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 "compat.h"
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "ioctl.h"
  50 #include "print-tree.h"
  51 #include "volumes.h"
  52 #include "locking.h"
  53 #include "inode-map.h"
  54 #include "backref.h"
  55
  56 /* Mask out flags that are inappropriate for the given type of inode. */
  57 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  58 {
  59         if (S_ISDIR(mode))
  60                 return flags;
  61         else if (S_ISREG(mode))
  62                 return flags & ~FS_DIRSYNC_FL;
  63         else
  64                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  65 }
  66
  67 /*
  68  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  69  */
  70 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  71 {
  72         unsigned int iflags = 0;
  73
  74         if (flags & BTRFS_INODE_SYNC)
  75                 iflags |= FS_SYNC_FL;
  76         if (flags & BTRFS_INODE_IMMUTABLE)
  77                 iflags |= FS_IMMUTABLE_FL;
  78         if (flags & BTRFS_INODE_APPEND)
  79                 iflags |= FS_APPEND_FL;
  80         if (flags & BTRFS_INODE_NODUMP)
  81                 iflags |= FS_NODUMP_FL;
  82         if (flags & BTRFS_INODE_NOATIME)
  83                 iflags |= FS_NOATIME_FL;
  84         if (flags & BTRFS_INODE_DIRSYNC)
  85                 iflags |= FS_DIRSYNC_FL;
  86         if (flags & BTRFS_INODE_NODATACOW)
  87                 iflags |= FS_NOCOW_FL;
  88
  89         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
  90                 iflags |= FS_COMPR_FL;
  91         else if (flags & BTRFS_INODE_NOCOMPRESS)
  92                 iflags |= FS_NOCOMP_FL;
  93
  94         return iflags;
  95 }
  96
  97 /*
  98  * Update inode->i_flags based on the btrfs internal flags.
  99  */
 100 void btrfs_update_iflags(struct inode *inode)
 101 {
 102         struct btrfs_inode *ip = BTRFS_I(inode);
 103
 104         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 105
 106         if (ip->flags & BTRFS_INODE_SYNC)
 107                 inode->i_flags |= S_SYNC;
 108         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 109                 inode->i_flags |= S_IMMUTABLE;
 110         if (ip->flags & BTRFS_INODE_APPEND)
 111                 inode->i_flags |= S_APPEND;
 112         if (ip->flags & BTRFS_INODE_NOATIME)
 113                 inode->i_flags |= S_NOATIME;
 114         if (ip->flags & BTRFS_INODE_DIRSYNC)
 115                 inode->i_flags |= S_DIRSYNC;
 116 }
 117
 118 /*
 119  * Inherit flags from the parent inode.
 120  *
 121  * Currently only the compression flags and the cow flags are inherited.
 122  */
 123 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 124 {
 125         unsigned int flags;
 126
 127         if (!dir)
 128                 return;
 129
 130         flags = BTRFS_I(dir)->flags;
 131
 132         if (flags & BTRFS_INODE_NOCOMPRESS) {
 133                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 134                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 135         } else if (flags & BTRFS_INODE_COMPRESS) {
 136                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 137                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 138         }
 139
 140         if (flags & BTRFS_INODE_NODATACOW)
 141                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 142
 143         btrfs_update_iflags(inode);
 144 }
 145
 146 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 147 {
 148         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 149         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 150
 151         if (copy_to_user(arg, &flags, sizeof(flags)))
 152                 return -EFAULT;
 153         return 0;
 154 }
 155
 156 static int check_flags(unsigned int flags)
 157 {
 158         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 159                       FS_NOATIME_FL | FS_NODUMP_FL | \
 160                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 161                       FS_NOCOMP_FL | FS_COMPR_FL |
 162                       FS_NOCOW_FL))
 163                 return -EOPNOTSUPP;
 164
 165         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 166                 return -EINVAL;
 167
 168         return 0;
 169 }
 170
 171 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 172 {
 173         struct inode *inode = file->f_path.dentry->d_inode;
 174         struct btrfs_inode *ip = BTRFS_I(inode);
 175         struct btrfs_root *root = ip->root;
 176         struct btrfs_trans_handle *trans;
 177         unsigned int flags, oldflags;
 178         int ret;
 179
 180         if (btrfs_root_readonly(root))
 181                 return -EROFS;
 182
 183         if (copy_from_user(&flags, arg, sizeof(flags)))
 184                 return -EFAULT;
 185
 186         ret = check_flags(flags);
 187         if (ret)
 188                 return ret;
 189
 190         if (!inode_owner_or_capable(inode))
 191                 return -EACCES;
 192
 193         mutex_lock(&inode->i_mutex);
 194
 195         flags = btrfs_mask_flags(inode->i_mode, flags);
 196         oldflags = btrfs_flags_to_ioctl(ip->flags);
 197         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 198                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 199                         ret = -EPERM;
 200                         goto out_unlock;
 201                 }
 202         }
 203
 204         ret = mnt_want_write(file->f_path.mnt);
 205         if (ret)
 206                 goto out_unlock;
 207
 208         if (flags & FS_SYNC_FL)
 209                 ip->flags |= BTRFS_INODE_SYNC;
 210         else
 211                 ip->flags &= ~BTRFS_INODE_SYNC;
 212         if (flags & FS_IMMUTABLE_FL)
 213                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 214         else
 215                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 216         if (flags & FS_APPEND_FL)
 217                 ip->flags |= BTRFS_INODE_APPEND;
 218         else
 219                 ip->flags &= ~BTRFS_INODE_APPEND;
 220         if (flags & FS_NODUMP_FL)
 221                 ip->flags |= BTRFS_INODE_NODUMP;
 222         else
 223                 ip->flags &= ~BTRFS_INODE_NODUMP;
 224         if (flags & FS_NOATIME_FL)
 225                 ip->flags |= BTRFS_INODE_NOATIME;
 226         else
 227                 ip->flags &= ~BTRFS_INODE_NOATIME;
 228         if (flags & FS_DIRSYNC_FL)
 229                 ip->flags |= BTRFS_INODE_DIRSYNC;
 230         else
 231                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 232         if (flags & FS_NOCOW_FL)
 233                 ip->flags |= BTRFS_INODE_NODATACOW;
 234         else
 235                 ip->flags &= ~BTRFS_INODE_NODATACOW;
 236
 237         /*
 238          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 239          * flag may be changed automatically if compression code won't make
 240          * things smaller.
 241          */
 242         if (flags & FS_NOCOMP_FL) {
 243                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 244                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 245         } else if (flags & FS_COMPR_FL) {
 246                 ip->flags |= BTRFS_INODE_COMPRESS;
 247                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 248         } else {
 249                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 250         }
 251
 252         trans = btrfs_join_transaction(root);
 253         BUG_ON(IS_ERR(trans));
 254
 255         btrfs_update_iflags(inode);
 256         inode->i_ctime = CURRENT_TIME;
 257         ret = btrfs_update_inode(trans, root, inode);
 258         BUG_ON(ret);
 259
 260         btrfs_end_transaction(trans, root);
 261
 262         mnt_drop_write(file->f_path.mnt);
 263
 264         ret = 0;
 265  out_unlock:
 266         mutex_unlock(&inode->i_mutex);
 267         return ret;
 268 }
 269
 270 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 271 {
 272         struct inode *inode = file->f_path.dentry->d_inode;
 273
 274         return put_user(inode->i_generation, arg);
 275 }
 276
 277 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 278 {
 279         struct btrfs_root *root = fdentry(file)->d_sb->s_fs_info;
 280         struct btrfs_fs_info *fs_info = root->fs_info;
 281         struct btrfs_device *device;
 282         struct request_queue *q;
 283         struct fstrim_range range;
 284         u64 minlen = ULLONG_MAX;
 285         u64 num_devices = 0;
 286         u64 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
 287         int ret;
 288
 289         if (!capable(CAP_SYS_ADMIN))
 290                 return -EPERM;
 291
 292         rcu_read_lock();
 293         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 294                                 dev_list) {
 295                 if (!device->bdev)
 296                         continue;
 297                 q = bdev_get_queue(device->bdev);
 298                 if (blk_queue_discard(q)) {
 299                         num_devices++;
 300                         minlen = min((u64)q->limits.discard_granularity,
 301                                      minlen);
 302                 }
 303         }
 304         rcu_read_unlock();
 305
 306         if (!num_devices)
 307                 return -EOPNOTSUPP;
 308         if (copy_from_user(&range, arg, sizeof(range)))
 309                 return -EFAULT;
 310         if (range.start > total_bytes)
 311                 return -EINVAL;
 312
 313         range.len = min(range.len, total_bytes - range.start);
 314         range.minlen = max(range.minlen, minlen);
 315         ret = btrfs_trim_fs(root, &range);
 316         if (ret < 0)
 317                 return ret;
 318
 319         if (copy_to_user(arg, &range, sizeof(range)))
 320                 return -EFAULT;
 321
 322         return 0;
 323 }
 324
 325 static noinline int create_subvol(struct btrfs_root *root,
 326                                   struct dentry *dentry,
 327                                   char *name, int namelen,
 328                                   u64 *async_transid)
 329 {
 330         struct btrfs_trans_handle *trans;
 331         struct btrfs_key key;
 332         struct btrfs_root_item root_item;
 333         struct btrfs_inode_item *inode_item;
 334         struct extent_buffer *leaf;
 335         struct btrfs_root *new_root;
 336         struct dentry *parent = dentry->d_parent;
 337         struct inode *dir;
 338         int ret;
 339         int err;
 340         u64 objectid;
 341         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 342         u64 index = 0;
 343
 344         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 345         if (ret)
 346                 return ret;
 347
 348         dir = parent->d_inode;
 349
 350         /*
 351          * 1 - inode item
 352          * 2 - refs
 353          * 1 - root item
 354          * 2 - dir items
 355          */
 356         trans = btrfs_start_transaction(root, 6);
 357         if (IS_ERR(trans))
 358                 return PTR_ERR(trans);
 359
 360         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 361                                       0, objectid, NULL, 0, 0, 0);
 362         if (IS_ERR(leaf)) {
 363                 ret = PTR_ERR(leaf);
 364                 goto fail;
 365         }
 366
 367         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 368         btrfs_set_header_bytenr(leaf, leaf->start);
 369         btrfs_set_header_generation(leaf, trans->transid);
 370         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 371         btrfs_set_header_owner(leaf, objectid);
 372
 373         write_extent_buffer(leaf, root->fs_info->fsid,
 374                             (unsigned long)btrfs_header_fsid(leaf),
 375                             BTRFS_FSID_SIZE);
 376         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 377                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 378                             BTRFS_UUID_SIZE);
 379         btrfs_mark_buffer_dirty(leaf);
 380
 381         inode_item = &root_item.inode;
 382         memset(inode_item, 0, sizeof(*inode_item));
 383         inode_item->generation = cpu_to_le64(1);
 384         inode_item->size = cpu_to_le64(3);
 385         inode_item->nlink = cpu_to_le32(1);
 386         inode_item->nbytes = cpu_to_le64(root->leafsize);
 387         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 388
 389         root_item.flags = 0;
 390         root_item.byte_limit = 0;
 391         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
 392
 393         btrfs_set_root_bytenr(&root_item, leaf->start);
 394         btrfs_set_root_generation(&root_item, trans->transid);
 395         btrfs_set_root_level(&root_item, 0);
 396         btrfs_set_root_refs(&root_item, 1);
 397         btrfs_set_root_used(&root_item, leaf->len);
 398         btrfs_set_root_last_snapshot(&root_item, 0);
 399
 400         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 401         root_item.drop_level = 0;
 402
 403         btrfs_tree_unlock(leaf);
 404         free_extent_buffer(leaf);
 405         leaf = NULL;
 406
 407         btrfs_set_root_dirid(&root_item, new_dirid);
 408
 409         key.objectid = objectid;
 410         key.offset = 0;
 411         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 412         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 413                                 &root_item);
 414         if (ret)
 415                 goto fail;
 416
 417         key.offset = (u64)-1;
 418         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 419         BUG_ON(IS_ERR(new_root));
 420
 421         btrfs_record_root_in_trans(trans, new_root);
 422
 423         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
 424         /*
 425          * insert the directory item
 426          */
 427         ret = btrfs_set_inode_index(dir, &index);
 428         BUG_ON(ret);
 429
 430         ret = btrfs_insert_dir_item(trans, root,
 431                                     name, namelen, dir, &key,
 432                                     BTRFS_FT_DIR, index);
 433         if (ret)
 434                 goto fail;
 435
 436         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 437         ret = btrfs_update_inode(trans, root, dir);
 438         BUG_ON(ret);
 439
 440         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 441                                  objectid, root->root_key.objectid,
 442                                  btrfs_ino(dir), index, name, namelen);
 443
 444         BUG_ON(ret);
 445
 446         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 447 fail:
 448         if (async_transid) {
 449                 *async_transid = trans->transid;
 450                 err = btrfs_commit_transaction_async(trans, root, 1);
 451         } else {
 452                 err = btrfs_commit_transaction(trans, root);
 453         }
 454         if (err && !ret)
 455                 ret = err;
 456         return ret;
 457 }
 458
 459 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 460                            char *name, int namelen, u64 *async_transid,
 461                            bool readonly)
 462 {
 463         struct inode *inode;
 464         struct btrfs_pending_snapshot *pending_snapshot;
 465         struct btrfs_trans_handle *trans;
 466         int ret;
 467
 468         if (!root->ref_cows)
 469                 return -EINVAL;
 470
 471         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 472         if (!pending_snapshot)
 473                 return -ENOMEM;
 474
 475         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
 476         pending_snapshot->dentry = dentry;
 477         pending_snapshot->root = root;
 478         pending_snapshot->readonly = readonly;
 479
 480         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
 481         if (IS_ERR(trans)) {
 482                 ret = PTR_ERR(trans);
 483                 goto fail;
 484         }
 485
 486         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 487         BUG_ON(ret);
 488
 489         spin_lock(&root->fs_info->trans_lock);
 490         list_add(&pending_snapshot->list,
 491                  &trans->transaction->pending_snapshots);
 492         spin_unlock(&root->fs_info->trans_lock);
 493         if (async_transid) {
 494                 *async_transid = trans->transid;
 495                 ret = btrfs_commit_transaction_async(trans,
 496                                      root->fs_info->extent_root, 1);
 497         } else {
 498                 ret = btrfs_commit_transaction(trans,
 499                                                root->fs_info->extent_root);
 500         }
 501         BUG_ON(ret);
 502
 503         ret = pending_snapshot->error;
 504         if (ret)
 505                 goto fail;
 506
 507         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 508         if (ret)
 509                 goto fail;
 510
 511         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 512         if (IS_ERR(inode)) {
 513                 ret = PTR_ERR(inode);
 514                 goto fail;
 515         }
 516         BUG_ON(!inode);
 517         d_instantiate(dentry, inode);
 518         ret = 0;
 519 fail:
 520         kfree(pending_snapshot);
 521         return ret;
 522 }
 523
 524 /*  copy of check_sticky in fs/namei.c()
 525 * It's inline, so penalty for filesystems that don't use sticky bit is
 526 * minimal.
