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 (btrfs_root_readonly(root))
 196                 return -EROFS;
 197
 198         if (copy_from_user(&flags, arg, sizeof(flags)))
 199                 return -EFAULT;
 200
 201         ret = check_flags(flags);
 202         if (ret)
 203                 return ret;
 204
 205         if (!inode_owner_or_capable(inode))
 206                 return -EACCES;
 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                                     int 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         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1198         int 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         ret = mnt_want_write_file(file);
1701         if (ret)
1702                 goto out;
1703
1704         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1705                 ret = -EINVAL;
1706                 goto out_drop_write;
1707         }
1708
1709         if (copy_from_user(&flags, arg, sizeof(flags))) {
1710                 ret = -EFAULT;
1711                 goto out_drop_write;
1712         }
1713
1714         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1715                 ret = -EINVAL;
1716                 goto out_drop_write;
1717         }
1718
1719         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1720                 ret = -EOPNOTSUPP;
1721                 goto out_drop_write;
1722         }
1723
1724         if (!inode_owner_or_capable(inode)) {
1725                 ret = -EACCES;
1726                 goto out_drop_write;
1727         }
1728
1729         down_write(&root->fs_info->subvol_sem);
1730
1731         /* nothing to do */
1732         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1733                 goto out_drop_sem;
1734
1735         root_flags = btrfs_root_flags(&root->root_item);
1736         if (flags & BTRFS_SUBVOL_RDONLY) {
1737                 btrfs_set_root_flags(&root->root_item,
1738                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1739         } else {
1740                 /*
1741                  * Block RO -> RW transition if this subvolume is involved in
1742                  * send
1743                  */
1744                 spin_lock(&root->root_item_lock);
1745                 if (root->send_in_progress == 0) {
1746                         btrfs_set_root_flags(&root->root_item,
1747                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1748                         spin_unlock(&root->root_item_lock);
1749                 } else {
1750                         spin_unlock(&root->root_item_lock);
1751                         btrfs_warn(root->fs_info,
1752                         "Attempt to set subvolume %llu read-write during send",
1753                                         root->root_key.objectid);
1754                         ret = -EPERM;
1755                         goto out_drop_sem;
1756                 }
1757         }
1758
1759         trans = btrfs_start_transaction(root, 1);
1760         if (IS_ERR(trans)) {
1761                 ret = PTR_ERR(trans);
1762                 goto out_reset;
1763         }
1764
1765         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1766                                 &root->root_key, &root->root_item);
1767
1768         btrfs_commit_transaction(trans, root);
1769 out_reset:
1770         if (ret)
1771                 btrfs_set_root_flags(&root->root_item, root_flags);
1772 out_drop_sem:
1773         up_write(&root->fs_info->subvol_sem);
1774 out_drop_write:
1775         mnt_drop_write_file(file);
1776 out:
1777         return ret;
1778 }
1779
1780 /*
1781  * helper to check if the subvolume references other subvolumes
1782  */
1783 static noinline int may_destroy_subvol(struct btrfs_root *root)
1784 {
1785         struct btrfs_path *path;
1786         struct btrfs_dir_item *di;
1787         struct btrfs_key key;
1788         u64 dir_id;
1789         int ret;
1790
1791         path = btrfs_alloc_path();
1792         if (!path)
1793                 return -ENOMEM;
1794
1795         /* Make sure this root isn't set as the default subvol */
1796         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1797         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1798                                    dir_id, "default", 7, 0);
1799         if (di && !IS_ERR(di)) {
1800                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1801                 if (key.objectid == root->root_key.objectid) {
1802                         ret = -ENOTEMPTY;
1803                         goto out;
1804                 }
1805                 btrfs_release_path(path);
1806         }
1807
1808         key.objectid = root->root_key.objectid;
1809         key.type = BTRFS_ROOT_REF_KEY;
1810         key.offset = (u64)-1;
1811
1812         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1813                                 &key, path, 0, 0);
1814         if (ret < 0)
1815                 goto out;
1816         BUG_ON(ret == 0);
1817
1818         ret = 0;
1819         if (path->slots[0] > 0) {
1820                 path->slots[0]--;
1821                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1822                 if (key.objectid == root->root_key.objectid &&
1823                     key.type == BTRFS_ROOT_REF_KEY)
1824                         ret = -ENOTEMPTY;
1825         }
1826 out:
1827         btrfs_free_path(path);
1828         return ret;
1829 }
1830
1831 static noinline int key_in_sk(struct btrfs_key *key,
1832                               struct btrfs_ioctl_search_key *sk)
1833 {
1834         struct btrfs_key test;
1835         int ret;
1836
1837         test.objectid = sk->min_objectid;
1838         test.type = sk->min_type;
1839         test.offset = sk->min_offset;
1840
1841         ret = btrfs_comp_cpu_keys(key, &test);
1842         if (ret < 0)
1843                 return 0;
1844
1845         test.objectid = sk->max_objectid;
1846         test.type = sk->max_type;
1847         test.offset = sk->max_offset;
1848
1849         ret = btrfs_comp_cpu_keys(key, &test);
1850         if (ret > 0)
1851                 return 0;
1852         return 1;
1853 }
1854
1855 static noinline int copy_to_sk(struct btrfs_root *root,
1856                                struct btrfs_path *path,
1857                                struct btrfs_key *key,
1858                                struct btrfs_ioctl_search_key *sk,
1859                                char *buf,
1860                                unsigned long *sk_offset,
1861                                int *num_found)
1862 {
1863         u64 found_transid;
1864         struct extent_buffer *leaf;
1865         struct btrfs_ioctl_search_header sh;
1866         unsigned long item_off;
1867         unsigned long item_len;
1868         int nritems;
1869         int i;
1870         int slot;
1871         int ret = 0;
1872
1873         leaf = path->nodes[0];
1874         slot = path->slots[0];
1875         nritems = btrfs_header_nritems(leaf);
1876
1877         if (btrfs_header_generation(leaf) > sk->max_transid) {
1878                 i = nritems;
1879                 goto advance_key;
1880         }
1881         found_transid = btrfs_header_generation(leaf);
1882
1883         for (i = slot; i < nritems; i++) {
1884                 item_off = btrfs_item_ptr_offset(leaf, i);
1885                 item_len = btrfs_item_size_nr(leaf, i);
1886
1887                 btrfs_item_key_to_cpu(leaf, key, i);
1888                 if (!key_in_sk(key, sk))
1889                         continue;
1890
1891                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1892                         item_len = 0;
1893
1894                 if (sizeof(sh) + item_len + *sk_offset >
1895                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1896                         ret = 1;
1897                         goto overflow;
1898                 }
1899
1900                 sh.objectid = key->objectid;
1901                 sh.offset = key->offset;
1902                 sh.type = key->type;
1903                 sh.len = item_len;
1904                 sh.transid = found_transid;
1905
1906                 /* copy search result header */
1907                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1908                 *sk_offset += sizeof(sh);
1909
1910                 if (item_len) {
1911                         char *p = buf + *sk_offset;
1912                         /* copy the item */
1913                         read_extent_buffer(leaf, p,
1914                                            item_off, item_len);
1915                         *sk_offset += item_len;
1916                 }
1917                 (*num_found)++;
1918
1919                 if (*num_found >= sk->nr_items)
1920                         break;
1921         }
1922 advance_key:
1923         ret = 0;
1924         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1925                 key->offset++;
1926         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1927                 key->offset = 0;
1928                 key->type++;
1929         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1930                 key->offset = 0;
1931                 key->type = 0;
1932                 key->objectid++;
