes_left -= c;
3219                 inodes->val[inodes->elem_cnt] = inum;
3220                 inodes->val[inodes->elem_cnt + 1] = offset;
3221                 inodes->val[inodes->elem_cnt + 2] = root;
3222                 inodes->elem_cnt += 3;
3223         } else {
3224                 inodes->bytes_missing += c - inodes->bytes_left;
3225                 inodes->bytes_left = 0;
3226                 inodes->elem_missed += 3;
3227         }
3228
3229         return 0;
3230 }
3231
3232 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3233                                         void __user *arg)
3234 {
3235         int ret = 0;
3236         int size;
3237         struct btrfs_ioctl_logical_ino_args *loi;
3238         struct btrfs_data_container *inodes = NULL;
3239         struct btrfs_path *path = NULL;
3240
3241         if (!capable(CAP_SYS_ADMIN))
3242                 return -EPERM;
3243
3244         loi = memdup_user(arg, sizeof(*loi));
3245         if (IS_ERR(loi)) {
3246                 ret = PTR_ERR(loi);
3247                 loi = NULL;
3248                 goto out;
3249         }
3250
3251         path = btrfs_alloc_path();
3252         if (!path) {
3253                 ret = -ENOMEM;
3254                 goto out;
3255         }
3256
3257         size = min_t(u32, loi->size, 64 * 1024);
3258         inodes = init_data_container(size);
3259         if (IS_ERR(inodes)) {
3260                 ret = PTR_ERR(inodes);
3261                 inodes = NULL;
3262                 goto out;
3263         }
3264
3265         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3266                                           build_ino_list, inodes);
3267         if (ret == -EINVAL)
3268                 ret = -ENOENT;
3269         if (ret < 0)
3270                 goto out;
3271
3272         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3273                            (void *)(unsigned long)inodes, size);
3274         if (ret)
3275                 ret = -EFAULT;
3276
3277 out:
3278         btrfs_free_path(path);
3279         vfree(inodes);
3280         kfree(loi);
3281
3282         return ret;
3283 }
3284
3285 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3286                                struct btrfs_ioctl_balance_args *bargs)
3287 {
3288         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3289
3290         bargs->flags = bctl->flags;
3291
3292         if (atomic_read(&fs_info->balance_running))
3293                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3294         if (atomic_read(&fs_info->balance_pause_req))
3295                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3296         if (atomic_read(&fs_info->balance_cancel_req))
3297                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3298
3299         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3300         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3301         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3302
3303         if (lock) {
3304                 spin_lock(&fs_info->balance_lock);
3305                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3306                 spin_unlock(&fs_info->balance_lock);
3307         } else {
3308                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3309         }
3310 }
3311
3312 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3313 {
3314         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3315         struct btrfs_fs_info *fs_info = root->fs_info;
3316         struct btrfs_ioctl_balance_args *bargs;
3317         struct btrfs_balance_control *bctl;
3318         int ret;
3319
3320         if (!capable(CAP_SYS_ADMIN))
3321                 return -EPERM;
3322
3323         ret = mnt_want_write_file(file);
3324         if (ret)
3325                 return ret;
3326
3327         mutex_lock(&fs_info->volume_mutex);
3328         mutex_lock(&fs_info->balance_mutex);
3329
3330         if (arg) {
3331                 bargs = memdup_user(arg, sizeof(*bargs));
3332                 if (IS_ERR(bargs)) {
3333                         ret = PTR_ERR(bargs);
3334                         goto out;
3335                 }
3336
3337                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3338                         if (!fs_info->balance_ctl) {