 527 */
 528 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 529 {
 530         uid_t fsuid = current_fsuid();
 531
 532         if (!(dir->i_mode & S_ISVTX))
 533                 return 0;
 534         if (inode->i_uid == fsuid)
 535                 return 0;
 536         if (dir->i_uid == fsuid)
 537                 return 0;
 538         return !capable(CAP_FOWNER);
 539 }
 540
 541 /*  copy of may_delete in fs/namei.c()
 542  *      Check whether we can remove a link victim from directory dir, check
 543  *  whether the type of victim is right.
 544  *  1. We can't do it if dir is read-only (done in permission())
 545  *  2. We should have write and exec permissions on dir
 546  *  3. We can't remove anything from append-only dir
 547  *  4. We can't do anything with immutable dir (done in permission())
 548  *  5. If the sticky bit on dir is set we should either
 549  *      a. be owner of dir, or
 550  *      b. be owner of victim, or
 551  *      c. have CAP_FOWNER capability
 552  *  6. If the victim is append-only or immutable we can't do antyhing with
 553  *     links pointing to it.
 554  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 555  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 556  *  9. We can't remove a root or mountpoint.
 557  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 558  *     nfs_async_unlink().
 559  */
 560
 561 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 562 {
 563         int error;
 564
 565         if (!victim->d_inode)
 566                 return -ENOENT;
 567
 568         BUG_ON(victim->d_parent->d_inode != dir);
 569         audit_inode_child(victim, dir);
 570
 571         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 572         if (error)
 573                 return error;
 574         if (IS_APPEND(dir))
 575                 return -EPERM;
 576         if (btrfs_check_sticky(dir, victim->d_inode)||
 577                 IS_APPEND(victim->d_inode)||
 578             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 579                 return -EPERM;
 580         if (isdir) {
 581                 if (!S_ISDIR(victim->d_inode->i_mode))
 582                         return -ENOTDIR;
 583                 if (IS_ROOT(victim))
 584                         return -EBUSY;
 585         } else if (S_ISDIR(victim->d_inode->i_mode))
 586                 return -EISDIR;
 587         if (IS_DEADDIR(dir))
 588                 return -ENOENT;
 589         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 590                 return -EBUSY;
 591         return 0;
 592 }
 593
 594 /* copy of may_create in fs/namei.c() */
 595 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 596 {
 597         if (child->d_inode)
 598                 return -EEXIST;
 599         if (IS_DEADDIR(dir))
 600                 return -ENOENT;
 601         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 602 }
 603
 604 /*
 605  * Create a new subvolume below @parent.  This is largely modeled after
 606  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 607  * inside this filesystem so it's quite a bit simpler.
 608  */
 609 static noinline int btrfs_mksubvol(struct path *parent,
 610                                    char *name, int namelen,
 611                                    struct btrfs_root *snap_src,
 612                                    u64 *async_transid, bool readonly)
 613 {
 614         struct inode *dir  = parent->dentry->d_inode;
 615         struct dentry *dentry;
 616         int error;
 617
 618         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 619
 620         dentry = lookup_one_len(name, parent->dentry, namelen);
 621         error = PTR_ERR(dentry);
 622         if (IS_ERR(dentry))
 623                 goto out_unlock;
 624
 625         error = -EEXIST;
 626         if (dentry->d_inode)
 627                 goto out_dput;
 628
 629         error = mnt_want_write(parent->mnt);
 630         if (error)
 631                 goto out_dput;
 632
 633         error = btrfs_may_create(dir, dentry);
 634         if (error)
 635                 goto out_drop_write;
 636
 637         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 638
 639         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 640                 goto out_up_read;
 641
 642         if (snap_src) {
 643                 error = create_snapshot(snap_src, dentry,
 644                                         name, namelen, async_transid, readonly);
 645         } else {
 646                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 647                                       name, namelen, async_transid);
 648         }
 649         if (!error)
 650                 fsnotify_mkdir(dir, dentry);
 651 out_up_read:
 652         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 653 out_drop_write:
 654         mnt_drop_write(parent->mnt);
 655 out_dput:
 656         dput(dentry);
 657 out_unlock:
 658         mutex_unlock(&dir->i_mutex);
 659         return error;
 660 }
 661
 662 /*
 663  * When we're defragging a range, we don't want to kick it off again
 664  * if it is really just waiting for delalloc to send it down.
 665  * If we find a nice big extent or delalloc range for the bytes in the
 666  * file you want to defrag, we return 0 to let you know to skip this
 667  * part of the file
 668  */
 669 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 670 {
 671         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 672         struct extent_map *em = NULL;
 673         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 674         u64 end;
 675
 676         read_lock(&em_tree->lock);
 677         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 678         read_unlock(&em_tree->lock);
 679
 680         if (em) {
 681                 end = extent_map_end(em);
 682                 free_extent_map(em);
 683                 if (end - offset > thresh)
 684                         return 0;
 685         }
 686         /* if we already have a nice delalloc here, just stop */
 687         thresh /= 2;
 688         end = count_range_bits(io_tree, &offset, offset + thresh,
 689                                thresh, EXTENT_DELALLOC, 1);
 690         if (end >= thresh)
 691                 return 0;
 692         return 1;
 693 }
 694
 695 /*
 696  * helper function to walk through a file and find extents
 697  * newer than a specific transid, and smaller than thresh.
 698  *
 699  * This is used by the defragging code to find new and small
 700  * extents
 701  */
 702 static int find_new_extents(struct btrfs_root *root,
 703                             struct inode *inode, u64 newer_than,
 704                             u64 *off, int thresh)
 705 {
 706         struct btrfs_path *path;
 707         struct btrfs_key min_key;
 708         struct btrfs_key max_key;
 709         struct extent_buffer *leaf;
 710         struct btrfs_file_extent_item *extent;
 711         int type;
 712         int ret;
 713         u64 ino = btrfs_ino(inode);
 714
 715         path = btrfs_alloc_path();
 716         if (!path)
 717                 return -ENOMEM;
 718
 719         min_key.objectid = ino;
 720         min_key.type = BTRFS_EXTENT_DATA_KEY;
 721         min_key.offset = *off;
 722
 723         max_key.objectid = ino;
 724         max_key.type = (u8)-1;
 725         max_key.offset = (u64)-1;
 726
 727         path->keep_locks = 1;
 728
 729         while(1) {
 730                 ret = btrfs_search_forward(root, &min_key, &max_key,
 731                                            path, 0, newer_than);
 732                 if (ret != 0)
 733                         goto none;
 734                 if (min_key.objectid != ino)
 735                         goto none;
 736                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 737                         goto none;
 738
 739                 leaf = path->nodes[0];
 740                 extent = btrfs_item_ptr(leaf, path->slots[0],
 741                                         struct btrfs_file_extent_item);
 742
 743                 type = btrfs_file_extent_type(leaf, extent);
 744                 if (type == BTRFS_FILE_EXTENT_REG &&
 745                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 746                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 747                         *off = min_key.offset;
 748                         btrfs_free_path(path);
 749                         return 0;
 750                 }
 751
 752                 if (min_key.offset == (u64)-1)
 753                         goto none;
 754
 755                 min_key.offset++;
 756                 btrfs_release_path(path);
 757         }
 758 none:
 759         btrfs_free_path(path);
 760         return -ENOENT;
 761 }
 762
 763 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
 764                                int thresh, u64 *last_len, u64 *skip,
 765                                u64 *defrag_end)
 766 {
 767         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 768         struct extent_map *em = NULL;
 769         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 770         int ret = 1;
 771
 772         /*
 773          * make sure that once we start defragging an extent, we keep on
 774          * defragging it
 775          */
 776         if (start < *defrag_end)
 777                 return 1;
 778
 779         *skip = 0;
 780
 781         /*
 782          * hopefully we have this extent in the tree already, try without
 783          * the full extent lock
 784          */
 785         read_lock(&em_tree->lock);
 786         em = lookup_extent_mapping(em_tree, start, len);
 787         read_unlock(&em_tree->lock);
 788
 789         if (!em) {
 790                 /* get the big lock and read metadata off disk */
 791                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 792                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 793                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 794
 795                 if (IS_ERR(em))
 796                         return 0;
 797         }
 798
 799         /* this will cover holes, and inline extents */
 800         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
 801                 ret = 0;
 802
 803         /*
 804          * we hit a real extent, if it is big don't bother defragging it again
 805          */
 806         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
 807                 ret = 0;
 808
 809         /*
 810          * last_len ends up being a counter of how many bytes we've defragged.
 811          * every time we choose not to defrag an extent, we reset *last_len
 812          * so that the next tiny extent will force a defrag.
 813          *
 814          * The end result of this is that tiny extents before a single big
 815          * extent will force at least part of that big extent to be defragged.
 816          */
 817         if (ret) {
 818                 *defrag_end = extent_map_end(em);
 819         } else {
 820                 *last_len = 0;
 821                 *skip = extent_map_end(em);
 822                 *defrag_end = 0;
 823         }
 824
 825         free_extent_map(em);
 826         return ret;
 827 }
 828
 829 /*
 830  * it doesn't do much good to defrag one or two pages
 831  * at a time.  This pulls in a nice chunk of pages
 832  * to COW and defrag.