1933         } else
1934                 ret = 1;
1935 overflow:
1936         return ret;
1937 }
1938
1939 static noinline int search_ioctl(struct inode *inode,
1940                                  struct btrfs_ioctl_search_args *args)
1941 {
1942         struct btrfs_root *root;
1943         struct btrfs_key key;
1944         struct btrfs_path *path;
1945         struct btrfs_ioctl_search_key *sk = &args->key;
1946         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1947         int ret;
1948         int num_found = 0;
1949         unsigned long sk_offset = 0;
1950
1951         path = btrfs_alloc_path();
1952         if (!path)
1953                 return -ENOMEM;
1954
1955         if (sk->tree_id == 0) {
1956                 /* search the root of the inode that was passed */
1957                 root = BTRFS_I(inode)->root;
1958         } else {
1959                 key.objectid = sk->tree_id;
1960                 key.type = BTRFS_ROOT_ITEM_KEY;
1961                 key.offset = (u64)-1;
1962                 root = btrfs_read_fs_root_no_name(info, &key);
1963                 if (IS_ERR(root)) {
1964                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
1965                                sk->tree_id);
1966                         btrfs_free_path(path);
1967                         return -ENOENT;
1968                 }
1969         }
1970
1971         key.objectid = sk->min_objectid;
1972         key.type = sk->min_type;
1973         key.offset = sk->min_offset;
1974
1975         path->keep_locks = 1;
1976
1977         while (1) {
1978                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
1979                 if (ret != 0) {
1980                         if (ret > 0)
1981                                 ret = 0;
1982                         goto err;
1983                 }
1984                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1985                                  &sk_offset, &num_found);
1986                 btrfs_release_path(path);
1987                 if (ret || num_found >= sk->nr_items)
1988                         break;
1989
1990         }
1991         ret = 0;
1992 err:
1993         sk->nr_items = num_found;
1994         btrfs_free_path(path);
1995         return ret;
1996 }
1997
1998 static noinline int btrfs_ioctl_tree_search(struct file *file,
1999                                            void __user *argp)
2000 {
2001          struct btrfs_ioctl_search_args *args;
2002          struct inode *inode;
2003          int ret;
2004
2005         if (!capable(CAP_SYS_ADMIN))
2006                 return -EPERM;
2007
2008         args = memdup_user(argp, sizeof(*args));
2009         if (IS_ERR(args))
2010                 return PTR_ERR(args);
2011
2012         inode = file_inode(file);
2013         ret = search_ioctl(inode, args);
2014         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2015                 ret = -EFAULT;
2016         kfree(args);
2017         return ret;
2018 }
2019
2020 /*
2021  * Search INODE_REFs to identify path name of 'dirid' directory
2022  * in a 'tree_id' tree. and sets path name to 'name'.
2023  */
2024 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2025                                 u64 tree_id, u64 dirid, char *name)
2026 {
2027         struct btrfs_root *root;
2028         struct btrfs_key key;
2029         char *ptr;
2030         int ret = -1;
2031         int slot;
2032         int len;
2033         int total_len = 0;
2034         struct btrfs_inode_ref *iref;
2035         struct extent_buffer *l;
2036         struct btrfs_path *path;
2037
2038         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2039                 name[0]='\0';
2040                 return 0;
2041         }
2042
2043         path = btrfs_alloc_path();
2044         if (!path)
2045                 return -ENOMEM;
2046
2047         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2048
2049         key.objectid = tree_id;
2050         key.type = BTRFS_ROOT_ITEM_KEY;
2051         key.offset = (u64)-1;
2052         root = btrfs_read_fs_root_no_name(info, &key);
2053         if (IS_ERR(root)) {
2054                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2055                 ret = -ENOENT;
2056                 goto out;
2057         }
2058
2059         key.objectid = dirid;
2060         key.type = BTRFS_INODE_REF_KEY;
2061         key.offset = (u64)-1;
2062
2063         while (1) {
2064                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2065                 if (ret < 0)
2066                         goto out;
2067                 else if (ret > 0) {
2068                         ret = btrfs_previous_item(root, path, dirid,
2069                                                   BTRFS_INODE_REF_KEY);
2070                         if (ret < 0)
2071                                 goto out;
2072                         else if (ret > 0) {
2073                                 ret = -ENOENT;
2074                                 goto out;
2075                         }
2076                 }
2077
2078                 l = path->nodes[0];
2079                 slot = path->slots[0];
2080                 btrfs_item_key_to_cpu(l, &key, slot);
2081
2082                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2083                 len = btrfs_inode_ref_name_len(l, iref);
2084                 ptr -= len + 1;
2085                 total_len += len + 1;
2086                 if (ptr < name) {
2087                         ret = -ENAMETOOLONG;
2088                         goto out;
2089                 }
2090
2091                 *(ptr + len) = '/';
2092                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2093
2094                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2095                         break;
2096
2097                 btrfs_release_path(path);
2098                 key.objectid = key.offset;
2099                 key.offset = (u64)-1;
2100                 dirid = key.objectid;
2101         }
2102         memmove(name, ptr, total_len);
2103         name[total_len] = '\0';
2104         ret = 0;
2105 out:
2106         btrfs_free_path(path);
2107         return ret;
2108 }
2109
2110 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2111                                            void __user *argp)
2112 {
2113          struct btrfs_ioctl_ino_lookup_args *args;
2114          struct inode *inode;
2115          int ret;
2116
2117         if (!capable(CAP_SYS_ADMIN))
2118                 return -EPERM;
2119
2120         args = memdup_user(argp, sizeof(*args));
2121         if (IS_ERR(args))
2122                 return PTR_ERR(args);
2123
2124         inode = file_inode(file);
2125
2126         if (args->treeid == 0)
2127                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2128
2129         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2130                                         args->treeid, args->objectid,
2131                                         args->name);
2132
2133         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2134                 ret = -EFAULT;
2135
2136         kfree(args);
2137         return ret;
2138 }
2139
2140 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2141                                              void __user *arg)
2142 {
2143         struct dentry *parent = file->f_path.dentry;
2144         struct dentry *dentry;
2145         struct inode *dir = parent->d_inode;
2146         struct inode *inode;
2147         struct btrfs_root *root = BTRFS_I(dir)->root;
2148         struct btrfs_root *dest = NULL;
2149         struct btrfs_ioctl_vol_args *vol_args;
2150         struct btrfs_trans_handle *trans;
2151         struct btrfs_block_rsv block_rsv;
2152         u64 qgroup_reserved;
2153         int namelen;
2154         int ret;
2155         int err = 0;
2156
2157         vol_args = memdup_user(arg, sizeof(*vol_args));
2158         if (IS_ERR(vol_args))
2159                 return PTR_ERR(vol_args);
2160
2161         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2162         namelen = strlen(vol_args->name);
2163         if (strchr(vol_args->name, '/') ||
2164             strncmp(vol_args->name, "..", namelen) == 0) {
2165                 err = -EINVAL;
2166                 goto out;
2167         }
2168
2169         err = mnt_want_write_file(file);
2170         if (err)
2171                 goto out;
2172
2173         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2174         if (err == -EINTR)
2175                 goto out_drop_write;
2176         dentry = lookup_one_len(vol_args->name, parent, namelen);
2177         if (IS_ERR(dentry)) {
2178                 err = PTR_ERR(dentry);
2179                 goto out_unlock_dir;
2180         }
2181
2182         if (!dentry->d_inode) {
2183                 err = -ENOENT;
2184                 goto out_dput;
2185         }
2186
2187         inode = dentry->d_inode;
2188         dest = BTRFS_I(inode)->root;
2189         if (!capable(CAP_SYS_ADMIN)) {
2190                 /*
2191                  * Regular user.  Only allow this with a special mount
2192                  * option, when the user has write+exec access to the
2193                  * subvol root, and when rmdir(2) would have been
2194                  * allowed.
