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 <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61
  62 static int btrfs_clone(struct inode *src, struct inode *inode,
  63                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  64
  65 /* Mask out flags that are inappropriate for the given type of inode. */
  66 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  67 {
  68         if (S_ISDIR(mode))
  69                 return flags;
  70         else if (S_ISREG(mode))
  71                 return flags & ~FS_DIRSYNC_FL;
  72         else
  73                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  74 }
  75
  76 /*
  77  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  78  */
  79 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  80 {
  81         unsigned int iflags = 0;
  82
  83         if (flags & BTRFS_INODE_SYNC)
  84                 iflags |= FS_SYNC_FL;
  85         if (flags & BTRFS_INODE_IMMUTABLE)
  86                 iflags |= FS_IMMUTABLE_FL;
  87         if (flags & BTRFS_INODE_APPEND)
  88                 iflags |= FS_APPEND_FL;
  89         if (flags & BTRFS_INODE_NODUMP)
  90                 iflags |= FS_NODUMP_FL;
  91         if (flags & BTRFS_INODE_NOATIME)
  92                 iflags |= FS_NOATIME_FL;
  93         if (flags & BTRFS_INODE_DIRSYNC)
  94                 iflags |= FS_DIRSYNC_FL;
  95         if (flags & BTRFS_INODE_NODATACOW)
  96                 iflags |= FS_NOCOW_FL;
  97
  98         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
  99                 iflags |= FS_COMPR_FL;
 100         else if (flags & BTRFS_INODE_NOCOMPRESS)
 101                 iflags |= FS_NOCOMP_FL;
 102
 103         return iflags;
 104 }
 105
 106 /*
 107  * Update inode->i_flags based on the btrfs internal flags.
 108  */
 109 void btrfs_update_iflags(struct inode *inode)
 110 {
 111         struct btrfs_inode *ip = BTRFS_I(inode);
 112
 113         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 114
 115         if (ip->flags & BTRFS_INODE_SYNC)
 116                 inode->i_flags |= S_SYNC;
 117         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 118                 inode->i_flags |= S_IMMUTABLE;
 119         if (ip->flags & BTRFS_INODE_APPEND)
 120                 inode->i_flags |= S_APPEND;
 121         if (ip->flags & BTRFS_INODE_NOATIME)
 122                 inode->i_flags |= S_NOATIME;
 123         if (ip->flags & BTRFS_INODE_DIRSYNC)
 124                 inode->i_flags |= S_DIRSYNC;
 125 }
 126
 127 /*
 128  * Inherit flags from the parent inode.
 129  *
 130  * Currently only the compression flags and the cow flags are inherited.
 131  */
 132 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 133 {
 134         unsigned int flags;
 135
 136         if (!dir)
 137                 return;
 138
 139         flags = BTRFS_I(dir)->flags;
 140
 141         if (flags & BTRFS_INODE_NOCOMPRESS) {
 142                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 143                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 144         } else if (flags & BTRFS_INODE_COMPRESS) {
 145                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 146                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 147         }
 148
 149         if (flags & BTRFS_INODE_NODATACOW) {
 150                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 151                 if (S_ISREG(inode->i_mode))
 152                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 153         }
 154
 155         btrfs_update_iflags(inode);
 156 }
 157
 158 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 159 {
 160         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 161         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 162
 163         if (copy_to_user(arg, &flags, sizeof(flags)))
 164                 return -EFAULT;
 165         return 0;
 166 }
 167
 168 static int check_flags(unsigned int flags)
 169 {
 170         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 171                       FS_NOATIME_FL | FS_NODUMP_FL | \
 172                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 173                       FS_NOCOMP_FL | FS_COMPR_FL |
 174                       FS_NOCOW_FL))
 175                 return -EOPNOTSUPP;
 176
 177         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 178                 return -EINVAL;
 179
 180         return 0;
 181 }
 182
 183 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 184 {
 185         struct inode *inode = file_inode(file);
 186         struct btrfs_inode *ip = BTRFS_I(inode);
 187         struct btrfs_root *root = ip->root;
 188         struct btrfs_trans_handle *trans;
 189         unsigned int flags, oldflags;
 190         int ret;
 191         u64 ip_oldflags;
 192         unsigned int i_oldflags;
 193         umode_t mode;
 194
 195         if (!inode_owner_or_capable(inode))
 196                 return -EPERM;
 197
 198         if (btrfs_root_readonly(root))
 199                 return -EROFS;
 200
 201         if (copy_from_user(&flags, arg, sizeof(flags)))
 202                 return -EFAULT;
 203
 204         ret = check_flags(flags);
 205         if (ret)
 206                 return ret;
 207
 208         ret = mnt_want_write_file(file);
 209         if (ret)
 210                 return ret;
 211
 212         mutex_lock(&inode->i_mutex);
 213
 214         ip_oldflags = ip->flags;
 215         i_oldflags = inode->i_flags;
 216         mode = inode->i_mode;
 217
 218         flags = btrfs_mask_flags(inode->i_mode, flags);
 219         oldflags = btrfs_flags_to_ioctl(ip->flags);
 220         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 221                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 222                         ret = -EPERM;
 223                         goto out_unlock;
 224                 }
 225         }
 226
 227         if (flags & FS_SYNC_FL)
 228                 ip->flags |= BTRFS_INODE_SYNC;
 229         else
 230                 ip->flags &= ~BTRFS_INODE_SYNC;
 231         if (flags & FS_IMMUTABLE_FL)
 232                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 233         else
 234                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 235         if (flags & FS_APPEND_FL)
 236                 ip->flags |= BTRFS_INODE_APPEND;
 237         else
 238                 ip->flags &= ~BTRFS_INODE_APPEND;
 239         if (flags & FS_NODUMP_FL)
 240                 ip->flags |= BTRFS_INODE_NODUMP;
 241         else
 242                 ip->flags &= ~BTRFS_INODE_NODUMP;
 243         if (flags & FS_NOATIME_FL)
 244                 ip->flags |= BTRFS_INODE_NOATIME;
 245         else
 246                 ip->flags &= ~BTRFS_INODE_NOATIME;
 247         if (flags & FS_DIRSYNC_FL)
 248                 ip->flags |= BTRFS_INODE_DIRSYNC;
 249         else
 250                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 251         if (flags & FS_NOCOW_FL) {
 252                 if (S_ISREG(mode)) {
 253                         /*
 254                          * It's safe to turn csums off here, no extents exist.
 255                          * Otherwise we want the flag to reflect the real COW
 256                          * status of the file and will not set it.
 257                          */
 258                         if (inode->i_size == 0)
 259                                 ip->flags |= BTRFS_INODE_NODATACOW
 260                                            | BTRFS_INODE_NODATASUM;
 261                 } else {
 262                         ip->flags |= BTRFS_INODE_NODATACOW;
 263                 }
 264         } else {
 265                 /*
 266                  * Revert back under same assuptions as above
 267                  */
 268                 if (S_ISREG(mode)) {
 269                         if (inode->i_size == 0)
 270                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 271                                              | BTRFS_INODE_NODATASUM);
 272                 } else {
 273                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 274                 }
 275         }
 276
 277         /*
 278          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 279          * flag may be changed automatically if compression code won't make
 280          * things smaller.
 281          */
 282         if (flags & FS_NOCOMP_FL) {
 283                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 284                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 285
 286                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 287                 if (ret && ret != -ENODATA)
 288                         goto out_drop;
 289         } else if (flags & FS_COMPR_FL) {
 290                 const char *comp;
 291
 292                 ip->flags |= BTRFS_INODE_COMPRESS;
 293                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 294
 295                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 296                         comp = "lzo";
 297                 else
 298                         comp = "zlib";
 299                 ret = btrfs_set_prop(inode, "btrfs.compression",
 300                                      comp, strlen(comp), 0);
 301                 if (ret)
 302                         goto out_drop;
 303
 304         } else {
 305                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 306         }
 307
 308         trans = btrfs_start_transaction(root, 1);
 309         if (IS_ERR(trans)) {
 310                 ret = PTR_ERR(trans);
 311                 goto out_drop;
 312         }
 313
 314         btrfs_update_iflags(inode);
 315         inode_inc_iversion(inode);
 316         inode->i_ctime = CURRENT_TIME;
 317         ret = btrfs_update_inode(trans, root, inode);
 318
 319         btrfs_end_transaction(trans, root);
 320  out_drop:
 321         if (ret) {
 322                 ip->flags = ip_oldflags;
 323                 inode->i_flags = i_oldflags;
 324         }
 325
 326  out_unlock:
 327         mutex_unlock(&inode->i_mutex);
 328         mnt_drop_write_file(file);
 329         return ret;
 330 }
 331
 332 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 333 {
 334         struct inode *inode = file_inode(file);
 335
 336         return put_user(inode->i_generation, arg);
 337 }
 338
 339 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 340 {
 341         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 342         struct btrfs_device *device;
 343         struct request_queue *q;
 344         struct fstrim_range range;
 345         u64 minlen = ULLONG_MAX;
 346         u64 num_devices = 0;
 347         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 348         int ret;
 349
 350         if (!capable(CAP_SYS_ADMIN))
 351                 return -EPERM;
 352
 353         rcu_read_lock();
 354         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 355                                 dev_list) {
 356                 if (!device->bdev)
 357                         continue;
 358                 q = bdev_get_queue(device->bdev);
 359                 if (blk_queue_discard(q)) {
 360                         num_devices++;
 361                         minlen = min((u64)q->limits.discard_granularity,
 362                                      minlen);
 363                 }
 364         }
 365         rcu_read_unlock();
 366
 367         if (!num_devices)
 368                 return -EOPNOTSUPP;
 369         if (copy_from_user(&range, arg, sizeof(range)))
 370                 return -EFAULT;
 371         if (range.start > total_bytes ||
 372             range.len < fs_info->sb->s_blocksize)
 373                 return -EINVAL;
 374
 375         range.len = min(range.len, total_bytes - range.start);
 376         range.minlen = max(range.minlen, minlen);
 377         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 378         if (ret < 0)
 379                 return ret;
 380
 381         if (copy_to_user(arg, &range, sizeof(range)))
 382                 return -EFAULT;
 383
 384         return 0;
 385 }
 386
 387 int btrfs_is_empty_uuid(u8 *uuid)
 388 {
 389         int i;
 390
 391         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 392                 if (uuid[i])
 393                         return 0;
 394         }
 395         return 1;
 396 }
 397
 398 static noinline int create_subvol(struct inode *dir,
 399                                   struct dentry *dentry,
 400                                   char *name, int namelen,
 401                                   u64 *async_transid,
 402                                   struct btrfs_qgroup_inherit *inherit)
 403 {
 404         struct btrfs_trans_handle *trans;
 405         struct btrfs_key key;
 406         struct btrfs_root_item root_item;
 407         struct btrfs_inode_item *inode_item;
 408         struct extent_buffer *leaf;
 409         struct btrfs_root *root = BTRFS_I(dir)->root;
 410         struct btrfs_root *new_root;
 411         struct btrfs_block_rsv block_rsv;
 412         struct timespec cur_time = CURRENT_TIME;
 413         struct inode *inode;
 414         int ret;
 415         int err;
 416         u64 objectid;
 417         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 418         u64 index = 0;
 419         u64 qgroup_reserved;
 420         uuid_le new_uuid;
 421
 422         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 423         if (ret)
 424                 return ret;
 425
 426         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 427         /*
 428          * The same as the snapshot creation, please see the comment
 429          * of create_snapshot().
 430          */
 431         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 432                                                8, &qgroup_reserved, false);
 433         if (ret)
 434                 return ret;
 435
 436         trans = btrfs_start_transaction(root, 0);
 437         if (IS_ERR(trans)) {
 438                 ret = PTR_ERR(trans);
 439                 btrfs_subvolume_release_metadata(root, &block_rsv,
 440                                                  qgroup_reserved);
 441                 return ret;
 442         }
 443         trans->block_rsv = &block_rsv;
 444         trans->bytes_reserved = block_rsv.size;
 445
 446         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 447         if (ret)
 448                 goto fail;
 449
 450         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 451                                       0, objectid, NULL, 0, 0, 0);
 452         if (IS_ERR(leaf)) {
 453                 ret = PTR_ERR(leaf);
 454                 goto fail;
 455         }
 456
 457         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 458         btrfs_set_header_bytenr(leaf, leaf->start);
 459         btrfs_set_header_generation(leaf, trans->transid);
 460         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 461         btrfs_set_header_owner(leaf, objectid);
 462
 463         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 464                             BTRFS_FSID_SIZE);
 465         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 466                             btrfs_header_chunk_tree_uuid(leaf),
 467                             BTRFS_UUID_SIZE);
 468         btrfs_mark_buffer_dirty(leaf);
 469
 470         memset(&root_item, 0, sizeof(root_item));
 471
 472         inode_item = &root_item.inode;
 473         btrfs_set_stack_inode_generation(inode_item, 1);
 474         btrfs_set_stack_inode_size(inode_item, 3);
 475         btrfs_set_stack_inode_nlink(inode_item, 1);
 476         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
 477         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 478
 479         btrfs_set_root_flags(&root_item, 0);
 480         btrfs_set_root_limit(&root_item, 0);
 481         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 482
 483         btrfs_set_root_bytenr(&root_item, leaf->start);
 484         btrfs_set_root_generation(&root_item, trans->transid);
 485         btrfs_set_root_level(&root_item, 0);
 486         btrfs_set_root_refs(&root_item, 1);
 487         btrfs_set_root_used(&root_item, leaf->len);
 488         btrfs_set_root_last_snapshot(&root_item, 0);
 489
 490         btrfs_set_root_generation_v2(&root_item,
 491                         btrfs_root_generation(&root_item));
 492         uuid_le_gen(&new_uuid);
 493         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 494         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 495         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 496         root_item.ctime = root_item.otime;
 497         btrfs_set_root_ctransid(&root_item, trans->transid);
 498         btrfs_set_root_otransid(&root_item, trans->transid);
 499
 500         btrfs_tree_unlock(leaf);
 501         free_extent_buffer(leaf);
 502         leaf = NULL;
 503
 504         btrfs_set_root_dirid(&root_item, new_dirid);
 505
 506         key.objectid = objectid;
 507         key.offset = 0;
 508         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 509         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 510                                 &root_item);
 511         if (ret)
 512                 goto fail;
 513
 514         key.offset = (u64)-1;
 515         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 516         if (IS_ERR(new_root)) {
 517                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 518                 ret = PTR_ERR(new_root);
 519                 goto fail;
 520         }
 521
 522         btrfs_record_root_in_trans(trans, new_root);
 523
 524         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 525         if (ret) {
 526                 /* We potentially lose an unused inode item here */
 527                 btrfs_abort_transaction(trans, root, ret);
 528                 goto fail;
 529         }
 530
 531         /*
 532          * insert the directory item
 533          */
 534         ret = btrfs_set_inode_index(dir, &index);
 535         if (ret) {
 536                 btrfs_abort_transaction(trans, root, ret);
 537                 goto fail;
 538         }
 539
 540         ret = btrfs_insert_dir_item(trans, root,
 541                                     name, namelen, dir, &key,
 542                                     BTRFS_FT_DIR, index);
 543         if (ret) {
 544                 btrfs_abort_transaction(trans, root, ret);
 545                 goto fail;
 546         }
 547
 548         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 549         ret = btrfs_update_inode(trans, root, dir);
 550         BUG_ON(ret);
 551
 552         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 553                                  objectid, root->root_key.objectid,
 554                                  btrfs_ino(dir), index, name, namelen);
 555         BUG_ON(ret);
 556
 557         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 558                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 559                                   objectid);
 560         if (ret)
 561                 btrfs_abort_transaction(trans, root, ret);
 562
 563 fail:
 564         trans->block_rsv = NULL;
 565         trans->bytes_reserved = 0;
 566         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 567
 568         if (async_transid) {
 569                 *async_transid = trans->transid;
 570                 err = btrfs_commit_transaction_async(trans, root, 1);
 571                 if (err)
 572                         err = btrfs_commit_transaction(trans, root);
 573         } else {
 574                 err = btrfs_commit_transaction(trans, root);
 575         }
 576         if (err && !ret)
 577                 ret = err;
 578
 579         if (!ret) {
 580                 inode = btrfs_lookup_dentry(dir, dentry);
 581                 if (IS_ERR(inode))
 582                         return PTR_ERR(inode);
 583                 d_instantiate(dentry, inode);
 584         }
 585         return ret;
 586 }
 587
 588 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 589                            struct dentry *dentry, char *name, int namelen,
 590                            u64 *async_transid, bool readonly,
 591                            struct btrfs_qgroup_inherit *inherit)
 592 {
 593         struct inode *inode;
 594         struct btrfs_pending_snapshot *pending_snapshot;
 595         struct btrfs_trans_handle *trans;
 596         int ret;
 597
 598         if (!root->ref_cows)
 599                 return -EINVAL;
 600
 601         ret = btrfs_start_delalloc_inodes(root, 0);
 602         if (ret)
 603                 return ret;
 604
 605         btrfs_wait_ordered_extents(root, -1);
 606
 607         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 608         if (!pending_snapshot)
 609                 return -ENOMEM;
 610
 611         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 612                              BTRFS_BLOCK_RSV_TEMP);
 613         /*
 614          * 1 - parent dir inode
 615          * 2 - dir entries
 616          * 1 - root item
 617          * 2 - root ref/backref
 618          * 1 - root of snapshot
 619          * 1 - UUID item
 620          */
 621         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 622                                         &pending_snapshot->block_rsv, 8,
 623                                         &pending_snapshot->qgroup_reserved,
 624                                         false);
 625         if (ret)
 626                 goto out;
 627
 628         pending_snapshot->dentry = dentry;
 629         pending_snapshot->root = root;
 630         pending_snapshot->readonly = readonly;
 631         pending_snapshot->dir = dir;
 632         pending_snapshot->inherit = inherit;
 633
 634         trans = btrfs_start_transaction(root, 0);
 635         if (IS_ERR(trans)) {
 636                 ret = PTR_ERR(trans);
 637                 goto fail;
 638         }
 639
 640         spin_lock(&root->fs_info->trans_lock);
 641         list_add(&pending_snapshot->list,
 642                  &trans->transaction->pending_snapshots);
 643         spin_unlock(&root->fs_info->trans_lock);
 644         if (async_transid) {
 645                 *async_transid = trans->transid;
 646                 ret = btrfs_commit_transaction_async(trans,
 647                                      root->fs_info->extent_root, 1);
 648                 if (ret)
 649                         ret = btrfs_commit_transaction(trans, root);
 650         } else {
 651                 ret = btrfs_commit_transaction(trans,
 652                                                root->fs_info->extent_root);
 653         }
 654         if (ret)
 655                 goto fail;
 656
 657         ret = pending_snapshot->error;
 658         if (ret)
 659                 goto fail;
 660
 661         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 662         if (ret)
 663                 goto fail;
 664
 665         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 666         if (IS_ERR(inode)) {
 667                 ret = PTR_ERR(inode);
 668                 goto fail;
 669         }
 670
 671         d_instantiate(dentry, inode);
 672         ret = 0;
 673 fail:
 674         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 675                                          &pending_snapshot->block_rsv,
 676                                          pending_snapshot->qgroup_reserved);
 677 out:
 678         kfree(pending_snapshot);
 679         return ret;
 680 }
 681
 682 /*  copy of check_sticky in fs/namei.c()
 683 * It's inline, so penalty for filesystems that don't use sticky bit is
 684 * minimal.