 833  *
 834  * It also makes sure the delalloc code has enough
 835  * dirty data to avoid making new small extents as part
 836  * of the defrag
 837  *
 838  * It's a good idea to start RA on this range
 839  * before calling this.
 840  */
 841 static int cluster_pages_for_defrag(struct inode *inode,
 842                                     struct page **pages,
 843                                     unsigned long start_index,
 844                                     int num_pages)
 845 {
 846         unsigned long file_end;
 847         u64 isize = i_size_read(inode);
 848         u64 page_start;
 849         u64 page_end;
 850         int ret;
 851         int i;
 852         int i_done;
 853         struct btrfs_ordered_extent *ordered;
 854         struct extent_state *cached_state = NULL;
 855         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 856
 857         if (isize == 0)
 858                 return 0;
 859         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 860
 861         mutex_lock(&inode->i_mutex);
 862         ret = btrfs_delalloc_reserve_space(inode,
 863                                            num_pages << PAGE_CACHE_SHIFT);
 864         mutex_unlock(&inode->i_mutex);
 865         if (ret)
 866                 return ret;
 867 again:
 868         ret = 0;
 869         i_done = 0;
 870
 871         /* step one, lock all the pages */
 872         for (i = 0; i < num_pages; i++) {
 873                 struct page *page;
 874                 page = find_or_create_page(inode->i_mapping,
 875                                             start_index + i, mask);
 876                 if (!page)
 877                         break;
 878
 879                 if (!PageUptodate(page)) {
 880                         btrfs_readpage(NULL, page);
 881                         lock_page(page);
 882                         if (!PageUptodate(page)) {
 883                                 unlock_page(page);
 884                                 page_cache_release(page);
 885                                 ret = -EIO;
 886                                 break;
 887                         }
 888                 }
 889                 isize = i_size_read(inode);
 890                 file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 891                 if (!isize || page->index > file_end ||
 892                     page->mapping != inode->i_mapping) {
 893                         /* whoops, we blew past eof, skip this page */
 894                         unlock_page(page);
 895                         page_cache_release(page);
 896                         break;
 897                 }
 898                 pages[i] = page;
 899                 i_done++;
 900         }
 901         if (!i_done || ret)
 902                 goto out;
 903
 904         if (!(inode->i_sb->s_flags & MS_ACTIVE))
 905                 goto out;
 906
 907         /*
 908          * so now we have a nice long stream of locked
 909          * and up to date pages, lets wait on them
 910          */
 911         for (i = 0; i < i_done; i++)
 912                 wait_on_page_writeback(pages[i]);
 913
 914         page_start = page_offset(pages[0]);
 915         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
 916
 917         lock_extent_bits(&BTRFS_I(inode)->io_tree,
 918                          page_start, page_end - 1, 0, &cached_state,
 919                          GFP_NOFS);
 920         ordered = btrfs_lookup_first_ordered_extent(inode, page_end - 1);
 921         if (ordered &&
 922             ordered->file_offset + ordered->len > page_start &&
 923             ordered->file_offset < page_end) {
 924                 btrfs_put_ordered_extent(ordered);
 925                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 926                                      page_start, page_end - 1,
 927                                      &cached_state, GFP_NOFS);
 928                 for (i = 0; i < i_done; i++) {
 929                         unlock_page(pages[i]);
 930                         page_cache_release(pages[i]);
 931                 }
 932                 btrfs_wait_ordered_range(inode, page_start,
 933                                          page_end - page_start);
 934                 goto again;
 935         }
 936         if (ordered)
 937                 btrfs_put_ordered_extent(ordered);
 938
 939         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
 940                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
 941                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
 942                           GFP_NOFS);
 943
 944         if (i_done != num_pages) {
 945                 spin_lock(&BTRFS_I(inode)->lock);
 946                 BTRFS_I(inode)->outstanding_extents++;
 947                 spin_unlock(&BTRFS_I(inode)->lock);
 948                 btrfs_delalloc_release_space(inode,
 949                                      (num_pages - i_done) << PAGE_CACHE_SHIFT);
 950         }
 951
 952
 953         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
 954                                   &cached_state);
 955
 956         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 957                              page_start, page_end - 1, &cached_state,
 958                              GFP_NOFS);
 959
 960         for (i = 0; i < i_done; i++) {
 961                 clear_page_dirty_for_io(pages[i]);
 962                 ClearPageChecked(pages[i]);
 963                 set_page_extent_mapped(pages[i]);
 964                 set_page_dirty(pages[i]);
 965                 unlock_page(pages[i]);
 966                 page_cache_release(pages[i]);
 967         }
 968         return i_done;
 969 out:
 970         for (i = 0; i < i_done; i++) {
 971                 unlock_page(pages[i]);
 972                 page_cache_release(pages[i]);
 973         }
 974         btrfs_delalloc_release_space(inode, num_pages << PAGE_CACHE_SHIFT);
 975         return ret;
 976
 977 }
 978
 979 int btrfs_defrag_file(struct inode *inode, struct file *file,
 980                       struct btrfs_ioctl_defrag_range_args *range,
 981                       u64 newer_than, unsigned long max_to_defrag)
 982 {
 983         struct btrfs_root *root = BTRFS_I(inode)->root;
 984         struct btrfs_super_block *disk_super;
 985         struct file_ra_state *ra = NULL;
 986         unsigned long last_index;
 987         u64 isize = i_size_read(inode);
 988         u64 features;
 989         u64 last_len = 0;
 990         u64 skip = 0;
 991         u64 defrag_end = 0;
 992         u64 newer_off = range->start;
 993         unsigned long i;
 994         unsigned long ra_index = 0;
 995         int ret;
 996         int defrag_count = 0;
 997         int compress_type = BTRFS_COMPRESS_ZLIB;
 998         int extent_thresh = range->extent_thresh;
 999         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1000         int cluster = max_cluster;
1001         u64 new_align = ~((u64)128 * 1024 - 1);
1002         struct page **pages = NULL;
1003
1004         if (extent_thresh == 0)
1005                 extent_thresh = 256 * 1024;
1006
1007         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1008                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1009                         return -EINVAL;
1010                 if (range->compress_type)
1011                         compress_type = range->compress_type;
1012         }
1013
1014         if (isize == 0)
1015                 return 0;
1016
1017         /*
1018          * if we were not given a file, allocate a readahead
1019          * context
1020          */
1021         if (!file) {
1022                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1023                 if (!ra)
1024                         return -ENOMEM;
1025                 file_ra_state_init(ra, inode->i_mapping);
1026         } else {
1027                 ra = &file->f_ra;
1028         }
1029
1030         pages = kmalloc(sizeof(struct page *) * max_cluster,
1031                         GFP_NOFS);
1032         if (!pages) {
1033                 ret = -ENOMEM;
1034                 goto out_ra;
1035         }
1036
1037         /* find the last page to defrag */
1038         if (range->start + range->len > range->start) {
1039                 last_index = min_t(u64, isize - 1,
1040                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1041         } else {
1042                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1043         }
1044
1045         if (newer_than) {
1046                 ret = find_new_extents(root, inode, newer_than,
1047                                        &newer_off, 64 * 1024);
1048                 if (!ret) {
1049                         range->start = newer_off;
1050                         /*
1051                          * we always align our defrag to help keep
1052                          * the extents in the file evenly spaced
1053                          */
1054                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1055                 } else
1056                         goto out_ra;
1057         } else {
1058                 i = range->start >> PAGE_CACHE_SHIFT;
1059         }
1060         if (!max_to_defrag)
1061                 max_to_defrag = last_index;
1062
1063         /*
1064          * make writeback starts from i, so the defrag range can be
1065          * written sequentially.
1066          */
1067         if (i < inode->i_mapping->writeback_index)
1068                 inode->i_mapping->writeback_index = i;
1069
1070         while (i <= last_index && defrag_count < max_to_defrag &&
1071                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1072                 PAGE_CACHE_SHIFT)) {
1073                 /*
1074                  * make sure we stop running if someone unmounts
1075                  * the FS
1076                  */
1077                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1078                         break;
1079
1080                 if (!newer_than &&
1081                     !should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1082                                         PAGE_CACHE_SIZE,
1083                                         extent_thresh,
1084                                         &last_len, &skip,
1085                                         &defrag_end)) {
1086                         unsigned long next;
1087                         /*
1088                          * the should_defrag function tells us how much to skip
1089                          * bump our counter by the suggested amount
1090                          */
1091                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1092                         i = max(i + 1, next);
1093                         continue;
1094                 }
1095
1096                 if (!newer_than) {
1097                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1098                                    PAGE_CACHE_SHIFT) - i;
1099                         cluster = min(cluster, max_cluster);
1100                 } else {
1101                         cluster = max_cluster;
1102                 }
1103
1104                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1105                         BTRFS_I(inode)->force_compress = compress_type;
1106
1107                 if (i + cluster > ra_index) {
1108                         ra_index = max(i, ra_index);
1109                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1110                                        cluster);
1111                         ra_index += max_cluster;
1112                 }
1113
1114                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1115                 if (ret < 0)
1116                         goto out_ra;
1117
1118                 defrag_count += ret;
1119                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1120
1121                 if (newer_than) {
1122                         if (newer_off == (u64)-1)
1123                                 break;
1124
1125                         newer_off = max(newer_off + 1,
1126                                         (u64)i << PAGE_CACHE_SHIFT);
1127
1128                         ret = find_new_extents(root, inode,
1129                                                newer_than, &newer_off,
1130                                                64 * 1024);
1131                         if (!ret) {
1132                                 range->start = newer_off;
1133                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1134                         } else {
1135                                 break;
1136                         }
1137                 } else {
1138                         if (ret > 0) {
1139                                 i += ret;
1140                                 last_len += ret << PAGE_CACHE_SHIFT;
1141                         } else {
1142                                 i++;
1143                                 last_len = 0;
1144                         }
1145                 }
1146         }
1147
1148         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1149                 filemap_flush(inode->i_mapping);
1150
1151         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1152                 /* the filemap_flush will queue IO into the worker threads, but
1153                  * we have to make sure the IO is actually started and that
1154                  * ordered extents get created before we return
1155                  */
1156                 atomic_inc(&root->fs_info->async_submit_draining);
1157                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1158                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1159                         wait_event(root->fs_info->async_submit_wait,
1160                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1161                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1162                 }
1163                 atomic_dec(&root->fs_info->async_submit_draining);
1164
1165                 mutex_lock(&inode->i_mutex);
1166                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1167                 mutex_unlock(&inode->i_mutex);
1168         }
1169
1170         disk_super = root->fs_info->super_copy;
1171         features = btrfs_super_incompat_flags(disk_super);
1172         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1173                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1174                 btrfs_set_super_incompat_flags(disk_super, features);
1175         }
1176
1177         ret = defrag_count;
1178
1179 out_ra:
1180         if (!file)
1181                 kfree(ra);
1182         kfree(pages);
1183         return ret;
1184 }
1185
1186 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1187                                         void __user *arg)
1188 {
1189         u64 new_size;
1190         u64 old_size;
1191         u64 devid = 1;
1192         struct btrfs_ioctl_vol_args *vol_args;
1193         struct btrfs_trans_handle *trans;
1194         struct btrfs_device *device = NULL;
1195         char *sizestr;
1196         char *devstr = NULL;
1197         int ret = 0;
1198         int mod = 0;
1199
1200         if (root->fs_info->sb->s_flags & MS_RDONLY)
1201                 return -EROFS;
1202
1203         if (!capable(CAP_SYS_ADMIN))
1204                 return -EPERM;
1205
1206         vol_args = memdup_user(arg, sizeof(*vol_args));
1207         if (IS_ERR(vol_args))
1208                 return PTR_ERR(vol_args);
1209
1210         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1211
1212         mutex_lock(&root->fs_info->volume_mutex);
1213         sizestr = vol_args->name;
1214         devstr = strchr(sizestr, ':');
1215         if (devstr) {
1216                 char *end;
1217                 sizestr = devstr + 1;
1218                 *devstr = '\0';
1219                 devstr = vol_args->name;
1220                 devid = simple_strtoull(devstr, &end, 10);
1221                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1222                        (unsigned long long)devid);
1223         }
1224         device = btrfs_find_device(root, devid, NULL, NULL);
1225         if (!device) {
1226                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1227                        (unsigned long long)devid);
1228                 ret = -EINVAL;
1229                 goto out_unlock;
1230         }
1231         if (!strcmp(sizestr, "max"))
1232                 new_size = device->bdev->bd_inode->i_size;
1233         else {
1234                 if (sizestr[0] == '-') {
1235                         mod = -1;
1236                         sizestr++;
1237                 } else if (sizestr[0] == '+') {
1238                         mod = 1;
1239                         sizestr++;
1240                 }
1241                 new_size = memparse(sizestr, NULL);
1242                 if (new_size == 0) {
1243                         ret = -EINVAL;
1244                         goto out_unlock;
1245                 }
1246         }
1247
1248         old_size = device->total_bytes;
1249
1250         if (mod < 0) {