2195                  *
2196                  * Note that this is _not_ check that the subvol is
2197                  * empty or doesn't contain data that we wouldn't
2198                  * otherwise be able to delete.
2199                  *
2200                  * Users who want to delete empty subvols should try
2201                  * rmdir(2).
2202                  */
2203                 err = -EPERM;
2204                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2205                         goto out_dput;
2206
2207                 /*
2208                  * Do not allow deletion if the parent dir is the same
2209                  * as the dir to be deleted.  That means the ioctl
2210                  * must be called on the dentry referencing the root
2211                  * of the subvol, not a random directory contained
2212                  * within it.
2213                  */
2214                 err = -EINVAL;
2215                 if (root == dest)
2216                         goto out_dput;
2217
2218                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2219                 if (err)
2220                         goto out_dput;
2221         }
2222
2223         /* check if subvolume may be deleted by a user */
2224         err = btrfs_may_delete(dir, dentry, 1);
2225         if (err)
2226                 goto out_dput;
2227
2228         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2229                 err = -EINVAL;
2230                 goto out_dput;
2231         }
2232
2233         mutex_lock(&inode->i_mutex);
2234         err = d_invalidate(dentry);
2235         if (err)
2236                 goto out_unlock;
2237
2238         down_write(&root->fs_info->subvol_sem);
2239
2240         err = may_destroy_subvol(dest);
2241         if (err)
2242                 goto out_up_write;
2243
2244         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2245         /*
2246          * One for dir inode, two for dir entries, two for root
2247          * ref/backref.
2248          */
2249         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2250                                                5, &qgroup_reserved, true);
2251         if (err)
2252                 goto out_up_write;
2253
2254         trans = btrfs_start_transaction(root, 0);
2255         if (IS_ERR(trans)) {
2256                 err = PTR_ERR(trans);
2257                 goto out_release;
2258         }
2259         trans->block_rsv = &block_rsv;
2260         trans->bytes_reserved = block_rsv.size;
2261
2262         ret = btrfs_unlink_subvol(trans, root, dir,
2263                                 dest->root_key.objectid,
2264                                 dentry->d_name.name,
2265                                 dentry->d_name.len);
2266         if (ret) {
2267                 err = ret;
2268                 btrfs_abort_transaction(trans, root, ret);
2269                 goto out_end_trans;
2270         }
2271
2272         btrfs_record_root_in_trans(trans, dest);
2273
2274         memset(&dest->root_item.drop_progress, 0,
2275                 sizeof(dest->root_item.drop_progress));
2276         dest->root_item.drop_level = 0;
2277         btrfs_set_root_refs(&dest->root_item, 0);
2278
2279         if (!xchg(&dest->orphan_item_inserted, 1)) {
2280                 ret = btrfs_insert_orphan_item(trans,
2281                                         root->fs_info->tree_root,
2282                                         dest->root_key.objectid);
2283                 if (ret) {
2284                         btrfs_abort_transaction(trans, root, ret);
2285                         err = ret;
2286                         goto out_end_trans;
2287                 }
2288         }
2289
2290         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2291                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2292                                   dest->root_key.objectid);
2293         if (ret && ret != -ENOENT) {
2294                 btrfs_abort_transaction(trans, root, ret);
2295                 err = ret;
2296                 goto out_end_trans;
2297         }
2298         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2299                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2300                                           dest->root_item.received_uuid,
2301                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2302                                           dest->root_key.objectid);
2303                 if (ret && ret != -ENOENT) {
2304                         btrfs_abort_transaction(trans, root, ret);
2305                         err = ret;
2306                         goto out_end_trans;
2307                 }
2308         }
2309
2310 out_end_trans:
2311         trans->block_rsv = NULL;
2312         trans->bytes_reserved = 0;
2313         ret = btrfs_end_transaction(trans, root);
2314         if (ret && !err)
2315                 err = ret;
2316         inode->i_flags |= S_DEAD;
2317 out_release:
2318         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2319 out_up_write:
2320         up_write(&root->fs_info->subvol_sem);
2321 out_unlock:
2322         mutex_unlock(&inode->i_mutex);
2323         if (!err) {
2324                 shrink_dcache_sb(root->fs_info->sb);
2325                 btrfs_invalidate_inodes(dest);
2326                 d_delete(dentry);
2327
2328                 /* the last ref */
2329                 if (dest->cache_inode) {
2330                         iput(dest->cache_inode);
2331                         dest->cache_inode = NULL;
2332                 }
2333         }
2334 out_dput:
2335         dput(dentry);
2336 out_unlock_dir:
2337         mutex_unlock(&dir->i_mutex);
2338 out_drop_write:
2339         mnt_drop_write_file(file);
2340 out:
2341         kfree(vol_args);
2342         return err;
2343 }
2344
2345 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2346 {
2347         struct inode *inode = file_inode(file);
2348         struct btrfs_root *root = BTRFS_I(inode)->root;
2349         struct btrfs_ioctl_defrag_range_args *range;
2350         int ret;
2351
2352         ret = mnt_want_write_file(file);
2353         if (ret)
2354                 return ret;
2355
2356         if (btrfs_root_readonly(root)) {
2357                 ret = -EROFS;
2358                 goto out;
2359         }
2360
2361         switch (inode->i_mode & S_IFMT) {
2362         case S_IFDIR:
2363                 if (!capable(CAP_SYS_ADMIN)) {
2364                         ret = -EPERM;
2365                         goto out;
2366                 }
2367                 ret = btrfs_defrag_root(root);
2368                 if (ret)
2369                         goto out;
2370                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2371                 break;
2372         case S_IFREG:
2373                 if (!(file->f_mode & FMODE_WRITE)) {
2374                         ret = -EINVAL;
2375                         goto out;
2376                 }
2377
2378                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2379                 if (!range) {
2380                         ret = -ENOMEM;
2381                         goto out;
2382                 }
2383
2384                 if (argp) {
2385                         if (copy_from_user(range, argp,
2386                                            sizeof(*range))) {
2387                                 ret = -EFAULT;
2388                                 kfree(range);
2389                                 goto out;
2390                         }
2391                         /* compression requires us to start the IO */
2392                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2393                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2394                                 range->extent_thresh = (u32)-1;
2395                         }
2396                 } else {
2397                         /* the rest are all set to zero by kzalloc */
2398                         range->len = (u64)-1;
2399                 }
2400                 ret = btrfs_defrag_file(file_inode(file), file,
2401                                         range, 0, 0);
2402                 if (ret > 0)
2403                         ret = 0;
2404                 kfree(range);
2405                 break;
2406         default:
2407                 ret = -EINVAL;
2408         }
2409 out:
2410         mnt_drop_write_file(file);
2411         return ret;
2412 }
2413
2414 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2415 {
2416         struct btrfs_ioctl_vol_args *vol_args;
2417         int ret;
2418
2419         if (!capable(CAP_SYS_ADMIN))
2420                 return -EPERM;
2421
2422         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2423                         1)) {
2424                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2425         }
2426
2427         mutex_lock(&root->fs_info->volume_mutex);
2428         vol_args = memdup_user(arg, sizeof(*vol_args));
2429         if (IS_ERR(vol_args)) {
2430                 ret = PTR_ERR(vol_args);
2431                 goto out;
2432         }
2433
2434         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2435         ret = btrfs_init_new_device(root, vol_args->name);
2436
2437         kfree(vol_args);
2438 out:
2439         mutex_unlock(&root->fs_info->volume_mutex);
2440         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2441         return ret;
