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