 685 */
 686 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 687 {
 688         kuid_t fsuid = current_fsuid();
 689
 690         if (!(dir->i_mode & S_ISVTX))
 691                 return 0;
 692         if (uid_eq(inode->i_uid, fsuid))
 693                 return 0;
 694         if (uid_eq(dir->i_uid, fsuid))
 695                 return 0;
 696         return !capable(CAP_FOWNER);
 697 }
 698
 699 /*  copy of may_delete in fs/namei.c()
 700  *      Check whether we can remove a link victim from directory dir, check
 701  *  whether the type of victim is right.
 702  *  1. We can't do it if dir is read-only (done in permission())
 703  *  2. We should have write and exec permissions on dir
 704  *  3. We can't remove anything from append-only dir
 705  *  4. We can't do anything with immutable dir (done in permission())
 706  *  5. If the sticky bit on dir is set we should either
 707  *      a. be owner of dir, or
 708  *      b. be owner of victim, or
 709  *      c. have CAP_FOWNER capability
 710  *  6. If the victim is append-only or immutable we can't do antyhing with
 711  *     links pointing to it.
 712  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 713  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 714  *  9. We can't remove a root or mountpoint.
 715  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 716  *     nfs_async_unlink().
 717  */
 718
 719 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 720 {
 721         int error;
 722
 723         if (!victim->d_inode)
 724                 return -ENOENT;
 725
 726         BUG_ON(victim->d_parent->d_inode != dir);
 727         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 728
 729         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 730         if (error)
 731                 return error;
 732         if (IS_APPEND(dir))
 733                 return -EPERM;
 734         if (btrfs_check_sticky(dir, victim->d_inode)||
 735                 IS_APPEND(victim->d_inode)||
 736             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 737                 return -EPERM;
 738         if (isdir) {
 739                 if (!S_ISDIR(victim->d_inode->i_mode))
 740                         return -ENOTDIR;
 741                 if (IS_ROOT(victim))
 742                         return -EBUSY;
 743         } else if (S_ISDIR(victim->d_inode->i_mode))
 744                 return -EISDIR;
 745         if (IS_DEADDIR(dir))
 746                 return -ENOENT;
 747         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 748                 return -EBUSY;
 749         return 0;
 750 }
 751
 752 /* copy of may_create in fs/namei.c() */
 753 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 754 {
 755         if (child->d_inode)
 756                 return -EEXIST;
 757         if (IS_DEADDIR(dir))
 758                 return -ENOENT;
 759         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 760 }
 761
 762 /*
 763  * Create a new subvolume below @parent.  This is largely modeled after
 764  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 765  * inside this filesystem so it's quite a bit simpler.
 766  */
 767 static noinline int btrfs_mksubvol(struct path *parent,
 768                                    char *name, int namelen,
 769                                    struct btrfs_root *snap_src,
 770                                    u64 *async_transid, bool readonly,
 771                                    struct btrfs_qgroup_inherit *inherit)
 772 {
 773         struct inode *dir  = parent->dentry->d_inode;
 774         struct dentry *dentry;
 775         int error;
 776
 777         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 778         if (error == -EINTR)
 779                 return error;
 780
 781         dentry = lookup_one_len(name, parent->dentry, namelen);
 782         error = PTR_ERR(dentry);
 783         if (IS_ERR(dentry))
 784                 goto out_unlock;
 785
 786         error = -EEXIST;
 787         if (dentry->d_inode)
 788                 goto out_dput;
 789
 790         error = btrfs_may_create(dir, dentry);
 791         if (error)
 792                 goto out_dput;
 793
 794         /*
 795          * even if this name doesn't exist, we may get hash collisions.
 796          * check for them now when we can safely fail
 797          */
 798         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 799                                                dir->i_ino, name,
 800                                                namelen);
 801         if (error)
 802                 goto out_dput;
 803
 804         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 805
 806         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 807                 goto out_up_read;
 808
 809         if (snap_src) {
 810                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 811                                         async_transid, readonly, inherit);
 812         } else {
 813                 error = create_subvol(dir, dentry, name, namelen,
 814                                       async_transid, inherit);
 815         }
 816         if (!error)
 817                 fsnotify_mkdir(dir, dentry);
 818 out_up_read:
 819         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 820 out_dput:
 821         dput(dentry);
 822 out_unlock:
 823         mutex_unlock(&dir->i_mutex);
 824         return error;
 825 }
 826
 827 /*
 828  * When we're defragging a range, we don't want to kick it off again
 829  * if it is really just waiting for delalloc to send it down.
 830  * If we find a nice big extent or delalloc range for the bytes in the
 831  * file you want to defrag, we return 0 to let you know to skip this
 832  * part of the file
 833  */
 834 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 835 {
 836         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 837         struct extent_map *em = NULL;
 838         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 839         u64 end;
 840
 841         read_lock(&em_tree->lock);
 842         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 843         read_unlock(&em_tree->lock);
 844
 845         if (em) {
 846                 end = extent_map_end(em);
 847                 free_extent_map(em);
 848                 if (end - offset > thresh)
 849                         return 0;
 850         }
 851         /* if we already have a nice delalloc here, just stop */
 852         thresh /= 2;
 853         end = count_range_bits(io_tree, &offset, offset + thresh,
 854                                thresh, EXTENT_DELALLOC, 1);
 855         if (end >= thresh)
 856                 return 0;
 857         return 1;
 858 }
 859
 860 /*
 861  * helper function to walk through a file and find extents
 862  * newer than a specific transid, and smaller than thresh.
 863  *
 864  * This is used by the defragging code to find new and small
 865  * extents
 866  */
 867 static int find_new_extents(struct btrfs_root *root,
 868                             struct inode *inode, u64 newer_than,
 869                             u64 *off, int thresh)
 870 {
 871         struct btrfs_path *path;
 872         struct btrfs_key min_key;
 873         struct extent_buffer *leaf;
 874         struct btrfs_file_extent_item *extent;
 875         int type;
 876         int ret;
 877         u64 ino = btrfs_ino(inode);
 878
 879         path = btrfs_alloc_path();
 880         if (!path)
 881                 return -ENOMEM;
 882
 883         min_key.objectid = ino;
 884         min_key.type = BTRFS_EXTENT_DATA_KEY;
 885         min_key.offset = *off;
 886
 887         path->keep_locks = 1;
 888
 889         while (1) {
 890                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 891                 if (ret != 0)
 892                         goto none;
 893                 if (min_key.objectid != ino)
 894                         goto none;
 895                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 896                         goto none;
 897
 898                 leaf = path->nodes[0];
 899                 extent = btrfs_item_ptr(leaf, path->slots[0],
 900                                         struct btrfs_file_extent_item);
 901
 902                 type = btrfs_file_extent_type(leaf, extent);
 903                 if (type == BTRFS_FILE_EXTENT_REG &&
 904                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 905                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 906                         *off = min_key.offset;
 907                         btrfs_free_path(path);
 908                         return 0;
 909                 }
 910
 911                 if (min_key.offset == (u64)-1)
 912                         goto none;
 913
 914                 min_key.offset++;
 915                 btrfs_release_path(path);
 916         }
 917 none:
 918         btrfs_free_path(path);
 919         return -ENOENT;
 920 }
 921
 922 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 923 {
 924         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 925         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 926         struct extent_map *em;
 927         u64 len = PAGE_CACHE_SIZE;
 928
 929         /*
 930          * hopefully we have this extent in the tree already, try without
 931          * the full extent lock
 932          */
 933         read_lock(&em_tree->lock);
 934         em = lookup_extent_mapping(em_tree, start, len);
 935         read_unlock(&em_tree->lock);
 936
 937         if (!em) {
 938                 /* get the big lock and read metadata off disk */
 939                 lock_extent(io_tree, start, start + len - 1);
 940                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 941                 unlock_extent(io_tree, start, start + len - 1);
 942
 943                 if (IS_ERR(em))
 944                         return NULL;
 945         }
 946
 947         return em;
 948 }
 949
 950 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 951 {
 952         struct extent_map *next;
 953         bool ret = true;
 954
 955         /* this is the last extent */
 956         if (em->start + em->len >= i_size_read(inode))
 957                 return false;
 958
 959         next = defrag_lookup_extent(inode, em->start + em->len);
 960         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 961                 ret = false;
 962
 963         free_extent_map(next);
 964         return ret;
 965 }
 966
 967 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
 968                                u64 *last_len, u64 *skip, u64 *defrag_end,
 969                                int compress)
 970 {
 971         struct extent_map *em;
 972         int ret = 1;
 973         bool next_mergeable = true;
 974
 975         /*
 976          * make sure that once we start defragging an extent, we keep on
 977          * defragging it
 978          */
 979         if (start < *defrag_end)
 980                 return 1;
 981
 982         *skip = 0;
 983
 984         em = defrag_lookup_extent(inode, start);
 985         if (!em)
 986                 return 0;
 987
 988         /* this will cover holes, and inline extents */
 989         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 990                 ret = 0;
 991                 goto out;
 992         }
 993
 994         next_mergeable = defrag_check_next_extent(inode, em);
 995
 996         /*
 997          * we hit a real extent, if it is big or the next extent is not a
 998          * real extent, don't bother defragging it
 999          */
1000         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1001             (em->len >= thresh || !next_mergeable))
1002                 ret = 0;
1003 out:
1004         /*
1005          * last_len ends up being a counter of how many bytes we've defragged.
1006          * every time we choose not to defrag an extent, we reset *last_len
1007          * so that the next tiny extent will force a defrag.
1008          *
1009          * The end result of this is that tiny extents before a single big
1010          * extent will force at least part of that big extent to be defragged.
1011          */
1012         if (ret) {
1013                 *defrag_end = extent_map_end(em);
1014         } else {
1015                 *last_len = 0;
1016                 *skip = extent_map_end(em);
1017                 *defrag_end = 0;
1018         }
1019
1020         free_extent_map(em);
1021         return ret;
1022 }
1023
1024 /*
1025  * it doesn't do much good to defrag one or two pages
1026  * at a time.  This pulls in a nice chunk of pages
1027  * to COW and defrag.
1028  *
1029  * It also makes sure the delalloc code has enough
1030  * dirty data to avoid making new small extents as part
1031  * of the defrag
1032  *
1033  * It's a good idea to start RA on this range
1034  * before calling this.
1035  */
1036 static int cluster_pages_for_defrag(struct inode *inode,
1037                                     struct page **pages,
1038                                     unsigned long start_index,
1039                                     unsigned long num_pages)
1040 {
1041         unsigned long file_end;
1042         u64 isize = i_size_read(inode);
1043         u64 page_start;
1044         u64 page_end;
1045         u64 page_cnt;
1046         int ret;
1047         int i;
1048         int i_done;
1049         struct btrfs_ordered_extent *ordered;
1050         struct extent_state *cached_state = NULL;
1051         struct extent_io_tree *tree;
1052         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1053
1054         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1055         if (!isize || start_index > file_end)
1056                 return 0;
1057
1058         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1059
1060         ret = btrfs_delalloc_reserve_space(inode,
1061                                            page_cnt << PAGE_CACHE_SHIFT);
1062         if (ret)
1063                 return ret;
1064         i_done = 0;
1065         tree = &BTRFS_I(inode)->io_tree;
1066
1067         /* step one, lock all the pages */
1068         for (i = 0; i < page_cnt; i++) {
1069                 struct page *page;
1070 again:
1071                 page = find_or_create_page(inode->i_mapping,
1072                                            start_index + i, mask);
1073                 if (!page)
1074                         break;
1075
1076                 page_start = page_offset(page);
1077                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1078                 while (1) {
1079                         lock_extent(tree, page_start, page_end);
1080                         ordered = btrfs_lookup_ordered_extent(inode,
1081                                                               page_start);
1082                         unlock_extent(tree, page_start, page_end);
1083                         if (!ordered)
1084                                 break;
1085
1086                         unlock_page(page);
1087                         btrfs_start_ordered_extent(inode, ordered, 1);
1088                         btrfs_put_ordered_extent(ordered);
1089                         lock_page(page);
1090                         /*
1091                          * we unlocked the page above, so we need check if
1092                          * it was released or not.
1093                          */
1094                         if (page->mapping != inode->i_mapping) {
1095                                 unlock_page(page);
1096                                 page_cache_release(page);
1097                                 goto again;
1098                         }
1099                 }
1100
1101                 if (!PageUptodate(page)) {
1102                         btrfs_readpage(NULL, page);
1103                         lock_page(page);
1104                         if (!PageUptodate(page)) {
1105                                 unlock_page(page);
1106                                 page_cache_release(page);
1107                                 ret = -EIO;
1108                                 break;
1109                         }
1110                 }
1111
1112                 if (page->mapping != inode->i_mapping) {
1113                         unlock_page(page);
1114                         page_cache_release(page);
1115                         goto again;
1116                 }
1117
1118                 pages[i] = page;
1119                 i_done++;
1120         }
1121         if (!i_done || ret)
1122                 goto out;
1123
1124         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1125                 goto out;
1126
1127         /*
1128          * so now we have a nice long stream of locked
1129          * and up to date pages, lets wait on them
1130          */
1131         for (i = 0; i < i_done; i++)
1132                 wait_on_page_writeback(pages[i]);
1133
1134         page_start = page_offset(pages[0]);
1135         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1136
1137         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1138                          page_start, page_end - 1, 0, &cached_state);
1139         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1140                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1141                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1142                           &cached_state, GFP_NOFS);
1143
1144         if (i_done != page_cnt) {
1145                 spin_lock(&BTRFS_I(inode)->lock);
1146                 BTRFS_I(inode)->outstanding_extents++;
1147                 spin_unlock(&BTRFS_I(inode)->lock);
1148                 btrfs_delalloc_release_space(inode,
1149                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1150         }
1151
1152
1153         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1154                           &cached_state, GFP_NOFS);
1155
1156         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1157                              page_start, page_end - 1, &cached_state,
1158                              GFP_NOFS);
1159
1160         for (i = 0; i < i_done; i++) {
1161                 clear_page_dirty_for_io(pages[i]);
1162                 ClearPageChecked(pages[i]);
1163                 set_page_extent_mapped(pages[i]);
1164                 set_page_dirty(pages[i]);
1165                 unlock_page(pages[i]);
1166                 page_cache_release(pages[i]);
1167         }
1168         return i_done;
1169 out:
1170         for (i = 0; i < i_done; i++) {
1171                 unlock_page(pages[i]);
1172                 page_cache_release(pages[i]);
1173         }
1174         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1175         return ret;
1176
1177 }
1178
1179 int btrfs_defrag_file(struct inode *inode, struct file *file,
1180                       struct btrfs_ioctl_defrag_range_args *range,
1181                       u64 newer_than, unsigned long max_to_defrag)
1182 {
1183         struct btrfs_root *root = BTRFS_I(inode)->root;
1184         struct file_ra_state *ra = NULL;
1185         unsigned long last_index;
1186         u64 isize = i_size_read(inode);
1187         u64 last_len = 0;
1188         u64 skip = 0;
1189         u64 defrag_end = 0;
1190         u64 newer_off = range->start;
1191         unsigned long i;
1192         unsigned long ra_index = 0;
1193         int ret;
1194         int defrag_count = 0;
1195         int compress_type = BTRFS_COMPRESS_ZLIB;
1196         int extent_thresh = range->extent_thresh;
1197         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1198         unsigned long cluster = max_cluster;
1199         u64 new_align = ~((u64)128 * 1024 - 1);
1200         struct page **pages = NULL;
1201
1202         if (isize == 0)
1203                 return 0;
1204
1205         if (range->start >= isize)
1206                 return -EINVAL;
1207
1208         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1209                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1210                         return -EINVAL;
1211                 if (range->compress_type)
1212                         compress_type = range->compress_type;
1213         }
1214
1215         if (extent_thresh == 0)
1216                 extent_thresh = 256 * 1024;
1217
1218         /*
1219          * if we were not given a file, allocate a readahead
1220          * context
1221          */
1222         if (!file) {
1223                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1224                 if (!ra)
1225                         return -ENOMEM;
1226                 file_ra_state_init(ra, inode->i_mapping);
1227         } else {
1228                 ra = &file->f_ra;
1229         }
1230
1231         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1232                         GFP_NOFS);
1233         if (!pages) {
1234                 ret = -ENOMEM;
1235                 goto out_ra;
1236         }
1237
1238         /* find the last page to defrag */
1239         if (range->start + range->len > range->start) {
1240                 last_index = min_t(u64, isize - 1,
1241                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1242         } else {
1243                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1244         }
1245
1246         if (newer_than) {
1247                 ret = find_new_extents(root, inode, newer_than,
1248                                        &newer_off, 64 * 1024);
1249                 if (!ret) {
1250                         range->start = newer_off;
1251                         /*
1252                          * we always align our defrag to help keep
1253                          * the extents in the file evenly spaced
1254                          */
1255                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1256                 } else
1257                         goto out_ra;
1258         } else {
1259                 i = range->start >> PAGE_CACHE_SHIFT;
1260         }
1261         if (!max_to_defrag)
1262                 max_to_defrag = last_index + 1;
1263
1264         /*
1265          * make writeback starts from i, so the defrag range can be
1266          * written sequentially.