2442 }
2443
2444 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2445 {
2446         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2447         struct btrfs_ioctl_vol_args *vol_args;
2448         int ret;
2449
2450         if (!capable(CAP_SYS_ADMIN))
2451                 return -EPERM;
2452
2453         ret = mnt_want_write_file(file);
2454         if (ret)
2455                 return ret;
2456
2457         vol_args = memdup_user(arg, sizeof(*vol_args));
2458         if (IS_ERR(vol_args)) {
2459                 ret = PTR_ERR(vol_args);
2460                 goto out;
2461         }
2462
2463         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2464
2465         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2466                         1)) {
2467                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2468                 goto out;
2469         }
2470
2471         mutex_lock(&root->fs_info->volume_mutex);
2472         ret = btrfs_rm_device(root, vol_args->name);
2473         mutex_unlock(&root->fs_info->volume_mutex);
2474         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2475
2476 out:
2477         kfree(vol_args);
2478         mnt_drop_write_file(file);
2479         return ret;
2480 }
2481
2482 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2483 {
2484         struct btrfs_ioctl_fs_info_args *fi_args;
2485         struct btrfs_device *device;
2486         struct btrfs_device *next;
2487         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2488         int ret = 0;
2489
2490         if (!capable(CAP_SYS_ADMIN))
2491                 return -EPERM;
2492
2493         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2494         if (!fi_args)
2495                 return -ENOMEM;
2496
2497         mutex_lock(&fs_devices->device_list_mutex);
2498         fi_args->num_devices = fs_devices->num_devices;
2499         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2500
2501         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2502                 if (device->devid > fi_args->max_id)
2503                         fi_args->max_id = device->devid;
2504         }
2505         mutex_unlock(&fs_devices->device_list_mutex);
2506
2507         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2508                 ret = -EFAULT;
2509
2510         kfree(fi_args);
2511         return ret;
2512 }
2513
2514 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2515 {
2516         struct btrfs_ioctl_dev_info_args *di_args;
2517         struct btrfs_device *dev;
2518         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2519         int ret = 0;
2520         char *s_uuid = NULL;
2521
2522         if (!capable(CAP_SYS_ADMIN))
2523                 return -EPERM;
2524
2525         di_args = memdup_user(arg, sizeof(*di_args));
2526         if (IS_ERR(di_args))
2527                 return PTR_ERR(di_args);
2528
2529         if (!btrfs_is_empty_uuid(di_args->uuid))
2530                 s_uuid = di_args->uuid;
2531
2532         mutex_lock(&fs_devices->device_list_mutex);
2533         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2534
2535         if (!dev) {
2536                 ret = -ENODEV;
2537                 goto out;
2538         }
2539
2540         di_args->devid = dev->devid;
2541         di_args->bytes_used = dev->bytes_used;
2542         di_args->total_bytes = dev->total_bytes;
2543         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2544         if (dev->name) {
2545                 struct rcu_string *name;
2546
2547                 rcu_read_lock();
2548                 name = rcu_dereference(dev->name);
2549                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2550                 rcu_read_unlock();
2551                 di_args->path[sizeof(di_args->path) - 1] = 0;
2552         } else {
2553                 di_args->path[0] = '\0';
2554         }
2555
2556 out:
2557         mutex_unlock(&fs_devices->device_list_mutex);
2558         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2559                 ret = -EFAULT;
2560
2561         kfree(di_args);
2562         return ret;
2563 }
2564
2565 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2566 {
2567         struct page *page;
2568         pgoff_t index;
2569         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2570
2571         index = off >> PAGE_CACHE_SHIFT;
2572
2573         page = grab_cache_page(inode->i_mapping, index);
2574         if (!page)
2575                 return NULL;
2576
2577         if (!PageUptodate(page)) {
2578                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2579                                                  0))
2580                         return NULL;
2581                 lock_page(page);
2582                 if (!PageUptodate(page)) {
2583                         unlock_page(page);
2584                         page_cache_release(page);
2585                         return NULL;
2586                 }
2587         }
2588         unlock_page(page);
2589
2590         return page;
2591 }
2592
2593 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2594 {
2595         /* do any pending delalloc/csum calc on src, one way or
2596            another, and lock file content */
2597         while (1) {
2598                 struct btrfs_ordered_extent *ordered;
2599                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2600                 ordered = btrfs_lookup_first_ordered_extent(inode,
2601                                                             off + len - 1);
2602                 if (!ordered &&
2603                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2604                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2605                         break;
2606                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2607                 if (ordered)
2608                         btrfs_put_ordered_extent(ordered);
2609                 btrfs_wait_ordered_range(inode, off, len);
2610         }
2611 }
2612
2613 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2614                                 struct inode *inode2, u64 loff2, u64 len)
2615 {
2616         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2617         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2618
2619         mutex_unlock(&inode1->i_mutex);
2620         mutex_unlock(&inode2->i_mutex);
2621 }
2622
2623 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2624                               struct inode *inode2, u64 loff2, u64 len)
2625 {
2626         if (inode1 < inode2) {
2627                 swap(inode1, inode2);
2628                 swap(loff1, loff2);
2629         }
2630
2631         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2632         lock_extent_range(inode1, loff1, len);
2633         if (inode1 != inode2) {
2634                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2635                 lock_extent_range(inode2, loff2, len);
2636         }
2637 }
2638
2639 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2640                           u64 dst_loff, u64 len)
2641 {
2642         int ret = 0;
2643         struct page *src_page, *dst_page;
2644         unsigned int cmp_len = PAGE_CACHE_SIZE;
2645         void *addr, *dst_addr;
2646
2647         while (len) {
2648                 if (len < PAGE_CACHE_SIZE)
2649                         cmp_len = len;
2650
2651                 src_page = extent_same_get_page(src, loff);
2652                 if (!src_page)
2653                         return -EINVAL;
2654                 dst_page = extent_same_get_page(dst, dst_loff);
2655                 if (!dst_page) {
2656                         page_cache_release(src_page);
2657                         return -EINVAL;
2658                 }
2659                 addr = kmap_atomic(src_page);
2660                 dst_addr = kmap_atomic(dst_page);
2661
2662                 flush_dcache_page(src_page);
2663                 flush_dcache_page(dst_page);
2664
2665                 if (memcmp(addr, dst_addr, cmp_len))
2666                         ret = BTRFS_SAME_DATA_DIFFERS;
2667
2668                 kunmap_atomic(addr);
2669                 kunmap_atomic(dst_addr);
2670                 page_cache_release(src_page);
2671                 page_cache_release(dst_page);
2672
2673                 if (ret)
2674                         break;
2675
2676                 loff += cmp_len;
2677                 dst_loff += cmp_len;
2678                 len -= cmp_len;
2679         }
2680
2681         return ret;
2682 }
2683
2684 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2685 {
2686         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2687
2688         if (off + len > inode->i_size || off + len < off)
2689                 return -EINVAL;
2690         /* Check that we are block aligned - btrfs_clone() requires this */
2691         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2692                 return -EINVAL;
2693
2694         return 0;
2695 }
2696
2697 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2698                              struct inode *dst, u64 dst_loff)
2699 {
2700         int ret;
2701
2702         /*
2703          * btrfs_clone() can't handle extents in the same file
2704          * yet. Once that works, we can drop this check and replace it
2705          * with a check for the same inode, but overlapping extents.
