fs/btrfs/ioctl.c

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