fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include "compat.h"
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "ioctl.h"
  50 #include "print-tree.h"
  51 #include "volumes.h"
  52 #include "locking.h"
  53 #include "inode-map.h"
  54 #include "backref.h"
  55
  56 /* Mask out flags that are inappropriate for the given type of inode. */
  57 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  58 {
  59         if (S_ISDIR(mode))
  60                 return flags;
  61         else if (S_ISREG(mode))
  62                 return flags & ~FS_DIRSYNC_FL;
  63         else
  64                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  65 }
  66
  67 /*
  68  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  69  */
  70 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  71 {
  72         unsigned int iflags = 0;
  73
  74         if (flags & BTRFS_INODE_SYNC)
  75                 iflags |= FS_SYNC_FL;
  76         if (flags & BTRFS_INODE_IMMUTABLE)
  77                 iflags |= FS_IMMUTABLE_FL;
  78         if (flags & BTRFS_INODE_APPEND)
  79                 iflags |= FS_APPEND_FL;
  80         if (flags & BTRFS_INODE_NODUMP)
  81                 iflags |= FS_NODUMP_FL;
  82         if (flags & BTRFS_INODE_NOATIME)
  83                 iflags |= FS_NOATIME_FL;
  84         if (flags & BTRFS_INODE_DIRSYNC)
  85                 iflags |= FS_DIRSYNC_FL;
  86         if (flags & BTRFS_INODE_NODATACOW)
  87                 iflags |= FS_NOCOW_FL;
  88
  89         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
  90                 iflags |= FS_COMPR_FL;
  91         else if (flags & BTRFS_INODE_NOCOMPRESS)
  92                 iflags |= FS_NOCOMP_FL;
  93
  94         return iflags;
  95 }
  96
  97 /*
  98  * Update inode->i_flags based on the btrfs internal flags.
  99  */
 100 void btrfs_update_iflags(struct inode *inode)
 101 {
 102         struct btrfs_inode *ip = BTRFS_I(inode);
 103
 104         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 105
 106         if (ip->flags & BTRFS_INODE_SYNC)
 107                 inode->i_flags |= S_SYNC;
 108         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 109                 inode->i_flags |= S_IMMUTABLE;
 110         if (ip->flags & BTRFS_INODE_APPEND)
 111                 inode->i_flags |= S_APPEND;
 112         if (ip->flags & BTRFS_INODE_NOATIME)
 113                 inode->i_flags |= S_NOATIME;
 114         if (ip->flags & BTRFS_INODE_DIRSYNC)
 115                 inode->i_flags |= S_DIRSYNC;
 116 }
 117
 118 /*
 119  * Inherit flags from the parent inode.
 120  *
 121  * Unlike extN we don't have any flags we don't want to inherit currently.
 122  */
 123 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 124 {
 125         unsigned int flags;
 126
 127         if (!dir)
 128                 return;
 129
 130         flags = BTRFS_I(dir)->flags;
 131
 132         if (S_ISREG(inode->i_mode))
 133                 flags &= ~BTRFS_INODE_DIRSYNC;
 134         else if (!S_ISDIR(inode->i_mode))
 135                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
 136
 137         BTRFS_I(inode)->flags = flags;
 138         btrfs_update_iflags(inode);
 139 }
 140
 141 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 142 {
 143         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 144         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 145
 146         if (copy_to_user(arg, &flags, sizeof(flags)))
 147                 return -EFAULT;
 148         return 0;
 149 }
 150
 151 static int check_flags(unsigned int flags)
 152 {
 153         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 154                       FS_NOATIME_FL | FS_NODUMP_FL | \
 155                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 156                       FS_NOCOMP_FL | FS_COMPR_FL |
 157                       FS_NOCOW_FL))
 158                 return -EOPNOTSUPP;
 159
 160         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 161                 return -EINVAL;
 162
 163         return 0;
 164 }
 165
 166 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 167 {
 168         struct inode *inode = file->f_path.dentry->d_inode;
 169         struct btrfs_inode *ip = BTRFS_I(inode);
 170         struct btrfs_root *root = ip->root;
 171         struct btrfs_trans_handle *trans;
 172         unsigned int flags, oldflags;
 173         int ret;
 174
 175         if (btrfs_root_readonly(root))
 176                 return -EROFS;
 177
 178         if (copy_from_user(&flags, arg, sizeof(flags)))
 179                 return -EFAULT;
 180
 181         ret = check_flags(flags);
 182         if (ret)
 183                 return ret;
 184
 185         if (!inode_owner_or_capable(inode))
 186                 return -EACCES;
 187
 188         mutex_lock(&inode->i_mutex);
 189
 190         flags = btrfs_mask_flags(inode->i_mode, flags);
 191         oldflags = btrfs_flags_to_ioctl(ip->flags);
 192         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 193                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 194                         ret = -EPERM;
 195                         goto out_unlock;
 196                 }
 197         }
 198
 199         ret = mnt_want_write(file->f_path.mnt);
 200         if (ret)
 201                 goto out_unlock;
 202
 203         if (flags & FS_SYNC_FL)
 204                 ip->flags |= BTRFS_INODE_SYNC;
 205         else
 206                 ip->flags &= ~BTRFS_INODE_SYNC;
 207         if (flags & FS_IMMUTABLE_FL)
 208                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 209         else
 210                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 211         if (flags & FS_APPEND_FL)
 212                 ip->flags |= BTRFS_INODE_APPEND;
 213         else
 214                 ip->flags &= ~BTRFS_INODE_APPEND;
 215         if (flags & FS_NODUMP_FL)
 216                 ip->flags |= BTRFS_INODE_NODUMP;
 217         else
 218                 ip->flags &= ~BTRFS_INODE_NODUMP;
 219         if (flags & FS_NOATIME_FL)
 220                 ip->flags |= BTRFS_INODE_NOATIME;
 221         else
 222                 ip->flags &= ~BTRFS_INODE_NOATIME;
 223         if (flags & FS_DIRSYNC_FL)
 224                 ip->flags |= BTRFS_INODE_DIRSYNC;
 225         else
 226                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 227         if (flags & FS_NOCOW_FL)
 228                 ip->flags |= BTRFS_INODE_NODATACOW;
 229         else
 230                 ip->flags &= ~BTRFS_INODE_NODATACOW;
 231
 232         /*
 233          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 234          * flag may be changed automatically if compression code won't make
 235          * things smaller.
 236          */
 237         if (flags & FS_NOCOMP_FL) {
 238                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 239                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 240         } else if (flags & FS_COMPR_FL) {
 241                 ip->flags |= BTRFS_INODE_COMPRESS;
 242                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 243         } else {
 244                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 245         }
 246
 247         trans = btrfs_join_transaction(root);
 248         BUG_ON(IS_ERR(trans));
 249
 250         ret = btrfs_update_inode(trans, root, inode);
 251         BUG_ON(ret);
 252
 253         btrfs_update_iflags(inode);
 254         inode->i_ctime = CURRENT_TIME;
 255         btrfs_end_transaction(trans, root);
 256
 257         mnt_drop_write(file->f_path.mnt);
 258
 259         ret = 0;
 260  out_unlock:
 261         mutex_unlock(&inode->i_mutex);
 262         return ret;
 263 }
 264
 265 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 266 {
 267         struct inode *inode = file->f_path.dentry->d_inode;
 268
 269         return put_user(inode->i_generation, arg);
 270 }
 271
 272 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 273 {
 274         struct btrfs_root *root = fdentry(file)->d_sb->s_fs_info;
 275         struct btrfs_fs_info *fs_info = root->fs_info;
 276         struct btrfs_device *device;
 277         struct request_queue *q;
 278         struct fstrim_range range;
 279         u64 minlen = ULLONG_MAX;
 280         u64 num_devices = 0;
 281         int ret;
 282
 283         if (!capable(CAP_SYS_ADMIN))
 284                 return -EPERM;
 285
 286         rcu_read_lock();
 287         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 288                                 dev_list) {
 289                 if (!device->bdev)
 290                         continue;
 291                 q = bdev_get_queue(device->bdev);
 292                 if (blk_queue_discard(q)) {
 293                         num_devices++;
 294                         minlen = min((u64)q->limits.discard_granularity,
 295                                      minlen);
 296                 }
 297         }
 298         rcu_read_unlock();
 299         if (!num_devices)
 300                 return -EOPNOTSUPP;
 301
 302         if (copy_from_user(&range, arg, sizeof(range)))
 303                 return -EFAULT;
 304
 305         range.minlen = max(range.minlen, minlen);
 306         ret = btrfs_trim_fs(root, &range);
 307         if (ret < 0)
 308                 return ret;
 309
 310         if (copy_to_user(arg, &range, sizeof(range)))
 311                 return -EFAULT;
 312
 313         return 0;
 314 }
 315
 316 static noinline int create_subvol(struct btrfs_root *root,
 317                                   struct dentry *dentry,
 318                                   char *name, int namelen,
 319                                   u64 *async_transid)
 320 {
 321         struct btrfs_trans_handle *trans;
 322         struct btrfs_key key;
 323         struct btrfs_root_item root_item;
 324         struct btrfs_inode_item *inode_item;
 325         struct extent_buffer *leaf;
 326         struct btrfs_root *new_root;
 327         struct dentry *parent = dentry->d_parent;
 328         struct inode *dir;
 329         int ret;
 330         int err;
 331         u64 objectid;
 332         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 333         u64 index = 0;
 334
 335         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 336         if (ret)
 337                 return ret;
 338
 339         dir = parent->d_inode;
 340
 341         /*
 342          * 1 - inode item
 343          * 2 - refs
 344          * 1 - root item
 345          * 2 - dir items
 346          */
 347         trans = btrfs_start_transaction(root, 6);
 348         if (IS_ERR(trans))
 349                 return PTR_ERR(trans);
 350
 351         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 352                                       0, objectid, NULL, 0, 0, 0);
 353         if (IS_ERR(leaf)) {
 354                 ret = PTR_ERR(leaf);
 355                 goto fail;
 356         }
 357
 358         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 359         btrfs_set_header_bytenr(leaf, leaf->start);
 360         btrfs_set_header_generation(leaf, trans->transid);
 361         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 362         btrfs_set_header_owner(leaf, objectid);
 363
 364         write_extent_buffer(leaf, root->fs_info->fsid,
 365                             (unsigned long)btrfs_header_fsid(leaf),
 366                             BTRFS_FSID_SIZE);
 367         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 368                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 369                             BTRFS_UUID_SIZE);
 370         btrfs_mark_buffer_dirty(leaf);
 371
 372         inode_item = &root_item.inode;
 373         memset(inode_item, 0, sizeof(*inode_item));
 374         inode_item->generation = cpu_to_le64(1);
 375         inode_item->size = cpu_to_le64(3);
 376         inode_item->nlink = cpu_to_le32(1);
 377         inode_item->nbytes = cpu_to_le64(root->leafsize);
 378         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 379
 380         root_item.flags = 0;
 381         root_item.byte_limit = 0;
 382         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
 383
 384         btrfs_set_root_bytenr(&root_item, leaf->start);
 385         btrfs_set_root_generation(&root_item, trans->transid);
 386         btrfs_set_root_level(&root_item, 0);
 387         btrfs_set_root_refs(&root_item, 1);
 388         btrfs_set_root_used(&root_item, leaf->len);
 389         btrfs_set_root_last_snapshot(&root_item, 0);
 390
 391         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 392         root_item.drop_level = 0;
 393
 394         btrfs_tree_unlock(leaf);
 395         free_extent_buffer(leaf);
 396         leaf = NULL;
 397
 398         btrfs_set_root_dirid(&root_item, new_dirid);
 399
 400         key.objectid = objectid;
 401         key.offset = 0;
 402         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 403         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 404                                 &root_item);
 405         if (ret)
 406                 goto fail;
 407
 408         key.offset = (u64)-1;
 409         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 410         BUG_ON(IS_ERR(new_root));
 411
 412         btrfs_record_root_in_trans(trans, new_root);
 413
 414         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
 415         /*
 416          * insert the directory item
 417          */
 418         ret = btrfs_set_inode_index(dir, &index);
 419         BUG_ON(ret);
 420
 421         ret = btrfs_insert_dir_item(trans, root,
 422                                     name, namelen, dir, &key,
 423                                     BTRFS_FT_DIR, index);
 424         if (ret)
 425                 goto fail;
 426
 427         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 428         ret = btrfs_update_inode(trans, root, dir);
 429         BUG_ON(ret);
 430
 431         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 432                                  objectid, root->root_key.objectid,
 433                                  btrfs_ino(dir), index, name, namelen);
 434
 435         BUG_ON(ret);
 436
 437         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 438 fail:
 439         if (async_transid) {
 440                 *async_transid = trans->transid;
 441                 err = btrfs_commit_transaction_async(trans, root, 1);
 442         } else {
 443                 err = btrfs_commit_transaction(trans, root);
 444         }
 445         if (err && !ret)
 446                 ret = err;
 447         return ret;
 448 }
 449
 450 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 451                            char *name, int namelen, u64 *async_transid,
 452                            bool readonly)
 453 {
 454         struct inode *inode;
 455         struct btrfs_pending_snapshot *pending_snapshot;
 456         struct btrfs_trans_handle *trans;
 457         int ret;
 458
 459         if (!root->ref_cows)
 460                 return -EINVAL;
 461
 462         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 463         if (!pending_snapshot)
 464                 return -ENOMEM;
 465
 466         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
 467         pending_snapshot->dentry = dentry;
 468         pending_snapshot->root = root;
 469         pending_snapshot->readonly = readonly;
 470
 471         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
 472         if (IS_ERR(trans)) {
 473                 ret = PTR_ERR(trans);
 474                 goto fail;
 475         }
 476
 477         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 478         BUG_ON(ret);
 479
 480         spin_lock(&root->fs_info->trans_lock);
 481         list_add(&pending_snapshot->list,
 482                  &trans->transaction->pending_snapshots);
 483         spin_unlock(&root->fs_info->trans_lock);
 484         if (async_transid) {
 485                 *async_transid = trans->transid;
 486                 ret = btrfs_commit_transaction_async(trans,
 487                                      root->fs_info->extent_root, 1);
 488         } else {
 489                 ret = btrfs_commit_transaction(trans,
 490                                                root->fs_info->extent_root);
 491         }
 492         BUG_ON(ret);
 493
 494         ret = pending_snapshot->error;
 495         if (ret)
 496                 goto fail;
 497
 498         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 499         if (ret)
 500                 goto fail;
 501
 502         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 503         if (IS_ERR(inode)) {
 504                 ret = PTR_ERR(inode);
 505                 goto fail;
 506         }
 507         BUG_ON(!inode);
 508         d_instantiate(dentry, inode);
 509         ret = 0;
 510 fail:
 511         kfree(pending_snapshot);
 512         return ret;
 513 }
 514
 515 /*  copy of check_sticky in fs/namei.c()
 516 * It's inline, so penalty for filesystems that don't use sticky bit is
 517 * minimal.