3339                                 ret = -ENOTCONN;
3340                                 goto out_bargs;
3341                         }
3342
3343                         bctl = fs_info->balance_ctl;
3344                         spin_lock(&fs_info->balance_lock);
3345                         bctl->flags |= BTRFS_BALANCE_RESUME;
3346                         spin_unlock(&fs_info->balance_lock);
3347
3348                         goto do_balance;
3349                 }
3350         } else {
3351                 bargs = NULL;
3352         }
3353
3354         if (fs_info->balance_ctl) {
3355                 ret = -EINPROGRESS;
3356                 goto out_bargs;
3357         }
3358
3359         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3360         if (!bctl) {
3361                 ret = -ENOMEM;
3362                 goto out_bargs;
3363         }
3364
3365         bctl->fs_info = fs_info;
3366         if (arg) {
3367                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3368                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3369                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3370
3371                 bctl->flags = bargs->flags;
3372         } else {
3373                 /* balance everything - no filters */
3374                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3375         }
3376
3377 do_balance:
3378         ret = btrfs_balance(bctl, bargs);
3379         /*
3380          * bctl is freed in __cancel_balance or in free_fs_info if
3381          * restriper was paused all the way until unmount
3382          */
3383         if (arg) {
3384                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3385                         ret = -EFAULT;
3386         }
3387
3388 out_bargs:
3389         kfree(bargs);
3390 out:
3391         mutex_unlock(&fs_info->balance_mutex);
3392         mutex_unlock(&fs_info->volume_mutex);
3393         mnt_drop_write_file(file);
3394         return ret;
3395 }
3396
3397 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3398 {
3399         if (!capable(CAP_SYS_ADMIN))
3400                 return -EPERM;
3401
3402         switch (cmd) {
3403         case BTRFS_BALANCE_CTL_PAUSE:
3404                 return btrfs_pause_balance(root->fs_info);
3405         case BTRFS_BALANCE_CTL_CANCEL:
3406                 return btrfs_cancel_balance(root->fs_info);
3407         }
3408
3409         return -EINVAL;
3410 }
3411
3412 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3413                                          void __user *arg)
3414 {
3415         struct btrfs_fs_info *fs_info = root->fs_info;
3416         struct btrfs_ioctl_balance_args *bargs;
3417         int ret = 0;
3418
3419         if (!capable(CAP_SYS_ADMIN))
3420                 return -EPERM;
3421
3422         mutex_lock(&fs_info->balance_mutex);
3423         if (!fs_info->balance_ctl) {
3424                 ret = -ENOTCONN;
3425                 goto out;
3426         }
3427
3428         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3429         if (!bargs) {
3430                 ret = -ENOMEM;
3431                 goto out;
3432         }
3433
3434         update_ioctl_balance_args(fs_info, 1, bargs);
3435
3436         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3437                 ret = -EFAULT;
3438
3439         kfree(bargs);
3440 out:
3441         mutex_unlock(&fs_info->balance_mutex);
3442         return ret;
3443 }
3444
3445 static long btrfs_ioctl_quota_ctl(struct btrfs_root *root, void __user *arg)
3446 {
3447         struct btrfs_ioctl_quota_ctl_args *sa;
3448         struct btrfs_trans_handle *trans = NULL;
3449         int ret;
3450         int err;
3451
3452         if (!capable(CAP_SYS_ADMIN))
3453                 return -EPERM;
3454
3455         if (root->fs_info->sb->s_flags & MS_RDONLY)
3456                 return -EROFS;
3457
3458         sa = memdup_user(arg, sizeof(*sa));
3459         if (IS_ERR(sa))
3460                 return PTR_ERR(sa);
3461
3462         if (sa->cmd != BTRFS_QUOTA_CTL_RESCAN) {
3463                 trans = btrfs_start_transaction(root, 2);
3464                 if (IS_ERR(trans)) {
3465                         ret = PTR_ERR(trans);
