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