1251                 if (new_size > old_size) {
1252                         ret = -EINVAL;
1253                         goto out_unlock;
1254                 }
1255                 new_size = old_size - new_size;
1256         } else if (mod > 0) {
1257                 new_size = old_size + new_size;
1258         }
1259
1260         if (new_size < 256 * 1024 * 1024) {
1261                 ret = -EINVAL;
1262                 goto out_unlock;
1263         }
1264         if (new_size > device->bdev->bd_inode->i_size) {
1265                 ret = -EFBIG;
1266                 goto out_unlock;
1267         }
1268
1269         do_div(new_size, root->sectorsize);
1270         new_size *= root->sectorsize;
1271
1272         printk(KERN_INFO "btrfs: new size for %s is %llu\n",
1273                 device->name, (unsigned long long)new_size);
1274
1275         if (new_size > old_size) {
1276                 trans = btrfs_start_transaction(root, 0);
1277                 if (IS_ERR(trans)) {
1278                         ret = PTR_ERR(trans);
1279                         goto out_unlock;
1280                 }
1281                 ret = btrfs_grow_device(trans, device, new_size);
1282                 btrfs_commit_transaction(trans, root);
1283         } else if (new_size < old_size) {
1284                 ret = btrfs_shrink_device(device, new_size);
1285         }
1286
1287 out_unlock:
1288         mutex_unlock(&root->fs_info->volume_mutex);
1289         kfree(vol_args);
1290         return ret;
1291 }
1292
1293 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1294                                                     char *name,
1295                                                     unsigned long fd,
1296                                                     int subvol,
1297                                                     u64 *transid,
1298                                                     bool readonly)
1299 {
1300         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1301         struct file *src_file;
1302         int namelen;
1303         int ret = 0;
1304
1305         if (root->fs_info->sb->s_flags & MS_RDONLY)
1306                 return -EROFS;
1307
1308         namelen = strlen(name);
1309         if (strchr(name, '/')) {
1310                 ret = -EINVAL;
1311                 goto out;
1312         }
1313
1314         if (subvol) {
1315                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1316                                      NULL, transid, readonly);
1317         } else {
1318                 struct inode *src_inode;
1319                 src_file = fget(fd);
1320                 if (!src_file) {
1321                         ret = -EINVAL;
1322                         goto out;
1323                 }
1324
1325                 src_inode = src_file->f_path.dentry->d_inode;
1326                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1327                         printk(KERN_INFO "btrfs: Snapshot src from "
1328                                "another FS\n");
1329                         ret = -EINVAL;
1330                 } else if (!inode_owner_or_capable(src_inode)) {
1331                         /*
1332                          * Subvolume creation is not restricted, but snapshots
1333                          * are limited to own subvolumes only
1334                          */
1335                         ret = -EPERM;
1336                 } else {
1337                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1338                                              BTRFS_I(src_inode)->root,
1339                                              transid, readonly);
1340                 }
1341                 fput(src_file);
1342         }
1343 out:
1344         return ret;
1345 }
1346
1347 static noinline int btrfs_ioctl_snap_create(struct file *file,
1348                                             void __user *arg, int subvol)
1349 {
1350         struct btrfs_ioctl_vol_args *vol_args;
1351         int ret;
1352
1353         vol_args = memdup_user(arg, sizeof(*vol_args));
1354         if (IS_ERR(vol_args))
1355                 return PTR_ERR(vol_args);
1356         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1357
1358         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1359                                               vol_args->fd, subvol,
1360                                               NULL, false);
1361
1362         kfree(vol_args);
1363         return ret;
1364 }
1365
1366 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1367                                                void __user *arg, int subvol)
1368 {
1369         struct btrfs_ioctl_vol_args_v2 *vol_args;
1370         int ret;
1371         u64 transid = 0;
1372         u64 *ptr = NULL;
1373         bool readonly = false;
1374
1375         vol_args = memdup_user(arg, sizeof(*vol_args));
1376         if (IS_ERR(vol_args))
1377                 return PTR_ERR(vol_args);
1378         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1379
1380         if (vol_args->flags &
1381             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1382                 ret = -EOPNOTSUPP;
1383                 goto out;
1384         }
1385
1386         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1387                 ptr = &transid;
1388         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1389                 readonly = true;
1390
1391         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1392                                               vol_args->fd, subvol,
1393                                               ptr, readonly);
1394
1395         if (ret == 0 && ptr &&
1396             copy_to_user(arg +
1397                          offsetof(struct btrfs_ioctl_vol_args_v2,
1398                                   transid), ptr, sizeof(*ptr)))
1399                 ret = -EFAULT;
1400 out:
1401         kfree(vol_args);
1402         return ret;
1403 }
1404
1405 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1406                                                 void __user *arg)
1407 {
1408         struct inode *inode = fdentry(file)->d_inode;
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410         int ret = 0;
1411         u64 flags = 0;
1412
1413         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1414                 return -EINVAL;
1415
1416         down_read(&root->fs_info->subvol_sem);
1417         if (btrfs_root_readonly(root))
1418                 flags |= BTRFS_SUBVOL_RDONLY;
1419         up_read(&root->fs_info->subvol_sem);
1420
1421         if (copy_to_user(arg, &flags, sizeof(flags)))
1422                 ret = -EFAULT;
1423
1424         return ret;
1425 }
1426
1427 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1428                                               void __user *arg)
1429 {
1430         struct inode *inode = fdentry(file)->d_inode;
1431         struct btrfs_root *root = BTRFS_I(inode)->root;
1432         struct btrfs_trans_handle *trans;
1433         u64 root_flags;
1434         u64 flags;
1435         int ret = 0;
1436
1437         if (root->fs_info->sb->s_flags & MS_RDONLY)
1438                 return -EROFS;
1439
1440         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1441                 return -EINVAL;
1442
1443         if (copy_from_user(&flags, arg, sizeof(flags)))
1444                 return -EFAULT;
1445
1446         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1447                 return -EINVAL;
1448
1449         if (flags & ~BTRFS_SUBVOL_RDONLY)
1450                 return -EOPNOTSUPP;
1451
1452         if (!inode_owner_or_capable(inode))
1453                 return -EACCES;
1454
1455         down_write(&root->fs_info->subvol_sem);
1456
1457         /* nothing to do */
1458         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1459                 goto out;
1460
1461         root_flags = btrfs_root_flags(&root->root_item);
1462         if (flags & BTRFS_SUBVOL_RDONLY)
1463                 btrfs_set_root_flags(&root->root_item,
1464                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1465         else
1466                 btrfs_set_root_flags(&root->root_item,
1467                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1468
1469         trans = btrfs_start_transaction(root, 1);
1470         if (IS_ERR(trans)) {
1471                 ret = PTR_ERR(trans);
1472                 goto out_reset;
1473         }
1474
1475         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1476                                 &root->root_key, &root->root_item);
1477
1478         btrfs_commit_transaction(trans, root);
1479 out_reset:
1480         if (ret)
1481                 btrfs_set_root_flags(&root->root_item, root_flags);
1482 out:
1483         up_write(&root->fs_info->subvol_sem);
1484         return ret;
1485 }
1486
1487 /*
1488  * helper to check if the subvolume references other subvolumes
1489  */
1490 static noinline int may_destroy_subvol(struct btrfs_root *root)
1491 {
1492         struct btrfs_path *path;
1493         struct btrfs_key key;
1494         int ret;
1495
1496         path = btrfs_alloc_path();
1497         if (!path)
1498                 return -ENOMEM;
1499
1500         key.objectid = root->root_key.objectid;
1501         key.type = BTRFS_ROOT_REF_KEY;
1502         key.offset = (u64)-1;
1503
1504         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1505                                 &key, path, 0, 0);
1506         if (ret < 0)
1507                 goto out;
1508         BUG_ON(ret == 0);
1509
1510         ret = 0;
1511         if (path->slots[0] > 0) {
1512                 path->slots[0]--;
1513                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1514                 if (key.objectid == root->root_key.objectid &&
1515                     key.type == BTRFS_ROOT_REF_KEY)
1516                         ret = -ENOTEMPTY;
1517         }
1518 out:
1519         btrfs_free_path(path);
1520         return ret;
1521 }
1522
1523 static noinline int key_in_sk(struct btrfs_key *key,
1524                               struct btrfs_ioctl_search_key *sk)
1525 {
1526         struct btrfs_key test;
1527         int ret;
1528
1529         test.objectid = sk->min_objectid;
1530         test.type = sk->min_type;
1531         test.offset = sk->min_offset;
1532
1533         ret = btrfs_comp_cpu_keys(key, &test);
1534         if (ret < 0)
1535                 return 0;
1536
1537         test.objectid = sk->max_objectid;
1538         test.type = sk->max_type;
1539         test.offset = sk->max_offset;
1540
1541         ret = btrfs_comp_cpu_keys(key, &test);
1542         if (ret > 0)
1543                 return 0;
1544         return 1;
1545 }
1546
1547 static noinline int copy_to_sk(struct btrfs_root *root,
1548                                struct btrfs_path *path,
1549                                struct btrfs_key *key,
1550                                struct btrfs_ioctl_search_key *sk,
1551                                char *buf,
1552                                unsigned long *sk_offset,
1553                                int *num_found)
1554 {
1555         u64 found_transid;
1556         struct extent_buffer *leaf;
1557         struct btrfs_ioctl_search_header sh;
1558         unsigned long item_off;
1559         unsigned long item_len;
1560         int nritems;
1561         int i;
1562         int slot;
1563         int ret = 0;
1564
1565         leaf = path->nodes[0];
1566         slot = path->slots[0];
1567         nritems = btrfs_header_nritems(leaf);
1568
1569         if (btrfs_header_generation(leaf) > sk->max_transid) {
1570                 i = nritems;
1571                 goto advance_key;
1572         }
1573         found_transid = btrfs_header_generation(leaf);
1574
1575         for (i = slot; i < nritems; i++) {
1576                 item_off = btrfs_item_ptr_offset(leaf, i);
1577                 item_len = btrfs_item_size_nr(leaf, i);
1578
1579                 btrfs_item_key_to_cpu(leaf, key, i);
1580                 if (!key_in_sk(key, sk))
1581                         continue;
1582
1583                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1584                         item_len = 0;
1585
1586                 if (sizeof(sh) + item_len + *sk_offset >
1587                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1588                         ret = 1;
1589                         goto overflow;
1590                 }
1591
1592                 sh.objectid = key->objectid;
1593                 sh.offset = key->offset;
1594                 sh.type = key->type;
1595                 sh.len = item_len;
1596                 sh.transid = found_transid;
1597
1598                 /* copy search result header */
1599                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1600                 *sk_offset += sizeof(sh);
1601
1602                 if (item_len) {
1603                         char *p = buf + *sk_offset;
1604                         /* copy the item */
1605                         read_extent_buffer(leaf, p,
1606                                            item_off, item_len);
1607                         *sk_offset += item_len;
1608                 }
1609                 (*num_found)++;
1610
1611                 if (*num_found >= sk->nr_items)
1612                         break;
1613         }
1614 advance_key:
1615         ret = 0;
1616         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1617                 key->offset++;
1618         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1619                 key->offset = 0;
1620                 key->type++;
1621         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1622                 key->offset = 0;
1623                 key->type = 0;
1624                 key->objectid++;
1625         } else
1626                 ret = 1;
1627 overflow:
1628         return ret;
1629 }
1630
1631 static noinline int search_ioctl(struct inode *inode,
1632                                  struct btrfs_ioctl_search_args *args)
1633 {
1634         struct btrfs_root *root;
1635         struct btrfs_key key;
1636         struct btrfs_key max_key;
1637         struct btrfs_path *path;
1638         struct btrfs_ioctl_search_key *sk = &args->key;
1639         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1640         int ret;
1641         int num_found = 0;
1642         unsigned long sk_offset = 0;
1643
1644         path = btrfs_alloc_path();
1645         if (!path)
1646                 return -ENOMEM;
1647
1648         if (sk->tree_id == 0) {
1649                 /* search the root of the inode that was passed */
1650                 root = BTRFS_I(inode)->root;
1651         } else {
1652                 key.objectid = sk->tree_id;
1653                 key.type = BTRFS_ROOT_ITEM_KEY;
1654                 key.offset = (u64)-1;
1655                 root = btrfs_read_fs_root_no_name(info, &key);
1656                 if (IS_ERR(root)) {
1657                         printk(KERN_ERR "could not find root %llu\n",
1658                                sk->tree_id);
1659                         btrfs_free_path(path);
1660                         return -ENOENT;
1661                 }
1662         }
1663
1664         key.objectid = sk->min_objectid;
1665         key.type = sk->min_type;
1666         key.offset = sk->min_offset;
1667
1668         max_key.objectid = sk->max_objectid;
1669         max_key.type = sk->max_type;
1670         max_key.offset = sk->max_offset;
1671
1672         path->keep_locks = 1;
1673
1674         while(1) {
1675                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1676                                            sk->min_transid);
1677                 if (ret != 0) {
1678                         if (ret > 0)
1679                                 ret = 0;
1680                         goto err;
1681                 }
1682                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1683                                  &sk_offset, &num_found);
1684                 btrfs_release_path(path);
1685                 if (ret || num_found >= sk->nr_items)
1686                         break;
1687
1688         }
1689         ret = 0;
1690 err:
1691         sk->nr_items = num_found;
1692         btrfs_free_path(path);
1693         return ret;
1694 }
1695
1696 static noinline int btrfs_ioctl_tree_search(struct file *file,
1697                                            void __user *argp)
1698 {
1699          struct btrfs_ioctl_search_args *args;
1700          struct inode *inode;
1701          int ret;
1702
1703         if (!capable(CAP_SYS_ADMIN))
1704                 return -EPERM;
1705
1706         args = memdup_user(argp, sizeof(*args));
1707         if (IS_ERR(args))
1708                 return PTR_ERR(args);
1709
1710         inode = fdentry(file)->d_inode;
1711         ret = search_ioctl(inode, args);
1712         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1713                 ret = -EFAULT;
1714         kfree(args);
1715         return ret;
1716 }
1717
1718 /*
1719  * Search INODE_REFs to identify path name of 'dirid' directory
1720  * in a 'tree_id' tree. and sets path name to 'name'.