1267          */
1268         if (i < inode->i_mapping->writeback_index)
1269                 inode->i_mapping->writeback_index = i;
1270
1271         while (i <= last_index && defrag_count < max_to_defrag &&
1272                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1273                 PAGE_CACHE_SHIFT)) {
1274                 /*
1275                  * make sure we stop running if someone unmounts
1276                  * the FS
1277                  */
1278                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1279                         break;
1280
1281                 if (btrfs_defrag_cancelled(root->fs_info)) {
1282                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1283                         ret = -EAGAIN;
1284                         break;
1285                 }
1286
1287                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1288                                          extent_thresh, &last_len, &skip,
1289                                          &defrag_end, range->flags &
1290                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1291                         unsigned long next;
1292                         /*
1293                          * the should_defrag function tells us how much to skip
1294                          * bump our counter by the suggested amount
1295                          */
1296                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1297                         i = max(i + 1, next);
1298                         continue;
1299                 }
1300
1301                 if (!newer_than) {
1302                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1303                                    PAGE_CACHE_SHIFT) - i;
1304                         cluster = min(cluster, max_cluster);
1305                 } else {
1306                         cluster = max_cluster;
1307                 }
1308
1309                 if (i + cluster > ra_index) {
1310                         ra_index = max(i, ra_index);
1311                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1312                                        cluster);
1313                         ra_index += max_cluster;
1314                 }
1315
1316                 mutex_lock(&inode->i_mutex);
1317                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1318                         BTRFS_I(inode)->force_compress = compress_type;
1319                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1320                 if (ret < 0) {
1321                         mutex_unlock(&inode->i_mutex);
1322                         goto out_ra;
1323                 }
1324
1325                 defrag_count += ret;
1326                 balance_dirty_pages_ratelimited(inode->i_mapping);
1327                 mutex_unlock(&inode->i_mutex);
1328
1329                 if (newer_than) {
1330                         if (newer_off == (u64)-1)
1331                                 break;
1332
1333                         if (ret > 0)
1334                                 i += ret;
1335
1336                         newer_off = max(newer_off + 1,
1337                                         (u64)i << PAGE_CACHE_SHIFT);
1338
1339                         ret = find_new_extents(root, inode,
1340                                                newer_than, &newer_off,
1341                                                64 * 1024);
1342                         if (!ret) {
1343                                 range->start = newer_off;
1344                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1345                         } else {
1346                                 break;
1347                         }
1348                 } else {
1349                         if (ret > 0) {
1350                                 i += ret;
1351                                 last_len += ret << PAGE_CACHE_SHIFT;
1352                         } else {
1353                                 i++;
1354                                 last_len = 0;
1355                         }
1356                 }
1357         }
1358
1359         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1360                 filemap_flush(inode->i_mapping);
1361
1362         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1363                 /* the filemap_flush will queue IO into the worker threads, but
1364                  * we have to make sure the IO is actually started and that
1365                  * ordered extents get created before we return
1366                  */
1367                 atomic_inc(&root->fs_info->async_submit_draining);
1368                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1369                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1370                         wait_event(root->fs_info->async_submit_wait,
1371                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1372                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1373                 }
1374                 atomic_dec(&root->fs_info->async_submit_draining);
1375         }
1376
1377         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1378                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1379         }
1380
1381         ret = defrag_count;
1382
1383 out_ra:
1384         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1385                 mutex_lock(&inode->i_mutex);
1386                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1387                 mutex_unlock(&inode->i_mutex);
1388         }
1389         if (!file)
1390                 kfree(ra);
1391         kfree(pages);
1392         return ret;
1393 }
1394
1395 static noinline int btrfs_ioctl_resize(struct file *file,
1396                                         void __user *arg)
1397 {
1398         u64 new_size;
1399         u64 old_size;
1400         u64 devid = 1;
1401         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1402         struct btrfs_ioctl_vol_args *vol_args;
1403         struct btrfs_trans_handle *trans;
1404         struct btrfs_device *device = NULL;
1405         char *sizestr;
1406         char *devstr = NULL;
1407         int ret = 0;
1408         int mod = 0;
1409
1410         if (!capable(CAP_SYS_ADMIN))
1411                 return -EPERM;
1412
1413         ret = mnt_want_write_file(file);
1414         if (ret)
1415                 return ret;
1416
1417         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1418                         1)) {
1419                 mnt_drop_write_file(file);
1420                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1421         }
1422
1423         mutex_lock(&root->fs_info->volume_mutex);
1424         vol_args = memdup_user(arg, sizeof(*vol_args));
1425         if (IS_ERR(vol_args)) {
1426                 ret = PTR_ERR(vol_args);
1427                 goto out;
1428         }
1429
1430         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1431
1432         sizestr = vol_args->name;
1433         devstr = strchr(sizestr, ':');
1434         if (devstr) {
1435                 char *end;
1436                 sizestr = devstr + 1;
1437                 *devstr = '\0';
1438                 devstr = vol_args->name;
1439                 devid = simple_strtoull(devstr, &end, 10);
1440                 if (!devid) {
1441                         ret = -EINVAL;
1442                         goto out_free;
1443                 }
1444                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1445         }
1446
1447         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1448         if (!device) {
1449                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1450                        devid);
1451                 ret = -ENODEV;
1452                 goto out_free;
1453         }
1454
1455         if (!device->writeable) {
1456                 btrfs_info(root->fs_info,
1457                            "resizer unable to apply on readonly device %llu",
1458                        devid);
1459                 ret = -EPERM;
1460                 goto out_free;
1461         }
1462
1463         if (!strcmp(sizestr, "max"))
1464                 new_size = device->bdev->bd_inode->i_size;
1465         else {
1466                 if (sizestr[0] == '-') {
1467                         mod = -1;
1468                         sizestr++;
1469                 } else if (sizestr[0] == '+') {
1470                         mod = 1;
1471                         sizestr++;
1472                 }
1473                 new_size = memparse(sizestr, NULL);
1474                 if (new_size == 0) {
1475                         ret = -EINVAL;
1476                         goto out_free;
1477                 }
1478         }
1479
1480         if (device->is_tgtdev_for_dev_replace) {
1481                 ret = -EPERM;
1482                 goto out_free;
1483         }
1484
1485         old_size = device->total_bytes;
1486
1487         if (mod < 0) {
1488                 if (new_size > old_size) {
1489                         ret = -EINVAL;
1490                         goto out_free;
1491                 }
1492                 new_size = old_size - new_size;
1493         } else if (mod > 0) {
1494                 if (new_size > ULLONG_MAX - old_size) {
1495                         ret = -EINVAL;
1496                         goto out_free;
1497                 }
1498                 new_size = old_size + new_size;
1499         }
1500
1501         if (new_size < 256 * 1024 * 1024) {
1502                 ret = -EINVAL;
1503                 goto out_free;
1504         }
1505         if (new_size > device->bdev->bd_inode->i_size) {
1506                 ret = -EFBIG;
1507                 goto out_free;
1508         }
1509
1510         do_div(new_size, root->sectorsize);
1511         new_size *= root->sectorsize;
1512
1513         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1514                       rcu_str_deref(device->name), new_size);
1515
1516         if (new_size > old_size) {
1517                 trans = btrfs_start_transaction(root, 0);
1518                 if (IS_ERR(trans)) {
1519                         ret = PTR_ERR(trans);
1520                         goto out_free;
1521                 }
1522                 ret = btrfs_grow_device(trans, device, new_size);
1523                 btrfs_commit_transaction(trans, root);
1524         } else if (new_size < old_size) {
1525                 ret = btrfs_shrink_device(device, new_size);
1526         } /* equal, nothing need to do */
1527
1528 out_free:
1529         kfree(vol_args);
1530 out:
1531         mutex_unlock(&root->fs_info->volume_mutex);
1532         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1533         mnt_drop_write_file(file);
1534         return ret;
1535 }
1536
1537 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1538                                 char *name, unsigned long fd, int subvol,
1539                                 u64 *transid, bool readonly,
1540                                 struct btrfs_qgroup_inherit *inherit)
1541 {
1542         int namelen;
1543         int ret = 0;
1544
1545         ret = mnt_want_write_file(file);
1546         if (ret)
1547                 goto out;
1548
1549         namelen = strlen(name);
1550         if (strchr(name, '/')) {
1551                 ret = -EINVAL;
1552                 goto out_drop_write;
1553         }
1554
1555         if (name[0] == '.' &&
1556            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1557                 ret = -EEXIST;
1558                 goto out_drop_write;
1559         }
1560
1561         if (subvol) {
1562                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1563                                      NULL, transid, readonly, inherit);
1564         } else {
1565                 struct fd src = fdget(fd);
1566                 struct inode *src_inode;
1567                 if (!src.file) {
1568                         ret = -EINVAL;
1569                         goto out_drop_write;
1570                 }
1571
1572                 src_inode = file_inode(src.file);
1573                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1574                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1575                                    "Snapshot src from another FS");
1576                         ret = -EINVAL;
1577                 } else if (!inode_owner_or_capable(src_inode)) {
1578                         /*
1579                          * Subvolume creation is not restricted, but snapshots
1580                          * are limited to own subvolumes only
1581                          */
1582                         ret = -EPERM;
1583                 } else {
1584                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1585                                              BTRFS_I(src_inode)->root,
1586                                              transid, readonly, inherit);
1587                 }
1588                 fdput(src);
1589         }
1590 out_drop_write:
1591         mnt_drop_write_file(file);
1592 out:
1593         return ret;
1594 }
1595
1596 static noinline int btrfs_ioctl_snap_create(struct file *file,
1597                                             void __user *arg, int subvol)
1598 {
1599         struct btrfs_ioctl_vol_args *vol_args;
1600         int ret;
1601
1602         vol_args = memdup_user(arg, sizeof(*vol_args));
1603         if (IS_ERR(vol_args))
1604                 return PTR_ERR(vol_args);
1605         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1606
1607         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1608                                               vol_args->fd, subvol,
1609                                               NULL, false, NULL);
1610
1611         kfree(vol_args);
1612         return ret;
1613 }
1614
1615 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1616                                                void __user *arg, int subvol)
1617 {
1618         struct btrfs_ioctl_vol_args_v2 *vol_args;
1619         int ret;
1620         u64 transid = 0;
1621         u64 *ptr = NULL;
1622         bool readonly = false;
1623         struct btrfs_qgroup_inherit *inherit = NULL;
1624
1625         vol_args = memdup_user(arg, sizeof(*vol_args));
1626         if (IS_ERR(vol_args))
1627                 return PTR_ERR(vol_args);
1628         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1629
1630         if (vol_args->flags &
1631             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1632               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1633                 ret = -EOPNOTSUPP;
1634                 goto out;
1635         }
1636
1637         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1638                 ptr = &transid;
1639         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1640                 readonly = true;
1641         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1642                 if (vol_args->size > PAGE_CACHE_SIZE) {
1643                         ret = -EINVAL;
1644                         goto out;
1645                 }
1646                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1647                 if (IS_ERR(inherit)) {
1648                         ret = PTR_ERR(inherit);
1649                         goto out;
1650                 }
1651         }
1652
1653         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1654                                               vol_args->fd, subvol, ptr,
1655                                               readonly, inherit);
1656
1657         if (ret == 0 && ptr &&
1658             copy_to_user(arg +
1659                          offsetof(struct btrfs_ioctl_vol_args_v2,
1660                                   transid), ptr, sizeof(*ptr)))
1661                 ret = -EFAULT;
1662 out:
1663         kfree(vol_args);
1664         kfree(inherit);
1665         return ret;
1666 }
1667
1668 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1669                                                 void __user *arg)
1670 {
1671         struct inode *inode = file_inode(file);
1672         struct btrfs_root *root = BTRFS_I(inode)->root;
1673         int ret = 0;
1674         u64 flags = 0;
1675
1676         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1677                 return -EINVAL;
1678
1679         down_read(&root->fs_info->subvol_sem);
1680         if (btrfs_root_readonly(root))
1681                 flags |= BTRFS_SUBVOL_RDONLY;
1682         up_read(&root->fs_info->subvol_sem);
1683
1684         if (copy_to_user(arg, &flags, sizeof(flags)))
1685                 ret = -EFAULT;
1686
1687         return ret;
1688 }
1689
1690 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1691                                               void __user *arg)
1692 {
1693         struct inode *inode = file_inode(file);
1694         struct btrfs_root *root = BTRFS_I(inode)->root;
1695         struct btrfs_trans_handle *trans;
1696         u64 root_flags;
1697         u64 flags;
1698         int ret = 0;
1699
1700         if (!inode_owner_or_capable(inode))
1701                 return -EPERM;
1702
1703         ret = mnt_want_write_file(file);
1704         if (ret)
1705                 goto out;
1706
1707         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1708                 ret = -EINVAL;
1709                 goto out_drop_write;
1710         }
1711
1712         if (copy_from_user(&flags, arg, sizeof(flags))) {
1713                 ret = -EFAULT;
1714                 goto out_drop_write;
1715         }
1716
1717         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1718                 ret = -EINVAL;
1719                 goto out_drop_write;
1720         }
1721
1722         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1723                 ret = -EOPNOTSUPP;
1724                 goto out_drop_write;
1725         }
1726
1727         down_write(&root->fs_info->subvol_sem);
1728
1729         /* nothing to do */
1730         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1731                 goto out_drop_sem;
1732
1733         root_flags = btrfs_root_flags(&root->root_item);
1734         if (flags & BTRFS_SUBVOL_RDONLY) {
1735                 btrfs_set_root_flags(&root->root_item,
1736                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1737         } else {
1738                 /*
1739                  * Block RO -> RW transition if this subvolume is involved in
1740                  * send
1741                  */
1742                 spin_lock(&root->root_item_lock);
1743                 if (root->send_in_progress == 0) {
1744                         btrfs_set_root_flags(&root->root_item,
1745                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1746                         spin_unlock(&root->root_item_lock);
1747                 } else {
1748                         spin_unlock(&root->root_item_lock);
1749                         btrfs_warn(root->fs_info,
1750                         "Attempt to set subvolume %llu read-write during send",
1751                                         root->root_key.objectid);
1752                         ret = -EPERM;
1753                         goto out_drop_sem;
1754                 }
1755         }
1756
1757         trans = btrfs_start_transaction(root, 1);
1758         if (IS_ERR(trans)) {
1759                 ret = PTR_ERR(trans);
1760                 goto out_reset;
1761         }
1762
1763         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1764                                 &root->root_key, &root->root_item);
1765
1766         btrfs_commit_transaction(trans, root);
1767 out_reset:
1768         if (ret)
1769                 btrfs_set_root_flags(&root->root_item, root_flags);
1770 out_drop_sem:
1771         up_write(&root->fs_info->subvol_sem);
1772 out_drop_write:
1773         mnt_drop_write_file(file);
1774 out:
1775         return ret;
1776 }
1777
1778 /*
1779  * helper to check if the subvolume references other subvolumes
1780  */
1781 static noinline int may_destroy_subvol(struct btrfs_root *root)
1782 {
1783         struct btrfs_path *path;
1784         struct btrfs_dir_item *di;
1785         struct btrfs_key key;
1786         u64 dir_id;
1787         int ret;
1788
1789         path = btrfs_alloc_path();
1790         if (!path)
1791                 return -ENOMEM;
1792
1793         /* Make sure this root isn't set as the default subvol */
1794         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1795         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1796                                    dir_id, "default", 7, 0);
1797         if (di && !IS_ERR(di)) {
1798                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1799                 if (key.objectid == root->root_key.objectid) {
1800                         ret = -ENOTEMPTY;
1801                         goto out;
1802                 }
1803                 btrfs_release_path(path);
1804         }
1805
1806         key.objectid = root->root_key.objectid;
1807         key.type = BTRFS_ROOT_REF_KEY;
1808         key.offset = (u64)-1;
1809
1810         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1811                                 &key, path, 0, 0);
1812         if (ret < 0)
1813                 goto out;
1814         BUG_ON(ret == 0);
1815
1816         ret = 0;
1817         if (path->slots[0] > 0) {
1818                 path->slots[0]--;
1819                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1820                 if (key.objectid == root->root_key.objectid &&
1821                     key.type == BTRFS_ROOT_REF_KEY)
1822                         ret = -ENOTEMPTY;
1823         }
1824 out:
1825         btrfs_free_path(path);
1826         return ret;
1827 }
1828
1829 static noinline int key_in_sk(struct btrfs_key *key,
1830                               struct btrfs_ioctl_search_key *sk)
1831 {
1832         struct btrfs_key test;
1833         int ret;
1834
1835         test.objectid = sk->min_objectid;
1836         test.type = sk->min_type;
1837         test.offset = sk->min_offset;
1838
1839         ret = btrfs_comp_cpu_keys(key, &test);
1840         if (ret < 0)
1841                 return 0;
1842
1843         test.objectid = sk->max_objectid;
1844         test.type = sk->max_type;
1845         test.offset = sk->max_offset;
1846
1847         ret = btrfs_comp_cpu_keys(key, &test);
1848         if (ret > 0)
1849                 return 0;
1850         return 1;
1851 }
1852
1853 static noinline int copy_to_sk(struct btrfs_root *root,
1854                                struct btrfs_path *path,
1855                                struct btrfs_key *key,
1856                                struct btrfs_ioctl_search_key *sk,
1857                                char *buf,
1858                                unsigned long *sk_offset,
1859                                int *num_found)
1860 {
1861         u64 found_transid;
1862         struct extent_buffer *leaf;
1863         struct btrfs_ioctl_search_header sh;
1864         unsigned long item_off;
1865         unsigned long item_len;
1866         int nritems;
1867         int i;
1868         int slot;
1869         int ret = 0;
1870
1871         leaf = path->nodes[0];
1872         slot = path->slots[0];
1873         nritems = btrfs_header_nritems(leaf);
1874
1875         if (btrfs_header_generation(leaf) > sk->max_transid) {
1876                 i = nritems;
1877                 goto advance_key;
1878         }
1879         found_transid = btrfs_header_generation(leaf);
1880
1881         for (i = slot; i < nritems; i++) {
1882                 item_off = btrfs_item_ptr_offset(leaf, i);
1883                 item_len = btrfs_item_size_nr(leaf, i);
1884
1885                 btrfs_item_key_to_cpu(leaf, key, i);
1886                 if (!key_in_sk(key, sk))
1887                         continue;
1888
1889                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1890                         item_len = 0;
1891
1892                 if (sizeof(sh) + item_len + *sk_offset >
1893                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1894                         ret = 1;
1895                         goto overflow;
1896                 }
1897
1898                 sh.objectid = key->objectid;
1899                 sh.offset = key->offset;
1900                 sh.type = key->type;
1901                 sh.len = item_len;
1902                 sh.transid = found_transid;
1903
1904                 /* copy search result header */
1905                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1906                 *sk_offset += sizeof(sh);
1907
1908                 if (item_len) {
1909                         char *p = buf + *sk_offset;
1910                         /* copy the item */
1911                         read_extent_buffer(leaf, p,
1912                                            item_off, item_len);
1913                         *sk_offset += item_len;
1914                 }
1915                 (*num_found)++;
1916
1917                 if (*num_found >= sk->nr_items)
1918                         break;
1919         }
1920 advance_key:
1921         ret = 0;
1922         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1923                 key->offset++;
1924         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1925                 key->offset = 0;
1926                 key->type++;
1927         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1928                 key->offset = 0;
1929                 key->type = 0;
1930                 key->objectid++;
1931         } else
1932                 ret = 1;
1933 overflow:
1934         return ret;
1935 }
1936
1937 static noinline int search_ioctl(struct inode *inode,
1938                                  struct btrfs_ioctl_search_args *args)
1939 {
1940         struct btrfs_root *root;
1941         struct btrfs_key key;
1942         struct btrfs_path *path;
1943         struct btrfs_ioctl_search_key *sk = &args->key;
1944         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1945         int ret;
1946         int num_found = 0;
1947         unsigned long sk_offset = 0;
1948
1949         path = btrfs_alloc_path();
1950         if (!path)
1951                 return -ENOMEM;
1952
1953         if (sk->tree_id == 0) {
1954                 /* search the root of the inode that was passed */
1955                 root = BTRFS_I(inode)->root;
1956         } else {
1957                 key.objectid = sk->tree_id;
1958                 key.type = BTRFS_ROOT_ITEM_KEY;
1959                 key.offset = (u64)-1;
1960                 root = btrfs_read_fs_root_no_name(info, &key);
1961                 if (IS_ERR(root)) {
1962                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
1963                                sk->tree_id);
1964                         btrfs_free_path(path);
1965                         return -ENOENT;
1966                 }
1967         }
1968
1969         key.objectid = sk->min_objectid;
1970         key.type = sk->min_type;
1971         key.offset = sk->min_offset;
1972
1973         path->keep_locks = 1;
1974
1975         while (1) {
1976                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
1977                 if (ret != 0) {
1978                         if (ret > 0)
1979                                 ret = 0;
1980                         goto err;
1981                 }
1982                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1983                                  &sk_offset, &num_found);
1984                 btrfs_release_path(path);
1985                 if (ret || num_found >= sk->nr_items)
1986                         break;
1987
1988         }
1989         ret = 0;
1990 err:
1991         sk->nr_items = num_found;
1992         btrfs_free_path(path);
1993         return ret;
1994 }
1995
1996 static noinline int btrfs_ioctl_tree_search(struct file *file,
1997                                            void __user *argp)
1998 {
1999          struct btrfs_ioctl_search_args *args;
2000          struct inode *inode;
2001          int ret;
2002
2003         if (!capable(CAP_SYS_ADMIN))
2004                 return -EPERM;
2005
2006         args = memdup_user(argp, sizeof(*args));
2007         if (IS_ERR(args))
2008                 return PTR_ERR(args);
2009
2010         inode = file_inode(file);
2011         ret = search_ioctl(inode, args);
2012         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2013                 ret = -EFAULT;
2014         kfree(args);
2015         return ret;
2016 }
2017
2018 /*
2019  * Search INODE_REFs to identify path name of 'dirid' directory
2020  * in a 'tree_id' tree. and sets path name to 'name'.