2706          */
2707         if (src == dst)
2708                 return -EINVAL;
2709
2710         btrfs_double_lock(src, loff, dst, dst_loff, len);
2711
2712         ret = extent_same_check_offsets(src, loff, len);
2713         if (ret)
2714                 goto out_unlock;
2715
2716         ret = extent_same_check_offsets(dst, dst_loff, len);
2717         if (ret)
2718                 goto out_unlock;
2719
2720         /* don't make the dst file partly checksummed */
2721         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2722             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2723                 ret = -EINVAL;
2724                 goto out_unlock;
2725         }
2726
2727         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2728         if (ret == 0)
2729                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2730
2731 out_unlock:
2732         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2733
2734         return ret;
2735 }
2736
2737 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2738
2739 static long btrfs_ioctl_file_extent_same(struct file *file,
2740                                          void __user *argp)
2741 {
2742         struct btrfs_ioctl_same_args tmp;
2743         struct btrfs_ioctl_same_args *same;
2744         struct btrfs_ioctl_same_extent_info *info;
2745         struct inode *src = file->f_dentry->d_inode;
2746         struct file *dst_file = NULL;
2747         struct inode *dst;
2748         u64 off;
2749         u64 len;
2750         int i;
2751         int ret;
2752         unsigned long size;
2753         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2754         bool is_admin = capable(CAP_SYS_ADMIN);
2755
2756         if (!(file->f_mode & FMODE_READ))
2757                 return -EINVAL;
2758
2759         ret = mnt_want_write_file(file);
2760         if (ret)
2761                 return ret;
2762
2763         if (copy_from_user(&tmp,
2764                            (struct btrfs_ioctl_same_args __user *)argp,
2765                            sizeof(tmp))) {
2766                 ret = -EFAULT;
2767                 goto out;
2768         }
2769
2770         size = sizeof(tmp) +
2771                 tmp.dest_count * sizeof(struct btrfs_ioctl_same_extent_info);
2772
2773         same = memdup_user((struct btrfs_ioctl_same_args __user *)argp, size);
2774
2775         if (IS_ERR(same)) {
2776                 ret = PTR_ERR(same);
2777                 goto out;
2778         }
2779
2780         off = same->logical_offset;
2781         len = same->length;
2782
2783         /*
2784          * Limit the total length we will dedupe for each operation.
2785          * This is intended to bound the total time spent in this
2786          * ioctl to something sane.
2787          */
2788         if (len > BTRFS_MAX_DEDUPE_LEN)
2789                 len = BTRFS_MAX_DEDUPE_LEN;
2790
2791         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2792                 /*
2793                  * Btrfs does not support blocksize < page_size. As a
2794                  * result, btrfs_cmp_data() won't correctly handle
2795                  * this situation without an update.
2796                  */
2797                 ret = -EINVAL;
2798                 goto out;
2799         }
2800
2801         ret = -EISDIR;
2802         if (S_ISDIR(src->i_mode))
2803                 goto out;
2804
2805         ret = -EACCES;
2806         if (!S_ISREG(src->i_mode))
2807                 goto out;
2808
2809         /* pre-format output fields to sane values */
2810         for (i = 0; i < same->dest_count; i++) {
2811                 same->info[i].bytes_deduped = 0ULL;
2812                 same->info[i].status = 0;
2813         }
2814
2815         ret = 0;
2816         for (i = 0; i < same->dest_count; i++) {
2817                 info = &same->info[i];
2818
2819                 dst_file = fget(info->fd);
2820                 if (!dst_file) {
2821                         info->status = -EBADF;
2822                         goto next;
2823                 }
2824
2825                 if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2826                         info->status = -EINVAL;
2827                         goto next;
2828                 }
2829
2830                 info->status = -EXDEV;
2831                 if (file->f_path.mnt != dst_file->f_path.mnt)
2832                         goto next;
2833
2834                 dst = dst_file->f_dentry->d_inode;
2835                 if (src->i_sb != dst->i_sb)
2836                         goto next;
2837
2838                 if (S_ISDIR(dst->i_mode)) {
2839                         info->status = -EISDIR;
2840                         goto next;
2841                 }
2842
2843                 if (!S_ISREG(dst->i_mode)) {
2844                         info->status = -EACCES;
2845                         goto next;
2846                 }
2847
2848                 info->status = btrfs_extent_same(src, off, len, dst,
2849                                                 info->logical_offset);
2850                 if (info->status == 0)
2851                         info->bytes_deduped += len;
2852
2853 next:
2854                 if (dst_file)
2855                         fput(dst_file);
2856         }
2857
2858         ret = copy_to_user(argp, same, size);
2859         if (ret)
2860                 ret = -EFAULT;
2861
2862 out:
2863         mnt_drop_write_file(file);
2864         return ret;
2865 }
2866
2867 /**
2868  * btrfs_clone() - clone a range from inode file to another
2869  *
2870  * @src: Inode to clone from
2871  * @inode: Inode to clone to
2872  * @off: Offset within source to start clone from
2873  * @olen: Original length, passed by user, of range to clone
2874  * @olen_aligned: Block-aligned value of olen, extent_same uses
2875  *               identical values here
2876  * @destoff: Offset within @inode to start clone
2877  */
2878 static int btrfs_clone(struct inode *src, struct inode *inode,
2879                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2880 {
2881         struct btrfs_root *root = BTRFS_I(inode)->root;
2882         struct btrfs_path *path = NULL;
2883         struct extent_buffer *leaf;
2884         struct btrfs_trans_handle *trans;
2885         char *buf = NULL;
2886         struct btrfs_key key;
2887         u32 nritems;
2888         int slot;
2889         int ret;
2890         u64 len = olen_aligned;
2891
2892         ret = -ENOMEM;
2893         buf = vmalloc(btrfs_level_size(root, 0));
2894         if (!buf)
2895                 return ret;
2896
2897         path = btrfs_alloc_path();
2898         if (!path) {
2899                 vfree(buf);
2900                 return ret;
2901         }
2902
2903         path->reada = 2;
2904         /* clone data */
2905         key.objectid = btrfs_ino(src);
2906         key.type = BTRFS_EXTENT_DATA_KEY;
2907         key.offset = 0;
2908
2909         while (1) {
2910                 /*
2911                  * note the key will change type as we walk through the
2912                  * tree.