1721  */
1722 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1723                                 u64 tree_id, u64 dirid, char *name)
1724 {
1725         struct btrfs_root *root;
1726         struct btrfs_key key;
1727         char *ptr;
1728         int ret = -1;
1729         int slot;
1730         int len;
1731         int total_len = 0;
1732         struct btrfs_inode_ref *iref;
1733         struct extent_buffer *l;
1734         struct btrfs_path *path;
1735
1736         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1737                 name[0]='\0';
1738                 return 0;
1739         }
1740
1741         path = btrfs_alloc_path();
1742         if (!path)
1743                 return -ENOMEM;
1744
1745         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1746
1747         key.objectid = tree_id;
1748         key.type = BTRFS_ROOT_ITEM_KEY;
1749         key.offset = (u64)-1;
1750         root = btrfs_read_fs_root_no_name(info, &key);
1751         if (IS_ERR(root)) {
1752                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1753                 ret = -ENOENT;
1754                 goto out;
1755         }
1756
1757         key.objectid = dirid;
1758         key.type = BTRFS_INODE_REF_KEY;
1759         key.offset = (u64)-1;
1760
1761         while(1) {
1762                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1763                 if (ret < 0)
1764                         goto out;
1765
1766                 l = path->nodes[0];
1767                 slot = path->slots[0];
1768                 if (ret > 0 && slot > 0)
1769                         slot--;
1770                 btrfs_item_key_to_cpu(l, &key, slot);
1771
1772                 if (ret > 0 && (key.objectid != dirid ||
1773                                 key.type != BTRFS_INODE_REF_KEY)) {
1774                         ret = -ENOENT;
1775                         goto out;
1776                 }
1777
1778                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1779                 len = btrfs_inode_ref_name_len(l, iref);
1780                 ptr -= len + 1;
1781                 total_len += len + 1;
1782                 if (ptr < name)
1783                         goto out;
1784
1785                 *(ptr + len) = '/';
1786                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1787
1788                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1789                         break;
1790
1791                 btrfs_release_path(path);
1792                 key.objectid = key.offset;
1793                 key.offset = (u64)-1;
1794                 dirid = key.objectid;
1795         }
1796         if (ptr < name)
1797                 goto out;
1798         memmove(name, ptr, total_len);
1799         name[total_len]='\0';
1800         ret = 0;
1801 out:
1802         btrfs_free_path(path);
1803         return ret;
1804 }
1805
1806 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1807                                            void __user *argp)
1808 {
1809          struct btrfs_ioctl_ino_lookup_args *args;
1810          struct inode *inode;
1811          int ret;
1812
1813         if (!capable(CAP_SYS_ADMIN))
1814                 return -EPERM;
1815
1816         args = memdup_user(argp, sizeof(*args));
1817         if (IS_ERR(args))
1818                 return PTR_ERR(args);
1819
1820         inode = fdentry(file)->d_inode;
1821
1822         if (args->treeid == 0)
1823                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1824
1825         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1826                                         args->treeid, args->objectid,
1827                                         args->name);
1828
1829         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1830                 ret = -EFAULT;
1831
1832         kfree(args);
1833         return ret;
1834 }
1835
1836 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1837                                              void __user *arg)
1838 {
1839         struct dentry *parent = fdentry(file);
1840         struct dentry *dentry;
1841         struct inode *dir = parent->d_inode;
1842         struct inode *inode;
1843         struct btrfs_root *root = BTRFS_I(dir)->root;
1844         struct btrfs_root *dest = NULL;
1845         struct btrfs_ioctl_vol_args *vol_args;
1846         struct btrfs_trans_handle *trans;
1847         int namelen;
1848         int ret;
1849         int err = 0;
1850
1851         vol_args = memdup_user(arg, sizeof(*vol_args));
1852         if (IS_ERR(vol_args))
1853                 return PTR_ERR(vol_args);
1854
1855         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1856         namelen = strlen(vol_args->name);
1857         if (strchr(vol_args->name, '/') ||
1858             strncmp(vol_args->name, "..", namelen) == 0) {
1859                 err = -EINVAL;
1860                 goto out;
1861         }
1862
1863         err = mnt_want_write(file->f_path.mnt);
1864         if (err)
1865                 goto out;
1866
1867         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1868         dentry = lookup_one_len(vol_args->name, parent, namelen);
1869         if (IS_ERR(dentry)) {
1870                 err = PTR_ERR(dentry);
1871                 goto out_unlock_dir;
1872         }
1873
1874         if (!dentry->d_inode) {
1875                 err = -ENOENT;
1876                 goto out_dput;
1877         }
1878
1879         inode = dentry->d_inode;
1880         dest = BTRFS_I(inode)->root;
1881         if (!capable(CAP_SYS_ADMIN)){
1882                 /*
1883                  * Regular user.  Only allow this with a special mount
1884                  * option, when the user has write+exec access to the
1885                  * subvol root, and when rmdir(2) would have been
1886                  * allowed.
1887                  *
1888                  * Note that this is _not_ check that the subvol is
1889                  * empty or doesn't contain data that we wouldn't
1890                  * otherwise be able to delete.
1891                  *
1892                  * Users who want to delete empty subvols should try
1893                  * rmdir(2).
1894                  */
1895                 err = -EPERM;
1896                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1897                         goto out_dput;
1898
1899                 /*
1900                  * Do not allow deletion if the parent dir is the same
1901                  * as the dir to be deleted.  That means the ioctl
1902                  * must be called on the dentry referencing the root
1903                  * of the subvol, not a random directory contained
1904                  * within it.