2021  */
2022 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2023                                 u64 tree_id, u64 dirid, char *name)
2024 {
2025         struct btrfs_root *root;
2026         struct btrfs_key key;
2027         char *ptr;
2028         int ret = -1;
2029         int slot;
2030         int len;
2031         int total_len = 0;
2032         struct btrfs_inode_ref *iref;
2033         struct extent_buffer *l;
2034         struct btrfs_path *path;
2035
2036         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2037                 name[0]='\0';
2038                 return 0;
2039         }
2040
2041         path = btrfs_alloc_path();
2042         if (!path)
2043                 return -ENOMEM;
2044
2045         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2046
2047         key.objectid = tree_id;
2048         key.type = BTRFS_ROOT_ITEM_KEY;
2049         key.offset = (u64)-1;
2050         root = btrfs_read_fs_root_no_name(info, &key);
2051         if (IS_ERR(root)) {
2052                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2053                 ret = -ENOENT;
2054                 goto out;
2055         }
2056
2057         key.objectid = dirid;
2058         key.type = BTRFS_INODE_REF_KEY;
2059         key.offset = (u64)-1;
2060
2061         while (1) {
2062                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2063                 if (ret < 0)
2064                         goto out;
2065                 else if (ret > 0) {
2066                         ret = btrfs_previous_item(root, path, dirid,
2067                                                   BTRFS_INODE_REF_KEY);
2068                         if (ret < 0)
2069                                 goto out;
2070                         else if (ret > 0) {
2071                                 ret = -ENOENT;
2072                                 goto out;
2073                         }
2074                 }
2075
2076                 l = path->nodes[0];
2077                 slot = path->slots[0];
2078                 btrfs_item_key_to_cpu(l, &key, slot);
2079
2080                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2081                 len = btrfs_inode_ref_name_len(l, iref);
2082                 ptr -= len + 1;
2083                 total_len += len + 1;
2084                 if (ptr < name) {
2085                         ret = -ENAMETOOLONG;
2086                         goto out;
2087                 }
2088
2089                 *(ptr + len) = '/';
2090                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2091
2092                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2093                         break;
2094
2095                 btrfs_release_path(path);
2096                 key.objectid = key.offset;
2097                 key.offset = (u64)-1;
2098                 dirid = key.objectid;
2099         }
2100         memmove(name, ptr, total_len);
2101         name[total_len] = '\0';
2102         ret = 0;
2103 out:
2104         btrfs_free_path(path);
2105         return ret;
2106 }
2107
2108 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2109                                            void __user *argp)
2110 {
2111          struct btrfs_ioctl_ino_lookup_args *args;
2112          struct inode *inode;
2113          int ret;
2114
2115         if (!capable(CAP_SYS_ADMIN))
2116                 return -EPERM;
2117
2118         args = memdup_user(argp, sizeof(*args));
2119         if (IS_ERR(args))
2120                 return PTR_ERR(args);
2121
2122         inode = file_inode(file);
2123
2124         if (args->treeid == 0)
2125                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2126
2127         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2128                                         args->treeid, args->objectid,
2129                                         args->name);
2130
2131         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2132                 ret = -EFAULT;
2133
2134         kfree(args);
2135         return ret;
2136 }
2137
2138 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2139                                              void __user *arg)
2140 {
2141         struct dentry *parent = file->f_path.dentry;
2142         struct dentry *dentry;
2143         struct inode *dir = parent->d_inode;
2144         struct inode *inode;
2145         struct btrfs_root *root = BTRFS_I(dir)->root;
2146         struct btrfs_root *dest = NULL;
2147         struct btrfs_ioctl_vol_args *vol_args;
2148         struct btrfs_trans_handle *trans;
2149         struct btrfs_block_rsv block_rsv;
2150         u64 qgroup_reserved;
2151         int namelen;
2152         int ret;
2153         int err = 0;
2154
2155         vol_args = memdup_user(arg, sizeof(*vol_args));
2156         if (IS_ERR(vol_args))
2157                 return PTR_ERR(vol_args);
2158
2159         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2160         namelen = strlen(vol_args->name);
2161         if (strchr(vol_args->name, '/') ||
2162             strncmp(vol_args->name, "..", namelen) == 0) {
2163                 err = -EINVAL;
2164                 goto out;
2165         }
2166
2167         err = mnt_want_write_file(file);
2168         if (err)
2169                 goto out;
2170
2171         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2172         if (err == -EINTR)
2173                 goto out_drop_write;
2174         dentry = lookup_one_len(vol_args->name, parent, namelen);
2175         if (IS_ERR(dentry)) {
2176                 err = PTR_ERR(dentry);
2177                 goto out_unlock_dir;
2178         }
2179
2180         if (!dentry->d_inode) {
2181                 err = -ENOENT;
2182                 goto out_dput;
2183         }
2184
2185         inode = dentry->d_inode;
2186         dest = BTRFS_I(inode)->root;
2187         if (!capable(CAP_SYS_ADMIN)) {
2188                 /*
2189                  * Regular user.  Only allow this with a special mount
2190                  * option, when the user has write+exec access to the
2191                  * subvol root, and when rmdir(2) would have been
2192                  * allowed.
2193                  *
2194                  * Note that this is _not_ check that the subvol is
2195                  * empty or doesn't contain data that we wouldn't
2196                  * otherwise be able to delete.
2197                  *
2198                  * Users who want to delete empty subvols should try
2199                  * rmdir(2).
2200                  */
2201                 err = -EPERM;
2202                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2203                         goto out_dput;
2204
2205                 /*
2206                  * Do not allow deletion if the parent dir is the same
2207                  * as the dir to be deleted.  That means the ioctl
2208                  * must be called on the dentry referencing the root
2209                  * of the subvol, not a random directory contained
2210                  * within it.
2211                  */
2212                 err = -EINVAL;
2213                 if (root == dest)
2214                         goto out_dput;
2215
2216                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2217                 if (err)
2218                         goto out_dput;
2219         }
2220
2221         /* check if subvolume may be deleted by a user */
2222         err = btrfs_may_delete(dir, dentry, 1);
2223         if (err)
2224                 goto out_dput;
2225
2226         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2227                 err = -EINVAL;
2228                 goto out_dput;
2229         }
2230
2231         mutex_lock(&inode->i_mutex);
2232         err = d_invalidate(dentry);
2233         if (err)
2234                 goto out_unlock;
2235
2236         down_write(&root->fs_info->subvol_sem);
2237
2238         err = may_destroy_subvol(dest);
2239         if (err)
2240                 goto out_up_write;
2241
2242         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2243         /*
2244          * One for dir inode, two for dir entries, two for root
2245          * ref/backref.
2246          */
2247         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2248                                                5, &qgroup_reserved, true);
2249         if (err)
2250                 goto out_up_write;
2251
2252         trans = btrfs_start_transaction(root, 0);
2253         if (IS_ERR(trans)) {
2254                 err = PTR_ERR(trans);
2255                 goto out_release;
2256         }
2257         trans->block_rsv = &block_rsv;
2258         trans->bytes_reserved = block_rsv.size;
2259
2260         ret = btrfs_unlink_subvol(trans, root, dir,
2261                                 dest->root_key.objectid,
2262                                 dentry->d_name.name,
2263                                 dentry->d_name.len);
2264         if (ret) {
2265                 err = ret;
2266                 btrfs_abort_transaction(trans, root, ret);
2267                 goto out_end_trans;
2268         }
2269
2270         btrfs_record_root_in_trans(trans, dest);
2271
2272         memset(&dest->root_item.drop_progress, 0,
2273                 sizeof(dest->root_item.drop_progress));
2274         dest->root_item.drop_level = 0;
2275         btrfs_set_root_refs(&dest->root_item, 0);
2276
2277         if (!xchg(&dest->orphan_item_inserted, 1)) {
2278                 ret = btrfs_insert_orphan_item(trans,
2279                                         root->fs_info->tree_root,
2280                                         dest->root_key.objectid);
2281                 if (ret) {
2282                         btrfs_abort_transaction(trans, root, ret);
2283                         err = ret;
2284                         goto out_end_trans;
2285                 }
2286         }
2287
2288         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2289                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2290                                   dest->root_key.objectid);
2291         if (ret && ret != -ENOENT) {
2292                 btrfs_abort_transaction(trans, root, ret);
2293                 err = ret;
2294                 goto out_end_trans;
2295         }
2296         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2297                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2298                                           dest->root_item.received_uuid,
2299                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2300                                           dest->root_key.objectid);
2301                 if (ret && ret != -ENOENT) {
2302                         btrfs_abort_transaction(trans, root, ret);
2303                         err = ret;
2304                         goto out_end_trans;
2305                 }
2306         }
2307
2308 out_end_trans:
2309         trans->block_rsv = NULL;
2310         trans->bytes_reserved = 0;
2311         ret = btrfs_end_transaction(trans, root);
2312         if (ret && !err)
2313                 err = ret;
2314         inode->i_flags |= S_DEAD;
2315 out_release:
2316         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2317 out_up_write:
2318         up_write(&root->fs_info->subvol_sem);
2319 out_unlock:
2320         mutex_unlock(&inode->i_mutex);
2321         if (!err) {
2322                 shrink_dcache_sb(root->fs_info->sb);
2323                 btrfs_invalidate_inodes(dest);
2324                 d_delete(dentry);
2325
2326                 /* the last ref */
2327                 if (dest->cache_inode) {
2328                         iput(dest->cache_inode);
2329                         dest->cache_inode = NULL;
2330                 }
2331         }
2332 out_dput:
2333         dput(dentry);
2334 out_unlock_dir:
2335         mutex_unlock(&dir->i_mutex);
2336 out_drop_write:
2337         mnt_drop_write_file(file);
2338 out:
2339         kfree(vol_args);
2340         return err;
2341 }
2342
2343 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2344 {
2345         struct inode *inode = file_inode(file);
2346         struct btrfs_root *root = BTRFS_I(inode)->root;
2347         struct btrfs_ioctl_defrag_range_args *range;
2348         int ret;
2349
2350         ret = mnt_want_write_file(file);
2351         if (ret)
2352                 return ret;
2353
2354         if (btrfs_root_readonly(root)) {
2355                 ret = -EROFS;
2356                 goto out;
2357         }
2358
2359         switch (inode->i_mode & S_IFMT) {
2360         case S_IFDIR:
2361                 if (!capable(CAP_SYS_ADMIN)) {
2362                         ret = -EPERM;
2363                         goto out;
2364                 }
2365                 ret = btrfs_defrag_root(root);
2366                 if (ret)
2367                         goto out;
2368                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2369                 break;
2370         case S_IFREG:
2371                 if (!(file->f_mode & FMODE_WRITE)) {
2372                         ret = -EINVAL;
2373                         goto out;
2374                 }
2375
2376                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2377                 if (!range) {
2378                         ret = -ENOMEM;
2379                         goto out;
2380                 }
2381
2382                 if (argp) {
2383                         if (copy_from_user(range, argp,
2384                                            sizeof(*range))) {
2385                                 ret = -EFAULT;
2386                                 kfree(range);
2387                                 goto out;
2388                         }
2389                         /* compression requires us to start the IO */
2390                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2391                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2392                                 range->extent_thresh = (u32)-1;
2393                         }
2394                 } else {
2395                         /* the rest are all set to zero by kzalloc */
2396                         range->len = (u64)-1;
2397                 }
2398                 ret = btrfs_defrag_file(file_inode(file), file,
2399                                         range, 0, 0);
2400                 if (ret > 0)
2401                         ret = 0;
2402                 kfree(range);
2403                 break;
2404         default:
2405                 ret = -EINVAL;
2406         }
2407 out:
2408         mnt_drop_write_file(file);
2409         return ret;
2410 }
2411
2412 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2413 {
2414         struct btrfs_ioctl_vol_args *vol_args;
2415         int ret;
2416
2417         if (!capable(CAP_SYS_ADMIN))
2418                 return -EPERM;
2419
2420         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2421                         1)) {
2422                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2423         }
2424
2425         mutex_lock(&root->fs_info->volume_mutex);
2426         vol_args = memdup_user(arg, sizeof(*vol_args));
2427         if (IS_ERR(vol_args)) {
2428                 ret = PTR_ERR(vol_args);
2429                 goto out;
2430         }
2431
2432         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2433         ret = btrfs_init_new_device(root, vol_args->name);
2434
2435         kfree(vol_args);
2436 out:
2437         mutex_unlock(&root->fs_info->volume_mutex);
2438         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2439         return ret;
2440 }
2441
2442 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2443 {
2444         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2445         struct btrfs_ioctl_vol_args *vol_args;
2446         int ret;
2447
2448         if (!capable(CAP_SYS_ADMIN))
2449                 return -EPERM;
2450
2451         ret = mnt_want_write_file(file);
2452         if (ret)
2453                 return ret;
2454
2455         vol_args = memdup_user(arg, sizeof(*vol_args));
2456         if (IS_ERR(vol_args)) {
2457                 ret = PTR_ERR(vol_args);
2458                 goto out;
2459         }
2460
2461         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2462
2463         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2464                         1)) {
2465                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2466                 goto out;
2467         }
2468
2469         mutex_lock(&root->fs_info->volume_mutex);
2470         ret = btrfs_rm_device(root, vol_args->name);
2471         mutex_unlock(&root->fs_info->volume_mutex);
2472         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2473
2474 out:
2475         kfree(vol_args);
2476         mnt_drop_write_file(file);
2477         return ret;
2478 }
2479
2480 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2481 {
2482         struct btrfs_ioctl_fs_info_args *fi_args;
2483         struct btrfs_device *device;
2484         struct btrfs_device *next;
2485         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2486         int ret = 0;
2487
2488         if (!capable(CAP_SYS_ADMIN))
2489                 return -EPERM;
2490
2491         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2492         if (!fi_args)
2493                 return -ENOMEM;
2494
2495         mutex_lock(&fs_devices->device_list_mutex);
2496         fi_args->num_devices = fs_devices->num_devices;
2497         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2498
2499         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2500                 if (device->devid > fi_args->max_id)
2501                         fi_args->max_id = device->devid;
2502         }
2503         mutex_unlock(&fs_devices->device_list_mutex);
2504
2505         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2506                 ret = -EFAULT;
2507
2508         kfree(fi_args);
2509         return ret;
2510 }
2511
2512 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2513 {
2514         struct btrfs_ioctl_dev_info_args *di_args;
2515         struct btrfs_device *dev;
2516         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2517         int ret = 0;
2518         char *s_uuid = NULL;
2519
2520         if (!capable(CAP_SYS_ADMIN))
2521                 return -EPERM;
2522
2523         di_args = memdup_user(arg, sizeof(*di_args));
2524         if (IS_ERR(di_args))
2525                 return PTR_ERR(di_args);
2526
2527         if (!btrfs_is_empty_uuid(di_args->uuid))
2528                 s_uuid = di_args->uuid;
2529
2530         mutex_lock(&fs_devices->device_list_mutex);
2531         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2532
2533         if (!dev) {
2534                 ret = -ENODEV;
2535                 goto out;
2536         }
2537
2538         di_args->devid = dev->devid;
2539         di_args->bytes_used = dev->bytes_used;
2540         di_args->total_bytes = dev->total_bytes;
2541         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2542         if (dev->name) {
2543                 struct rcu_string *name;
2544
2545                 rcu_read_lock();
2546                 name = rcu_dereference(dev->name);
2547                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2548                 rcu_read_unlock();
2549                 di_args->path[sizeof(di_args->path) - 1] = 0;
2550         } else {
2551                 di_args->path[0] = '\0';
2552         }
2553
2554 out:
2555         mutex_unlock(&fs_devices->device_list_mutex);
2556         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2557                 ret = -EFAULT;
2558
2559         kfree(di_args);
2560         return ret;
2561 }
2562
2563 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2564 {
2565         struct page *page;
2566         pgoff_t index;
2567         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2568
2569         index = off >> PAGE_CACHE_SHIFT;
2570
2571         page = grab_cache_page(inode->i_mapping, index);
2572         if (!page)
2573                 return NULL;
2574
2575         if (!PageUptodate(page)) {
2576                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2577                                                  0))
2578                         return NULL;
2579                 lock_page(page);
2580                 if (!PageUptodate(page)) {
2581                         unlock_page(page);
2582                         page_cache_release(page);
2583                         return NULL;
2584                 }
2585         }
2586         unlock_page(page);
2587
2588         return page;
2589 }
2590
2591 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2592 {
2593         /* do any pending delalloc/csum calc on src, one way or
2594            another, and lock file content */
2595         while (1) {
2596                 struct btrfs_ordered_extent *ordered;
2597                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2598                 ordered = btrfs_lookup_first_ordered_extent(inode,
2599                                                             off + len - 1);
2600                 if (!ordered &&
2601                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2602                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2603                         break;
2604                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2605                 if (ordered)
2606                         btrfs_put_ordered_extent(ordered);
2607                 btrfs_wait_ordered_range(inode, off, len);
2608         }
2609 }
2610
2611 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2612                                 struct inode *inode2, u64 loff2, u64 len)
2613 {
2614         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2615         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2616
2617         mutex_unlock(&inode1->i_mutex);
2618         mutex_unlock(&inode2->i_mutex);
2619 }
2620
2621 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2622                               struct inode *inode2, u64 loff2, u64 len)
2623 {
2624         if (inode1 < inode2) {
2625                 swap(inode1, inode2);
2626                 swap(loff1, loff2);
2627         }
2628
2629         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2630         lock_extent_range(inode1, loff1, len);
2631         if (inode1 != inode2) {
2632                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2633                 lock_extent_range(inode2, loff2, len);
2634         }
2635 }
2636
2637 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2638                           u64 dst_loff, u64 len)
2639 {
2640         int ret = 0;
2641         struct page *src_page, *dst_page;