2913                  */
2914                 path->leave_spinning = 1;
2915                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2916                                 0, 0);
2917                 if (ret < 0)
2918                         goto out;
2919
2920                 nritems = btrfs_header_nritems(path->nodes[0]);
2921 process_slot:
2922                 if (path->slots[0] >= nritems) {
2923                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2924                         if (ret < 0)
2925                                 goto out;
2926                         if (ret > 0)
2927                                 break;
2928                         nritems = btrfs_header_nritems(path->nodes[0]);
2929                 }
2930                 leaf = path->nodes[0];
2931                 slot = path->slots[0];
2932
2933                 btrfs_item_key_to_cpu(leaf, &key, slot);
2934                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2935                     key.objectid != btrfs_ino(src))
2936                         break;
2937
2938                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2939                         struct btrfs_file_extent_item *extent;
2940                         int type;
2941                         u32 size;
2942                         struct btrfs_key new_key;
2943                         u64 disko = 0, diskl = 0;
2944                         u64 datao = 0, datal = 0;
2945                         u8 comp;
2946                         u64 endoff;
2947
2948                         extent = btrfs_item_ptr(leaf, slot,
2949                                                 struct btrfs_file_extent_item);
2950                         comp = btrfs_file_extent_compression(leaf, extent);
2951                         type = btrfs_file_extent_type(leaf, extent);
2952                         if (type == BTRFS_FILE_EXTENT_REG ||
2953                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2954                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2955                                                                       extent);
2956                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2957                                                                  extent);
2958                                 datao = btrfs_file_extent_offset(leaf, extent);
2959                                 datal = btrfs_file_extent_num_bytes(leaf,
2960                                                                     extent);
2961                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2962                                 /* take upper bound, may be compressed */
2963                                 datal = btrfs_file_extent_ram_bytes(leaf,
2964                                                                     extent);
2965                         }
2966
2967                         if (key.offset + datal <= off ||
2968                             key.offset >= off + len - 1) {
2969                                 path->slots[0]++;
2970                                 goto process_slot;
2971                         }
2972
2973                         size = btrfs_item_size_nr(leaf, slot);
2974                         read_extent_buffer(leaf, buf,
2975                                            btrfs_item_ptr_offset(leaf, slot),
2976                                            size);
2977
2978                         btrfs_release_path(path);
2979                         path->leave_spinning = 0;
2980
2981                         memcpy(&new_key, &key, sizeof(new_key));
2982                         new_key.objectid = btrfs_ino(inode);
2983                         if (off <= key.offset)
2984                                 new_key.offset = key.offset + destoff - off;
2985                         else
2986                                 new_key.offset = destoff;
2987
2988                         /*
2989                          * 1 - adjusting old extent (we may have to split it)
2990                          * 1 - add new extent
2991                          * 1 - inode update
2992                          */
2993                         trans = btrfs_start_transaction(root, 3);
2994                         if (IS_ERR(trans)) {
2995                                 ret = PTR_ERR(trans);
2996                                 goto out;
2997                         }
2998
2999                         if (type == BTRFS_FILE_EXTENT_REG ||
3000                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3001                                 /*
3002                                  *    a  | --- range to clone ---|  b
3003                                  * | ------------- extent ------------- |
3004                                  */
3005
3006                                 /* substract range b */
3007                                 if (key.offset + datal > off + len)
3008                                         datal = off + len - key.offset;
3009
3010                                 /* substract range a */
3011                                 if (off > key.offset) {
3012                                         datao += off - key.offset;
3013                                         datal -= off - key.offset;
3014                                 }
3015
3016                                 ret = btrfs_drop_extents(trans, root, inode,
3017                                                          new_key.offset,
3018                                                          new_key.offset + datal,
3019                                                          1);
3020                                 if (ret) {
3021                                         btrfs_abort_transaction(trans, root,
3022                                                                 ret);
3023                                         btrfs_end_transaction(trans, root);
3024                                         goto out;
3025                                 }
3026
3027                                 ret = btrfs_insert_empty_item(trans, root, path,
3028                                                               &new_key, size);
3029                                 if (ret) {
3030                                         btrfs_abort_transaction(trans, root,
3031                                                                 ret);
3032                                         btrfs_end_transaction(trans, root);
3033                                         goto out;
3034                                 }
3035
3036                                 leaf = path->nodes[0];
3037                                 slot = path->slots[0];
3038                                 write_extent_buffer(leaf, buf,
3039                                             btrfs_item_ptr_offset(leaf, slot),
3040                                             size);
3041
3042                                 extent = btrfs_item_ptr(leaf, slot,
3043                                                 struct btrfs_file_extent_item);
3044
3045                                 /* disko == 0 means it's a hole */
3046                                 if (!disko)
3047                                         datao = 0;
3048
3049                                 btrfs_set_file_extent_offset(leaf, extent,
3050                                                              datao);
3051                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3052                                                                 datal);
3053                                 if (disko) {
3054                                         inode_add_bytes(inode, datal);
3055                                         ret = btrfs_inc_extent_ref(trans, root,
3056                                                         disko, diskl, 0,
3057                                                         root->root_key.objectid,
3058                                                         btrfs_ino(inode),
3059                                                         new_key.offset - datao,
3060                                                         0);
3061                                         if (ret) {
3062                                                 btrfs_abort_transaction(trans,
3063                                                                         root,
3064                                                                         ret);
3065                                                 btrfs_end_transaction(trans,
3066                                                                       root);
3067                                                 goto out;
3068
3069                                         }
3070                                 }
3071                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3072                                 u64 skip = 0;
3073                                 u64 trim = 0;
3074                                 if (off > key.offset) {
3075                                         skip = off - key.offset;
3076                                         new_key.offset += skip;
3077                                 }
3078
3079                                 if (key.offset + datal > off + len)
3080                                         trim = key.offset + datal - (off + len);
3081
3082                                 if (comp && (skip || trim)) {
3083                                         ret = -EINVAL;
3084                                         btrfs_end_transaction(trans, root);
3085                                         goto out;
3086                                 }
3087                                 size -= skip + trim;
3088                                 datal -= skip + trim;
3089
3090                                 ret = btrfs_drop_extents(trans, root, inode,
3091                                                          new_key.offset,
3092                                                          new_key.offset + datal,
3093                                                          1);
3094                                 if (ret) {
3095                                         btrfs_abort_transaction(trans, root,
3096                                                                 ret);
3097                                         btrfs_end_transaction(trans, root);
3098                                         goto out;
3099                                 }
3100
3101                                 ret = btrfs_insert_empty_item(trans, root, path,
3102                                                               &new_key, size);
3103                                 if (ret) {
3104                                         btrfs_abort_transaction(trans, root,
3105                                                                 ret);
3106                                         btrfs_end_transaction(trans, root);
3107                                         goto out;
3108                                 }
3109
3110                                 if (skip) {
3111                                         u32 start =
3112                                           btrfs_file_extent_calc_inline_size(0);
3113                                         memmove(buf+start, buf+start+skip,
3114                                                 datal);
3115                                 }
3116
3117                                 leaf = path->nodes[0];
3118                                 slot = path->slots[0];
3119                                 write_extent_buffer(leaf, buf,
3120                                             btrfs_item_ptr_offset(leaf, slot),
3121                                             size);
3122                                 inode_add_bytes(inode, datal);
3123                         }
3124
3125                         btrfs_mark_buffer_dirty(leaf);
3126                         btrfs_release_path(path);
3127
3128                         inode_inc_iversion(inode);
3129                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3130
3131                         /*
3132                          * we round up to the block size at eof when
3133                          * determining which extents to clone above,
3134                          * but shouldn't round up the file size
3135                          */
3136                         endoff = new_key.offset + datal;
3137                         if (endoff > destoff+olen)
3138                                 endoff = destoff+olen;
3139                         if (endoff > inode->i_size)
3140                                 btrfs_i_size_write(inode, endoff);
3141
3142                         ret = btrfs_update_inode(trans, root, inode);
3143                         if (ret) {
3144                                 btrfs_abort_transaction(trans, root, ret);
3145                                 btrfs_end_transaction(trans, root);
3146                                 goto out;
3147                         }
3148                         ret = btrfs_end_transaction(trans, root);
3149                 }
3150                 btrfs_release_path(path);
3151                 key.offset++;
3152         }
3153         ret = 0;
3154
3155 out:
3156         btrfs_release_path(path);
3157         btrfs_free_path(path);
3158         vfree(buf);
3159         return ret;
3160 }
3161
3162 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3163                                        u64 off, u64 olen, u64 destoff)
3164 {
3165         struct inode *inode = file_inode(file);
3166         struct btrfs_root *root = BTRFS_I(inode)->root;
3167         struct fd src_file;
3168         struct inode *src;
3169         int ret;
3170         u64 len = olen;
3171         u64 bs = root->fs_info->sb->s_blocksize;
3172         int same_inode = 0;
3173
3174         /*
3175          * TODO:
3176          * - split compressed inline extents.  annoying: we need to
3177          *   decompress into destination's address_space (the file offset
3178          *   may change, so source mapping won't do), then recompress (or
3179          *   otherwise reinsert) a subrange.