 518 */
 519 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 520 {
 521         uid_t fsuid = current_fsuid();
 522
 523         if (!(dir->i_mode & S_ISVTX))
 524                 return 0;
 525         if (inode->i_uid == fsuid)
 526                 return 0;
 527         if (dir->i_uid == fsuid)
 528                 return 0;
 529         return !capable(CAP_FOWNER);
 530 }
 531
 532 /*  copy of may_delete in fs/namei.c()
 533  *      Check whether we can remove a link victim from directory dir, check
 534  *  whether the type of victim is right.
 535  *  1. We can't do it if dir is read-only (done in permission())
 536  *  2. We should have write and exec permissions on dir
 537  *  3. We can't remove anything from append-only dir
 538  *  4. We can't do anything with immutable dir (done in permission())
 539  *  5. If the sticky bit on dir is set we should either
 540  *      a. be owner of dir, or
 541  *      b. be owner of victim, or
 542  *      c. have CAP_FOWNER capability
 543  *  6. If the victim is append-only or immutable we can't do antyhing with
 544  *     links pointing to it.
 545  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 546  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 547  *  9. We can't remove a root or mountpoint.
 548  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 549  *     nfs_async_unlink().
 550  */
 551
 552 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 553 {
 554         int error;
 555
 556         if (!victim->d_inode)
 557                 return -ENOENT;
 558
 559         BUG_ON(victim->d_parent->d_inode != dir);
 560         audit_inode_child(victim, dir);
 561
 562         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 563         if (error)
 564                 return error;
 565         if (IS_APPEND(dir))
 566                 return -EPERM;
 567         if (btrfs_check_sticky(dir, victim->d_inode)||
 568                 IS_APPEND(victim->d_inode)||
 569             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 570                 return -EPERM;
 571         if (isdir) {
 572                 if (!S_ISDIR(victim->d_inode->i_mode))
 573                         return -ENOTDIR;
 574                 if (IS_ROOT(victim))
 575                         return -EBUSY;
 576         } else if (S_ISDIR(victim->d_inode->i_mode))
 577                 return -EISDIR;
 578         if (IS_DEADDIR(dir))
 579                 return -ENOENT;
 580         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 581                 return -EBUSY;
 582         return 0;
 583 }
 584
 585 /* copy of may_create in fs/namei.c() */
 586 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 587 {
 588         if (child->d_inode)
 589                 return -EEXIST;
 590         if (IS_DEADDIR(dir))
 591                 return -ENOENT;
 592         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 593 }
 594
 595 /*
 596  * Create a new subvolume below @parent.  This is largely modeled after
 597  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 598  * inside this filesystem so it's quite a bit simpler.
 599  */
 600 static noinline int btrfs_mksubvol(struct path *parent,
 601                                    char *name, int namelen,
 602                                    struct btrfs_root *snap_src,
 603                                    u64 *async_transid, bool readonly)
 604 {
 605         struct inode *dir  = parent->dentry->d_inode;
 606         struct dentry *dentry;
 607         int error;
 608
 609         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 610
 611         dentry = lookup_one_len(name, parent->dentry, namelen);
 612         error = PTR_ERR(dentry);
 613         if (IS_ERR(dentry))
 614                 goto out_unlock;
 615
 616         error = -EEXIST;
 617         if (dentry->d_inode)
 618                 goto out_dput;
 619
 620         error = mnt_want_write(parent->mnt);
 621         if (error)
 622                 goto out_dput;
 623
 624         error = btrfs_may_create(dir, dentry);
 625         if (error)
 626                 goto out_drop_write;
 627
 628         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 629
 630         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 631                 goto out_up_read;
 632
 633         if (snap_src) {
 634                 error = create_snapshot(snap_src, dentry,
 635                                         name, namelen, async_transid, readonly);
 636         } else {
 637                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 638                                       name, namelen, async_transid);
 639         }
 640         if (!error)
 641                 fsnotify_mkdir(dir, dentry);
 642 out_up_read:
 643         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 644 out_drop_write:
 645         mnt_drop_write(parent->mnt);
 646 out_dput:
 647         dput(dentry);
 648 out_unlock:
 649         mutex_unlock(&dir->i_mutex);
 650         return error;
 651 }
 652
 653 /*
 654  * When we're defragging a range, we don't want to kick it off again
 655  * if it is really just waiting for delalloc to send it down.
 656  * If we find a nice big extent or delalloc range for the bytes in the
 657  * file you want to defrag, we return 0 to let you know to skip this
 658  * part of the file
 659  */
 660 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 661 {
 662         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 663         struct extent_map *em = NULL;
 664         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 665         u64 end;
 666
 667         read_lock(&em_tree->lock);
 668         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 669         read_unlock(&em_tree->lock);
 670
 671         if (em) {
 672                 end = extent_map_end(em);
 673                 free_extent_map(em);
 674                 if (end - offset > thresh)
 675                         return 0;
 676         }
 677         /* if we already have a nice delalloc here, just stop */
 678         thresh /= 2;
 679         end = count_range_bits(io_tree, &offset, offset + thresh,
 680                                thresh, EXTENT_DELALLOC, 1);
 681         if (end >= thresh)
 682                 return 0;
 683         return 1;
 684 }
 685
 686 /*
 687  * helper function to walk through a file and find extents
 688  * newer than a specific transid, and smaller than thresh.
 689  *
 690  * This is used by the defragging code to find new and small
 691  * extents
 692  */
 693 static int find_new_extents(struct btrfs_root *root,
 694                             struct inode *inode, u64 newer_than,
 695                             u64 *off, int thresh)
 696 {
 697         struct btrfs_path *path;
 698         struct btrfs_key min_key;
 699         struct btrfs_key max_key;
 700         struct extent_buffer *leaf;
 701         struct btrfs_file_extent_item *extent;
 702         int type;
 703         int ret;
 704         u64 ino = btrfs_ino(inode);
 705
 706         path = btrfs_alloc_path();
 707         if (!path)
 708                 return -ENOMEM;
 709
 710         min_key.objectid = ino;
 711         min_key.type = BTRFS_EXTENT_DATA_KEY;
 712         min_key.offset = *off;
 713
 714         max_key.objectid = ino;
 715         max_key.type = (u8)-1;
 716         max_key.offset = (u64)-1;
 717
 718         path->keep_locks = 1;
 719
 720         while(1) {
 721                 ret = btrfs_search_forward(root, &min_key, &max_key,
 722                                            path, 0, newer_than);
 723                 if (ret != 0)
 724                         goto none;
 725                 if (min_key.objectid != ino)
 726                         goto none;
 727                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 728                         goto none;
 729
 730                 leaf = path->nodes[0];
 731                 extent = btrfs_item_ptr(leaf, path->slots[0],
 732                                         struct btrfs_file_extent_item);
 733
 734                 type = btrfs_file_extent_type(leaf, extent);
 735                 if (type == BTRFS_FILE_EXTENT_REG &&
 736                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 737                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 738                         *off = min_key.offset;
 739                         btrfs_free_path(path);
 740                         return 0;
 741                 }
 742
 743                 if (min_key.offset == (u64)-1)
 744                         goto none;
 745
 746                 min_key.offset++;
 747                 btrfs_release_path(path);
 748         }
 749 none:
 750         btrfs_free_path(path);
 751         return -ENOENT;
 752 }
 753
 754 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
 755                                int thresh, u64 *last_len, u64 *skip,
 756                                u64 *defrag_end)
 757 {
 758         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 759         struct extent_map *em = NULL;
 760         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 761         int ret = 1;
 762
 763         /*
 764          * make sure that once we start defragging and extent, we keep on
 765          * defragging it
 766          */
 767         if (start < *defrag_end)
 768                 return 1;
 769
 770         *skip = 0;
 771
 772         /*
 773          * hopefully we have this extent in the tree already, try without
 774          * the full extent lock
 775          */
 776         read_lock(&em_tree->lock);
 777         em = lookup_extent_mapping(em_tree, start, len);
 778         read_unlock(&em_tree->lock);
 779
 780         if (!em) {
 781                 /* get the big lock and read metadata off disk */
 782                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 783                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 784                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 785
 786                 if (IS_ERR(em))
 787                         return 0;
 788         }
 789
 790         /* this will cover holes, and inline extents */
 791         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
 792                 ret = 0;
 793
 794         /*
 795          * we hit a real extent, if it is big don't bother defragging it again
 796          */
 797         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
 798                 ret = 0;
 799
 800         /*
 801          * last_len ends up being a counter of how many bytes we've defragged.
 802          * every time we choose not to defrag an extent, we reset *last_len
 803          * so that the next tiny extent will force a defrag.
 804          *
 805          * The end result of this is that tiny extents before a single big
 806          * extent will force at least part of that big extent to be defragged.