1905                  */
1906                 err = -EINVAL;
1907                 if (root == dest)
1908                         goto out_dput;
1909
1910                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1911                 if (err)
1912                         goto out_dput;
1913
1914                 /* check if subvolume may be deleted by a non-root user */
1915                 err = btrfs_may_delete(dir, dentry, 1);
1916                 if (err)
1917                         goto out_dput;
1918         }
1919
1920         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1921                 err = -EINVAL;
1922                 goto out_dput;
1923         }
1924
1925         mutex_lock(&inode->i_mutex);
1926         err = d_invalidate(dentry);
1927         if (err)
1928                 goto out_unlock;
1929
1930         down_write(&root->fs_info->subvol_sem);
1931
1932         err = may_destroy_subvol(dest);
1933         if (err)
1934                 goto out_up_write;
1935
1936         trans = btrfs_start_transaction(root, 0);
1937         if (IS_ERR(trans)) {
1938                 err = PTR_ERR(trans);
1939                 goto out_up_write;
1940         }
1941         trans->block_rsv = &root->fs_info->global_block_rsv;
1942
1943         ret = btrfs_unlink_subvol(trans, root, dir,
1944                                 dest->root_key.objectid,
1945                                 dentry->d_name.name,
1946                                 dentry->d_name.len);
1947         BUG_ON(ret);
1948
1949         btrfs_record_root_in_trans(trans, dest);
1950
1951         memset(&dest->root_item.drop_progress, 0,
1952                 sizeof(dest->root_item.drop_progress));
1953         dest->root_item.drop_level = 0;
1954         btrfs_set_root_refs(&dest->root_item, 0);
1955
1956         if (!xchg(&dest->orphan_item_inserted, 1)) {
1957                 ret = btrfs_insert_orphan_item(trans,
1958                                         root->fs_info->tree_root,
1959                                         dest->root_key.objectid);
1960                 BUG_ON(ret);
1961         }
1962
1963         ret = btrfs_end_transaction(trans, root);
1964         BUG_ON(ret);
1965         inode->i_flags |= S_DEAD;
1966 out_up_write:
1967         up_write(&root->fs_info->subvol_sem);
1968 out_unlock:
1969         mutex_unlock(&inode->i_mutex);
1970         if (!err) {
1971                 shrink_dcache_sb(root->fs_info->sb);
1972                 btrfs_invalidate_inodes(dest);
1973                 d_delete(dentry);
1974         }
1975 out_dput:
1976         dput(dentry);
1977 out_unlock_dir:
1978         mutex_unlock(&dir->i_mutex);
1979         mnt_drop_write(file->f_path.mnt);
1980 out:
1981         kfree(vol_args);
1982         return err;
1983 }
1984
1985 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1986 {
1987         struct inode *inode = fdentry(file)->d_inode;
1988         struct btrfs_root *root = BTRFS_I(inode)->root;
1989         struct btrfs_ioctl_defrag_range_args *range;
1990         int ret;
1991
1992         if (btrfs_root_readonly(root))
1993                 return -EROFS;
1994
1995         ret = mnt_want_write(file->f_path.mnt);
1996         if (ret)
1997                 return ret;
1998
1999         switch (inode->i_mode & S_IFMT) {
2000         case S_IFDIR:
2001                 if (!capable(CAP_SYS_ADMIN)) {
2002                         ret = -EPERM;
2003                         goto out;
2004                 }
2005                 ret = btrfs_defrag_root(root, 0);
2006                 if (ret)
2007                         goto out;
2008                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2009                 break;
2010         case S_IFREG:
2011                 if (!(file->f_mode & FMODE_WRITE)) {
2012                         ret = -EINVAL;
2013                         goto out;
2014                 }
2015
2016                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2017                 if (!range) {
2018                         ret = -ENOMEM;
2019                         goto out;
2020                 }
2021
2022                 if (argp) {
2023                         if (copy_from_user(range, argp,
2024                                            sizeof(*range))) {
2025                                 ret = -EFAULT;
2026                                 kfree(range);
2027                                 goto out;
2028                         }
2029                         /* compression requires us to start the IO */
2030                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2031                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2032                                 range->extent_thresh = (u32)-1;
2033                         }
2034                 } else {
2035                         /* the rest are all set to zero by kzalloc */
2036                         range->len = (u64)-1;
2037                 }
2038                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2039                                         range, 0, 0);
2040                 if (ret > 0)
2041                         ret = 0;
2042                 kfree(range);
2043                 break;
2044         default:
2045                 ret = -EINVAL;
2046         }
2047 out:
2048         mnt_drop_write(file->f_path.mnt);
2049         return ret;
2050 }
2051
2052 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2053 {
2054         struct btrfs_ioctl_vol_args *vol_args;
2055         int ret;
2056
2057         if (!capable(CAP_SYS_ADMIN))
2058                 return -EPERM;
2059
2060         vol_args = memdup_user(arg, sizeof(*vol_args));
2061         if (IS_ERR(vol_args))
2062                 return PTR_ERR(vol_args);
2063
2064         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2065         ret = btrfs_init_new_device(root, vol_args->name);
2066
2067         kfree(vol_args);
2068         return ret;
2069 }
2070
2071 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2072 {
2073         struct btrfs_ioctl_vol_args *vol_args;
2074         int ret;
2075
2076         if (!capable(CAP_SYS_ADMIN))
2077                 return -EPERM;
2078
2079         if (root->fs_info->sb->s_flags & MS_RDONLY)
2080                 return -EROFS;
2081
2082         vol_args = memdup_user(arg, sizeof(*vol_args));
2083         if (IS_ERR(vol_args))
2084                 return PTR_ERR(vol_args);
2085
2086         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2087         ret = btrfs_rm_device(root, vol_args->name);
2088
2089         kfree(vol_args);
2090         return ret;
2091 }
2092
2093 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2094 {
2095         struct btrfs_ioctl_fs_info_args *fi_args;
2096         struct btrfs_device *device;
2097         struct btrfs_device *next;
2098         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2099         int ret = 0;
2100
2101         if (!capable(CAP_SYS_ADMIN))
2102                 return -EPERM;
2103
2104         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2105         if (!fi_args)
2106                 return -ENOMEM;
2107
2108         fi_args->num_devices = fs_devices->num_devices;
2109         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2110
2111         mutex_lock(&fs_devices->device_list_mutex);
2112         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2113                 if (device->devid > fi_args->max_id)
2114                         fi_args->max_id = device->devid;
2115         }
2116         mutex_unlock(&fs_devices->device_list_mutex);
2117
2118         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2119                 ret = -EFAULT;
2120
2121         kfree(fi_args);
2122         return ret;
2123 }
2124
2125 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2126 {
2127         struct btrfs_ioctl_dev_info_args *di_args;
2128         struct btrfs_device *dev;
2129         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2130         int ret = 0;
2131         char *s_uuid = NULL;
2132         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2133
2134         if (!capable(CAP_SYS_ADMIN))
2135                 return -EPERM;
2136
2137         di_args = memdup_user(arg, sizeof(*di_args));
2138         if (IS_ERR(di_args))
2139                 return PTR_ERR(di_args);
2140
2141         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2142                 s_uuid = di_args->uuid;
2143
2144         mutex_lock(&fs_devices->device_list_mutex);
2145         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2146         mutex_unlock(&fs_devices->device_list_mutex);
2147
2148         if (!dev) {
2149                 ret = -ENODEV;
2150                 goto out;
2151         }
2152
2153         di_args->devid = dev->devid;
2154         di_args->bytes_used = dev->bytes_used;
2155         di_args->total_bytes = dev->total_bytes;
2156         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2157         strncpy(di_args->path, dev->name, sizeof(di_args->path));
2158
2159 out:
2160         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2161                 ret = -EFAULT;
2162
2163         kfree(di_args);
2164         return ret;
2165 }
2166
2167 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2168                                        u64 off, u64 olen, u64 destoff)
2169 {
2170         struct inode *inode = fdentry(file)->d_inode;
2171         struct btrfs_root *root = BTRFS_I(inode)->root;
2172         struct file *src_file;
2173         struct inode *src;
2174         struct btrfs_trans_handle *trans;
2175         struct btrfs_path *path;
2176         struct extent_buffer *leaf;
2177         char *buf;
2178         struct btrfs_key key;
2179         u32 nritems;
2180         int slot;
2181         int ret;
2182         u64 len = olen;
2183         u64 bs = root->fs_info->sb->s_blocksize;
2184         u64 hint_byte;
2185
2186         /*
2187          * TODO:
2188          * - split compressed inline extents.  annoying: we need to
2189          *   decompress into destination's address_space (the file offset
2190          *   may change, so source mapping won't do), then recompress (or
2191          *   otherwise reinsert) a subrange.
2192          * - allow ranges within the same file to be cloned (provided
2193          *   they don't overlap)?
2194          */
2195
2196         /* the destination must be opened for writing */
2197         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2198                 return -EINVAL;
2199
2200         if (btrfs_root_readonly(root))
2201                 return -EROFS;
2202
2203         ret = mnt_want_write(file->f_path.mnt);
2204         if (ret)
2205                 return ret;
2206
2207         src_file = fget(srcfd);
2208         if (!src_file) {
2209                 ret = -EBADF;
2210                 goto out_drop_write;
2211         }
2212
2213         src = src_file->f_dentry->d_inode;
2214
2215         ret = -EINVAL;
2216         if (src == inode)
2217                 goto out_fput;
2218
2219         /* the src must be open for reading */
2220         if (!(src_file->f_mode & FMODE_READ))
2221                 goto out_fput;
2222
2223         /* don't make the dst file partly checksummed */
2224         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2225             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2226                 goto out_fput;
2227
2228         ret = -EISDIR;
2229         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2230                 goto out_fput;
2231
2232         ret = -EXDEV;
2233         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2234                 goto out_fput;
2235
2236         ret = -ENOMEM;
2237         buf = vmalloc(btrfs_level_size(root, 0));
2238         if (!buf)
2239                 goto out_fput;
2240
2241         path = btrfs_alloc_path();
2242         if (!path) {
2243                 vfree(buf);
2244                 goto out_fput;
2245         }
2246         path->reada = 2;
2247
2248         if (inode < src) {
2249                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2250                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2251         } else {
2252                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2253                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2254         }
2255
2256         /* determine range to clone */
2257         ret = -EINVAL;
2258         if (off + len > src->i_size || off + len < off)
2259                 goto out_unlock;
2260         if (len == 0)
2261                 olen = len = src->i_size - off;
2262         /* if we extend to eof, continue to block boundary */
2263         if (off + len == src->i_size)
2264                 len = ALIGN(src->i_size, bs) - off;
2265
2266         /* verify the end result is block aligned */
2267         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2268             !IS_ALIGNED(destoff, bs))
2269                 goto out_unlock;
2270
2271         if (destoff > inode->i_size) {
2272                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2273                 if (ret)
2274                         goto out_unlock;
2275         }
2276
2277         /* truncate page cache pages from target inode range */
2278         truncate_inode_pages_range(&inode->i_data, destoff,
2279                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2280
2281         /* do any pending delalloc/csum calc on src, one way or
2282            another, and lock file content */
2283         while (1) {
2284                 struct btrfs_ordered_extent *ordered;
2285                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2286                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2287                 if (!ordered &&
2288                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2289                                    EXTENT_DELALLOC, 0, NULL))
2290                         break;
2291                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2292                 if (ordered)
2293                         btrfs_put_ordered_extent(ordered);
2294                 btrfs_wait_ordered_range(src, off, len);
2295         }
2296
2297         /* clone data */
2298         key.objectid = btrfs_ino(src);
2299         key.type = BTRFS_EXTENT_DATA_KEY;
2300         key.offset = 0;
2301
2302         while (1) {
2303                 /*
2304                  * note the key will change type as we walk through the
2305                  * tree.