3180          * - allow ranges within the same file to be cloned (provided
3181          *   they don't overlap)?
3182          */
3183
3184         /* the destination must be opened for writing */
3185         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3186                 return -EINVAL;
3187
3188         if (btrfs_root_readonly(root))
3189                 return -EROFS;
3190
3191         ret = mnt_want_write_file(file);
3192         if (ret)
3193                 return ret;
3194
3195         src_file = fdget(srcfd);
3196         if (!src_file.file) {
3197                 ret = -EBADF;
3198                 goto out_drop_write;
3199         }
3200
3201         ret = -EXDEV;
3202         if (src_file.file->f_path.mnt != file->f_path.mnt)
3203                 goto out_fput;
3204
3205         src = file_inode(src_file.file);
3206
3207         ret = -EINVAL;
3208         if (src == inode)
3209                 same_inode = 1;
3210
3211         /* the src must be open for reading */
3212         if (!(src_file.file->f_mode & FMODE_READ))
3213                 goto out_fput;
3214
3215         /* don't make the dst file partly checksummed */
3216         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3217             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3218                 goto out_fput;
3219
3220         ret = -EISDIR;
3221         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3222                 goto out_fput;
3223
3224         ret = -EXDEV;
3225         if (src->i_sb != inode->i_sb)
3226                 goto out_fput;
3227
3228         if (!same_inode) {
3229                 if (inode < src) {
3230                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3231                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3232                 } else {
3233                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3234                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3235                 }
3236         } else {
3237                 mutex_lock(&src->i_mutex);
3238         }
3239
3240         /* determine range to clone */
3241         ret = -EINVAL;
3242         if (off + len > src->i_size || off + len < off)
3243                 goto out_unlock;
3244         if (len == 0)
3245                 olen = len = src->i_size - off;
3246         /* if we extend to eof, continue to block boundary */
3247         if (off + len == src->i_size)
3248                 len = ALIGN(src->i_size, bs) - off;
3249
3250         /* verify the end result is block aligned */
3251         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3252             !IS_ALIGNED(destoff, bs))
3253                 goto out_unlock;
3254
3255         /* verify if ranges are overlapped within the same file */
3256         if (same_inode) {
3257                 if (destoff + len > off && destoff < off + len)
3258                         goto out_unlock;
3259         }
3260
3261         if (destoff > inode->i_size) {
3262                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3263                 if (ret)
3264                         goto out_unlock;
3265         }
3266
3267         /* truncate page cache pages from target inode range */
3268         truncate_inode_pages_range(&inode->i_data, destoff,
3269                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3270
3271         lock_extent_range(src, off, len);
3272
3273         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3274
3275         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3276 out_unlock:
3277         if (!same_inode) {
3278                 if (inode < src) {
3279                         mutex_unlock(&src->i_mutex);
3280                         mutex_unlock(&inode->i_mutex);
3281                 } else {
3282                         mutex_unlock(&inode->i_mutex);
3283                         mutex_unlock(&src->i_mutex);
3284                 }
3285         } else {
3286                 mutex_unlock(&src->i_mutex);
3287         }
3288 out_fput:
3289         fdput(src_file);
3290 out_drop_write:
3291         mnt_drop_write_file(file);
3292         return ret;
3293 }
3294
3295 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3296 {
3297         struct btrfs_ioctl_clone_range_args args;
3298
3299         if (copy_from_user(&args, argp, sizeof(args)))
3300                 return -EFAULT;
3301         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3302                                  args.src_length, args.dest_offset);
3303 }
3304
3305 /*
3306  * there are many ways the trans_start and trans_end ioctls can lead
3307  * to deadlocks.  They should only be used by applications that
3308  * basically own the machine, and have a very in depth understanding
3309  * of all the possible deadlocks and enospc problems.
3310  */
3311 static long btrfs_ioctl_trans_start(struct file *file)
3312 {
3313         struct inode *inode = file_inode(file);
3314         struct btrfs_root *root = BTRFS_I(inode)->root;
3315         struct btrfs_trans_handle *trans;
3316         int ret;
3317
3318         ret = -EPERM;
3319         if (!capable(CAP_SYS_ADMIN))
3320                 goto out;
3321
3322         ret = -EINPROGRESS;
3323         if (file->private_data)
3324                 goto out;
3325
3326         ret = -EROFS;
3327         if (btrfs_root_readonly(root))
3328                 goto out;
3329
3330         ret = mnt_want_write_file(file);
3331         if (ret)
3332                 goto out;
3333
3334         atomic_inc(&root->fs_info->open_ioctl_trans);
3335
3336         ret = -ENOMEM;
3337         trans = btrfs_start_ioctl_transaction(root);
3338         if (IS_ERR(trans))
3339                 goto out_drop;
3340
3341         file->private_data = trans;
3342         return 0;
3343
3344 out_drop:
3345         atomic_dec(&root->fs_info->open_ioctl_trans);
3346         mnt_drop_write_file(file);
3347 out:
3348         return ret;
3349 }
3350
3351 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3352 {
3353         struct inode *inode = file_inode(file);
3354         struct btrfs_root *root = BTRFS_I(inode)->root;
3355         struct btrfs_root *new_root;
3356         struct btrfs_dir_item *di;
3357         struct btrfs_trans_handle *trans;
3358         struct btrfs_path *path;
3359         struct btrfs_key location;
3360         struct btrfs_disk_key disk_key;
3361         u64 objectid = 0;
3362         u64 dir_id;
3363         int ret;
3364
3365         if (!capable(CAP_SYS_ADMIN))
3366                 return -EPERM;
3367
3368         ret = mnt_want_write_file(file);
3369         if (ret)
3370                 return ret;
3371
3372         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3373                 ret = -EFAULT;
3374                 goto out;
3375         }
3376
3377         if (!objectid)
3378                 objectid = BTRFS_FS_TREE_OBJECTID;
3379
3380         location.objectid = objectid;
3381         location.type = BTRFS_ROOT_ITEM_KEY;
3382         location.offset = (u64)-1;
3383
3384         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3385         if (IS_ERR(new_root)) {
3386                 ret = PTR_ERR(new_root);
3387                 goto out;
3388         }
3389
3390         path = btrfs_alloc_path();
3391         if (!path) {
3392                 ret = -ENOMEM;
3393                 goto out;
3394         }
3395         path->leave_spinning = 1;
3396
3397         trans = btrfs_start_transaction(root, 1);
3398         if (IS_ERR(trans)) {
3399                 btrfs_free_path(path);
3400                 ret = PTR_ERR(trans);
3401                 goto out;
3402         }
3403
3404         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3405         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3406                                    dir_id, "default", 7, 1);
3407         if (IS_ERR_OR_NULL(di)) {
3408                 btrfs_free_path(path);
3409                 btrfs_end_transaction(trans, root);
3410                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3411                            "item, this isn't going to work");
3412                 ret = -ENOENT;
3413                 goto out;
3414         }
3415
3416         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3417         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3418         btrfs_mark_buffer_dirty(path->nodes[0]);
3419         btrfs_free_path(path);
3420
3421         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3422         btrfs_end_transaction(trans, root);
3423 out:
3424         mnt_drop_write_file(file);
3425         return ret;
3426 }
3427
3428 void btrfs_get_block_group_info(struct list_head *groups_list,
3429                                 struct btrfs_ioctl_space_info *space)
3430 {
3431         struct btrfs_block_group_cache *block_group;
3432
3433         space->total_bytes = 0;
3434         space->used_bytes = 0;
3435         space->flags = 0;
3436         list_for_each_entry(block_group, groups_list, list) {
3437                 space->flags = block_group->flags;
