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