2306                  */
2307                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2308                 if (ret < 0)
2309                         goto out;
2310
2311                 nritems = btrfs_header_nritems(path->nodes[0]);
2312                 if (path->slots[0] >= nritems) {
2313                         ret = btrfs_next_leaf(root, path);
2314                         if (ret < 0)
2315                                 goto out;
2316                         if (ret > 0)
2317                                 break;
2318                         nritems = btrfs_header_nritems(path->nodes[0]);
2319                 }
2320                 leaf = path->nodes[0];
2321                 slot = path->slots[0];
2322
2323                 btrfs_item_key_to_cpu(leaf, &key, slot);
2324                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2325                     key.objectid != btrfs_ino(src))
2326                         break;
2327
2328                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2329                         struct btrfs_file_extent_item *extent;
2330                         int type;
2331                         u32 size;
2332                         struct btrfs_key new_key;
2333                         u64 disko = 0, diskl = 0;
2334                         u64 datao = 0, datal = 0;
2335                         u8 comp;
2336                         u64 endoff;
2337
2338                         size = btrfs_item_size_nr(leaf, slot);
2339                         read_extent_buffer(leaf, buf,
2340                                            btrfs_item_ptr_offset(leaf, slot),
2341                                            size);
2342
2343                         extent = btrfs_item_ptr(leaf, slot,
2344                                                 struct btrfs_file_extent_item);
2345                         comp = btrfs_file_extent_compression(leaf, extent);
2346                         type = btrfs_file_extent_type(leaf, extent);
2347                         if (type == BTRFS_FILE_EXTENT_REG ||
2348                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2349                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2350                                                                       extent);
2351                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2352                                                                  extent);
2353                                 datao = btrfs_file_extent_offset(leaf, extent);
2354                                 datal = btrfs_file_extent_num_bytes(leaf,
2355                                                                     extent);
2356                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2357                                 /* take upper bound, may be compressed */
2358                                 datal = btrfs_file_extent_ram_bytes(leaf,
2359                                                                     extent);
2360                         }
2361                         btrfs_release_path(path);
2362
2363                         if (key.offset + datal <= off ||
2364                             key.offset >= off+len)
2365                                 goto next;
2366
2367                         memcpy(&new_key, &key, sizeof(new_key));
2368                         new_key.objectid = btrfs_ino(inode);
2369                         if (off <= key.offset)
2370                                 new_key.offset = key.offset + destoff - off;
2371                         else
2372                                 new_key.offset = destoff;
2373
2374                         /*
2375                          * 1 - adjusting old extent (we may have to split it)
2376                          * 1 - add new extent
2377                          * 1 - inode update
2378                          */
2379                         trans = btrfs_start_transaction(root, 3);
2380                         if (IS_ERR(trans)) {
2381                                 ret = PTR_ERR(trans);
2382                                 goto out;
2383                         }
2384
2385                         if (type == BTRFS_FILE_EXTENT_REG ||
2386                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2387                                 /*
2388                                  *    a  | --- range to clone ---|  b
2389                                  * | ------------- extent ------------- |
2390                                  */
2391
2392                                 /* substract range b */
2393                                 if (key.offset + datal > off + len)
2394                                         datal = off + len - key.offset;
2395
2396                                 /* substract range a */
2397                                 if (off > key.offset) {
2398                                         datao += off - key.offset;
2399                                         datal -= off - key.offset;
2400                                 }
2401
2402                                 ret = btrfs_drop_extents(trans, inode,
2403                                                          new_key.offset,
2404                                                          new_key.offset + datal,
2405                                                          &hint_byte, 1);
2406                                 BUG_ON(ret);
2407
2408                                 ret = btrfs_insert_empty_item(trans, root, path,
2409                                                               &new_key, size);
2410                                 BUG_ON(ret);
2411
2412                                 leaf = path->nodes[0];
2413                                 slot = path->slots[0];
2414                                 write_extent_buffer(leaf, buf,
2415                                             btrfs_item_ptr_offset(leaf, slot),
2416                                             size);
2417
2418                                 extent = btrfs_item_ptr(leaf, slot,
2419                                                 struct btrfs_file_extent_item);
2420
2421                                 /* disko == 0 means it's a hole */
2422                                 if (!disko)
2423                                         datao = 0;
2424
2425                                 btrfs_set_file_extent_offset(leaf, extent,
2426                                                              datao);
2427                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2428                                                                 datal);
2429                                 if (disko) {
2430                                         inode_add_bytes(inode, datal);
2431                                         ret = btrfs_inc_extent_ref(trans, root,
2432                                                         disko, diskl, 0,
2433                                                         root->root_key.objectid,
2434                                                         btrfs_ino(inode),
2435                                                         new_key.offset - datao);
2436                                         BUG_ON(ret);
2437                                 }
2438                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2439                                 u64 skip = 0;
2440                                 u64 trim = 0;
2441                                 if (off > key.offset) {
2442                                         skip = off - key.offset;
2443                                         new_key.offset += skip;
2444                                 }
2445
2446                                 if (key.offset + datal > off+len)
2447                                         trim = key.offset + datal - (off+len);
2448
2449                                 if (comp && (skip || trim)) {
2450                                         ret = -EINVAL;
2451                                         btrfs_end_transaction(trans, root);
2452                                         goto out;
2453                                 }
2454                                 size -= skip + trim;
2455                                 datal -= skip + trim;
2456
2457                                 ret = btrfs_drop_extents(trans, inode,
2458                                                          new_key.offset,
2459                                                          new_key.offset + datal,
2460                                                          &hint_byte, 1);
2461                                 BUG_ON(ret);
2462
2463                                 ret = btrfs_insert_empty_item(trans, root, path,
2464                                                               &new_key, size);
2465                                 BUG_ON(ret);
2466
2467                                 if (skip) {
2468                                         u32 start =
2469                                           btrfs_file_extent_calc_inline_size(0);
2470                                         memmove(buf+start, buf+start+skip,
2471                                                 datal);
2472                                 }
2473
2474                                 leaf = path->nodes[0];
2475                                 slot = path->slots[0];
2476                                 write_extent_buffer(leaf, buf,
2477                                             btrfs_item_ptr_offset(leaf, slot),
2478                                             size);
2479                                 inode_add_bytes(inode, datal);
2480                         }
2481
2482                         btrfs_mark_buffer_dirty(leaf);
2483                         btrfs_release_path(path);
2484
2485                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2486
2487                         /*
2488                          * we round up to the block size at eof when
2489                          * determining which extents to clone above,
2490                          * but shouldn't round up the file size
2491                          */
2492                         endoff = new_key.offset + datal;
2493                         if (endoff > destoff+olen)
2494                                 endoff = destoff+olen;
2495                         if (endoff > inode->i_size)
2496                                 btrfs_i_size_write(inode, endoff);
2497
2498                         ret = btrfs_update_inode(trans, root, inode);
2499                         BUG_ON(ret);
2500                         btrfs_end_transaction(trans, root);
2501                 }
2502 next:
2503                 btrfs_release_path(path);
2504                 key.offset++;
2505         }
2506         ret = 0;
2507 out:
2508         btrfs_release_path(path);
2509         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2510 out_unlock:
2511         mutex_unlock(&src->i_mutex);
2512         mutex_unlock(&inode->i_mutex);
2513         vfree(buf);
2514         btrfs_free_path(path);
2515 out_fput:
2516         fput(src_file);
2517 out_drop_write:
2518         mnt_drop_write(file->f_path.mnt);
2519         return ret;
2520 }
2521
2522 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2523 {
2524         struct btrfs_ioctl_clone_range_args args;
2525
2526         if (copy_from_user(&args, argp, sizeof(args)))
2527                 return -EFAULT;
2528         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2529                                  args.src_length, args.dest_offset);
2530 }
2531
2532 /*
2533  * there are many ways the trans_start and trans_end ioctls can lead
2534  * to deadlocks.  They should only be used by applications that
2535  * basically own the machine, and have a very in depth understanding
2536  * of all the possible deadlocks and enospc problems.
2537  */
2538 static long btrfs_ioctl_trans_start(struct file *file)
2539 {
2540         struct inode *inode = fdentry(file)->d_inode;
2541         struct btrfs_root *root = BTRFS_I(inode)->root;
2542         struct btrfs_trans_handle *trans;
2543         int ret;
2544
2545         ret = -EPERM;
2546         if (!capable(CAP_SYS_ADMIN))
2547                 goto out;
2548
2549         ret = -EINPROGRESS;
2550         if (file->private_data)
2551                 goto out;
2552
2553         ret = -EROFS;
2554         if (btrfs_root_readonly(root))
2555                 goto out;
2556
2557         ret = mnt_want_write(file->f_path.mnt);
2558         if (ret)
2559                 goto out;
2560
2561         atomic_inc(&root->fs_info->open_ioctl_trans);
2562
2563         ret = -ENOMEM;
2564         trans = btrfs_start_ioctl_transaction(root);
2565         if (IS_ERR(trans))
2566                 goto out_drop;
2567
2568         file->private_data = trans;
2569         return 0;
2570
2571 out_drop:
2572         atomic_dec(&root->fs_info->open_ioctl_trans);
2573         mnt_drop_write(file->f_path.mnt);
2574 out:
2575         return ret;
2576 }
2577
2578 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2579 {
2580         struct inode *inode = fdentry(file)->d_inode;
2581         struct btrfs_root *root = BTRFS_I(inode)->root;
2582         struct btrfs_root *new_root;
2583         struct btrfs_dir_item *di;
2584         struct btrfs_trans_handle *trans;
2585         struct btrfs_path *path;
2586         struct btrfs_key location;
2587         struct btrfs_disk_key disk_key;
2588         struct btrfs_super_block *disk_super;
2589         u64 features;
2590         u64 objectid = 0;
2591         u64 dir_id;
2592
2593         if (!capable(CAP_SYS_ADMIN))
2594                 return -EPERM;
2595
2596         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2597                 return -EFAULT;
2598
2599         if (!objectid)
2600                 objectid = root->root_key.objectid;
2601
2602         location.objectid = objectid;
2603         location.type = BTRFS_ROOT_ITEM_KEY;
2604         location.offset = (u64)-1;
2605
2606         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2607         if (IS_ERR(new_root))
2608                 return PTR_ERR(new_root);
2609
2610         if (btrfs_root_refs(&new_root->root_item) == 0)
2611                 return -ENOENT;
2612
2613         path = btrfs_alloc_path();
2614         if (!path)
2615                 return -ENOMEM;
2616         path->leave_spinning = 1;
2617
2618         trans = btrfs_start_transaction(root, 1);
2619         if (IS_ERR(trans)) {
2620                 btrfs_free_path(path);
2621                 return PTR_ERR(trans);
2622         }
2623
2624         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2625         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2626                                    dir_id, "default", 7, 1);
2627         if (IS_ERR_OR_NULL(di)) {
2628                 btrfs_free_path(path);
2629                 btrfs_end_transaction(trans, root);
2630                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2631                        "this isn't going to work\n");
2632                 return -ENOENT;
2633         }
2634
2635         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2636         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2637         btrfs_mark_buffer_dirty(path->nodes[0]);
2638         btrfs_free_path(path);
2639
2640         disk_super = root->fs_info->super_copy;
2641         features = btrfs_super_incompat_flags(disk_super);
2642         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2643                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2644                 btrfs_set_super_incompat_flags(disk_super, features);
2645         }
2646         btrfs_end_transaction(trans, root);
2647
2648         return 0;
2649 }
2650
2651 static void get_block_group_info(struct list_head *groups_list,
2652                                  struct btrfs_ioctl_space_info *space)
2653 {
2654         struct btrfs_block_group_cache *block_group;
2655
2656         space->total_bytes = 0;
2657         space->used_bytes = 0;
2658         space->flags = 0;
2659         list_for_each_entry(block_group, groups_list, list) {
2660                 space->flags = block_group->flags;
2661                 space->total_bytes += block_group->key.offset;
2662                 space->used_bytes +=
2663                         btrfs_block_group_used(&block_group->item);
2664         }
2665 }
2666
2667 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2668 {
2669         struct btrfs_ioctl_space_args space_args;
2670         struct btrfs_ioctl_space_info space;
2671         struct btrfs_ioctl_space_info *dest;
2672         struct btrfs_ioctl_space_info *dest_orig;
2673         struct btrfs_ioctl_space_info __user *user_dest;
2674         struct btrfs_space_info *info;
2675         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2676                        BTRFS_BLOCK_GROUP_SYSTEM,
2677                        BTRFS_BLOCK_GROUP_METADATA,
2678                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2679         int num_types = 4;
2680         int alloc_size;
2681         int ret = 0;
2682         u64 slot_count = 0;
2683         int i, c;
2684
2685         if (copy_from_user(&space_args,
2686                            (struct btrfs_ioctl_space_args __user *)arg,
2687                            sizeof(space_args)))
2688                 return -EFAULT;
2689
2690         for (i = 0; i < num_types; i++) {
2691                 struct btrfs_space_info *tmp;
2692
2693                 info = NULL;
2694                 rcu_read_lock();
2695                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2696                                         list) {
2697                         if (tmp->flags == types[i]) {
2698                                 info = tmp;
2699                                 break;
2700                         }
2701                 }
2702                 rcu_read_unlock();
2703
2704                 if (!info)
2705                         continue;
2706
2707                 down_read(&info->groups_sem);
2708                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2709                         if (!list_empty(&info->block_groups[c]))
2710                                 slot_count++;
2711                 }
2712                 up_read(&info->groups_sem);
2713         }
2714
2715         /* space_slots == 0 means they are asking for a count */
2716         if (space_args.space_slots == 0) {
2717                 space_args.total_spaces = slot_count;
2718                 goto out;
2719         }
2720
2721         slot_count = min_t(u64, space_args.space_slots, slot_count);
2722
2723         alloc_size = sizeof(*dest) * slot_count;
2724
2725         /* we generally have at most 6 or so space infos, one for each raid
2726          * level.  So, a whole page should be more than enough for everyone
2727          */
2728         if (alloc_size > PAGE_CACHE_SIZE)
2729                 return -ENOMEM;
2730
2731         space_args.total_spaces = 0;
2732         dest = kmalloc(alloc_size, GFP_NOFS);
2733         if (!dest)
2734                 return -ENOMEM;
2735         dest_orig = dest;
2736
2737         /* now we have a buffer to copy into */
2738         for (i = 0; i < num_types; i++) {
2739                 struct btrfs_space_info *tmp;
2740
2741                 if (!slot_count)
2742                         break;
2743
2744                 info = NULL;
2745                 rcu_read_lock();
2746                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2747                                         list) {
2748                         if (tmp->flags == types[i]) {
2749                                 info = tmp;
2750                                 break;
2751                         }
2752                 }
2753                 rcu_read_unlock();
2754
2755                 if (!info)
2756                         continue;
2757                 down_read(&info->groups_sem);
2758                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2759                         if (!list_empty(&info->block_groups[c])) {
2760                                 get_block_group_info(&info->block_groups[c],
2761                                                      &space);
2762                                 memcpy(dest, &space, sizeof(space));
2763                                 dest++;
2764                                 space_args.total_spaces++;
2765                                 slot_count--;
2766                         }
2767                         if (!slot_count)
2768                                 break;
2769                 }
2770                 up_read(&info->groups_sem);
2771         }
2772
2773         user_dest = (struct btrfs_ioctl_space_info *)
2774                 (arg + sizeof(struct btrfs_ioctl_space_args));
2775
2776         if (copy_to_user(user_dest, dest_orig, alloc_size))
2777                 ret = -EFAULT;
2778
2779         kfree(dest_orig);
2780 out:
2781         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2782                 ret = -EFAULT;
2783
2784         return ret;
2785 }
2786
2787 /*
2788  * there are many ways the trans_start and trans_end ioctls can lead
2789  * to deadlocks.  They should only be used by applications that
2790  * basically own the machine, and have a very in depth understanding
2791  * of all the possible deadlocks and enospc problems.