 807          */
 808         if (ret) {
 809                 *last_len += len;
 810                 *defrag_end = extent_map_end(em);
 811         } else {
 812                 *last_len = 0;
 813                 *skip = extent_map_end(em);
 814                 *defrag_end = 0;
 815         }
 816
 817         free_extent_map(em);
 818         return ret;
 819 }
 820
 821 /*
 822  * it doesn't do much good to defrag one or two pages
 823  * at a time.  This pulls in a nice chunk of pages
 824  * to COW and defrag.
 825  *
 826  * It also makes sure the delalloc code has enough
 827  * dirty data to avoid making new small extents as part
 828  * of the defrag
 829  *
 830  * It's a good idea to start RA on this range
 831  * before calling this.
 832  */
 833 static int cluster_pages_for_defrag(struct inode *inode,
 834                                     struct page **pages,
 835                                     unsigned long start_index,
 836                                     int num_pages)
 837 {
 838         unsigned long file_end;
 839         u64 isize = i_size_read(inode);
 840         u64 page_start;
 841         u64 page_end;
 842         int ret;
 843         int i;
 844         int i_done;
 845         struct btrfs_ordered_extent *ordered;
 846         struct extent_state *cached_state = NULL;
 847
 848         if (isize == 0)
 849                 return 0;
 850         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 851
 852         ret = btrfs_delalloc_reserve_space(inode,
 853                                            num_pages << PAGE_CACHE_SHIFT);
 854         if (ret)
 855                 return ret;
 856 again:
 857         ret = 0;
 858         i_done = 0;
 859
 860         /* step one, lock all the pages */
 861         for (i = 0; i < num_pages; i++) {
 862                 struct page *page;
 863                 page = find_or_create_page(inode->i_mapping,
 864                                             start_index + i, GFP_NOFS);
 865                 if (!page)
 866                         break;
 867
 868                 if (!PageUptodate(page)) {
 869                         btrfs_readpage(NULL, page);
 870                         lock_page(page);
 871                         if (!PageUptodate(page)) {
 872                                 unlock_page(page);
 873                                 page_cache_release(page);
 874                                 ret = -EIO;
 875                                 break;
 876                         }
 877                 }
 878                 isize = i_size_read(inode);
 879                 file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 880                 if (!isize || page->index > file_end ||
 881                     page->mapping != inode->i_mapping) {
 882                         /* whoops, we blew past eof, skip this page */
 883                         unlock_page(page);
 884                         page_cache_release(page);
 885                         break;
 886                 }
 887                 pages[i] = page;
 888                 i_done++;
 889         }
 890         if (!i_done || ret)
 891                 goto out;
 892
 893         if (!(inode->i_sb->s_flags & MS_ACTIVE))
 894                 goto out;
 895
 896         /*
 897          * so now we have a nice long stream of locked
 898          * and up to date pages, lets wait on them
 899          */
 900         for (i = 0; i < i_done; i++)
 901                 wait_on_page_writeback(pages[i]);
 902
 903         page_start = page_offset(pages[0]);
 904         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
 905
 906         lock_extent_bits(&BTRFS_I(inode)->io_tree,
 907                          page_start, page_end - 1, 0, &cached_state,
 908                          GFP_NOFS);
 909         ordered = btrfs_lookup_first_ordered_extent(inode, page_end - 1);
 910         if (ordered &&
 911             ordered->file_offset + ordered->len > page_start &&
 912             ordered->file_offset < page_end) {
 913                 btrfs_put_ordered_extent(ordered);
 914                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 915                                      page_start, page_end - 1,
 916                                      &cached_state, GFP_NOFS);
 917                 for (i = 0; i < i_done; i++) {
 918                         unlock_page(pages[i]);
 919                         page_cache_release(pages[i]);
 920                 }
 921                 btrfs_wait_ordered_range(inode, page_start,
 922                                          page_end - page_start);
 923                 goto again;
 924         }
 925         if (ordered)
 926                 btrfs_put_ordered_extent(ordered);
 927
 928         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
 929                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
 930                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
 931                           GFP_NOFS);
 932
 933         if (i_done != num_pages) {
 934                 spin_lock(&BTRFS_I(inode)->lock);
 935                 BTRFS_I(inode)->outstanding_extents++;
 936                 spin_unlock(&BTRFS_I(inode)->lock);
 937                 btrfs_delalloc_release_space(inode,
 938                                      (num_pages - i_done) << PAGE_CACHE_SHIFT);
 939         }
 940
 941
 942         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
 943                                   &cached_state);
 944
 945         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 946                              page_start, page_end - 1, &cached_state,
 947                              GFP_NOFS);
 948
 949         for (i = 0; i < i_done; i++) {
 950                 clear_page_dirty_for_io(pages[i]);
 951                 ClearPageChecked(pages[i]);
 952                 set_page_extent_mapped(pages[i]);
 953                 set_page_dirty(pages[i]);
 954                 unlock_page(pages[i]);
 955                 page_cache_release(pages[i]);
 956         }
 957         return i_done;
 958 out:
 959         for (i = 0; i < i_done; i++) {
 960                 unlock_page(pages[i]);
 961                 page_cache_release(pages[i]);
 962         }
 963         btrfs_delalloc_release_space(inode, num_pages << PAGE_CACHE_SHIFT);
 964         return ret;
 965
 966 }
 967
 968 int btrfs_defrag_file(struct inode *inode, struct file *file,
 969                       struct btrfs_ioctl_defrag_range_args *range,
 970                       u64 newer_than, unsigned long max_to_defrag)
 971 {
 972         struct btrfs_root *root = BTRFS_I(inode)->root;
 973         struct btrfs_super_block *disk_super;
 974         struct file_ra_state *ra = NULL;
 975         unsigned long last_index;
 976         u64 features;
 977         u64 last_len = 0;
 978         u64 skip = 0;
 979         u64 defrag_end = 0;
 980         u64 newer_off = range->start;
 981         int newer_left = 0;
 982         unsigned long i;
 983         int ret;
 984         int defrag_count = 0;
 985         int compress_type = BTRFS_COMPRESS_ZLIB;
 986         int extent_thresh = range->extent_thresh;
 987         int newer_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
 988         u64 new_align = ~((u64)128 * 1024 - 1);
 989         struct page **pages = NULL;
 990
 991         if (extent_thresh == 0)
 992                 extent_thresh = 256 * 1024;
 993
 994         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
 995                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
 996                         return -EINVAL;
 997                 if (range->compress_type)
 998                         compress_type = range->compress_type;
 999         }
1000
1001         if (inode->i_size == 0)
1002                 return 0;
1003
1004         /*
1005          * if we were not given a file, allocate a readahead
1006          * context
1007          */
1008         if (!file) {
1009                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1010                 if (!ra)
1011                         return -ENOMEM;
1012                 file_ra_state_init(ra, inode->i_mapping);
1013         } else {
1014                 ra = &file->f_ra;
1015         }
1016
1017         pages = kmalloc(sizeof(struct page *) * newer_cluster,
1018                         GFP_NOFS);
1019         if (!pages) {
1020                 ret = -ENOMEM;
1021                 goto out_ra;
1022         }
1023
1024         /* find the last page to defrag */
1025         if (range->start + range->len > range->start) {
1026                 last_index = min_t(u64, inode->i_size - 1,
1027                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1028         } else {
1029                 last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
1030         }
1031
1032         if (newer_than) {
1033                 ret = find_new_extents(root, inode, newer_than,
1034                                        &newer_off, 64 * 1024);
1035                 if (!ret) {
1036                         range->start = newer_off;
1037                         /*
1038                          * we always align our defrag to help keep
1039                          * the extents in the file evenly spaced
1040                          */
1041                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1042                         newer_left = newer_cluster;
1043                 } else
1044                         goto out_ra;
1045         } else {
1046                 i = range->start >> PAGE_CACHE_SHIFT;
1047         }
1048         if (!max_to_defrag)
1049                 max_to_defrag = last_index - 1;
1050
1051         while (i <= last_index && defrag_count < max_to_defrag) {
1052                 /*
1053                  * make sure we stop running if someone unmounts
1054                  * the FS
1055                  */
1056                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1057                         break;
1058
1059                 if (!newer_than &&
1060                     !should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1061                                         PAGE_CACHE_SIZE,
1062                                         extent_thresh,
1063                                         &last_len, &skip,
1064                                         &defrag_end)) {
1065                         unsigned long next;
1066                         /*
1067                          * the should_defrag function tells us how much to skip
1068                          * bump our counter by the suggested amount
1069                          */
1070                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1071                         i = max(i + 1, next);
1072                         continue;
1073                 }
1074                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1075                         BTRFS_I(inode)->force_compress = compress_type;
1076
1077                 btrfs_force_ra(inode->i_mapping, ra, file, i, newer_cluster);
1078
1079                 ret = cluster_pages_for_defrag(inode, pages, i, newer_cluster);
1080                 if (ret < 0)
1081                         goto out_ra;
1082
1083                 defrag_count += ret;
1084                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1085                 i += ret;
1086
1087                 if (newer_than) {
1088                         if (newer_off == (u64)-1)
1089                                 break;
1090
1091                         newer_off = max(newer_off + 1,
1092                                         (u64)i << PAGE_CACHE_SHIFT);
1093
1094                         ret = find_new_extents(root, inode,
1095                                                newer_than, &newer_off,
1096                                                64 * 1024);
1097                         if (!ret) {
1098                                 range->start = newer_off;
1099                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1100                                 newer_left = newer_cluster;
1101                         } else {
1102                                 break;
1103                         }
1104                 } else {
1105                         i++;
1106                 }
1107         }
1108
1109         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1110                 filemap_flush(inode->i_mapping);
1111
1112         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1113                 /* the filemap_flush will queue IO into the worker threads, but
1114                  * we have to make sure the IO is actually started and that
1115                  * ordered extents get created before we return
1116                  */
1117                 atomic_inc(&root->fs_info->async_submit_draining);
1118                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1119                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1120                         wait_event(root->fs_info->async_submit_wait,
1121                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1122                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1123                 }
1124                 atomic_dec(&root->fs_info->async_submit_draining);
1125
1126                 mutex_lock(&inode->i_mutex);
1127                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1128                 mutex_unlock(&inode->i_mutex);
1129         }
1130
1131         disk_super = &root->fs_info->super_copy;
1132         features = btrfs_super_incompat_flags(disk_super);
1133         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1134                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1135                 btrfs_set_super_incompat_flags(disk_super, features);
1136         }
1137
1138         if (!file)
1139                 kfree(ra);
1140         return defrag_count;
1141
1142 out_ra:
1143         if (!file)
1144                 kfree(ra);
1145         kfree(pages);
1146         return ret;
1147 }
1148
1149 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1150                                         void __user *arg)
1151 {
1152         u64 new_size;
1153         u64 old_size;
1154         u64 devid = 1;
1155         struct btrfs_ioctl_vol_args *vol_args;
1156         struct btrfs_trans_handle *trans;
1157         struct btrfs_device *device = NULL;
1158         char *sizestr;
1159         char *devstr = NULL;
1160         int ret = 0;
1161         int mod = 0;
1162
1163         if (root->fs_info->sb->s_flags & MS_RDONLY)
1164                 return -EROFS;
1165
1166         if (!capable(CAP_SYS_ADMIN))
1167                 return -EPERM;
1168
1169         vol_args = memdup_user(arg, sizeof(*vol_args));
1170         if (IS_ERR(vol_args))
1171                 return PTR_ERR(vol_args);
1172
1173         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1174
1175         mutex_lock(&root->fs_info->volume_mutex);
1176         sizestr = vol_args->name;
1177         devstr = strchr(sizestr, ':');
1178         if (devstr) {
1179                 char *end;
1180                 sizestr = devstr + 1;
1181                 *devstr = '\0';
1182                 devstr = vol_args->name;
1183                 devid = simple_strtoull(devstr, &end, 10);
1184                 printk(KERN_INFO "resizing devid %llu\n",
1185                        (unsigned long long)devid);
1186         }
1187         device = btrfs_find_device(root, devid, NULL, NULL);
1188         if (!device) {
1189                 printk(KERN_INFO "resizer unable to find device %llu\n",
1190                        (unsigned long long)devid);
1191                 ret = -EINVAL;
1192                 goto out_unlock;
1193         }
1194         if (!strcmp(sizestr, "max"))
1195                 new_size = device->bdev->bd_inode->i_size;
1196         else {
1197                 if (sizestr[0] == '-') {
1198                         mod = -1;
1199                         sizestr++;
1200                 } else if (sizestr[0] == '+') {
1201                         mod = 1;
1202                         sizestr++;
1203                 }
1204                 new_size = memparse(sizestr, NULL);
1205                 if (new_size == 0) {
1206                         ret = -EINVAL;
1207                         goto out_unlock;
1208                 }
1209         }
1210
1211         old_size = device->total_bytes;
1212
1213         if (mod < 0) {
1214                 if (new_size > old_size) {
1215                         ret = -EINVAL;
1216                         goto out_unlock;
1217                 }
1218                 new_size = old_size - new_size;