2642         unsigned int cmp_len = PAGE_CACHE_SIZE;
2643         void *addr, *dst_addr;
2644
2645         while (len) {
2646                 if (len < PAGE_CACHE_SIZE)
2647                         cmp_len = len;
2648
2649                 src_page = extent_same_get_page(src, loff);
2650                 if (!src_page)
2651                         return -EINVAL;
2652                 dst_page = extent_same_get_page(dst, dst_loff);
2653                 if (!dst_page) {
2654                         page_cache_release(src_page);
2655                         return -EINVAL;
2656                 }
2657                 addr = kmap_atomic(src_page);
2658                 dst_addr = kmap_atomic(dst_page);
2659
2660                 flush_dcache_page(src_page);
2661                 flush_dcache_page(dst_page);
2662
2663                 if (memcmp(addr, dst_addr, cmp_len))
2664                         ret = BTRFS_SAME_DATA_DIFFERS;
2665
2666                 kunmap_atomic(addr);
2667                 kunmap_atomic(dst_addr);
2668                 page_cache_release(src_page);
2669                 page_cache_release(dst_page);
2670
2671                 if (ret)
2672                         break;
2673
2674                 loff += cmp_len;
2675                 dst_loff += cmp_len;
2676                 len -= cmp_len;
2677         }
2678
2679         return ret;
2680 }
2681
2682 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2683 {
2684         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2685
2686         if (off + len > inode->i_size || off + len < off)
2687                 return -EINVAL;
2688         /* Check that we are block aligned - btrfs_clone() requires this */
2689         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2690                 return -EINVAL;
2691
2692         return 0;
2693 }
2694
2695 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2696                              struct inode *dst, u64 dst_loff)
2697 {
2698         int ret;
2699
2700         /*
2701          * btrfs_clone() can't handle extents in the same file
2702          * yet. Once that works, we can drop this check and replace it
2703          * with a check for the same inode, but overlapping extents.
2704          */
2705         if (src == dst)
2706                 return -EINVAL;
2707
2708         btrfs_double_lock(src, loff, dst, dst_loff, len);
2709
2710         ret = extent_same_check_offsets(src, loff, len);
2711         if (ret)
2712                 goto out_unlock;
2713
2714         ret = extent_same_check_offsets(dst, dst_loff, len);
2715         if (ret)
2716                 goto out_unlock;
2717
2718         /* don't make the dst file partly checksummed */
2719         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2720             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2721                 ret = -EINVAL;
2722                 goto out_unlock;
2723         }
2724
2725         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2726         if (ret == 0)
2727                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2728
2729 out_unlock:
2730         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2731
2732         return ret;
2733 }
2734
2735 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2736
2737 static long btrfs_ioctl_file_extent_same(struct file *file,
2738                         struct btrfs_ioctl_same_args __user *argp)
2739 {
2740         struct btrfs_ioctl_same_args *same;
2741         struct btrfs_ioctl_same_extent_info *info;
2742         struct inode *src = file_inode(file);
2743         u64 off;
2744         u64 len;
2745         int i;
2746         int ret;
2747         unsigned long size;
2748         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2749         bool is_admin = capable(CAP_SYS_ADMIN);
2750         u16 count;
2751
2752         if (!(file->f_mode & FMODE_READ))
2753                 return -EINVAL;
2754
2755         ret = mnt_want_write_file(file);
2756         if (ret)
2757                 return ret;
2758
2759         if (get_user(count, &argp->dest_count)) {
2760                 ret = -EFAULT;
2761                 goto out;
2762         }
2763
2764         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2765
2766         same = memdup_user(argp, size);
2767
2768         if (IS_ERR(same)) {
2769                 ret = PTR_ERR(same);
2770                 goto out;
2771         }
2772
2773         off = same->logical_offset;
2774         len = same->length;
2775
2776         /*
2777          * Limit the total length we will dedupe for each operation.
2778          * This is intended to bound the total time spent in this
2779          * ioctl to something sane.
2780          */
2781         if (len > BTRFS_MAX_DEDUPE_LEN)
2782                 len = BTRFS_MAX_DEDUPE_LEN;
2783
2784         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2785                 /*
2786                  * Btrfs does not support blocksize < page_size. As a
2787                  * result, btrfs_cmp_data() won't correctly handle
2788                  * this situation without an update.
2789                  */
2790                 ret = -EINVAL;
2791                 goto out;
2792         }
2793
2794         ret = -EISDIR;
2795         if (S_ISDIR(src->i_mode))
2796                 goto out;
2797
2798         ret = -EACCES;
2799         if (!S_ISREG(src->i_mode))
2800                 goto out;
2801
2802         /* pre-format output fields to sane values */
2803         for (i = 0; i < count; i++) {
2804                 same->info[i].bytes_deduped = 0ULL;
2805                 same->info[i].status = 0;
2806         }
2807
2808         for (i = 0, info = same->info; i < count; i++, info++) {
2809                 struct inode *dst;
2810                 struct fd dst_file = fdget(info->fd);
2811                 if (!dst_file.file) {
2812                         info->status = -EBADF;
2813                         continue;
2814                 }
2815                 dst = file_inode(dst_file.file);
2816
2817                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
2818                         info->status = -EINVAL;
2819                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
2820                         info->status = -EXDEV;
2821                 } else if (S_ISDIR(dst->i_mode)) {
2822                         info->status = -EISDIR;
2823                 } else if (!S_ISREG(dst->i_mode)) {
2824                         info->status = -EACCES;
2825                 } else {
2826                         info->status = btrfs_extent_same(src, off, len, dst,
2827                                                         info->logical_offset);
2828                         if (info->status == 0)
2829                                 info->bytes_deduped += len;
2830                 }
2831                 fdput(dst_file);
2832         }
2833
2834         ret = copy_to_user(argp, same, size);
2835         if (ret)
2836                 ret = -EFAULT;
2837
2838 out:
2839         mnt_drop_write_file(file);
2840         return ret;
2841 }
2842
2843 /**
2844  * btrfs_clone() - clone a range from inode file to another
2845  *
2846  * @src: Inode to clone from
2847  * @inode: Inode to clone to
2848  * @off: Offset within source to start clone from
2849  * @olen: Original length, passed by user, of range to clone
2850  * @olen_aligned: Block-aligned value of olen, extent_same uses
2851  *               identical values here
2852  * @destoff: Offset within @inode to start clone
2853  */
2854 static int btrfs_clone(struct inode *src, struct inode *inode,
2855                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2856 {
2857         struct btrfs_root *root = BTRFS_I(inode)->root;
2858         struct btrfs_path *path = NULL;
2859         struct extent_buffer *leaf;
2860         struct btrfs_trans_handle *trans;
2861         char *buf = NULL;
2862         struct btrfs_key key;
2863         u32 nritems;
2864         int slot;
2865         int ret;
2866         u64 len = olen_aligned;
2867
2868         ret = -ENOMEM;
2869         buf = vmalloc(btrfs_level_size(root, 0));
2870         if (!buf)
2871                 return ret;
2872
2873         path = btrfs_alloc_path();
2874         if (!path) {
2875                 vfree(buf);
2876                 return ret;
2877         }
2878
2879         path->reada = 2;
2880         /* clone data */
2881         key.objectid = btrfs_ino(src);
2882         key.type = BTRFS_EXTENT_DATA_KEY;
2883         key.offset = 0;
2884
2885         while (1) {
2886                 /*
2887                  * note the key will change type as we walk through the
2888                  * tree.
2889                  */
2890                 path->leave_spinning = 1;
2891                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2892                                 0, 0);
2893                 if (ret < 0)
2894                         goto out;
2895
2896                 nritems = btrfs_header_nritems(path->nodes[0]);
2897 process_slot:
2898                 if (path->slots[0] >= nritems) {
2899                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2900                         if (ret < 0)
2901                                 goto out;
2902                         if (ret > 0)
2903                                 break;
2904                         nritems = btrfs_header_nritems(path->nodes[0]);
2905                 }
2906                 leaf = path->nodes[0];
2907                 slot = path->slots[0];
2908
2909                 btrfs_item_key_to_cpu(leaf, &key, slot);
2910                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2911                     key.objectid != btrfs_ino(src))
2912                         break;
2913
2914                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2915                         struct btrfs_file_extent_item *extent;
2916                         int type;
2917                         u32 size;
2918                         struct btrfs_key new_key;
2919                         u64 disko = 0, diskl = 0;
2920                         u64 datao = 0, datal = 0;
2921                         u8 comp;
2922                         u64 endoff;
2923
2924                         extent = btrfs_item_ptr(leaf, slot,
2925                                                 struct btrfs_file_extent_item);
2926                         comp = btrfs_file_extent_compression(leaf, extent);
2927                         type = btrfs_file_extent_type(leaf, extent);
2928                         if (type == BTRFS_FILE_EXTENT_REG ||
2929                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2930                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2931                                                                       extent);
2932                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2933                                                                  extent);
2934                                 datao = btrfs_file_extent_offset(leaf, extent);
2935                                 datal = btrfs_file_extent_num_bytes(leaf,
2936                                                                     extent);
2937                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2938                                 /* take upper bound, may be compressed */
2939                                 datal = btrfs_file_extent_ram_bytes(leaf,
2940                                                                     extent);
2941                         }
2942
2943                         if (key.offset + datal <= off ||
2944                             key.offset >= off + len - 1) {
2945                                 path->slots[0]++;
2946                                 goto process_slot;
2947                         }
2948
2949                         size = btrfs_item_size_nr(leaf, slot);
2950                         read_extent_buffer(leaf, buf,
2951                                            btrfs_item_ptr_offset(leaf, slot),
2952                                            size);
2953
2954                         btrfs_release_path(path);
2955                         path->leave_spinning = 0;
2956
2957                         memcpy(&new_key, &key, sizeof(new_key));
2958                         new_key.objectid = btrfs_ino(inode);
2959                         if (off <= key.offset)
2960                                 new_key.offset = key.offset + destoff - off;
2961                         else
2962                                 new_key.offset = destoff;
2963
2964                         /*
2965                          * 1 - adjusting old extent (we may have to split it)
2966                          * 1 - add new extent
2967                          * 1 - inode update
2968                          */
2969                         trans = btrfs_start_transaction(root, 3);
2970                         if (IS_ERR(trans)) {
2971                                 ret = PTR_ERR(trans);
2972                                 goto out;
2973                         }
2974
2975                         if (type == BTRFS_FILE_EXTENT_REG ||
2976                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2977                                 /*
2978                                  *    a  | --- range to clone ---|  b
2979                                  * | ------------- extent ------------- |
2980                                  */
2981
2982                                 /* substract range b */
2983                                 if (key.offset + datal > off + len)
2984                                         datal = off + len - key.offset;
2985
2986                                 /* substract range a */
2987                                 if (off > key.offset) {
2988                                         datao += off - key.offset;
2989                                         datal -= off - key.offset;
2990                                 }
2991
2992                                 ret = btrfs_drop_extents(trans, root, inode,
2993                                                          new_key.offset,
2994                                                          new_key.offset + datal,
2995                                                          1);
2996                                 if (ret) {
2997                                         btrfs_abort_transaction(trans, root,
2998                                                                 ret);
2999                                         btrfs_end_transaction(trans, root);
3000                                         goto out;
3001                                 }
3002
3003                                 ret = btrfs_insert_empty_item(trans, root, path,
3004                                                               &new_key, size);
3005                                 if (ret) {
3006                                         btrfs_abort_transaction(trans, root,
3007                                                                 ret);
3008                                         btrfs_end_transaction(trans, root);
3009                                         goto out;
3010                                 }
3011
3012                                 leaf = path->nodes[0];
3013                                 slot = path->slots[0];
3014                                 write_extent_buffer(leaf, buf,
3015                                             btrfs_item_ptr_offset(leaf, slot),
3016                                             size);
3017
3018                                 extent = btrfs_item_ptr(leaf, slot,
3019                                                 struct btrfs_file_extent_item);
3020
3021                                 /* disko == 0 means it's a hole */
3022                                 if (!disko)
3023                                         datao = 0;
3024
3025                                 btrfs_set_file_extent_offset(leaf, extent,
3026                                                              datao);
3027                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3028                                                                 datal);
3029                                 if (disko) {
3030                                         inode_add_bytes(inode, datal);
3031                                         ret = btrfs_inc_extent_ref(trans, root,
3032                                                         disko, diskl, 0,
3033                                                         root->root_key.objectid,
3034                                                         btrfs_ino(inode),
3035                                                         new_key.offset - datao,
3036                                                         0);
3037                                         if (ret) {
3038                                                 btrfs_abort_transaction(trans,
3039                                                                         root,
3040                                                                         ret);
3041                                                 btrfs_end_transaction(trans,
3042                                                                       root);
3043                                                 goto out;
3044
3045                                         }
3046                                 }
3047                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3048                                 u64 skip = 0;
3049                                 u64 trim = 0;
3050                                 if (off > key.offset) {
3051                                         skip = off - key.offset;
3052                                         new_key.offset += skip;
3053                                 }
3054
3055                                 if (key.offset + datal > off + len)
3056                                         trim = key.offset + datal - (off + len);
3057
3058                                 if (comp && (skip || trim)) {
3059                                         ret = -EINVAL;
3060                                         btrfs_end_transaction(trans, root);
3061                                         goto out;
3062                                 }
3063                                 size -= skip + trim;
3064                                 datal -= skip + trim;
3065
3066                                 ret = btrfs_drop_extents(trans, root, inode,
3067                                                          new_key.offset,
3068                                                          new_key.offset + datal,
3069                                                          1);
3070                                 if (ret) {
3071                                         btrfs_abort_transaction(trans, root,
3072                                                                 ret);
3073                                         btrfs_end_transaction(trans, root);
3074                                         goto out;
3075                                 }
3076
3077                                 ret = btrfs_insert_empty_item(trans, root, path,
3078                                                               &new_key, size);
3079                                 if (ret) {
3080                                         btrfs_abort_transaction(trans, root,
3081                                                                 ret);
3082                                         btrfs_end_transaction(trans, root);
3083                                         goto out;
3084                                 }
3085
3086                                 if (skip) {
3087                                         u32 start =
3088                                           btrfs_file_extent_calc_inline_size(0);
3089                                         memmove(buf+start, buf+start+skip,
3090                                                 datal);
3091                                 }
3092
3093                                 leaf = path->nodes[0];
3094                                 slot = path->slots[0];
3095                                 write_extent_buffer(leaf, buf,
3096                                             btrfs_item_ptr_offset(leaf, slot),
3097                                             size);
3098                                 inode_add_bytes(inode, datal);
3099                         }
3100
3101                         btrfs_mark_buffer_dirty(leaf);
3102                         btrfs_release_path(path);
3103
3104                         inode_inc_iversion(inode);
3105                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3106
3107                         /*
3108                          * we round up to the block size at eof when
3109                          * determining which extents to clone above,
3110                          * but shouldn't round up the file size
3111                          */
3112                         endoff = new_key.offset + datal;
3113                         if (endoff > destoff+olen)
3114                                 endoff = destoff+olen;
3115                         if (endoff > inode->i_size)
3116                                 btrfs_i_size_write(inode, endoff);
3117
3118                         ret = btrfs_update_inode(trans, root, inode);
3119                         if (ret) {
3120                                 btrfs_abort_transaction(trans, root, ret);
3121                                 btrfs_end_transaction(trans, root);
3122                                 goto out;
3123                         }
3124                         ret = btrfs_end_transaction(trans, root);
3125                 }
3126                 btrfs_release_path(path);
3127                 key.offset++;
3128         }
3129         ret = 0;
3130
3131 out:
3132         btrfs_release_path(path);
3133         btrfs_free_path(path);
3134         vfree(buf);
3135         return ret;
3136 }
3137
3138 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3139                                        u64 off, u64 olen, u64 destoff)
3140 {
3141         struct inode *inode = file_inode(file);
3142         struct btrfs_root *root = BTRFS_I(inode)->root;
3143         struct fd src_file;
3144         struct inode *src;
3145         int ret;
3146         u64 len = olen;
3147         u64 bs = root->fs_info->sb->s_blocksize;
3148         int same_inode = 0;
3149
3150         /*
3151          * TODO:
3152          * - split compressed inline extents.  annoying: we need to
3153          *   decompress into destination's address_space (the file offset
3154          *   may change, so source mapping won't do), then recompress (or
3155          *   otherwise reinsert) a subrange.