2792  */
2793 long btrfs_ioctl_trans_end(struct file *file)
2794 {
2795         struct inode *inode = fdentry(file)->d_inode;
2796         struct btrfs_root *root = BTRFS_I(inode)->root;
2797         struct btrfs_trans_handle *trans;
2798
2799         trans = file->private_data;
2800         if (!trans)
2801                 return -EINVAL;
2802         file->private_data = NULL;
2803
2804         btrfs_end_transaction(trans, root);
2805
2806         atomic_dec(&root->fs_info->open_ioctl_trans);
2807
2808         mnt_drop_write(file->f_path.mnt);
2809         return 0;
2810 }
2811
2812 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2813 {
2814         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2815         struct btrfs_trans_handle *trans;
2816         u64 transid;
2817         int ret;
2818
2819         trans = btrfs_start_transaction(root, 0);
2820         if (IS_ERR(trans))
2821                 return PTR_ERR(trans);
2822         transid = trans->transid;
2823         ret = btrfs_commit_transaction_async(trans, root, 0);
2824         if (ret) {
2825                 btrfs_end_transaction(trans, root);
2826                 return ret;
2827         }
2828
2829         if (argp)
2830                 if (copy_to_user(argp, &transid, sizeof(transid)))
2831                         return -EFAULT;
2832         return 0;
2833 }
2834
2835 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2836 {
2837         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2838         u64 transid;
2839
2840         if (argp) {
2841                 if (copy_from_user(&transid, argp, sizeof(transid)))
2842                         return -EFAULT;
2843         } else {
2844                 transid = 0;  /* current trans */
2845         }
2846         return btrfs_wait_for_commit(root, transid);
2847 }
2848
2849 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
2850 {
2851         int ret;
2852         struct btrfs_ioctl_scrub_args *sa;
2853
2854         if (!capable(CAP_SYS_ADMIN))
2855                 return -EPERM;
2856
2857         sa = memdup_user(arg, sizeof(*sa));
2858         if (IS_ERR(sa))
2859                 return PTR_ERR(sa);
2860
2861         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
2862                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
2863
2864         if (copy_to_user(arg, sa, sizeof(*sa)))
2865                 ret = -EFAULT;
2866
2867         kfree(sa);
2868         return ret;
2869 }
2870
2871 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
2872 {
2873         if (!capable(CAP_SYS_ADMIN))
2874                 return -EPERM;
2875
2876         return btrfs_scrub_cancel(root);
2877 }
2878
2879 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
2880                                        void __user *arg)
2881 {
2882         struct btrfs_ioctl_scrub_args *sa;
2883         int ret;
2884
2885         if (!capable(CAP_SYS_ADMIN))
2886                 return -EPERM;
2887
2888         sa = memdup_user(arg, sizeof(*sa));
2889         if (IS_ERR(sa))
2890                 return PTR_ERR(sa);
2891
2892         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
2893
2894         if (copy_to_user(arg, sa, sizeof(*sa)))
2895                 ret = -EFAULT;
2896
2897         kfree(sa);
2898         return ret;
2899 }
2900
2901 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
2902 {
2903         int ret = 0;
2904         int i;
2905         u64 rel_ptr;
2906         int size;
2907         struct btrfs_ioctl_ino_path_args *ipa = NULL;
2908         struct inode_fs_paths *ipath = NULL;
2909         struct btrfs_path *path;
2910
2911         if (!capable(CAP_SYS_ADMIN))
2912                 return -EPERM;
2913
2914         path = btrfs_alloc_path();
2915         if (!path) {
2916                 ret = -ENOMEM;
2917                 goto out;
2918         }
2919
2920         ipa = memdup_user(arg, sizeof(*ipa));
2921         if (IS_ERR(ipa)) {
2922                 ret = PTR_ERR(ipa);
2923                 ipa = NULL;
2924                 goto out;
2925         }
2926
2927         size = min_t(u32, ipa->size, 4096);
2928         ipath = init_ipath(size, root, path);
2929         if (IS_ERR(ipath)) {
2930                 ret = PTR_ERR(ipath);
2931                 ipath = NULL;
2932                 goto out;
2933         }
2934
2935         ret = paths_from_inode(ipa->inum, ipath);
2936         if (ret < 0)
2937                 goto out;
2938
2939         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
2940                 rel_ptr = ipath->fspath->val[i] -
2941                           (u64)(unsigned long)ipath->fspath->val;
2942                 ipath->fspath->val[i] = rel_ptr;
2943         }
2944
2945         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
2946                            (void *)(unsigned long)ipath->fspath, size);
2947         if (ret) {
2948                 ret = -EFAULT;
2949                 goto out;
2950         }
2951
2952 out:
2953         btrfs_free_path(path);
2954         free_ipath(ipath);
2955         kfree(ipa);
2956
2957         return ret;
2958 }
2959
2960 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
2961 {
2962         struct btrfs_data_container *inodes = ctx;
2963         const size_t c = 3 * sizeof(u64);
2964
2965         if (inodes->bytes_left >= c) {
2966                 inodes->bytes_left -= c;
2967                 inodes->val[inodes->elem_cnt] = inum;
2968                 inodes->val[inodes->elem_cnt + 1] = offset;
2969                 inodes->val[inodes->elem_cnt + 2] = root;
2970                 inodes->elem_cnt += 3;
2971         } else {
2972                 inodes->bytes_missing += c - inodes->bytes_left;
2973                 inodes->bytes_left = 0;
2974                 inodes->elem_missed += 3;
2975         }
2976
2977         return 0;
2978 }
2979
2980 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
2981                                         void __user *arg)
2982 {
2983         int ret = 0;
2984         int size;
2985         u64 extent_offset;
2986         struct btrfs_ioctl_logical_ino_args *loi;
2987         struct btrfs_data_container *inodes = NULL;
2988         struct btrfs_path *path = NULL;
2989         struct btrfs_key key;
2990
2991         if (!capable(CAP_SYS_ADMIN))
2992                 return -EPERM;
2993
2994         loi = memdup_user(arg, sizeof(*loi));
2995         if (IS_ERR(loi)) {
2996                 ret = PTR_ERR(loi);
2997                 loi = NULL;
2998                 goto out;
2999         }
3000
3001         path = btrfs_alloc_path();
3002         if (!path) {
3003                 ret = -ENOMEM;
3004                 goto out;
3005         }
3006
3007         size = min_t(u32, loi->size, 4096);
3008         inodes = init_data_container(size);
3009         if (IS_ERR(inodes)) {
3010                 ret = PTR_ERR(inodes);
3011                 inodes = NULL;
3012                 goto out;
3013         }
3014
3015         ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
3016
3017         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
3018                 ret = -ENOENT;
3019         if (ret < 0)
3020                 goto out;
3021
3022         extent_offset = loi->logical - key.objectid;
3023         ret = iterate_extent_inodes(root->fs_info, path, key.objectid,
3024                                         extent_offset, build_ino_list, inodes);
3025
3026         if (ret < 0)
3027                 goto out;
3028
3029         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3030                            (void *)(unsigned long)inodes, size);
3031         if (ret)
3032                 ret = -EFAULT;
3033
3034 out:
3035         btrfs_free_path(path);
3036         kfree(inodes);
3037         kfree(loi);
3038
3039         return ret;
3040 }
3041
3042 long btrfs_ioctl(struct file *file, unsigned int
3043                 cmd, unsigned long arg)
3044 {
3045         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3046         void __user *argp = (void __user *)arg;
3047
3048         switch (cmd) {
3049         case FS_IOC_GETFLAGS:
3050                 return btrfs_ioctl_getflags(file, argp);
3051         case FS_IOC_SETFLAGS:
3052                 return btrfs_ioctl_setflags(file, argp);
3053         case FS_IOC_GETVERSION:
3054                 return btrfs_ioctl_getversion(file, argp);
3055         case FITRIM:
3056                 return btrfs_ioctl_fitrim(file, argp);
3057         case BTRFS_IOC_SNAP_CREATE:
3058                 return btrfs_ioctl_snap_create(file, argp, 0);
3059         case BTRFS_IOC_SNAP_CREATE_V2:
3060                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3061         case BTRFS_IOC_SUBVOL_CREATE:
3062                 return btrfs_ioctl_snap_create(file, argp, 1);
3063         case BTRFS_IOC_SNAP_DESTROY:
3064                 return btrfs_ioctl_snap_destroy(file, argp);
3065         case BTRFS_IOC_SUBVOL_GETFLAGS:
3066                 return btrfs_ioctl_subvol_getflags(file, argp);
3067         case BTRFS_IOC_SUBVOL_SETFLAGS:
3068                 return btrfs_ioctl_subvol_setflags(file, argp);
3069         case BTRFS_IOC_DEFAULT_SUBVOL:
3070                 return btrfs_ioctl_default_subvol(file, argp);
3071         case BTRFS_IOC_DEFRAG:
3072                 return btrfs_ioctl_defrag(file, NULL);
3073         case BTRFS_IOC_DEFRAG_RANGE:
3074                 return btrfs_ioctl_defrag(file, argp);
3075         case BTRFS_IOC_RESIZE:
3076                 return btrfs_ioctl_resize(root, argp);
3077         case BTRFS_IOC_ADD_DEV:
3078                 return btrfs_ioctl_add_dev(root, argp);
3079         case BTRFS_IOC_RM_DEV:
3080                 return btrfs_ioctl_rm_dev(root, argp);
3081         case BTRFS_IOC_FS_INFO:
3082                 return btrfs_ioctl_fs_info(root, argp);
3083         case BTRFS_IOC_DEV_INFO:
3084                 return btrfs_ioctl_dev_info(root, argp);
3085         case BTRFS_IOC_BALANCE:
3086                 return btrfs_balance(root->fs_info->dev_root);
3087         case BTRFS_IOC_CLONE:
3088                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3089         case BTRFS_IOC_CLONE_RANGE:
3090                 return btrfs_ioctl_clone_range(file, argp);
3091         case BTRFS_IOC_TRANS_START:
3092                 return btrfs_ioctl_trans_start(file);
3093         case BTRFS_IOC_TRANS_END:
3094                 return btrfs_ioctl_trans_end(file);
3095         case BTRFS_IOC_TREE_SEARCH:
3096                 return btrfs_ioctl_tree_search(file, argp);
3097         case BTRFS_IOC_INO_LOOKUP:
3098                 return btrfs_ioctl_ino_lookup(file, argp);
3099         case BTRFS_IOC_INO_PATHS:
3100                 return btrfs_ioctl_ino_to_path(root, argp);
3101         case BTRFS_IOC_LOGICAL_INO:
3102                 return btrfs_ioctl_logical_to_ino(root, argp);
3103         case BTRFS_IOC_SPACE_INFO:
3104                 return btrfs_ioctl_space_info(root, argp);
3105         case BTRFS_IOC_SYNC:
3106                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3107                 return 0;
3108         case BTRFS_IOC_START_SYNC:
3109                 return btrfs_ioctl_start_sync(file, argp);
3110         case BTRFS_IOC_WAIT_SYNC:
3111                 return btrfs_ioctl_wait_sync(file, argp);
3112         case BTRFS_IOC_SCRUB:
3113                 return btrfs_ioctl_scrub(root, argp);
3114         case BTRFS_IOC_SCRUB_CANCEL:
3115                 return btrfs_ioctl_scrub_cancel(root, argp);
3116         case BTRFS_IOC_SCRUB_PROGRESS:
3117                 return btrfs_ioctl_scrub_progress(root, argp);
3118         }
3119
3120         return -ENOTTY;
3121 }