1219         } else if (mod > 0) {
1220                 new_size = old_size + new_size;
1221         }
1222
1223         if (new_size < 256 * 1024 * 1024) {
1224                 ret = -EINVAL;
1225                 goto out_unlock;
1226         }
1227         if (new_size > device->bdev->bd_inode->i_size) {
1228                 ret = -EFBIG;
1229                 goto out_unlock;
1230         }
1231
1232         do_div(new_size, root->sectorsize);
1233         new_size *= root->sectorsize;
1234
1235         printk(KERN_INFO "new size for %s is %llu\n",
1236                 device->name, (unsigned long long)new_size);
1237
1238         if (new_size > old_size) {
1239                 trans = btrfs_start_transaction(root, 0);
1240                 if (IS_ERR(trans)) {
1241                         ret = PTR_ERR(trans);
1242                         goto out_unlock;
1243                 }
1244                 ret = btrfs_grow_device(trans, device, new_size);
1245                 btrfs_commit_transaction(trans, root);
1246         } else {
1247                 ret = btrfs_shrink_device(device, new_size);
1248         }
1249
1250 out_unlock:
1251         mutex_unlock(&root->fs_info->volume_mutex);
1252         kfree(vol_args);
1253         return ret;
1254 }
1255
1256 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1257                                                     char *name,
1258                                                     unsigned long fd,
1259                                                     int subvol,
1260                                                     u64 *transid,
1261                                                     bool readonly)
1262 {
1263         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1264         struct file *src_file;
1265         int namelen;
1266         int ret = 0;
1267
1268         if (root->fs_info->sb->s_flags & MS_RDONLY)
1269                 return -EROFS;
1270
1271         namelen = strlen(name);
1272         if (strchr(name, '/')) {
1273                 ret = -EINVAL;
1274                 goto out;
1275         }
1276
1277         if (subvol) {
1278                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1279                                      NULL, transid, readonly);
1280         } else {
1281                 struct inode *src_inode;
1282                 src_file = fget(fd);
1283                 if (!src_file) {
1284                         ret = -EINVAL;
1285                         goto out;
1286                 }
1287
1288                 src_inode = src_file->f_path.dentry->d_inode;
1289                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1290                         printk(KERN_INFO "btrfs: Snapshot src from "
1291                                "another FS\n");
1292                         ret = -EINVAL;
1293                         fput(src_file);
1294                         goto out;
1295                 }
1296                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1297                                      BTRFS_I(src_inode)->root,
1298                                      transid, readonly);
1299                 fput(src_file);
1300         }
1301 out:
1302         return ret;
1303 }
1304
1305 static noinline int btrfs_ioctl_snap_create(struct file *file,
1306                                             void __user *arg, int subvol)
1307 {
1308         struct btrfs_ioctl_vol_args *vol_args;
1309         int ret;
1310
1311         vol_args = memdup_user(arg, sizeof(*vol_args));
1312         if (IS_ERR(vol_args))
1313                 return PTR_ERR(vol_args);
1314         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1315
1316         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1317                                               vol_args->fd, subvol,
1318                                               NULL, false);
1319
1320         kfree(vol_args);
1321         return ret;
1322 }
1323
1324 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1325                                                void __user *arg, int subvol)
1326 {
1327         struct btrfs_ioctl_vol_args_v2 *vol_args;
1328         int ret;
1329         u64 transid = 0;
1330         u64 *ptr = NULL;
1331         bool readonly = false;
1332
1333         vol_args = memdup_user(arg, sizeof(*vol_args));
1334         if (IS_ERR(vol_args))
1335                 return PTR_ERR(vol_args);
1336         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1337
1338         if (vol_args->flags &
1339             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1340                 ret = -EOPNOTSUPP;
1341                 goto out;
1342         }
1343
1344         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1345                 ptr = &transid;
1346         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1347                 readonly = true;
1348
1349         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1350                                               vol_args->fd, subvol,
1351                                               ptr, readonly);
1352
1353         if (ret == 0 && ptr &&
1354             copy_to_user(arg +
1355                          offsetof(struct btrfs_ioctl_vol_args_v2,
1356                                   transid), ptr, sizeof(*ptr)))
1357                 ret = -EFAULT;
1358 out:
1359         kfree(vol_args);
1360         return ret;
1361 }
1362
1363 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1364                                                 void __user *arg)
1365 {
1366         struct inode *inode = fdentry(file)->d_inode;
1367         struct btrfs_root *root = BTRFS_I(inode)->root;
1368         int ret = 0;
1369         u64 flags = 0;
1370
1371         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1372                 return -EINVAL;
1373
1374         down_read(&root->fs_info->subvol_sem);
1375         if (btrfs_root_readonly(root))
1376                 flags |= BTRFS_SUBVOL_RDONLY;
1377         up_read(&root->fs_info->subvol_sem);
1378
1379         if (copy_to_user(arg, &flags, sizeof(flags)))
1380                 ret = -EFAULT;
1381
1382         return ret;
1383 }
1384
1385 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1386                                               void __user *arg)
1387 {
1388         struct inode *inode = fdentry(file)->d_inode;
1389         struct btrfs_root *root = BTRFS_I(inode)->root;
1390         struct btrfs_trans_handle *trans;
1391         u64 root_flags;
1392         u64 flags;
1393         int ret = 0;
1394
1395         if (root->fs_info->sb->s_flags & MS_RDONLY)
1396                 return -EROFS;
1397
1398         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1399                 return -EINVAL;
1400
1401         if (copy_from_user(&flags, arg, sizeof(flags)))
1402                 return -EFAULT;
1403
1404         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1405                 return -EINVAL;
1406
1407         if (flags & ~BTRFS_SUBVOL_RDONLY)
1408                 return -EOPNOTSUPP;
1409
1410         if (!inode_owner_or_capable(inode))
1411                 return -EACCES;
1412
1413         down_write(&root->fs_info->subvol_sem);
1414
1415         /* nothing to do */
1416         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1417                 goto out;
1418
1419         root_flags = btrfs_root_flags(&root->root_item);
1420         if (flags & BTRFS_SUBVOL_RDONLY)
1421                 btrfs_set_root_flags(&root->root_item,
1422                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1423         else
1424                 btrfs_set_root_flags(&root->root_item,
1425                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1426
1427         trans = btrfs_start_transaction(root, 1);
1428         if (IS_ERR(trans)) {
1429                 ret = PTR_ERR(trans);
1430                 goto out_reset;
1431         }
1432
1433         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1434                                 &root->root_key, &root->root_item);
1435
1436         btrfs_commit_transaction(trans, root);
1437 out_reset:
1438         if (ret)
1439                 btrfs_set_root_flags(&root->root_item, root_flags);
1440 out:
1441         up_write(&root->fs_info->subvol_sem);
1442         return ret;
1443 }
1444
1445 /*
1446  * helper to check if the subvolume references other subvolumes
1447  */
1448 static noinline int may_destroy_subvol(struct btrfs_root *root)
1449 {
1450         struct btrfs_path *path;
1451         struct btrfs_key key;
1452         int ret;
1453
1454         path = btrfs_alloc_path();
1455         if (!path)
1456                 return -ENOMEM;
1457
1458         key.objectid = root->root_key.objectid;
1459         key.type = BTRFS_ROOT_REF_KEY;
1460         key.offset = (u64)-1;
1461
1462         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1463                                 &key, path, 0, 0);
1464         if (ret < 0)
1465                 goto out;
1466         BUG_ON(ret == 0);
1467
1468         ret = 0;
1469         if (path->slots[0] > 0) {
1470                 path->slots[0]--;
1471                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1472                 if (key.objectid == root->root_key.objectid &&
1473                     key.type == BTRFS_ROOT_REF_KEY)
1474                         ret = -ENOTEMPTY;
1475         }
1476 out:
1477         btrfs_free_path(path);
1478         return ret;
1479 }
1480
1481 static noinline int key_in_sk(struct btrfs_key *key,
1482                               struct btrfs_ioctl_search_key *sk)
1483 {
1484         struct btrfs_key test;
1485         int ret;
1486
1487         test.objectid = sk->min_objectid;
1488         test.type = sk->min_type;
1489         test.offset = sk->min_offset;
1490
1491         ret = btrfs_comp_cpu_keys(key, &test);
1492         if (ret < 0)
1493                 return 0;
1494
1495         test.objectid = sk->max_objectid;
1496         test.type = sk->max_type;
1497         test.offset = sk->max_offset;
1498
1499         ret = btrfs_comp_cpu_keys(key, &test);
1500         if (ret > 0)
1501                 return 0;
1502         return 1;
1503 }
1504
1505 static noinline int copy_to_sk(struct btrfs_root *root,
1506                                struct btrfs_path *path,
1507                                struct btrfs_key *key,
1508                                struct btrfs_ioctl_search_key *sk,
1509                                char *buf,
1510                                unsigned long *sk_offset,
1511                                int *num_found)
1512 {
1513         u64 found_transid;
1514         struct extent_buffer *leaf;
1515         struct btrfs_ioctl_search_header sh;
1516         unsigned long item_off;
1517         unsigned long item_len;
1518         int nritems;
1519         int i;
1520         int slot;
1521         int ret = 0;
1522
1523         leaf = path->nodes[0];
1524         slot = path->slots[0];
1525         nritems = btrfs_header_nritems(leaf);
1526
1527         if (btrfs_header_generation(leaf) > sk->max_transid) {
1528                 i = nritems;
1529                 goto advance_key;
1530         }
1531         found_transid = btrfs_header_generation(leaf);
1532
1533         for (i = slot; i < nritems; i++) {
1534                 item_off = btrfs_item_ptr_offset(leaf, i);
1535                 item_len = btrfs_item_size_nr(leaf, i);
1536
1537                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1538                         item_len = 0;
1539
1540                 if (sizeof(sh) + item_len + *sk_offset >
1541                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1542                         ret = 1;
1543                         goto overflow;
1544                 }
1545
1546                 btrfs_item_key_to_cpu(leaf, key, i);
1547                 if (!key_in_sk(key, sk))
1548                         continue;
1549
1550                 sh.objectid = key->objectid;
1551                 sh.offset = key->offset;
1552                 sh.type = key->type;
1553                 sh.len = item_len;
1554                 sh.transid = found_transid;
1555
1556                 /* copy search result header */
1557                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1558                 *sk_offset += sizeof(sh);
1559
1560                 if (item_len) {
1561                         char *p = buf + *sk_offset;
1562                         /* copy the item */
1563                         read_extent_buffer(leaf, p,
1564                                            item_off, item_len);
1565                         *sk_offset += item_len;
1566                 }
1567                 (*num_found)++;
1568
1569                 if (*num_found >= sk->nr_items)
1570                         break;
1571         }
1572 advance_key:
1573         ret = 0;
1574         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1575                 key->offset++;
1576         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1577                 key->offset = 0;
1578                 key->type++;
1579         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1580                 key->offset = 0;
1581                 key->type = 0;
1582                 key->objectid++;
1583         } else
1584                 ret = 1;
1585 overflow:
1586         return ret;
1587 }
1588
1589 static noinline int search_ioctl(struct inode *inode,
1590                                  struct btrfs_ioctl_search_args *args)
1591 {
1592         struct btrfs_root *root;
1593         struct btrfs_key key;
1594         struct btrfs_key max_key;
1595         struct btrfs_path *path;
1596         struct btrfs_ioctl_search_key *sk = &args->key;
1597         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1598         int ret;
1599         int num_found = 0;
1600         unsigned long sk_offset = 0;
1601
1602         path = btrfs_alloc_path();
1603         if (!path)
1604                 return -ENOMEM;
1605
1606         if (sk->tree_id == 0) {
1607                 /* search the root of the inode that was passed */
1608                 root = BTRFS_I(inode)->root;
1609         } else {
1610                 key.objectid = sk->tree_id;
1611                 key.type = BTRFS_ROOT_ITEM_KEY;
1612                 key.offset = (u64)-1;
1613                 root = btrfs_read_fs_root_no_name(info, &key);
1614                 if (IS_ERR(root)) {
1615                         printk(KERN_ERR "could not find root %llu\n",
1616                                sk->tree_id);
1617                         btrfs_free_path(path);
1618                         return -ENOENT;
1619                 }
1620         }
1621
1622         key.objectid = sk->min_objectid;
1623         key.type = sk->min_type;
1624         key.offset = sk->min_offset;
1625
1626         max_key.objectid = sk->max_objectid;
1627         max_key.type = sk->max_type;
1628         max_key.offset = sk->max_offset;
1629
1630         path->keep_locks = 1;
1631
1632         while(1) {
1633                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1634                                            sk->min_transid);
1635                 if (ret != 0) {
1636                         if (ret > 0)
1637                                 ret = 0;
1638                         goto err;
1639                 }
1640                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1641                                  &sk_offset, &num_found);
1642                 btrfs_release_path(path);
1643                 if (ret || num_found >= sk->nr_items)
1644                         break;
1645
1646         }
1647         ret = 0;
1648 err:
1649         sk->nr_items = num_found;
1650         btrfs_free_path(path);
1651         return ret;
1652 }
1653
1654 static noinline int btrfs_ioctl_tree_search(struct file *file,
1655                                            void __user *argp)
1656 {
1657          struct btrfs_ioctl_search_args *args;
1658          struct inode *inode;
1659          int ret;
1660
1661         if (!capable(CAP_SYS_ADMIN))
1662                 return -EPERM;
1663
1664         args = memdup_user(argp, sizeof(*args));
1665         if (IS_ERR(args))
1666                 return PTR_ERR(args);
1667
1668         inode = fdentry(file)->d_inode;
1669         ret = search_ioctl(inode, args);
1670         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1671                 ret = -EFAULT;
1672         kfree(args);
1673         return ret;
1674 }
1675
1676 /*
1677  * Search INODE_REFs to identify path name of 'dirid' directory
1678  * in a 'tree_id' tree. and sets path name to 'name'.