3156          * - allow ranges within the same file to be cloned (provided
3157          *   they don't overlap)?
3158          */
3159
3160         /* the destination must be opened for writing */
3161         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3162                 return -EINVAL;
3163
3164         if (btrfs_root_readonly(root))
3165                 return -EROFS;
3166
3167         ret = mnt_want_write_file(file);
3168         if (ret)
3169                 return ret;
3170
3171         src_file = fdget(srcfd);
3172         if (!src_file.file) {
3173                 ret = -EBADF;
3174                 goto out_drop_write;
3175         }
3176
3177         ret = -EXDEV;
3178         if (src_file.file->f_path.mnt != file->f_path.mnt)
3179                 goto out_fput;
3180
3181         src = file_inode(src_file.file);
3182
3183         ret = -EINVAL;
3184         if (src == inode)
3185                 same_inode = 1;
3186
3187         /* the src must be open for reading */
3188         if (!(src_file.file->f_mode & FMODE_READ))
3189                 goto out_fput;
3190
3191         /* don't make the dst file partly checksummed */
3192         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3193             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3194                 goto out_fput;
3195
3196         ret = -EISDIR;
3197         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3198                 goto out_fput;
3199
3200         ret = -EXDEV;
3201         if (src->i_sb != inode->i_sb)
3202                 goto out_fput;
3203
3204         if (!same_inode) {
3205                 if (inode < src) {
3206                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3207                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3208                 } else {
3209                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3210                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3211                 }
3212         } else {
3213                 mutex_lock(&src->i_mutex);
3214         }
3215
3216         /* determine range to clone */
3217         ret = -EINVAL;
3218         if (off + len > src->i_size || off + len < off)
3219                 goto out_unlock;
3220         if (len == 0)
3221                 olen = len = src->i_size - off;
3222         /* if we extend to eof, continue to block boundary */
3223         if (off + len == src->i_size)
3224                 len = ALIGN(src->i_size, bs) - off;
3225
3226         /* verify the end result is block aligned */
3227         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3228             !IS_ALIGNED(destoff, bs))
3229                 goto out_unlock;
3230
3231         /* verify if ranges are overlapped within the same file */
3232         if (same_inode) {
3233                 if (destoff + len > off && destoff < off + len)
3234                         goto out_unlock;
3235         }
3236
3237         if (destoff > inode->i_size) {
3238                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3239                 if (ret)
3240                         goto out_unlock;
3241         }
3242
3243         /* truncate page cache pages from target inode range */
3244         truncate_inode_pages_range(&inode->i_data, destoff,
3245                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3246
3247         lock_extent_range(src, off, len);
3248
3249         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3250
3251         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3252 out_unlock:
3253         if (!same_inode) {
3254                 if (inode < src) {
3255                         mutex_unlock(&src->i_mutex);
3256                         mutex_unlock(&inode->i_mutex);
3257                 } else {
3258                         mutex_unlock(&inode->i_mutex);
3259                         mutex_unlock(&src->i_mutex);
3260                 }
3261         } else {
3262                 mutex_unlock(&src->i_mutex);
3263         }
3264 out_fput:
3265         fdput(src_file);
3266 out_drop_write:
3267         mnt_drop_write_file(file);
3268         return ret;
3269 }
3270
3271 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3272 {
3273         struct btrfs_ioctl_clone_range_args args;
3274
3275         if (copy_from_user(&args, argp, sizeof(args)))
3276                 return -EFAULT;
3277         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3278                                  args.src_length, args.dest_offset);
3279 }
3280
3281 /*
3282  * there are many ways the trans_start and trans_end ioctls can lead
3283  * to deadlocks.  They should only be used by applications that
3284  * basically own the machine, and have a very in depth understanding
3285  * of all the possible deadlocks and enospc problems.
3286  */
3287 static long btrfs_ioctl_trans_start(struct file *file)
3288 {
3289         struct inode *inode = file_inode(file);
3290         struct btrfs_root *root = BTRFS_I(inode)->root;
3291         struct btrfs_trans_handle *trans;
3292         int ret;
3293
3294         ret = -EPERM;
3295         if (!capable(CAP_SYS_ADMIN))
3296                 goto out;
3297
3298         ret = -EINPROGRESS;
3299         if (file->private_data)
3300                 goto out;
3301
3302         ret = -EROFS;
3303         if (btrfs_root_readonly(root))
3304                 goto out;
3305
3306         ret = mnt_want_write_file(file);
3307         if (ret)
3308                 goto out;
3309
3310         atomic_inc(&root->fs_info->open_ioctl_trans);
3311
3312         ret = -ENOMEM;
3313         trans = btrfs_start_ioctl_transaction(root);
3314         if (IS_ERR(trans))
3315                 goto out_drop;
3316
3317         file->private_data = trans;
3318         return 0;
3319
3320 out_drop:
3321         atomic_dec(&root->fs_info->open_ioctl_trans);
3322         mnt_drop_write_file(file);
3323 out:
3324         return ret;
3325 }
3326
3327 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3328 {
3329         struct inode *inode = file_inode(file);
3330         struct btrfs_root *root = BTRFS_I(inode)->root;
3331         struct btrfs_root *new_root;
3332         struct btrfs_dir_item *di;
3333         struct btrfs_trans_handle *trans;
3334         struct btrfs_path *path;
3335         struct btrfs_key location;
3336         struct btrfs_disk_key disk_key;
3337         u64 objectid = 0;
3338         u64 dir_id;
3339         int ret;
3340
3341         if (!capable(CAP_SYS_ADMIN))
3342                 return -EPERM;
3343
3344         ret = mnt_want_write_file(file);
3345         if (ret)
3346                 return ret;
3347
3348         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3349                 ret = -EFAULT;
3350                 goto out;
3351         }
3352
3353         if (!objectid)
3354                 objectid = BTRFS_FS_TREE_OBJECTID;
3355
3356         location.objectid = objectid;
3357         location.type = BTRFS_ROOT_ITEM_KEY;
3358         location.offset = (u64)-1;
3359
3360         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3361         if (IS_ERR(new_root)) {
3362                 ret = PTR_ERR(new_root);
3363                 goto out;
3364         }
3365
3366         path = btrfs_alloc_path();
3367         if (!path) {
3368                 ret = -ENOMEM;
3369                 goto out;
3370         }
3371         path->leave_spinning = 1;
3372
3373         trans = btrfs_start_transaction(root, 1);
3374         if (IS_ERR(trans)) {
3375                 btrfs_free_path(path);
3376                 ret = PTR_ERR(trans);
3377                 goto out;
3378         }
3379
3380         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3381         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3382                                    dir_id, "default", 7, 1);
3383         if (IS_ERR_OR_NULL(di)) {
3384                 btrfs_free_path(path);
3385                 btrfs_end_transaction(trans, root);
3386                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3387                            "item, this isn't going to work");
3388                 ret = -ENOENT;
3389                 goto out;
3390         }
3391
3392         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3393         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3394         btrfs_mark_buffer_dirty(path->nodes[0]);
3395         btrfs_free_path(path);
3396
3397         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3398         btrfs_end_transaction(trans, root);
3399 out:
3400         mnt_drop_write_file(file);
3401         return ret;
3402 }
3403
3404 void btrfs_get_block_group_info(struct list_head *groups_list,
3405                                 struct btrfs_ioctl_space_info *space)
3406 {
3407         struct btrfs_block_group_cache *block_group;
3408
3409         space->total_bytes = 0;
3410         space->used_bytes = 0;
3411         space->flags = 0;
3412         list_for_each_entry(block_group, groups_list, list) {
3413                 space->flags = block_group->flags;
3414                 space->total_bytes += block_group->key.offset;
3415                 space->used_bytes +=
3416                         btrfs_block_group_used(&block_group->item);
3417         }
3418 }
3419
3420 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3421 {
3422         struct btrfs_ioctl_space_args space_args;
3423         struct btrfs_ioctl_space_info space;
3424         struct btrfs_ioctl_space_info *dest;
3425         struct btrfs_ioctl_space_info *dest_orig;
3426         struct btrfs_ioctl_space_info __user *user_dest;
3427         struct btrfs_space_info *info;
3428         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3429                        BTRFS_BLOCK_GROUP_SYSTEM,
3430                        BTRFS_BLOCK_GROUP_METADATA,
3431                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3432         int num_types = 4;
3433         int alloc_size;
3434         int ret = 0;
3435         u64 slot_count = 0;
3436         int i, c;
3437
3438         if (copy_from_user(&space_args,
3439                            (struct btrfs_ioctl_space_args __user *)arg,
3440                            sizeof(space_args)))
3441                 return -EFAULT;
3442
3443         for (i = 0; i < num_types; i++) {
3444                 struct btrfs_space_info *tmp;
3445
3446                 info = NULL;
3447                 rcu_read_lock();
3448                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3449                                         list) {
3450                         if (tmp->flags == types[i]) {
3451                                 info = tmp;
3452                                 break;
3453                         }
3454                 }
3455                 rcu_read_unlock();
3456
3457                 if (!info)
3458                         continue;
3459
3460                 down_read(&info->groups_sem);
3461                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3462                         if (!list_empty(&info->block_groups[c]))
3463                                 slot_count++;
3464                 }
3465                 up_read(&info->groups_sem);
3466         }
3467
3468         /* space_slots == 0 means they are asking for a count */
3469         if (space_args.space_slots == 0) {
3470                 space_args.total_spaces = slot_count;
3471                 goto out;
3472         }
3473
3474         slot_count = min_t(u64, space_args.space_slots, slot_count);
3475
3476         alloc_size = sizeof(*dest) * slot_count;
3477
3478         /* we generally have at most 6 or so space infos, one for each raid
3479          * level.  So, a whole page should be more than enough for everyone
3480          */
3481         if (alloc_size > PAGE_CACHE_SIZE)
3482                 return -ENOMEM;
3483
3484         space_args.total_spaces = 0;
3485         dest = kmalloc(alloc_size, GFP_NOFS);
3486         if (!dest)
3487                 return -ENOMEM;
3488         dest_orig = dest;
3489
3490         /* now we have a buffer to copy into */
3491         for (i = 0; i < num_types; i++) {
3492                 struct btrfs_space_info *tmp;
3493
3494                 if (!slot_count)
3495                         break;
3496
3497                 info = NULL;
3498                 rcu_read_lock();
3499                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3500                                         list) {
3501                         if (tmp->flags == types[i]) {
3502                                 info = tmp;
3503                                 break;
3504                         }
3505                 }
3506                 rcu_read_unlock();
3507
3508                 if (!info)
3509                         continue;
3510                 down_read(&info->groups_sem);
3511                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3512                         if (!list_empty(&info->block_groups[c])) {
3513                                 btrfs_get_block_group_info(
3514                                         &info->block_groups[c], &space);
3515                                 memcpy(dest, &space, sizeof(space));
3516                                 dest++;
3517                                 space_args.total_spaces++;
3518                                 slot_count--;
3519                         }
3520                         if (!slot_count)
3521                                 break;
3522                 }
3523                 up_read(&info->groups_sem);
3524         }
3525
3526         user_dest = (struct btrfs_ioctl_space_info __user *)
3527                 (arg + sizeof(struct btrfs_ioctl_space_args));
3528
3529         if (copy_to_user(user_dest, dest_orig, alloc_size))
3530                 ret = -EFAULT;
3531
3532         kfree(dest_orig);
3533 out:
3534         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3535                 ret = -EFAULT;
3536
3537         return ret;
3538 }
3539
3540 /*
3541  * there are many ways the trans_start and trans_end ioctls can lead
3542  * to deadlocks.  They should only be used by applications that
3543  * basically own the machine, and have a very in depth understanding
3544  * of all the possible deadlocks and enospc problems.