3466                         goto out;
3467                 }
3468         }
3469
3470         switch (sa->cmd) {
3471         case BTRFS_QUOTA_CTL_ENABLE:
3472                 ret = btrfs_quota_enable(trans, root->fs_info);
3473                 break;
3474         case BTRFS_QUOTA_CTL_DISABLE:
3475                 ret = btrfs_quota_disable(trans, root->fs_info);
3476                 break;
3477         case BTRFS_QUOTA_CTL_RESCAN:
3478                 ret = btrfs_quota_rescan(root->fs_info);
3479                 break;
3480         default:
3481                 ret = -EINVAL;
3482                 break;
3483         }
3484
3485         if (copy_to_user(arg, sa, sizeof(*sa)))
3486                 ret = -EFAULT;
3487
3488         if (trans) {
3489                 err = btrfs_commit_transaction(trans, root);
3490                 if (err && !ret)
3491                         ret = err;
3492         }
3493
3494 out:
3495         kfree(sa);
3496         return ret;
3497 }
3498
3499 static long btrfs_ioctl_qgroup_assign(struct btrfs_root *root, void __user *arg)
3500 {
3501         struct btrfs_ioctl_qgroup_assign_args *sa;
3502         struct btrfs_trans_handle *trans;
3503         int ret;
3504         int err;
3505
3506         if (!capable(CAP_SYS_ADMIN))
3507                 return -EPERM;
3508
3509         if (root->fs_info->sb->s_flags & MS_RDONLY)
3510                 return -EROFS;
3511
3512         sa = memdup_user(arg, sizeof(*sa));
3513         if (IS_ERR(sa))
3514                 return PTR_ERR(sa);
3515
3516         trans = btrfs_join_transaction(root);
3517         if (IS_ERR(trans)) {
3518                 ret = PTR_ERR(trans);
3519                 goto out;
3520         }
3521
3522         /* FIXME: check if the IDs really exist */
3523         if (sa->assign) {
3524                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
3525                                                 sa->src, sa->dst);
3526         } else {
3527                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
3528                                                 sa->src, sa->dst);
3529         }
3530
3531         err = btrfs_end_transaction(trans, root);
3532         if (err && !ret)
3533                 ret = err;
3534
3535 out:
3536         kfree(sa);
3537         return ret;
3538 }
3539
3540 static long btrfs_ioctl_qgroup_create(struct btrfs_root *root, void __user *arg)
3541 {
3542         struct btrfs_ioctl_qgroup_create_args *sa;
3543         struct btrfs_trans_handle *trans;
3544         int ret;
3545         int err;
3546
3547         if (!capable(CAP_SYS_ADMIN))
3548                 return -EPERM;
3549
3550         if (root->fs_info->sb->s_flags & MS_RDONLY)
3551                 return -EROFS;
3552
3553         sa = memdup_user(arg, sizeof(*sa));
3554         if (IS_ERR(sa))
3555                 return PTR_ERR(sa);
3556
3557         trans = btrfs_join_transaction(root);
3558         if (IS_ERR(trans)) {
3559                 ret = PTR_ERR(trans);
3560                 goto out;
3561         }
3562
3563         /* FIXME: check if the IDs really exist */
3564         if (sa->create) {
3565                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
3566                                           NULL);
3567         } else {
3568                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
3569         }
3570
3571         err = btrfs_end_transaction(trans, root);
3572         if (err && !ret)
3573                 ret = err;
3574
3575 out:
3576         kfree(sa);
3577         return ret;
3578 }
3579
3580 static long btrfs_ioctl_qgroup_limit(struct btrfs_root *root, void __user *arg)
3581 {
3582         struct btrfs_ioctl_qgroup_limit_args *sa;
3583         struct btrfs_trans_handle *trans;
3584         int ret;
3585         int err;
3586         u64 qgroupid;
3587
3588         if (!capable(CAP_SYS_ADMIN))
3589                 return -EPERM;
3590
3591         if (root->fs_info->sb->s_flags & MS_RDONLY)
3592                 return -EROFS;
3593
3594         sa = memdup_user(arg, sizeof(*sa));
3595         if (IS_ERR(sa))
3596                 return PTR_ERR(sa);
3597
3598         trans = btrfs_join_transaction(root);
3599         if (IS_ERR(trans)) {