1679  */
1680 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1681                                 u64 tree_id, u64 dirid, char *name)
1682 {
1683         struct btrfs_root *root;
1684         struct btrfs_key key;
1685         char *ptr;
1686         int ret = -1;
1687         int slot;
1688         int len;
1689         int total_len = 0;
1690         struct btrfs_inode_ref *iref;
1691         struct extent_buffer *l;
1692         struct btrfs_path *path;
1693
1694         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1695                 name[0]='\0';
1696                 return 0;
1697         }
1698
1699         path = btrfs_alloc_path();
1700         if (!path)
1701                 return -ENOMEM;
1702
1703         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1704
1705         key.objectid = tree_id;
1706         key.type = BTRFS_ROOT_ITEM_KEY;
1707         key.offset = (u64)-1;
1708         root = btrfs_read_fs_root_no_name(info, &key);
1709         if (IS_ERR(root)) {
1710                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1711                 ret = -ENOENT;
1712                 goto out;
1713         }
1714
1715         key.objectid = dirid;
1716         key.type = BTRFS_INODE_REF_KEY;
1717         key.offset = (u64)-1;
1718
1719         while(1) {
1720                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1721                 if (ret < 0)
1722                         goto out;
1723
1724                 l = path->nodes[0];
1725                 slot = path->slots[0];
1726                 if (ret > 0 && slot > 0)
1727                         slot--;
1728                 btrfs_item_key_to_cpu(l, &key, slot);
1729
1730                 if (ret > 0 && (key.objectid != dirid ||
1731                                 key.type != BTRFS_INODE_REF_KEY)) {
1732                         ret = -ENOENT;
1733                         goto out;
1734                 }
1735
1736                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1737                 len = btrfs_inode_ref_name_len(l, iref);
1738                 ptr -= len + 1;
1739                 total_len += len + 1;
1740                 if (ptr < name)
1741                         goto out;
1742
1743                 *(ptr + len) = '/';
1744                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1745
1746                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1747                         break;
1748
1749                 btrfs_release_path(path);
1750                 key.objectid = key.offset;
1751                 key.offset = (u64)-1;
1752                 dirid = key.objectid;
1753         }
1754         if (ptr < name)
1755                 goto out;
1756         memmove(name, ptr, total_len);
1757         name[total_len]='\0';
1758         ret = 0;
1759 out:
1760         btrfs_free_path(path);
1761         return ret;
1762 }
1763
1764 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1765                                            void __user *argp)
1766 {
1767          struct btrfs_ioctl_ino_lookup_args *args;
1768          struct inode *inode;
1769          int ret;
1770
1771         if (!capable(CAP_SYS_ADMIN))
1772                 return -EPERM;
1773
1774         args = memdup_user(argp, sizeof(*args));
1775         if (IS_ERR(args))
1776                 return PTR_ERR(args);
1777
1778         inode = fdentry(file)->d_inode;
1779
1780         if (args->treeid == 0)
1781                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1782
1783         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1784                                         args->treeid, args->objectid,
1785                                         args->name);
1786
1787         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1788                 ret = -EFAULT;
1789
1790         kfree(args);
1791         return ret;
1792 }
1793
1794 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1795                                              void __user *arg)
1796 {
1797         struct dentry *parent = fdentry(file);
1798         struct dentry *dentry;
1799         struct inode *dir = parent->d_inode;
1800         struct inode *inode;
1801         struct btrfs_root *root = BTRFS_I(dir)->root;
1802         struct btrfs_root *dest = NULL;
1803         struct btrfs_ioctl_vol_args *vol_args;
1804         struct btrfs_trans_handle *trans;
1805         int namelen;
1806         int ret;
1807         int err = 0;
1808
1809         vol_args = memdup_user(arg, sizeof(*vol_args));
1810         if (IS_ERR(vol_args))
1811                 return PTR_ERR(vol_args);
1812
1813         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1814         namelen = strlen(vol_args->name);
1815         if (strchr(vol_args->name, '/') ||
1816             strncmp(vol_args->name, "..", namelen) == 0) {
1817                 err = -EINVAL;
1818                 goto out;
1819         }
1820
1821         err = mnt_want_write(file->f_path.mnt);
1822         if (err)
1823                 goto out;
1824
1825         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1826         dentry = lookup_one_len(vol_args->name, parent, namelen);
1827         if (IS_ERR(dentry)) {
1828                 err = PTR_ERR(dentry);
1829                 goto out_unlock_dir;
1830         }
1831
1832         if (!dentry->d_inode) {
1833                 err = -ENOENT;
1834                 goto out_dput;
1835         }
1836
1837         inode = dentry->d_inode;
1838         dest = BTRFS_I(inode)->root;
1839         if (!capable(CAP_SYS_ADMIN)){
1840                 /*
1841                  * Regular user.  Only allow this with a special mount
1842                  * option, when the user has write+exec access to the
1843                  * subvol root, and when rmdir(2) would have been
1844                  * allowed.
1845                  *
1846                  * Note that this is _not_ check that the subvol is
1847                  * empty or doesn't contain data that we wouldn't
1848                  * otherwise be able to delete.
1849                  *
1850                  * Users who want to delete empty subvols should try
1851                  * rmdir(2).
1852                  */
1853                 err = -EPERM;
1854                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1855                         goto out_dput;
1856
1857                 /*
1858                  * Do not allow deletion if the parent dir is the same
1859                  * as the dir to be deleted.  That means the ioctl
1860                  * must be called on the dentry referencing the root
1861                  * of the subvol, not a random directory contained
1862                  * within it.