3545  */
3546 long btrfs_ioctl_trans_end(struct file *file)
3547 {
3548         struct inode *inode = file_inode(file);
3549         struct btrfs_root *root = BTRFS_I(inode)->root;
3550         struct btrfs_trans_handle *trans;
3551
3552         trans = file->private_data;
3553         if (!trans)
3554                 return -EINVAL;
3555         file->private_data = NULL;
3556
3557         btrfs_end_transaction(trans, root);
3558
3559         atomic_dec(&root->fs_info->open_ioctl_trans);
3560
3561         mnt_drop_write_file(file);
3562         return 0;
3563 }
3564
3565 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3566                                             void __user *argp)
3567 {
3568         struct btrfs_trans_handle *trans;
3569         u64 transid;
3570         int ret;
3571
3572         trans = btrfs_attach_transaction_barrier(root);
3573         if (IS_ERR(trans)) {
3574                 if (PTR_ERR(trans) != -ENOENT)
3575                         return PTR_ERR(trans);
3576
3577                 /* No running transaction, don't bother */
3578                 transid = root->fs_info->last_trans_committed;
3579                 goto out;
3580         }
3581         transid = trans->transid;
3582         ret = btrfs_commit_transaction_async(trans, root, 0);
3583         if (ret) {
3584                 btrfs_end_transaction(trans, root);
3585                 return ret;
3586         }
3587 out:
3588         if (argp)
3589                 if (copy_to_user(argp, &transid, sizeof(transid)))
3590                         return -EFAULT;
3591         return 0;
3592 }
3593
3594 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3595                                            void __user *argp)
3596 {
3597         u64 transid;
3598
3599         if (argp) {
3600                 if (copy_from_user(&transid, argp, sizeof(transid)))
3601                         return -EFAULT;
3602         } else {
3603                 transid = 0;  /* current trans */
3604         }
3605         return btrfs_wait_for_commit(root, transid);
3606 }
3607
3608 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3609 {
3610         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3611         struct btrfs_ioctl_scrub_args *sa;
3612         int ret;
3613
3614         if (!capable(CAP_SYS_ADMIN))
3615                 return -EPERM;
3616
3617         sa = memdup_user(arg, sizeof(*sa));
3618         if (IS_ERR(sa))
3619                 return PTR_ERR(sa);
3620
3621         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3622                 ret = mnt_want_write_file(file);
3623                 if (ret)
3624                         goto out;
3625         }
3626
3627         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3628                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3629                               0);
3630
3631         if (copy_to_user(arg, sa, sizeof(*sa)))
3632                 ret = -EFAULT;
3633
3634         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3635                 mnt_drop_write_file(file);
3636 out:
3637         kfree(sa);
3638         return ret;
3639 }
3640
3641 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3642 {
3643         if (!capable(CAP_SYS_ADMIN))
3644                 return -EPERM;
3645
3646         return btrfs_scrub_cancel(root->fs_info);
3647 }
3648
3649 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3650                                        void __user *arg)
3651 {
3652         struct btrfs_ioctl_scrub_args *sa;
3653         int ret;
3654
3655         if (!capable(CAP_SYS_ADMIN))
3656                 return -EPERM;
3657
3658         sa = memdup_user(arg, sizeof(*sa));
3659         if (IS_ERR(sa))
3660                 return PTR_ERR(sa);
3661
3662         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3663
3664         if (copy_to_user(arg, sa, sizeof(*sa)))
3665                 ret = -EFAULT;
3666
3667         kfree(sa);
3668         return ret;
3669 }
3670
3671 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3672                                       void __user *arg)
3673 {
3674         struct btrfs_ioctl_get_dev_stats *sa;
3675         int ret;
3676
3677         sa = memdup_user(arg, sizeof(*sa));
3678         if (IS_ERR(sa))
3679                 return PTR_ERR(sa);
3680
3681         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3682                 kfree(sa);
3683                 return -EPERM;
3684         }
3685
3686         ret = btrfs_get_dev_stats(root, sa);
3687
3688         if (copy_to_user(arg, sa, sizeof(*sa)))
3689                 ret = -EFAULT;
3690
3691         kfree(sa);
3692         return ret;
3693 }
3694
3695 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3696 {
3697         struct btrfs_ioctl_dev_replace_args *p;
3698         int ret;
3699
3700         if (!capable(CAP_SYS_ADMIN))
3701                 return -EPERM;
3702
3703         p = memdup_user(arg, sizeof(*p));
3704         if (IS_ERR(p))
3705                 return PTR_ERR(p);
3706
3707         switch (p->cmd) {
3708         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3709                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
3710                         ret = -EROFS;
3711                         goto out;
3712                 }
3713                 if (atomic_xchg(
3714                         &root->fs_info->mutually_exclusive_operation_running,
3715                         1)) {
3716                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3717                 } else {
3718                         ret = btrfs_dev_replace_start(root, p);
3719                         atomic_set(
3720                          &root->fs_info->mutually_exclusive_operation_running,
3721                          0);
3722                 }
3723                 break;
3724         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3725                 btrfs_dev_replace_status(root->fs_info, p);
3726                 ret = 0;
3727                 break;
3728         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3729                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3730                 break;
3731         default:
3732                 ret = -EINVAL;
3733                 break;
3734         }
3735
3736         if (copy_to_user(arg, p, sizeof(*p)))
3737                 ret = -EFAULT;
3738 out:
3739         kfree(p);
3740         return ret;
3741 }
3742
3743 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3744 {
3745         int ret = 0;
3746         int i;
3747         u64 rel_ptr;
3748         int size;
3749         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3750         struct inode_fs_paths *ipath = NULL;
3751         struct btrfs_path *path;
3752
3753         if (!capable(CAP_DAC_READ_SEARCH))
3754                 return -EPERM;
3755
3756         path = btrfs_alloc_path();
3757         if (!path) {
3758                 ret = -ENOMEM;
3759                 goto out;
3760         }
3761
3762         ipa = memdup_user(arg, sizeof(*ipa));
3763         if (IS_ERR(ipa)) {
3764                 ret = PTR_ERR(ipa);
3765                 ipa = NULL;
3766                 goto out;
3767         }
3768
3769         size = min_t(u32, ipa->size, 4096);
3770         ipath = init_ipath(size, root, path);
3771         if (IS_ERR(ipath)) {
3772                 ret = PTR_ERR(ipath);
3773                 ipath = NULL;
3774                 goto out;
3775         }
3776
3777         ret = paths_from_inode(ipa->inum, ipath);
3778         if (ret < 0)
3779                 goto out;
3780
3781         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3782                 rel_ptr = ipath->fspath->val[i] -
3783                           (u64)(unsigned long)ipath->fspath->val;
3784                 ipath->fspath->val[i] = rel_ptr;
3785         }
3786
3787         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3788                            (void *)(unsigned long)ipath->fspath, size);
3789         if (ret) {
3790                 ret = -EFAULT;
3791                 goto out;
3792         }
3793
3794 out:
3795         btrfs_free_path(path);
3796         free_ipath(ipath);
3797         kfree(ipa);
3798
3799         return ret;
3800 }
3801
3802 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3803 {
3804         struct btrfs_data_container *inodes = ctx;
3805         const size_t c = 3 * sizeof(u64);
3806
3807         if (inodes->bytes_left >= c) {
3808                 inodes->bytes_left -= c;
3809                 inodes->val[inodes->elem_cnt] = inum;
3810                 inodes->val[inodes->elem_cnt + 1] = offset;
3811                 inodes->val[inodes->elem_cnt + 2] = root;
3812                 inodes->elem_cnt += 3;
3813         } else {
3814                 inodes->bytes_missing += c - inodes->bytes_left;
3815                 inodes->bytes_left = 0;
3816                 inodes->elem_missed += 3;
3817         }
3818
3819         return 0;
3820 }
3821
3822 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3823                                         void __user *arg)
3824 {
3825         int ret = 0;
3826         int size;
3827         struct btrfs_ioctl_logical_ino_args *loi;
3828         struct btrfs_data_container *inodes = NULL;
3829         struct btrfs_path *path = NULL;
3830
3831         if (!capable(CAP_SYS_ADMIN))
3832                 return -EPERM;
3833
3834         loi = memdup_user(arg, sizeof(*loi));
3835         if (IS_ERR(loi)) {
3836                 ret = PTR_ERR(loi);
3837                 loi = NULL;
3838                 goto out;
3839         }
3840
3841         path = btrfs_alloc_path();
3842         if (!path) {
3843                 ret = -ENOMEM;
3844                 goto out;
3845         }
3846
3847         size = min_t(u32, loi->size, 64 * 1024);
3848         inodes = init_data_container(size);
3849         if (IS_ERR(inodes)) {
3850                 ret = PTR_ERR(inodes);
3851                 inodes = NULL;
3852                 goto out;
3853         }
3854
3855         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3856                                           build_ino_list, inodes);
3857         if (ret == -EINVAL)
3858                 ret = -ENOENT;
3859         if (ret < 0)
3860                 goto out;
3861
3862         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3863                            (void *)(unsigned long)inodes, size);
3864         if (ret)
3865                 ret = -EFAULT;
3866
3867 out:
3868         btrfs_free_path(path);
3869         vfree(inodes);
3870         kfree(loi);
3871
3872         return ret;
3873 }
3874
3875 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3876                                struct btrfs_ioctl_balance_args *bargs)
3877 {
3878         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3879
3880         bargs->flags = bctl->flags;
3881
3882         if (atomic_read(&fs_info->balance_running))
3883                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3884         if (atomic_read(&fs_info->balance_pause_req))
3885                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3886         if (atomic_read(&fs_info->balance_cancel_req))
3887                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3888
3889         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3890         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3891         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3892
3893         if (lock) {
3894                 spin_lock(&fs_info->balance_lock);
3895                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3896                 spin_unlock(&fs_info->balance_lock);
3897         } else {
3898                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3899         }
3900 }
3901
3902 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3903 {
3904         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3905         struct btrfs_fs_info *fs_info = root->fs_info;
3906         struct btrfs_ioctl_balance_args *bargs;
3907         struct btrfs_balance_control *bctl;
3908         bool need_unlock; /* for mut. excl. ops lock */
3909         int ret;
3910
3911         if (!capable(CAP_SYS_ADMIN))
3912                 return -EPERM;
3913
3914         ret = mnt_want_write_file(file);
3915         if (ret)
3916                 return ret;
3917
3918 again:
3919         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
3920                 mutex_lock(&fs_info->volume_mutex);
3921                 mutex_lock(&fs_info->balance_mutex);
3922                 need_unlock = true;
3923                 goto locked;
3924         }
3925
3926         /*
3927          * mut. excl. ops lock is locked.  Three possibilites:
3928          *   (1) some other op is running
3929          *   (2) balance is running
3930          *   (3) balance is paused -- special case (think resume)
3931          */
3932         mutex_lock(&fs_info->balance_mutex);
3933         if (fs_info->balance_ctl) {
3934                 /* this is either (2) or (3) */
3935                 if (!atomic_read(&fs_info->balance_running)) {
3936                         mutex_unlock(&fs_info->balance_mutex);
3937                         if (!mutex_trylock(&fs_info->volume_mutex))
3938                                 goto again;
3939                         mutex_lock(&fs_info->balance_mutex);
3940
3941                         if (fs_info->balance_ctl &&
3942                             !atomic_read(&fs_info->balance_running)) {
3943                                 /* this is (3) */
3944                                 need_unlock = false;
3945                                 goto locked;
3946                         }
3947
3948                         mutex_unlock(&fs_info->balance_mutex);
3949                         mutex_unlock(&fs_info->volume_mutex);
3950                         goto again;
3951                 } else {
3952                         /* this is (2) */
3953                         mutex_unlock(&fs_info->balance_mutex);
3954                         ret = -EINPROGRESS;
3955                         goto out;
3956                 }
3957         } else {
3958                 /* this is (1) */
3959                 mutex_unlock(&fs_info->balance_mutex);
3960                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3961                 goto out;
3962         }
3963
3964 locked:
3965         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
3966
3967         if (arg) {
3968                 bargs = memdup_user(arg, sizeof(*bargs));
3969                 if (IS_ERR(bargs)) {
3970                         ret = PTR_ERR(bargs);
3971                         goto out_unlock;
3972                 }
3973
3974                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3975                         if (!fs_info->balance_ctl) {
3976                                 ret = -ENOTCONN;
3977                                 goto out_bargs;
3978                         }
3979
3980                         bctl = fs_info->balance_ctl;
3981                         spin_lock(&fs_info->balance_lock);
3982                         bctl->flags |= BTRFS_BALANCE_RESUME;
3983                         spin_unlock(&fs_info->balance_lock);
3984
3985                         goto do_balance;
3986                 }
3987         } else {
3988                 bargs = NULL;
3989         }
3990
3991         if (fs_info->balance_ctl) {
3992                 ret = -EINPROGRESS;
3993                 goto out_bargs;
3994         }
3995
3996         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3997         if (!bctl) {
3998                 ret = -ENOMEM;
3999                 goto out_bargs;
4000         }
4001
4002         bctl->fs_info = fs_info;
4003         if (arg) {
4004                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4005                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4006                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4007
4008                 bctl->flags = bargs->flags;
4009         } else {
4010                 /* balance everything - no filters */
4011                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4012         }
4013
4014 do_balance:
4015         /*
4016          * Ownership of bctl and mutually_exclusive_operation_running
4017          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4018          * or, if restriper was paused all the way until unmount, in
4019          * free_fs_info.  mutually_exclusive_operation_running is
4020          * cleared in __cancel_balance.
4021          */
4022         need_unlock = false;
4023
4024         ret = btrfs_balance(bctl, bargs);
4025
4026         if (arg) {
4027                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4028                         ret = -EFAULT;
4029         }
4030
4031 out_bargs:
4032         kfree(bargs);
4033 out_unlock:
4034         mutex_unlock(&fs_info->balance_mutex);
4035         mutex_unlock(&fs_info->volume_mutex);
4036         if (need_unlock)
4037                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4038 out:
4039         mnt_drop_write_file(file);
4040         return ret;
4041 }
4042
4043 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4044 {
4045         if (!capable(CAP_SYS_ADMIN))
4046                 return -EPERM;
4047
4048         switch (cmd) {
4049         case BTRFS_BALANCE_CTL_PAUSE:
4050                 return btrfs_pause_balance(root->fs_info);
4051         case BTRFS_BALANCE_CTL_CANCEL:
4052                 return btrfs_cancel_balance(root->fs_info);
4053         }
4054
4055         return -EINVAL;
4056 }
4057
4058 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4059                                          void __user *arg)
4060 {
4061         struct btrfs_fs_info *fs_info = root->fs_info;
4062         struct btrfs_ioctl_balance_args *bargs;
4063         int ret = 0;
4064
4065         if (!capable(CAP_SYS_ADMIN))
4066                 return -EPERM;
4067
4068         mutex_lock(&fs_info->balance_mutex);
4069         if (!fs_info->balance_ctl) {
4070                 ret = -ENOTCONN;
4071                 goto out;
4072         }
4073
4074         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4075         if (!bargs) {
4076                 ret = -ENOMEM;
4077                 goto out;
4078         }
4079
4080         update_ioctl_balance_args(fs_info, 1, bargs);
4081
4082         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4083                 ret = -EFAULT;
4084
4085         kfree(bargs);
4086 out:
4087         mutex_unlock(&fs_info->balance_mutex);
4088         return ret;
4089 }
4090
4091 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4092 {
4093         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4094         struct btrfs_ioctl_quota_ctl_args *sa;
4095         struct btrfs_trans_handle *trans = NULL;
4096         int ret;
4097         int err;
4098
4099         if (!capable(CAP_SYS_ADMIN))
4100                 return -EPERM;
4101
4102         ret = mnt_want_write_file(file);
4103         if (ret)
4104                 return ret;
4105
4106         sa = memdup_user(arg, sizeof(*sa));
4107         if (IS_ERR(sa)) {
4108                 ret = PTR_ERR(sa);
4109                 goto drop_write;
4110         }
4111
4112         down_write(&root->fs_info->subvol_sem);
4113         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4114         if (IS_ERR(trans)) {
4115                 ret = PTR_ERR(trans);
4116                 goto out;
4117         }
4118
4119         switch (sa->cmd) {
4120         case BTRFS_QUOTA_CTL_ENABLE:
4121                 ret = btrfs_quota_enable(trans, root->fs_info);
4122                 break;
4123         case BTRFS_QUOTA_CTL_DISABLE:
4124                 ret = btrfs_quota_disable(trans, root->fs_info);
4125                 break;
4126         default:
4127                 ret = -EINVAL;
4128                 break;
4129         }
4130
4131         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4132         if (err && !ret)
4133                 ret = err;
4134 out:
4135         kfree(sa);
4136         up_write(&root->fs_info->subvol_sem);
4137 drop_write:
4138         mnt_drop_write_file(file);
4139         return ret;
4140 }
4141
4142 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4143 {
4144         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4145         struct btrfs_ioctl_qgroup_assign_args *sa;
4146         struct btrfs_trans_handle *trans;
4147         int ret;
4148         int err;
4149
4150         if (!capable(CAP_SYS_ADMIN))
4151                 return -EPERM;
4152
4153         ret = mnt_want_write_file(file);
4154         if (ret)
4155                 return ret;
4156
4157         sa = memdup_user(arg, sizeof(*sa));
4158         if (IS_ERR(sa)) {
4159                 ret = PTR_ERR(sa);
4160                 goto drop_write;
4161         }
4162
4163         trans = btrfs_join_transaction(root);
4164         if (IS_ERR(trans)) {
4165                 ret = PTR_ERR(trans);
4166                 goto out;
4167         }
4168
4169         /* FIXME: check if the IDs really exist */
4170         if (sa->assign) {
4171                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4172                                                 sa->src, sa->dst);
4173         } else {
4174                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4175                                                 sa->src, sa->dst);
4176         }
4177
4178         err = btrfs_end_transaction(trans, root);
4179         if (err && !ret)
4180                 ret = err;
4181
4182 out:
4183         kfree(sa);
4184 drop_write:
4185         mnt_drop_write_file(file);
4186         return ret;
4187 }
4188
4189 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4190 {
4191         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4192         struct btrfs_ioctl_qgroup_create_args *sa;
4193         struct btrfs_trans_handle *trans;
4194         int ret;
4195         int err;
4196
4197         if (!capable(CAP_SYS_ADMIN))
4198                 return -EPERM;
4199
4200         ret = mnt_want_write_file(file);
4201         if (ret)
4202                 return ret;
4203
4204         sa = memdup_user(arg, sizeof(*sa));
4205         if (IS_ERR(sa)) {
4206                 ret = PTR_ERR(sa);
4207                 goto drop_write;
4208         }
4209
4210         if (!sa->qgroupid) {
4211                 ret = -EINVAL;
4212                 goto out;
4213         }
4214
4215         trans = btrfs_join_transaction(root);
4216         if (IS_ERR(trans)) {
4217                 ret = PTR_ERR(trans);
4218                 goto out;
4219         }
4220
4221         /* FIXME: check if the IDs really exist */
4222         if (sa->create) {
4223                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4224                                           NULL);
4225         } else {
4226                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4227         }