3600                 ret = PTR_ERR(trans);
3601                 goto out;
3602         }
3603
3604         qgroupid = sa->qgroupid;
3605         if (!qgroupid) {
3606                 /* take the current subvol as qgroup */
3607                 qgroupid = root->root_key.objectid;
3608         }
3609
3610         /* FIXME: check if the IDs really exist */
3611         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
3612
3613         err = btrfs_end_transaction(trans, root);
3614         if (err && !ret)
3615                 ret = err;
3616
3617 out:
3618         kfree(sa);
3619         return ret;
3620 }
3621
3622 static long btrfs_ioctl_set_received_subvol(struct file *file,
3623                                             void __user *arg)
3624 {
3625         struct btrfs_ioctl_received_subvol_args *sa = NULL;
3626         struct inode *inode = fdentry(file)->d_inode;
3627         struct btrfs_root *root = BTRFS_I(inode)->root;
3628         struct btrfs_root_item *root_item = &root->root_item;
3629         struct btrfs_trans_handle *trans;
3630         struct timespec ct = CURRENT_TIME;
3631         int ret = 0;
3632
3633         ret = mnt_want_write_file(file);
3634         if (ret < 0)
3635                 return ret;
3636
3637         down_write(&root->fs_info->subvol_sem);
3638
3639         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
3640                 ret = -EINVAL;
3641                 goto out;
3642         }
3643
3644         if (btrfs_root_readonly(root)) {
3645                 ret = -EROFS;
3646                 goto out;
3647         }
3648
3649         if (!inode_owner_or_capable(inode)) {
3650                 ret = -EACCES;
3651                 goto out;
3652         }
3653
3654         sa = memdup_user(arg, sizeof(*sa));
3655         if (IS_ERR(sa)) {
3656                 ret = PTR_ERR(sa);
3657                 sa = NULL;
3658                 goto out;
3659         }
3660
3661         trans = btrfs_start_transaction(root, 1);
3662         if (IS_ERR(trans)) {
3663                 ret = PTR_ERR(trans);
3664                 trans = NULL;
3665                 goto out;
3666         }
3667
3668         sa->rtransid = trans->transid;
3669         sa->rtime.sec = ct.tv_sec;
3670         sa->rtime.nsec = ct.tv_nsec;
3671
3672         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
3673         btrfs_set_root_stransid(root_item, sa->stransid);
3674         btrfs_set_root_rtransid(root_item, sa->rtransid);
3675         root_item->stime.sec = cpu_to_le64(sa->stime.sec);
3676         root_item->stime.nsec = cpu_to_le32(sa->stime.nsec);
3677         root_item->rtime.sec = cpu_to_le64(sa->rtime.sec);
3678         root_item->rtime.nsec = cpu_to_le32(sa->rtime.nsec);
3679
3680         ret = btrfs_update_root(trans, root->fs_info->tree_root,
3681                                 &root->root_key, &root->root_item);
3682         if (ret < 0) {
3683                 btrfs_end_transaction(trans, root);
3684                 trans = NULL;
3685                 goto out;
3686         } else {
3687                 ret = btrfs_commit_transaction(trans, root);
3688                 if (ret < 0)
3689                         goto out;
3690         }
3691
3692         ret = copy_to_user(arg, sa, sizeof(*sa));
3693         if (ret)
3694                 ret = -EFAULT;
3695
3696 out:
3697         kfree(sa);
3698         up_write(&root->fs_info->subvol_sem);
3699         mnt_drop_write_file(file);
3700         return ret;
3701 }
3702
3703 long btrfs_ioctl(struct file *file, unsigned int
3704                 cmd, unsigned long arg)
3705 {
3706         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3707         void __user *argp = (void __user *)arg;
3708
3709         switch (cmd) {
3710         case FS_IOC_GETFLAGS:
3711                 return btrfs_ioctl_getflags(file, argp);
3712         case FS_IOC_SETFLAGS:
3713                 return btrfs_ioctl_setflags(file, argp);
3714         case FS_IOC_GETVERSION:
3715                 return btrfs_ioctl_getversion(file, argp);
3716         case FITRIM:
3717                 return btrfs_ioctl_fitrim(file, argp);
3718         case BTRFS_IOC_SNAP_CREATE:
3719                 return btrfs_ioctl_snap_create(file, argp, 0);
3720         case BTRFS_IOC_SNAP_CREATE_V2:
3721                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3722         case BTRFS_IOC_SUBVOL_CREATE:
3723                 return btrfs_ioctl_snap_create(file, argp, 1);
3724         case BTRFS_IOC_SUBVOL_CREATE_V2:
3725                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
3726         case BTRFS_IOC_SNAP_DESTROY:
3727                 return btrfs_ioctl_snap_destroy(file, argp);
3728         case BTRFS_IOC_SUBVOL_GETFLAGS:
3729                 return btrfs_ioctl_subvol_getflags(file, argp);
3730         case BTRFS_IOC_SUBVOL_SETFLAGS:
3731                 return btrfs_ioctl_subvol_setflags(file, argp);
3732         case BTRFS_IOC_DEFAULT_SUBVOL:
3733                 return btrfs_ioctl_default_subvol(file, argp);
3734         case BTRFS_IOC_DEFRAG:
3735                 return btrfs_ioctl_defrag(file, NULL);
3736         case BTRFS_IOC_DEFRAG_RANGE:
3737                 return btrfs_ioctl_defrag(file, argp);
3738         case BTRFS_IOC_RESIZE:
3739                 return btrfs_ioctl_resize(root, argp);
3740         case BTRFS_IOC_ADD_DEV:
3741                 return btrfs_ioctl_add_dev(root, argp);
3742         case BTRFS_IOC_RM_DEV:
3743                 return btrfs_ioctl_rm_dev(root, argp);
3744         case BTRFS_IOC_FS_INFO:
3745                 return btrfs_ioctl_fs_info(root, argp);
3746         case BTRFS_IOC_DEV_INFO:
3747                 return btrfs_ioctl_dev_info(root, argp);
3748         case BTRFS_IOC_BALANCE:
3749                 return btrfs_ioctl_balance(file, NULL);
3750         case BTRFS_IOC_CLONE:
3751                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3752         case BTRFS_IOC_CLONE_RANGE:
3753                 return btrfs_ioctl_clone_range(file, argp);
3754         case BTRFS_IOC_TRANS_START:
3755                 return btrfs_ioctl_trans_start(file);
3756         case BTRFS_IOC_TRANS_END:
3757                 return btrfs_ioctl_trans_end(file);
3758         case BTRFS_IOC_TREE_SEARCH:
3759                 return btrfs_ioctl_tree_search(file, argp);
3760         case BTRFS_IOC_INO_LOOKUP:
3761                 return btrfs_ioctl_ino_lookup(file, argp);
3762         case BTRFS_IOC_INO_PATHS:
3763                 return btrfs_ioctl_ino_to_path(root, argp);
3764         case BTRFS_IOC_LOGICAL_INO:
3765                 return btrfs_ioctl_logical_to_ino(root, argp);
3766         case BTRFS_IOC_SPACE_INFO:
3767                 return btrfs_ioctl_space_info(root, argp);
3768         case BTRFS_IOC_SYNC:
3769                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3770                 return 0;
3771         case BTRFS_IOC_START_SYNC:
3772                 return btrfs_ioctl_start_sync(file, argp);
3773         case BTRFS_IOC_WAIT_SYNC:
3774                 return btrfs_ioctl_wait_sync(file, argp);
3775         case BTRFS_IOC_SCRUB:
3776                 return btrfs_ioctl_scrub(root, argp);
3777         case BTRFS_IOC_SCRUB_CANCEL:
3778                 return btrfs_ioctl_scrub_cancel(root, argp);
3779         case BTRFS_IOC_SCRUB_PROGRESS:
3780                 return btrfs_ioctl_scrub_progress(root, argp);
3781         case BTRFS_IOC_BALANCE_V2:
3782                 return btrfs_ioctl_balance(file, argp);
3783         case BTRFS_IOC_BALANCE_CTL:
3784                 return btrfs_ioctl_balance_ctl(root, arg);
3785         case BTRFS_IOC_BALANCE_PROGRESS:
3786                 return btrfs_ioctl_balance_progress(root, argp);
3787         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
3788                 return btrfs_ioctl_set_received_subvol(file, argp);
3789         case BTRFS_IOC_SEND:
3790                 return btrfs_ioctl_send(file, argp);
3791         case BTRFS_IOC_GET_DEV_STATS:
3792                 return btrfs_ioctl_get_dev_stats(root, argp);
3793         case BTRFS_IOC_QUOTA_CTL:
3794                 return btrfs_ioctl_quota_ctl(root, argp);
3795         case BTRFS_IOC_QGROUP_ASSIGN:
3796                 return btrfs_ioctl_qgroup_assign(root, argp);
3797         case BTRFS_IOC_QGROUP_CREATE:
3798                 return btrfs_ioctl_qgroup_create(root, argp);
3799         case BTRFS_IOC_QGROUP_LIMIT:
3800                 return btrfs_ioctl_qgroup_limit(root, argp);
3801         }
3802
3803         return -ENOTTY;
3804 }