1863                  */
1864                 err = -EINVAL;
1865                 if (root == dest)
1866                         goto out_dput;
1867
1868                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1869                 if (err)
1870                         goto out_dput;
1871
1872                 /* check if subvolume may be deleted by a non-root user */
1873                 err = btrfs_may_delete(dir, dentry, 1);
1874                 if (err)
1875                         goto out_dput;
1876         }
1877
1878         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1879                 err = -EINVAL;
1880                 goto out_dput;
1881         }
1882
1883         mutex_lock(&inode->i_mutex);
1884         err = d_invalidate(dentry);
1885         if (err)
1886                 goto out_unlock;
1887
1888         down_write(&root->fs_info->subvol_sem);
1889
1890         err = may_destroy_subvol(dest);
1891         if (err)
1892                 goto out_up_write;
1893
1894         trans = btrfs_start_transaction(root, 0);
1895         if (IS_ERR(trans)) {
1896                 err = PTR_ERR(trans);
1897                 goto out_up_write;
1898         }
1899         trans->block_rsv = &root->fs_info->global_block_rsv;
1900
1901         ret = btrfs_unlink_subvol(trans, root, dir,
1902                                 dest->root_key.objectid,
1903                                 dentry->d_name.name,
1904                                 dentry->d_name.len);
1905         BUG_ON(ret);
1906
1907         btrfs_record_root_in_trans(trans, dest);
1908
1909         memset(&dest->root_item.drop_progress, 0,
1910                 sizeof(dest->root_item.drop_progress));
1911         dest->root_item.drop_level = 0;
1912         btrfs_set_root_refs(&dest->root_item, 0);
1913
1914         if (!xchg(&dest->orphan_item_inserted, 1)) {
1915                 ret = btrfs_insert_orphan_item(trans,
1916                                         root->fs_info->tree_root,
1917                                         dest->root_key.objectid);
1918                 BUG_ON(ret);
1919         }
1920
1921         ret = btrfs_end_transaction(trans, root);
1922         BUG_ON(ret);
1923         inode->i_flags |= S_DEAD;
1924 out_up_write:
1925         up_write(&root->fs_info->subvol_sem);
1926 out_unlock:
1927         mutex_unlock(&inode->i_mutex);
1928         if (!err) {
1929                 shrink_dcache_sb(root->fs_info->sb);
1930                 btrfs_invalidate_inodes(dest);
1931                 d_delete(dentry);
1932         }
1933 out_dput:
1934         dput(dentry);
1935 out_unlock_dir:
1936         mutex_unlock(&dir->i_mutex);
1937         mnt_drop_write(file->f_path.mnt);
1938 out:
1939         kfree(vol_args);
1940         return err;
1941 }
1942
1943 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1944 {
1945         struct inode *inode = fdentry(file)->d_inode;
1946         struct btrfs_root *root = BTRFS_I(inode)->root;
1947         struct btrfs_ioctl_defrag_range_args *range;
1948         int ret;
1949
1950         if (btrfs_root_readonly(root))
1951                 return -EROFS;
1952
1953         ret = mnt_want_write(file->f_path.mnt);
1954         if (ret)
1955                 return ret;
1956
1957         switch (inode->i_mode & S_IFMT) {
1958         case S_IFDIR:
1959                 if (!capable(CAP_SYS_ADMIN)) {
1960                         ret = -EPERM;
1961                         goto out;
1962                 }
1963                 ret = btrfs_defrag_root(root, 0);
1964                 if (ret)
1965                         goto out;
1966                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
1967                 break;
1968         case S_IFREG:
1969                 if (!(file->f_mode & FMODE_WRITE)) {
1970                         ret = -EINVAL;
1971                         goto out;
1972                 }
1973
1974                 range = kzalloc(sizeof(*range), GFP_KERNEL);
1975                 if (!range) {
1976                         ret = -ENOMEM;
1977                         goto out;
1978                 }
1979
1980                 if (argp) {
1981                         if (copy_from_user(range, argp,
1982                                            sizeof(*range))) {
1983                                 ret = -EFAULT;
1984                                 kfree(range);
1985                                 goto out;
1986                         }
1987                         /* compression requires us to start the IO */
1988                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1989                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
1990                                 range->extent_thresh = (u32)-1;
1991                         }
1992                 } else {
1993                         /* the rest are all set to zero by kzalloc */
1994                         range->len = (u64)-1;
1995                 }
1996                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
1997                                         range, 0, 0);
1998                 if (ret > 0)
1999                         ret = 0;
2000                 kfree(range);
2001                 break;
2002         default:
2003                 ret = -EINVAL;
2004         }
2005 out:
2006         mnt_drop_write(file->f_path.mnt);
2007         return ret;
2008 }
2009
2010 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2011 {
2012         struct btrfs_ioctl_vol_args *vol_args;
2013         int ret;
2014
2015         if (!capable(CAP_SYS_ADMIN))
2016                 return -EPERM;
2017
2018         vol_args = memdup_user(arg, sizeof(*vol_args));
2019         if (IS_ERR(vol_args))
2020                 return PTR_ERR(vol_args);
2021
2022         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2023         ret = btrfs_init_new_device(root, vol_args->name);
2024
2025         kfree(vol_args);
2026         return ret;
2027 }
2028
2029 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2030 {
2031         struct btrfs_ioctl_vol_args *vol_args;
2032         int ret;
2033
2034         if (!capable(CAP_SYS_ADMIN))
2035                 return -EPERM;
2036
2037         if (root->fs_info->sb->s_flags & MS_RDONLY)
2038                 return -EROFS;
2039
2040         vol_args = memdup_user(arg, sizeof(*vol_args));
2041         if (IS_ERR(vol_args))
2042                 return PTR_ERR(vol_args);
2043
2044         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2045         ret = btrfs_rm_device(root, vol_args->name);
2046
2047         kfree(vol_args);
2048         return ret;
2049 }
2050
2051 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2052 {
2053         struct btrfs_ioctl_fs_info_args *fi_args;
2054         struct btrfs_device *device;
2055         struct btrfs_device *next;
2056         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2057         int ret = 0;
2058
2059         if (!capable(CAP_SYS_ADMIN))
2060                 return -EPERM;
2061
2062         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2063         if (!fi_args)
2064                 return -ENOMEM;
2065
2066         fi_args->num_devices = fs_devices->num_devices;
2067         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2068
2069         mutex_lock(&fs_devices->device_list_mutex);
2070         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2071                 if (device->devid > fi_args->max_id)
2072                         fi_args->max_id = device->devid;
2073         }
2074         mutex_unlock(&fs_devices->device_list_mutex);
2075
2076         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2077                 ret = -EFAULT;
2078
2079         kfree(fi_args);
2080         return ret;
2081 }
2082
2083 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2084 {
2085         struct btrfs_ioctl_dev_info_args *di_args;
2086         struct btrfs_device *dev;
2087         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2088         int ret = 0;
2089         char *s_uuid = NULL;
2090         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2091
2092         if (!capable(CAP_SYS_ADMIN))
2093                 return -EPERM;
2094
2095         di_args = memdup_user(arg, sizeof(*di_args));
2096         if (IS_ERR(di_args))
2097                 return PTR_ERR(di_args);
2098
2099         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2100                 s_uuid = di_args->uuid;
2101
2102         mutex_lock(&fs_devices->device_list_mutex);
2103         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2104         mutex_unlock(&fs_devices->device_list_mutex);
2105
2106         if (!dev) {
2107                 ret = -ENODEV;
2108                 goto out;
2109         }
2110
2111         di_args->devid = dev->devid;
2112         di_args->bytes_used = dev->bytes_used;
2113         di_args->total_bytes = dev->total_bytes;
2114         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2115         strncpy(di_args->path, dev->name, sizeof(di_args->path));
2116
2117 out:
2118         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2119                 ret = -EFAULT;
2120
2121         kfree(di_args);
2122         return ret;
2123 }
2124
2125 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2126                                        u64 off, u64 olen, u64 destoff)
2127 {
2128         struct inode *inode = fdentry(file)->d_inode;
2129         struct btrfs_root *root = BTRFS_I(inode)->root;
2130         struct file *src_file;
2131         struct inode *src;
2132         struct btrfs_trans_handle *trans;
2133         struct btrfs_path *path;
2134         struct extent_buffer *leaf;
2135         char *buf;
2136         struct btrfs_key key;
2137         u32 nritems;
2138         int slot;
2139         int ret;
2140         u64 len = olen;
2141         u64 bs = root->fs_info->sb->s_blocksize;
2142         u64 hint_byte;
2143
2144         /*
2145          * TODO:
2146          * - split compressed inline extents.  annoying: we need to
2147          *   decompress into destination's address_space (the file offset
2148          *   may change, so source mapping won't do), then recompress (or
2149          *   otherwise reinsert) a subrange.
2150          * - allow ranges within the same file to be cloned (provided
2151          *   they don't overlap)?
2152          */
2153
2154         /* the destination must be opened for writing */
2155         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2156                 return -EINVAL;
2157
2158         if (btrfs_root_readonly(root))
2159                 return -EROFS;
2160
2161         ret = mnt_want_write(file->f_path.mnt);
2162         if (ret)
2163                 return ret;
2164
2165         src_file = fget(srcfd);
2166         if (!src_file) {
2167                 ret = -EBADF;
2168                 goto out_drop_write;
2169         }
2170
2171         src = src_file->f_dentry->d_inode;
2172
2173         ret = -EINVAL;
2174         if (src == inode)
2175                 goto out_fput;
2176
2177         /* the src must be open for reading */
2178         if (!(src_file->f_mode & FMODE_READ))
2179                 goto out_fput;
2180
2181         /* don't make the dst file partly checksummed */
2182         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2183             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2184                 goto out_fput;
2185
2186         ret = -EISDIR;
2187         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2188                 goto out_fput;
2189
2190         ret = -EXDEV;
2191         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2192                 goto out_fput;
2193
2194         ret = -ENOMEM;
2195         buf = vmalloc(btrfs_level_size(root, 0));
2196         if (!buf)
2197                 goto out_fput;
2198
2199         path = btrfs_alloc_path();
2200         if (!path) {
2201                 vfree(buf);
2202                 goto out_fput;
2203         }
2204         path->reada = 2;
2205
2206         if (inode < src) {
2207                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2208                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2209         } else {
2210                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2211                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2212         }
2213
2214         /* determine range to clone */
2215         ret = -EINVAL;
2216         if (off + len > src->i_size || off + len < off)
2217                 goto out_unlock;
2218         if (len == 0)
2219                 olen = len = src->i_size - off;
2220         /* if we extend to eof, continue to block boundary */
2221         if (off + len == src->i_size)
2222                 len = ALIGN(src->i_size, bs) - off;
2223
2224         /* verify the end result is block aligned */
2225         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2226             !IS_ALIGNED(destoff, bs))
2227                 goto out_unlock;
2228
2229         if (destoff > inode->i_size) {
2230                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2231                 if (ret)
2232                         goto out_unlock;
2233         }
2234
2235         /* truncate page cache pages from target inode range */
2236         truncate_inode_pages_range(&inode->i_data, destoff,
2237                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2238
2239         /* do any pending delalloc/csum calc on src, one way or
2240            another, and lock file content */
2241         while (1) {
2242                 struct btrfs_ordered_extent *ordered;
2243                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2244                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2245                 if (!ordered &&
2246                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2247                                    EXTENT_DELALLOC, 0, NULL))
2248                         break;
2249                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2250                 if (ordered)
2251                         btrfs_put_ordered_extent(ordered);
2252                 btrfs_wait_ordered_range(src, off, len);
2253         }
2254
2255         /* clone data */
2256         key.objectid = btrfs_ino(src);
2257         key.type = BTRFS_EXTENT_DATA_KEY;
2258         key.offset = 0;
2259
2260         while (1) {
2261                 /*
2262                  * note the key will change type as we walk through the
2263                  * tree.
2264                  */
2265                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2266                 if (ret < 0)
2267                         goto out;
2268
2269                 nritems = btrfs_header_nritems(path->nodes[0]);
2270                 if (path->slots[0] >= nritems) {
2271                         ret = btrfs_next_leaf(root, path);
2272                         if (ret < 0)
2273                                 goto out;
2274                         if (ret > 0)
2275                                 break;
2276                         nritems = btrfs_header_nritems(path->nodes[0]);
2277                 }
2278                 leaf = path->nodes[0];
2279                 slot = path->slots[0];
2280
2281                 btrfs_item_key_to_cpu(leaf, &key, slot);
2282                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2283                     key.objectid != btrfs_ino(src))
2284                         break;
2285
2286                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2287                         struct btrfs_file_extent_item *extent;
2288                         int type;
2289                         u32 size;
2290                         struct btrfs_key new_key;
2291                         u64 disko = 0, diskl = 0;
2292                         u64 datao = 0, datal = 0;
2293                         u8 comp;
2294                         u64 endoff;
2295
2296                         size = btrfs_item_size_nr(leaf, slot);
2297                         read_extent_buffer(leaf, buf,
2298                                            btrfs_item_ptr_offset(leaf, slot),
2299                                            size);
2300
2301                         extent = btrfs_item_ptr(leaf, slot,
2302                                                 struct btrfs_file_extent_item);
2303                         comp = btrfs_file_extent_compression(leaf, extent);
2304                         type = btrfs_file_extent_type(leaf, extent);
2305                         if (type == BTRFS_FILE_EXTENT_REG ||
2306                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2307                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2308                                                                       extent);
2309                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2310                                                                  extent);
2311                                 datao = btrfs_file_extent_offset(leaf, extent);
2312                                 datal = btrfs_file_extent_num_bytes(leaf,
2313                                                                     extent);
2314                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2315                                 /* take upper bound, may be compressed */
2316                                 datal = btrfs_file_extent_ram_bytes(leaf,
2317                                                                     extent);
2318                         }
2319                         btrfs_release_path(path);
2320
2321                         if (key.offset + datal <= off ||
2322                             key.offset >= off+len)
2323                                 goto next;
2324
2325                         memcpy(&new_key, &key, sizeof(new_key));
2326                         new_key.objectid = btrfs_ino(inode);
2327                         if (off <= key.offset)