4228
4229         err = btrfs_end_transaction(trans, root);
4230         if (err && !ret)
4231                 ret = err;
4232
4233 out:
4234         kfree(sa);
4235 drop_write:
4236         mnt_drop_write_file(file);
4237         return ret;
4238 }
4239
4240 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4241 {
4242         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4243         struct btrfs_ioctl_qgroup_limit_args *sa;
4244         struct btrfs_trans_handle *trans;
4245         int ret;
4246         int err;
4247         u64 qgroupid;
4248
4249         if (!capable(CAP_SYS_ADMIN))
4250                 return -EPERM;
4251
4252         ret = mnt_want_write_file(file);
4253         if (ret)
4254                 return ret;
4255
4256         sa = memdup_user(arg, sizeof(*sa));
4257         if (IS_ERR(sa)) {
4258                 ret = PTR_ERR(sa);
4259                 goto drop_write;
4260         }
4261
4262         trans = btrfs_join_transaction(root);
4263         if (IS_ERR(trans)) {
4264                 ret = PTR_ERR(trans);
4265                 goto out;
4266         }
4267
4268         qgroupid = sa->qgroupid;
4269         if (!qgroupid) {
4270                 /* take the current subvol as qgroup */
4271                 qgroupid = root->root_key.objectid;
4272         }
4273
4274         /* FIXME: check if the IDs really exist */
4275         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4276
4277         err = btrfs_end_transaction(trans, root);
4278         if (err && !ret)
4279                 ret = err;
4280
4281 out:
4282         kfree(sa);
4283 drop_write:
4284         mnt_drop_write_file(file);
4285         return ret;
4286 }
4287
4288 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4289 {
4290         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4291         struct btrfs_ioctl_quota_rescan_args *qsa;
4292         int ret;
4293
4294         if (!capable(CAP_SYS_ADMIN))
4295                 return -EPERM;
4296
4297         ret = mnt_want_write_file(file);
4298         if (ret)
4299                 return ret;
4300
4301         qsa = memdup_user(arg, sizeof(*qsa));
4302         if (IS_ERR(qsa)) {
4303                 ret = PTR_ERR(qsa);
4304                 goto drop_write;
4305         }
4306
4307         if (qsa->flags) {
4308                 ret = -EINVAL;
4309                 goto out;
4310         }
4311
4312         ret = btrfs_qgroup_rescan(root->fs_info);
4313
4314 out:
4315         kfree(qsa);
4316 drop_write:
4317         mnt_drop_write_file(file);
4318         return ret;
4319 }
4320
4321 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4322 {
4323         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4324         struct btrfs_ioctl_quota_rescan_args *qsa;
4325         int ret = 0;
4326
4327         if (!capable(CAP_SYS_ADMIN))
4328                 return -EPERM;
4329
4330         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4331         if (!qsa)
4332                 return -ENOMEM;
4333
4334         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4335                 qsa->flags = 1;
4336                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4337         }
4338
4339         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4340                 ret = -EFAULT;
4341
4342         kfree(qsa);
4343         return ret;
4344 }
4345
4346 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4347 {
4348         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4349
4350         if (!capable(CAP_SYS_ADMIN))
4351                 return -EPERM;
4352
4353         return btrfs_qgroup_wait_for_completion(root->fs_info);
4354 }
4355
4356 static long btrfs_ioctl_set_received_subvol(struct file *file,
4357                                             void __user *arg)
4358 {
4359         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4360         struct inode *inode = file_inode(file);
4361         struct btrfs_root *root = BTRFS_I(inode)->root;
4362         struct btrfs_root_item *root_item = &root->root_item;
4363         struct btrfs_trans_handle *trans;
4364         struct timespec ct = CURRENT_TIME;
4365         int ret = 0;
4366         int received_uuid_changed;
4367
4368         if (!inode_owner_or_capable(inode))
4369                 return -EPERM;
4370
4371         ret = mnt_want_write_file(file);
4372         if (ret < 0)
4373                 return ret;
4374
4375         down_write(&root->fs_info->subvol_sem);
4376
4377         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4378                 ret = -EINVAL;
4379                 goto out;
4380         }
4381
4382         if (btrfs_root_readonly(root)) {
4383                 ret = -EROFS;
4384                 goto out;
4385         }
4386
4387         sa = memdup_user(arg, sizeof(*sa));
4388         if (IS_ERR(sa)) {
4389                 ret = PTR_ERR(sa);
4390                 sa = NULL;
4391                 goto out;
4392         }
4393
4394         /*
4395          * 1 - root item
4396          * 2 - uuid items (received uuid + subvol uuid)
4397          */
4398         trans = btrfs_start_transaction(root, 3);
4399         if (IS_ERR(trans)) {
4400                 ret = PTR_ERR(trans);
4401                 trans = NULL;
4402                 goto out;
4403         }
4404
4405         sa->rtransid = trans->transid;
4406         sa->rtime.sec = ct.tv_sec;
4407         sa->rtime.nsec = ct.tv_nsec;
4408
4409         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4410                                        BTRFS_UUID_SIZE);
4411         if (received_uuid_changed &&
4412             !btrfs_is_empty_uuid(root_item->received_uuid))
4413                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4414                                     root_item->received_uuid,
4415                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4416                                     root->root_key.objectid);
4417         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4418         btrfs_set_root_stransid(root_item, sa->stransid);
4419         btrfs_set_root_rtransid(root_item, sa->rtransid);
4420         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4421         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4422         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4423         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4424
4425         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4426                                 &root->root_key, &root->root_item);
4427         if (ret < 0) {
4428                 btrfs_end_transaction(trans, root);
4429                 goto out;
4430         }
4431         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4432                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4433                                           sa->uuid,
4434                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4435                                           root->root_key.objectid);
4436                 if (ret < 0 && ret != -EEXIST) {
4437                         btrfs_abort_transaction(trans, root, ret);
4438                         goto out;
4439                 }
4440         }
4441         ret = btrfs_commit_transaction(trans, root);
4442         if (ret < 0) {
4443                 btrfs_abort_transaction(trans, root, ret);
4444                 goto out;
4445         }
4446
4447         ret = copy_to_user(arg, sa, sizeof(*sa));
4448         if (ret)
4449                 ret = -EFAULT;
4450
4451 out:
4452         kfree(sa);
4453         up_write(&root->fs_info->subvol_sem);
4454         mnt_drop_write_file(file);
4455         return ret;
4456 }
4457
4458 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4459 {
4460         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4461         size_t len;
4462         int ret;
4463         char label[BTRFS_LABEL_SIZE];
4464
4465         spin_lock(&root->fs_info->super_lock);
4466         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4467         spin_unlock(&root->fs_info->super_lock);
4468
4469         len = strnlen(label, BTRFS_LABEL_SIZE);
4470
4471         if (len == BTRFS_LABEL_SIZE) {
4472                 btrfs_warn(root->fs_info,
4473                         "label is too long, return the first %zu bytes", --len);
4474         }
4475
4476         ret = copy_to_user(arg, label, len);
4477
4478         return ret ? -EFAULT : 0;
4479 }
4480
4481 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4482 {
4483         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4484         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4485         struct btrfs_trans_handle *trans;
4486         char label[BTRFS_LABEL_SIZE];
4487         int ret;
4488
4489         if (!capable(CAP_SYS_ADMIN))
4490                 return -EPERM;
4491
4492         if (copy_from_user(label, arg, sizeof(label)))
4493                 return -EFAULT;
4494
4495         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4496                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
4497                        BTRFS_LABEL_SIZE - 1);
4498                 return -EINVAL;
4499         }
4500
4501         ret = mnt_want_write_file(file);
4502         if (ret)
4503                 return ret;
4504
4505         trans = btrfs_start_transaction(root, 0);
4506         if (IS_ERR(trans)) {
4507                 ret = PTR_ERR(trans);
4508                 goto out_unlock;
4509         }
4510
4511         spin_lock(&root->fs_info->super_lock);
4512         strcpy(super_block->label, label);
4513         spin_unlock(&root->fs_info->super_lock);
4514         ret = btrfs_commit_transaction(trans, root);
4515
4516 out_unlock:
4517         mnt_drop_write_file(file);
4518         return ret;
4519 }
4520
4521 #define INIT_FEATURE_FLAGS(suffix) \
4522         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4523           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4524           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4525
4526 static int btrfs_ioctl_get_supported_features(struct file *file,
4527                                               void __user *arg)
4528 {
4529         static struct btrfs_ioctl_feature_flags features[3] = {
4530                 INIT_FEATURE_FLAGS(SUPP),
4531                 INIT_FEATURE_FLAGS(SAFE_SET),
4532                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4533         };
4534
4535         if (copy_to_user(arg, &features, sizeof(features)))
4536                 return -EFAULT;
4537
4538         return 0;
4539 }
4540
4541 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
4542 {
4543         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4544         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4545         struct btrfs_ioctl_feature_flags features;
4546
4547         features.compat_flags = btrfs_super_compat_flags(super_block);
4548         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4549         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4550
4551         if (copy_to_user(arg, &features, sizeof(features)))
4552                 return -EFAULT;
4553
4554         return 0;
4555 }
4556
4557 static int check_feature_bits(struct btrfs_root *root,
4558                               enum btrfs_feature_set set,
4559                               u64 change_mask, u64 flags, u64 supported_flags,
4560                               u64 safe_set, u64 safe_clear)
4561 {
4562         const char *type = btrfs_feature_set_names[set];
4563         char *names;
4564         u64 disallowed, unsupported;
4565         u64 set_mask = flags & change_mask;
4566         u64 clear_mask = ~flags & change_mask;
4567
4568         unsupported = set_mask & ~supported_flags;
4569         if (unsupported) {
4570                 names = btrfs_printable_features(set, unsupported);
4571                 if (names) {
4572                         btrfs_warn(root->fs_info,
4573                            "this kernel does not support the %s feature bit%s",
4574                            names, strchr(names, ',') ? "s" : "");
4575                         kfree(names);
4576                 } else
4577                         btrfs_warn(root->fs_info,
4578                            "this kernel does not support %s bits 0x%llx",
4579                            type, unsupported);
4580                 return -EOPNOTSUPP;
4581         }
4582
4583         disallowed = set_mask & ~safe_set;
4584         if (disallowed) {
4585                 names = btrfs_printable_features(set, disallowed);
4586                 if (names) {
4587                         btrfs_warn(root->fs_info,
4588                            "can't set the %s feature bit%s while mounted",
4589                            names, strchr(names, ',') ? "s" : "");
4590                         kfree(names);
4591                 } else
4592                         btrfs_warn(root->fs_info,
4593                            "can't set %s bits 0x%llx while mounted",
4594                            type, disallowed);
4595                 return -EPERM;
4596         }
4597
4598         disallowed = clear_mask & ~safe_clear;
4599         if (disallowed) {
4600                 names = btrfs_printable_features(set, disallowed);
4601                 if (names) {
4602                         btrfs_warn(root->fs_info,
4603                            "can't clear the %s feature bit%s while mounted",
4604                            names, strchr(names, ',') ? "s" : "");
4605                         kfree(names);
4606                 } else
4607                         btrfs_warn(root->fs_info,
4608                            "can't clear %s bits 0x%llx while mounted",
4609                            type, disallowed);
4610                 return -EPERM;
4611         }
4612
4613         return 0;
4614 }
4615
4616 #define check_feature(root, change_mask, flags, mask_base)      \
4617 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
4618                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4619                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4620                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4621
4622 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4623 {
4624         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4625         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4626         struct btrfs_ioctl_feature_flags flags[2];
4627         struct btrfs_trans_handle *trans;
4628         u64 newflags;
4629         int ret;
4630
4631         if (!capable(CAP_SYS_ADMIN))
4632                 return -EPERM;
4633
4634         if (copy_from_user(flags, arg, sizeof(flags)))
4635                 return -EFAULT;
4636
4637         /* Nothing to do */
4638         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4639             !flags[0].incompat_flags)
4640                 return 0;
4641
4642         ret = check_feature(root, flags[0].compat_flags,
4643                             flags[1].compat_flags, COMPAT);
4644         if (ret)
4645                 return ret;
4646
4647         ret = check_feature(root, flags[0].compat_ro_flags,
4648                             flags[1].compat_ro_flags, COMPAT_RO);
4649         if (ret)
4650                 return ret;
4651
4652         ret = check_feature(root, flags[0].incompat_flags,
4653                             flags[1].incompat_flags, INCOMPAT);
4654         if (ret)
4655                 return ret;
4656
4657         trans = btrfs_start_transaction(root, 0);
4658         if (IS_ERR(trans))
4659                 return PTR_ERR(trans);
4660
4661         spin_lock(&root->fs_info->super_lock);
4662         newflags = btrfs_super_compat_flags(super_block);
4663         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4664         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4665         btrfs_set_super_compat_flags(super_block, newflags);
4666
4667         newflags = btrfs_super_compat_ro_flags(super_block);
4668         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4669         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4670         btrfs_set_super_compat_ro_flags(super_block, newflags);
4671
4672         newflags = btrfs_super_incompat_flags(super_block);
4673         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4674         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4675         btrfs_set_super_incompat_flags(super_block, newflags);
4676         spin_unlock(&root->fs_info->super_lock);
4677
4678         return btrfs_commit_transaction(trans, root);
4679 }
4680
4681 long btrfs_ioctl(struct file *file, unsigned int
4682                 cmd, unsigned long arg)
4683 {
4684         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4685         void __user *argp = (void __user *)arg;
4686
4687         switch (cmd) {
4688         case FS_IOC_GETFLAGS:
4689                 return btrfs_ioctl_getflags(file, argp);
4690         case FS_IOC_SETFLAGS:
4691                 return btrfs_ioctl_setflags(file, argp);
4692         case FS_IOC_GETVERSION:
4693                 return btrfs_ioctl_getversion(file, argp);
4694         case FITRIM:
4695                 return btrfs_ioctl_fitrim(file, argp);
4696         case BTRFS_IOC_SNAP_CREATE:
4697                 return btrfs_ioctl_snap_create(file, argp, 0);
4698         case BTRFS_IOC_SNAP_CREATE_V2:
4699                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4700         case BTRFS_IOC_SUBVOL_CREATE:
4701                 return btrfs_ioctl_snap_create(file, argp, 1);
4702         case BTRFS_IOC_SUBVOL_CREATE_V2:
4703                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4704         case BTRFS_IOC_SNAP_DESTROY:
4705                 return btrfs_ioctl_snap_destroy(file, argp);
4706         case BTRFS_IOC_SUBVOL_GETFLAGS:
4707                 return btrfs_ioctl_subvol_getflags(file, argp);
4708         case BTRFS_IOC_SUBVOL_SETFLAGS:
4709                 return btrfs_ioctl_subvol_setflags(file, argp);
4710         case BTRFS_IOC_DEFAULT_SUBVOL:
4711                 return btrfs_ioctl_default_subvol(file, argp);
4712         case BTRFS_IOC_DEFRAG:
4713                 return btrfs_ioctl_defrag(file, NULL);
4714         case BTRFS_IOC_DEFRAG_RANGE:
4715                 return btrfs_ioctl_defrag(file, argp);
4716         case BTRFS_IOC_RESIZE:
4717                 return btrfs_ioctl_resize(file, argp);
4718         case BTRFS_IOC_ADD_DEV:
4719                 return btrfs_ioctl_add_dev(root, argp);
4720         case BTRFS_IOC_RM_DEV:
4721                 return btrfs_ioctl_rm_dev(file, argp);
4722         case BTRFS_IOC_FS_INFO:
4723                 return btrfs_ioctl_fs_info(root, argp);
4724         case BTRFS_IOC_DEV_INFO:
4725                 return btrfs_ioctl_dev_info(root, argp);
4726         case BTRFS_IOC_BALANCE:
4727                 return btrfs_ioctl_balance(file, NULL);
4728         case BTRFS_IOC_CLONE:
4729                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4730         case BTRFS_IOC_CLONE_RANGE:
4731                 return btrfs_ioctl_clone_range(file, argp);
4732         case BTRFS_IOC_TRANS_START:
4733                 return btrfs_ioctl_trans_start(file);
4734         case BTRFS_IOC_TRANS_END:
4735                 return btrfs_ioctl_trans_end(file);
4736         case BTRFS_IOC_TREE_SEARCH:
4737                 return btrfs_ioctl_tree_search(file, argp);
4738         case BTRFS_IOC_INO_LOOKUP:
4739                 return btrfs_ioctl_ino_lookup(file, argp);
4740         case BTRFS_IOC_INO_PATHS:
4741                 return btrfs_ioctl_ino_to_path(root, argp);
4742         case BTRFS_IOC_LOGICAL_INO:
4743                 return btrfs_ioctl_logical_to_ino(root, argp);
4744         case BTRFS_IOC_SPACE_INFO:
4745                 return btrfs_ioctl_space_info(root, argp);
4746         case BTRFS_IOC_SYNC: {
4747                 int ret;
4748
4749                 ret = btrfs_start_delalloc_roots(root->fs_info, 0);
4750                 if (ret)
4751                         return ret;
4752                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
4753                 return ret;
4754         }
4755         case BTRFS_IOC_START_SYNC:
4756                 return btrfs_ioctl_start_sync(root, argp);
4757         case BTRFS_IOC_WAIT_SYNC:
4758                 return btrfs_ioctl_wait_sync(root, argp);
4759         case BTRFS_IOC_SCRUB:
4760                 return btrfs_ioctl_scrub(file, argp);
4761         case BTRFS_IOC_SCRUB_CANCEL:
4762                 return btrfs_ioctl_scrub_cancel(root, argp);
4763         case BTRFS_IOC_SCRUB_PROGRESS:
4764                 return btrfs_ioctl_scrub_progress(root, argp);
4765         case BTRFS_IOC_BALANCE_V2:
4766                 return btrfs_ioctl_balance(file, argp);
4767         case BTRFS_IOC_BALANCE_CTL:
4768                 return btrfs_ioctl_balance_ctl(root, arg);
4769         case BTRFS_IOC_BALANCE_PROGRESS:
4770                 return btrfs_ioctl_balance_progress(root, argp);
4771         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4772                 return btrfs_ioctl_set_received_subvol(file, argp);
4773         case BTRFS_IOC_SEND:
4774                 return btrfs_ioctl_send(file, argp);
4775         case BTRFS_IOC_GET_DEV_STATS:
4776                 return btrfs_ioctl_get_dev_stats(root, argp);
4777         case BTRFS_IOC_QUOTA_CTL:
4778                 return btrfs_ioctl_quota_ctl(file, argp);
4779         case BTRFS_IOC_QGROUP_ASSIGN:
4780                 return btrfs_ioctl_qgroup_assign(file, argp);
4781         case BTRFS_IOC_QGROUP_CREATE:
4782                 return btrfs_ioctl_qgroup_create(file, argp);
4783         case BTRFS_IOC_QGROUP_LIMIT:
4784                 return btrfs_ioctl_qgroup_limit(file, argp);
4785         case BTRFS_IOC_QUOTA_RESCAN:
4786                 return btrfs_ioctl_quota_rescan(file, argp);
4787         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4788                 return btrfs_ioctl_quota_rescan_status(file, argp);
4789         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4790                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4791         case BTRFS_IOC_DEV_REPLACE:
4792                 return btrfs_ioctl_dev_replace(root, argp);
4793         case BTRFS_IOC_GET_FSLABEL:
4794                 return btrfs_ioctl_get_fslabel(file, argp);
4795         case BTRFS_IOC_SET_FSLABEL:
4796                 return btrfs_ioctl_set_fslabel(file, argp);
4797         case BTRFS_IOC_FILE_EXTENT_SAME:
4798                 return btrfs_ioctl_file_extent_same(file, argp);
4799         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4800                 return btrfs_ioctl_get_supported_features(file, argp);
4801         case BTRFS_IOC_GET_FEATURES:
4802                 return btrfs_ioctl_get_features(file, argp);
4803         case BTRFS_IOC_SET_FEATURES:
4804                 return btrfs_ioctl_set_features(file, argp);
4805         }
4806
4807         return -ENOTTY;
4808 }