2328                                 new_key.offset = key.offset + destoff - off;
2329                         else
2330                                 new_key.offset = destoff;
2331
2332                         /*
2333                          * 1 - adjusting old extent (we may have to split it)
2334                          * 1 - add new extent
2335                          * 1 - inode update
2336                          */
2337                         trans = btrfs_start_transaction(root, 3);
2338                         if (IS_ERR(trans)) {
2339                                 ret = PTR_ERR(trans);
2340                                 goto out;
2341                         }
2342
2343                         if (type == BTRFS_FILE_EXTENT_REG ||
2344                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2345                                 /*
2346                                  *    a  | --- range to clone ---|  b
2347                                  * | ------------- extent ------------- |
2348                                  */
2349
2350                                 /* substract range b */
2351                                 if (key.offset + datal > off + len)
2352                                         datal = off + len - key.offset;
2353
2354                                 /* substract range a */
2355                                 if (off > key.offset) {
2356                                         datao += off - key.offset;
2357                                         datal -= off - key.offset;
2358                                 }
2359
2360                                 ret = btrfs_drop_extents(trans, inode,
2361                                                          new_key.offset,
2362                                                          new_key.offset + datal,
2363                                                          &hint_byte, 1);
2364                                 BUG_ON(ret);
2365
2366                                 ret = btrfs_insert_empty_item(trans, root, path,
2367                                                               &new_key, size);
2368                                 BUG_ON(ret);
2369
2370                                 leaf = path->nodes[0];
2371                                 slot = path->slots[0];
2372                                 write_extent_buffer(leaf, buf,
2373                                             btrfs_item_ptr_offset(leaf, slot),
2374                                             size);
2375
2376                                 extent = btrfs_item_ptr(leaf, slot,
2377                                                 struct btrfs_file_extent_item);
2378
2379                                 /* disko == 0 means it's a hole */
2380                                 if (!disko)
2381                                         datao = 0;
2382
2383                                 btrfs_set_file_extent_offset(leaf, extent,
2384                                                              datao);
2385                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2386                                                                 datal);
2387                                 if (disko) {
2388                                         inode_add_bytes(inode, datal);
2389                                         ret = btrfs_inc_extent_ref(trans, root,
2390                                                         disko, diskl, 0,
2391                                                         root->root_key.objectid,
2392                                                         btrfs_ino(inode),
2393                                                         new_key.offset - datao);
2394                                         BUG_ON(ret);
2395                                 }
2396                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2397                                 u64 skip = 0;
2398                                 u64 trim = 0;
2399                                 if (off > key.offset) {
2400                                         skip = off - key.offset;
2401                                         new_key.offset += skip;
2402                                 }
2403
2404                                 if (key.offset + datal > off+len)
2405                                         trim = key.offset + datal - (off+len);
2406
2407                                 if (comp && (skip || trim)) {
2408                                         ret = -EINVAL;
2409                                         btrfs_end_transaction(trans, root);
2410                                         goto out;
2411                                 }
2412                                 size -= skip + trim;
2413                                 datal -= skip + trim;
2414
2415                                 ret = btrfs_drop_extents(trans, inode,
2416                                                          new_key.offset,
2417                                                          new_key.offset + datal,
2418                                                          &hint_byte, 1);
2419                                 BUG_ON(ret);
2420
2421                                 ret = btrfs_insert_empty_item(trans, root, path,
2422                                                               &new_key, size);
2423                                 BUG_ON(ret);
2424
2425                                 if (skip) {
2426                                         u32 start =
2427                                           btrfs_file_extent_calc_inline_size(0);
2428                                         memmove(buf+start, buf+start+skip,
2429                                                 datal);
2430                                 }
2431
2432                                 leaf = path->nodes[0];
2433                                 slot = path->slots[0];
2434                                 write_extent_buffer(leaf, buf,
2435                                             btrfs_item_ptr_offset(leaf, slot),
2436                                             size);
2437                                 inode_add_bytes(inode, datal);
2438                         }
2439
2440                         btrfs_mark_buffer_dirty(leaf);
2441                         btrfs_release_path(path);
2442
2443                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2444
2445                         /*
2446                          * we round up to the block size at eof when
2447                          * determining which extents to clone above,
2448                          * but shouldn't round up the file size
2449                          */
2450                         endoff = new_key.offset + datal;
2451                         if (endoff > destoff+olen)
2452                                 endoff = destoff+olen;
2453                         if (endoff > inode->i_size)
2454                                 btrfs_i_size_write(inode, endoff);
2455
2456                         ret = btrfs_update_inode(trans, root, inode);
2457                         BUG_ON(ret);
2458                         btrfs_end_transaction(trans, root);
2459                 }
2460 next:
2461                 btrfs_release_path(path);
2462                 key.offset++;
2463         }
2464         ret = 0;
2465 out:
2466         btrfs_release_path(path);
2467         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2468 out_unlock:
2469         mutex_unlock(&src->i_mutex);
2470         mutex_unlock(&inode->i_mutex);
2471         vfree(buf);
2472         btrfs_free_path(path);
2473 out_fput:
2474         fput(src_file);
2475 out_drop_write:
2476         mnt_drop_write(file->f_path.mnt);
2477         return ret;
2478 }
2479
2480 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2481 {
2482         struct btrfs_ioctl_clone_range_args args;
2483
2484         if (copy_from_user(&args, argp, sizeof(args)))
2485                 return -EFAULT;
2486         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2487                                  args.src_length, args.dest_offset);
2488 }
2489
2490 /*
2491  * there are many ways the trans_start and trans_end ioctls can lead
2492  * to deadlocks.  They should only be used by applications that
2493  * basically own the machine, and have a very in depth understanding
2494  * of all the possible deadlocks and enospc problems.
2495  */
2496 static long btrfs_ioctl_trans_start(struct file *file)
2497 {
2498         struct inode *inode = fdentry(file)->d_inode;
2499         struct btrfs_root *root = BTRFS_I(inode)->root;
2500         struct btrfs_trans_handle *trans;
2501         int ret;
2502
2503         ret = -EPERM;
2504         if (!capable(CAP_SYS_ADMIN))
2505                 goto out;
2506
2507         ret = -EINPROGRESS;
2508         if (file->private_data)
2509                 goto out;
2510
2511         ret = -EROFS;
2512         if (btrfs_root_readonly(root))
2513                 goto out;
2514
2515         ret = mnt_want_write(file->f_path.mnt);
2516         if (ret)
2517                 goto out;
2518
2519         atomic_inc(&root->fs_info->open_ioctl_trans);
2520
2521         ret = -ENOMEM;
2522         trans = btrfs_start_ioctl_transaction(root);
2523         if (IS_ERR(trans))
2524                 goto out_drop;
2525
2526         file->private_data = trans;
2527         return 0;
2528
2529 out_drop:
2530         atomic_dec(&root->fs_info->open_ioctl_trans);
2531         mnt_drop_write(file->f_path.mnt);
2532 out:
2533         return ret;
2534 }
2535
2536 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2537 {
2538         struct inode *inode = fdentry(file)->d_inode;
2539         struct btrfs_root *root = BTRFS_I(inode)->root;
2540         struct btrfs_root *new_root;
2541         struct btrfs_dir_item *di;
2542         struct btrfs_trans_handle *trans;
2543         struct btrfs_path *path;
2544         struct btrfs_key location;
2545         struct btrfs_disk_key disk_key;
2546         struct btrfs_super_block *disk_super;
2547         u64 features;
2548         u64 objectid = 0;
2549         u64 dir_id;
2550
2551         if (!capable(CAP_SYS_ADMIN))
2552                 return -EPERM;
2553
2554         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2555                 return -EFAULT;
2556
2557         if (!objectid)
2558                 objectid = root->root_key.objectid;
2559
2560         location.objectid = objectid;
2561         location.type = BTRFS_ROOT_ITEM_KEY;
2562         location.offset = (u64)-1;
2563
2564         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2565         if (IS_ERR(new_root))
2566                 return PTR_ERR(new_root);
2567
2568         if (btrfs_root_refs(&new_root->root_item) == 0)
2569                 return -ENOENT;
2570
2571         path = btrfs_alloc_path();
2572         if (!path)
2573                 return -ENOMEM;
2574         path->leave_spinning = 1;
2575
2576         trans = btrfs_start_transaction(root, 1);
2577         if (IS_ERR(trans)) {
2578                 btrfs_free_path(path);
2579                 return PTR_ERR(trans);
2580         }
2581
2582         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
2583         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2584                                    dir_id, "default", 7, 1);
2585         if (IS_ERR_OR_NULL(di)) {
2586                 btrfs_free_path(path);
2587                 btrfs_end_transaction(trans, root);
2588                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2589                        "this isn't going to work\n");
2590                 return -ENOENT;
2591         }
2592
2593         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2594         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2595         btrfs_mark_buffer_dirty(path->nodes[0]);
2596         btrfs_free_path(path);
2597
2598         disk_super = &root->fs_info->super_copy;
2599         features = btrfs_super_incompat_flags(disk_super);
2600         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2601                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2602                 btrfs_set_super_incompat_flags(disk_super, features);
2603         }
2604         btrfs_end_transaction(trans, root);
2605
2606         return 0;
2607 }
2608
2609 static void get_block_group_info(struct list_head *groups_list,
2610                                  struct btrfs_ioctl_space_info *space)
2611 {
2612         struct btrfs_block_group_cache *block_group;
2613
2614         space->total_bytes = 0;
2615         space->used_bytes = 0;
2616         space->flags = 0;
2617         list_for_each_entry(block_group, groups_list, list) {
2618                 space->flags = block_group->flags;
2619                 space->total_bytes += block_group->key.offset;
2620                 space->used_bytes +=
2621                         btrfs_block_group_used(&block_group->item);
2622         }
2623 }
2624
2625 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2626 {
2627         struct btrfs_ioctl_space_args space_args;
2628         struct btrfs_ioctl_space_info space;
2629         struct btrfs_ioctl_space_info *dest;
2630         struct btrfs_ioctl_space_info *dest_orig;
2631         struct btrfs_ioctl_space_info __user *user_dest;
2632         struct btrfs_space_info *info;
2633         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2634                        BTRFS_BLOCK_GROUP_SYSTEM,
2635                        BTRFS_BLOCK_GROUP_METADATA,
2636                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2637         int num_types = 4;
2638         int alloc_size;
2639         int ret = 0;
2640         u64 slot_count = 0;
2641         int i, c;
2642
2643         if (copy_from_user(&space_args,
2644                            (struct btrfs_ioctl_space_args __user *)arg,
2645                            sizeof(space_args)))
2646                 return -EFAULT;
2647
2648         for (i = 0; i < num_types; i++) {
2649                 struct btrfs_space_info *tmp;
2650
2651                 info = NULL;
2652                 rcu_read_lock();
2653                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2654                                         list) {
2655                         if (tmp->flags == types[i]) {
2656                                 info = tmp;
2657                                 break;
2658                         }
2659                 }
2660                 rcu_read_unlock();
2661
2662                 if (!info)
2663                         continue;
2664
2665                 down_read(&info->groups_sem);
2666                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2667                         if (!list_empty(&info->block_groups[c]))
2668                                 slot_count++;
2669                 }
2670                 up_read(&info->groups_sem);
2671         }
2672
2673         /* space_slots == 0 means they are asking for a count */
2674         if (space_args.space_slots == 0) {
2675                 space_args.total_spaces = slot_count;
2676                 goto out;
2677         }
2678
2679         slot_count = min_t(u64, space_args.space_slots, slot_count);
2680
2681         alloc_size = sizeof(*dest) * slot_count;
2682
2683         /* we generally have at most 6 or so space infos, one for each raid
2684          * level.  So, a whole page should be more than enough for everyone
2685          */
2686         if (alloc_size > PAGE_CACHE_SIZE)
2687                 return -ENOMEM;
2688
2689         space_args.total_spaces = 0;
2690         dest = kmalloc(alloc_size, GFP_NOFS);
2691         if (!dest)
2692                 return -ENOMEM;
2693         dest_orig = dest;
2694
2695         /* now we have a buffer to copy into */
2696         for (i = 0; i < num_types; i++) {
2697                 struct btrfs_space_info *tmp;
2698
2699                 if (!slot_count)
2700                         break;
2701
2702                 info = NULL;
2703                 rcu_read_lock();
2704                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2705                                         list) {
2706                         if (tmp->flags == types[i]) {
2707                                 info = tmp;
2708                                 break;
2709                         }
2710                 }
2711                 rcu_read_unlock();
2712
2713                 if (!info)
2714                         continue;
2715                 down_read(&info->groups_sem);
2716                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2717                         if (!list_empty(&info->block_groups[c])) {
2718                                 get_block_group_info(&info->block_groups[c],
2719                                                      &space);
2720                                 memcpy(dest, &space, sizeof(space));
2721                                 dest++;
2722                                 space_args.total_spaces++;
2723                                 slot_count--;
2724                         }
2725                         if (!slot_count)
2726                                 break;
2727                 }
2728                 up_read(&info->groups_sem);
2729         }
2730
2731         user_dest = (struct btrfs_ioctl_space_info *)
2732                 (arg + sizeof(struct btrfs_ioctl_space_args));
2733
2734         if (copy_to_user(user_dest, dest_orig, alloc_size))
2735                 ret = -EFAULT;
2736
2737         kfree(dest_orig);
2738 out:
2739         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2740                 ret = -EFAULT;
2741
2742         return ret;
2743 }
2744
2745 /*
2746  * there are many ways the trans_start and trans_end ioctls can lead
2747  * to deadlocks.  They should only be used by applications that
2748  * basically own the machine, and have a very in depth understanding
2749  * of all the possible deadlocks and enospc problems.
2750  */
2751 long btrfs_ioctl_trans_end(struct file *file)
2752 {
2753         struct inode *inode = fdentry(file)->d_inode;
2754         struct btrfs_root *root = BTRFS_I(inode)->root;
2755         struct btrfs_trans_handle *trans;
2756
2757         trans = file->private_data;
2758         if (!trans)
2759                 return -EINVAL;
2760         file->private_data = NULL;
2761
2762         btrfs_end_transaction(trans, root);
2763
2764         atomic_dec(&root->fs_info->open_ioctl_trans);
2765
2766         mnt_drop_write(file->f_path.mnt);
2767         return 0;
2768 }
2769
2770 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2771 {
2772         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2773         struct btrfs_trans_handle *trans;
2774         u64 transid;
2775         int ret;
2776
2777         trans = btrfs_start_transaction(root, 0);
2778         if (IS_ERR(trans))
2779                 return PTR_ERR(trans);
2780         transid = trans->transid;
2781         ret = btrfs_commit_transaction_async(trans, root, 0);
2782         if (ret) {
2783                 btrfs_end_transaction(trans, root);
2784                 return ret;
2785         }
2786
2787         if (argp)
2788                 if (copy_to_user(argp, &transid, sizeof(transid)))
2789                         return -EFAULT;
2790         return 0;
2791 }
2792
2793 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2794 {
2795         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2796         u64 transid;
2797
2798         if (argp) {
2799                 if (copy_from_user(&transid, argp, sizeof(transid)))
2800                         return -EFAULT;
2801         } else {
2802                 transid = 0;  /* current trans */
2803         }
2804         return btrfs_wait_for_commit(root, transid);
2805 }
2806
2807 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
2808 {
2809         int ret;
2810         struct btrfs_ioctl_scrub_args *sa;
2811
2812         if (!capable(CAP_SYS_ADMIN))
2813                 return -EPERM;
2814
2815         sa = memdup_user(arg, sizeof(*sa));
2816         if (IS_ERR(sa))
2817                 return PTR_ERR(sa);
2818
2819         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
2820                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
2821
2822         if (copy_to_user(arg, sa, sizeof(*sa)))
2823                 ret = -EFAULT;
2824
2825         kfree(sa);
2826         return ret;
2827 }
2828
2829 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
2830 {
2831         if (!capable(CAP_SYS_ADMIN))
2832                 return -EPERM;
2833
2834         return btrfs_scrub_cancel(root);
2835 }
2836
2837 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
2838                                        void __user *arg)
2839 {
2840         struct btrfs_ioctl_scrub_args *sa;
2841         int ret;
2842
2843         if (!capable(CAP_SYS_ADMIN))
2844                 return -EPERM;
2845
2846         sa = memdup_user(arg, sizeof(*sa));
2847         if (IS_ERR(sa))
2848                 return PTR_ERR(sa);
2849
2850         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
2851
2852         if (copy_to_user(arg, sa, sizeof(*sa)))
2853                 ret = -EFAULT;
2854
2855         kfree(sa);
2856         return ret;
2857 }
2858
2859 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
2860 {
2861         int ret = 0;
2862         int i;
2863         unsigned long rel_ptr;
2864         int size;
2865         struct btrfs_ioctl_ino_path_args *ipa;
2866         struct inode_fs_paths *ipath = NULL;
2867         struct btrfs_path *path;
2868
2869         if (!capable(CAP_SYS_ADMIN))
2870                 return -EPERM;
2871
2872         path = btrfs_alloc_path();
2873         if (!path) {
2874                 ret = -ENOMEM;
2875                 goto out;
2876         }
2877
2878         ipa = memdup_user(arg, sizeof(*ipa));
2879         if (IS_ERR(ipa)) {
2880                 ret = PTR_ERR(ipa);
2881                 ipa = NULL;
2882                 goto out;
2883         }
2884
2885         size = min_t(u32, ipa->size, 4096);
2886         ipath = init_ipath(size, root, path);
2887         if (IS_ERR(ipath)) {
2888                 ret = PTR_ERR(ipath);
2889                 ipath = NULL;
2890                 goto out;
2891         }
2892
2893         ret = paths_from_inode(ipa->inum, ipath);
2894         if (ret < 0)
2895                 goto out;
2896
2897         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
2898                 rel_ptr = ipath->fspath->str[i] - (char *)ipath->fspath->str;
2899                 ipath->fspath->str[i] = (void *)rel_ptr;
2900         }
2901
2902         ret = copy_to_user(ipa->fspath, ipath->fspath, size);
2903         if (ret) {
2904                 ret = -EFAULT;
2905                 goto out;
2906         }
2907
2908 out:
2909         btrfs_free_path(path);
2910         free_ipath(ipath);
2911         kfree(ipa);
2912
2913         return ret;
2914 }
2915
2916 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
2917 {
2918         struct btrfs_data_container *inodes = ctx;
2919         const size_t c = 3 * sizeof(u64);
2920
2921         if (inodes->bytes_left >= c) {
2922                 inodes->bytes_left -= c;
2923                 inodes->val[inodes->elem_cnt] = inum;
2924                 inodes->val[inodes->elem_cnt + 1] = offset;
2925                 inodes->val[inodes->elem_cnt + 2] = root;
2926                 inodes->elem_cnt += 3;
2927         } else {
2928                 inodes->bytes_missing += c - inodes->bytes_left;
2929                 inodes->bytes_left = 0;
2930                 inodes->elem_missed += 3;
2931         }
2932
2933         return 0;
2934 }
2935
2936 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
2937                                         void __user *arg)
2938 {
2939         int ret = 0;
2940         int size;
2941         u64 extent_offset;
2942         struct btrfs_ioctl_logical_ino_args *loi;
2943         struct btrfs_data_container *inodes = NULL;
2944         struct btrfs_path *path = NULL;
2945         struct btrfs_key key;
2946
2947         if (!capable(CAP_SYS_ADMIN))
2948                 return -EPERM;
2949
2950         loi = memdup_user(arg, sizeof(*loi));
2951         if (IS_ERR(loi)) {
2952                 ret = PTR_ERR(loi);
2953                 loi = NULL;
2954                 goto out;
2955         }
2956
2957         path = btrfs_alloc_path();
2958         if (!path) {
2959                 ret = -ENOMEM;
2960                 goto out;
2961         }
2962
2963         size = min_t(u32, loi->size, 4096);
2964         inodes = init_data_container(size);
2965         if (IS_ERR(inodes)) {
2966                 ret = PTR_ERR(inodes);
2967                 inodes = NULL;
2968                 goto out;
2969         }
2970
2971         ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
2972
2973         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2974                 ret = -ENOENT;
2975         if (ret < 0)
2976                 goto out;
2977
2978         extent_offset = loi->logical - key.objectid;
2979         ret = iterate_extent_inodes(root->fs_info, path, key.objectid,
2980                                         extent_offset, build_ino_list, inodes);
2981
2982         if (ret < 0)
2983                 goto out;
2984
2985         ret = copy_to_user(loi->inodes, inodes, size);
2986         if (ret)
2987                 ret = -EFAULT;
2988
2989 out:
2990         btrfs_free_path(path);
2991         kfree(inodes);
2992         kfree(loi);
2993
2994         return ret;
2995 }
2996
2997 long btrfs_ioctl(struct file *file, unsigned int
2998                 cmd, unsigned long arg)
2999 {
3000         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3001         void __user *argp = (void __user *)arg;
3002
3003         switch (cmd) {
3004         case FS_IOC_GETFLAGS:
3005                 return btrfs_ioctl_getflags(file, argp);
3006         case FS_IOC_SETFLAGS:
3007                 return btrfs_ioctl_setflags(file, argp);
3008         case FS_IOC_GETVERSION:
3009                 return btrfs_ioctl_getversion(file, argp);
3010         case FITRIM:
3011                 return btrfs_ioctl_fitrim(file, argp);
3012         case BTRFS_IOC_SNAP_CREATE:
3013                 return btrfs_ioctl_snap_create(file, argp, 0);
3014         case BTRFS_IOC_SNAP_CREATE_V2:
3015                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3016         case BTRFS_IOC_SUBVOL_CREATE:
3017                 return btrfs_ioctl_snap_create(file, argp, 1);
3018         case BTRFS_IOC_SNAP_DESTROY:
3019                 return btrfs_ioctl_snap_destroy(file, argp);
3020         case BTRFS_IOC_SUBVOL_GETFLAGS:
3021                 return btrfs_ioctl_subvol_getflags(file, argp);
3022         case BTRFS_IOC_SUBVOL_SETFLAGS:
3023                 return btrfs_ioctl_subvol_setflags(file, argp);
3024         case BTRFS_IOC_DEFAULT_SUBVOL:
3025                 return btrfs_ioctl_default_subvol(file, argp);
3026         case BTRFS_IOC_DEFRAG:
3027                 return btrfs_ioctl_defrag(file, NULL);
3028         case BTRFS_IOC_DEFRAG_RANGE:
3029                 return btrfs_ioctl_defrag(file, argp);
3030         case BTRFS_IOC_RESIZE:
3031                 return btrfs_ioctl_resize(root, argp);
3032         case BTRFS_IOC_ADD_DEV:
3033                 return btrfs_ioctl_add_dev(root, argp);
3034         case BTRFS_IOC_RM_DEV:
3035                 return btrfs_ioctl_rm_dev(root, argp);
3036         case BTRFS_IOC_FS_INFO:
3037                 return btrfs_ioctl_fs_info(root, argp);
3038         case BTRFS_IOC_DEV_INFO:
3039                 return btrfs_ioctl_dev_info(root, argp);
3040         case BTRFS_IOC_BALANCE:
3041                 return btrfs_balance(root->fs_info->dev_root);
3042         case BTRFS_IOC_CLONE:
3043                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3044         case BTRFS_IOC_CLONE_RANGE:
3045                 return btrfs_ioctl_clone_range(file, argp);
3046         case BTRFS_IOC_TRANS_START:
3047                 return btrfs_ioctl_trans_start(file);
3048         case BTRFS_IOC_TRANS_END:
3049                 return btrfs_ioctl_trans_end(file);
3050         case BTRFS_IOC_TREE_SEARCH:
3051                 return btrfs_ioctl_tree_search(file, argp);
3052         case BTRFS_IOC_INO_LOOKUP:
3053                 return btrfs_ioctl_ino_lookup(file, argp);
3054         case BTRFS_IOC_INO_PATHS:
3055                 return btrfs_ioctl_ino_to_path(root, argp);
3056         case BTRFS_IOC_LOGICAL_INO:
3057                 return btrfs_ioctl_logical_to_ino(root, argp);
3058         case BTRFS_IOC_SPACE_INFO:
3059                 return btrfs_ioctl_space_info(root, argp);
3060         case BTRFS_IOC_SYNC:
3061                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3062                 return 0;
3063         case BTRFS_IOC_START_SYNC:
3064                 return btrfs_ioctl_start_sync(file, argp);
3065         case BTRFS_IOC_WAIT_SYNC:
3066                 return btrfs_ioctl_wait_sync(file, argp);
3067         case BTRFS_IOC_SCRUB:
3068                 return btrfs_ioctl_scrub(root, argp);
3069         case BTRFS_IOC_SCRUB_CANCEL:
3070                 return btrfs_ioctl_scrub_cancel(root, argp);
3071         case BTRFS_IOC_SCRUB_PROGRESS:
3072                 return btrfs_ioctl_scrub_progress(root, argp);
3073         }
3074
3075         return -ENOTTY;
3076 }