3438                 space->total_bytes += block_group->key.offset;
3439                 space->used_bytes +=
3440                         btrfs_block_group_used(&block_group->item);
3441         }
3442 }
3443
3444 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3445 {
3446         struct btrfs_ioctl_space_args space_args;
3447         struct btrfs_ioctl_space_info space;
3448         struct btrfs_ioctl_space_info *dest;
3449         struct btrfs_ioctl_space_info *dest_orig;
3450         struct btrfs_ioctl_space_info __user *user_dest;
3451         struct btrfs_space_info *info;
3452         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3453                        BTRFS_BLOCK_GROUP_SYSTEM,
3454                        BTRFS_BLOCK_GROUP_METADATA,
3455                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3456         int num_types = 4;
3457         int alloc_size;
3458         int ret = 0;
3459         u64 slot_count = 0;
3460         int i, c;
3461
3462         if (copy_from_user(&space_args,
3463                            (struct btrfs_ioctl_space_args __user *)arg,
3464                            sizeof(space_args)))
3465                 return -EFAULT;
3466
3467         for (i = 0; i < num_types; i++) {
3468                 struct btrfs_space_info *tmp;
3469
3470                 info = NULL;
3471                 rcu_read_lock();
3472                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3473                                         list) {
3474                         if (tmp->flags == types[i]) {
3475                                 info = tmp;
3476                                 break;
3477                         }
3478                 }
3479                 rcu_read_unlock();
3480
3481                 if (!info)
3482                         continue;
3483
3484                 down_read(&info->groups_sem);
3485                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3486                         if (!list_empty(&info->block_groups[c]))
3487                                 slot_count++;
3488                 }
3489                 up_read(&info->groups_sem);
3490         }
3491
3492         /* space_slots == 0 means they are asking for a count */
3493         if (space_args.space_slots == 0) {
3494                 space_args.total_spaces = slot_count;
3495                 goto out;
3496         }
3497
3498         slot_count = min_t(u64, space_args.space_slots, slot_count);
3499
3500         alloc_size = sizeof(*dest) * slot_count;
3501
3502         /* we generally have at most 6 or so space infos, one for each raid
3503          * level.  So, a whole page should be more than enough for everyone
3504          */
3505         if (alloc_size > PAGE_CACHE_SIZE)
3506                 return -ENOMEM;
3507
3508         space_args.total_spaces = 0;
3509         dest = kmalloc(alloc_size, GFP_NOFS);
3510         if (!dest)
3511                 return -ENOMEM;
3512         dest_orig = dest;
3513
3514         /* now we have a buffer to copy into */
3515         for (i = 0; i < num_types; i++) {
3516                 struct btrfs_space_info *tmp;
3517
3518                 if (!slot_count)
3519                         break;
3520
3521                 info = NULL;
3522                 rcu_read_lock();
3523                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3524                                         list) {
3525                         if (tmp->flags == types[i]) {
3526                                 info = tmp;
3527                                 break;
3528                         }
3529                 }
3530                 rcu_read_unlock();
3531
3532                 if (!info)
3533                         continue;
3534                 down_read(&info->groups_sem);
3535                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3536                         if (!list_empty(&info->block_groups[c])) {
3537                                 btrfs_get_block_group_info(
3538                                         &info->block_groups[c], &space);
3539                                 memcpy(dest, &space, sizeof(space));
3540                                 dest++;
3541                                 space_args.total_spaces++;
3542                                 slot_count--;
3543                         }
3544                         if (!slot_count)
3545                                 break;
3546                 }
3547                 up_read(&info->groups_sem);
3548         }
3549
3550         user_dest = (struct btrfs_ioctl_space_info __user *)
3551                 (arg + sizeof(struct btrfs_ioctl_space_args));
3552
3553         if (copy_to_user(user_dest, dest_orig, alloc_size))
3554                 ret = -EFAULT;
3555
3556         kfree(dest_orig);
3557 out:
3558         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3559                 ret = -EFAULT;
3560
3561         return ret;
3562 }
3563
3564 static long btrfs_ioctl_global_rsv(struct btrfs_root *root, void __user *arg)
3565 {
3566         struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3567         u64 reserved;
3568
3569         spin_lock(&block_rsv->lock);
3570         reserved = block_rsv->reserved;
3571         spin_unlock(&block_rsv->lock);
3572
3573         if (arg && copy_to_user(arg, &reserved, sizeof(reserved)))
3574                 return -EFAULT;
3575         return 0;
3576 }
3577
3578 /*
3579  * there are many ways the trans_start and trans_end ioctls can lead
3580  * to deadlocks.  They should only be used by applications that
3581  * basically own the machine, and have a very in depth understanding
3582  * of all the possible deadlocks and enospc problems.
3583  */
3584 long btrfs_ioctl_trans_end(struct file *file)
3585 {
3586         struct inode *inode = file_inode(file);
3587         struct btrfs_root *root = BTRFS_I(inode)->root;
3588         struct btrfs_trans_handle *trans;
3589
3590         trans = file->private_data;
3591         if (!trans)
3592                 return -EINVAL;
3593         file->private_data = NULL;
3594
3595         btrfs_end_transaction(trans, root);
3596
3597         atomic_dec(&root->fs_info->open_ioctl_trans);
3598
3599         mnt_drop_write_file(file);
3600         return 0;
3601 }
3602
3603 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3604                                             void __user *argp)
3605 {
3606         struct btrfs_trans_handle *trans;
3607         u64 transid;
3608         int ret;
3609
3610         trans = btrfs_attach_transaction_barrier(root);
3611         if (IS_ERR(trans)) {
3612                 if (PTR_ERR(trans) != -ENOENT)
3613                         return PTR_ERR(trans);
3614
3615                 /* No running transaction, don't bother */
3616                 transid = root->fs_info->last_trans_committed;
3617                 goto out;
3618         }
3619         transid = trans->transid;
3620         ret = btrfs_commit_transaction_async(trans, root, 0);
3621         if (ret) {
3622                 btrfs_end_transaction(trans, root);
3623                 return ret;
3624         }
3625 out:
3626         if (argp)
3627                 if (copy_to_user(argp, &transid, sizeof(transid)))
3628                         return -EFAULT;
3629         return 0;
3630 }
3631
3632 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3633                                            void __user *argp)
3634 {
3635         u64 transid;
3636
3637         if (argp) {
3638                 if (copy_from_user(&transid, argp, sizeof(transid)))
3639                         return -EFAULT;
3640         } else {
3641                 transid = 0;  /* current trans */
3642         }
3643         return btrfs_wait_for_commit(root, transid);
3644 }
3645
3646 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3647 {
3648         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3649         struct btrfs_ioctl_scrub_args *sa;
3650         int ret;
3651
3652         if (!capable(CAP_SYS_ADMIN))
3653                 return -EPERM;
3654
3655         sa = memdup_user(arg, sizeof(*sa));
3656         if (IS_ERR(sa))
3657                 return PTR_ERR(sa);
3658
3659         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3660                 ret = mnt_want_write_file(file);
3661                 if (ret)
3662                         goto out;
3663         }
3664
3665         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3666                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3667                               0);
3668
3669         if (copy_to_user(arg, sa, sizeof(*sa)))
3670                 ret = -EFAULT;
3671
3672         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3673                 mnt_drop_write_file(file);
3674 out:
3675         kfree(sa);
3676         return ret;
3677 }
3678
3679 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3680 {
3681         if (!capable(CAP_SYS_ADMIN))
3682                 return -EPERM;
3683
3684         return btrfs_scrub_cancel(root->fs_info);
3685 }
3686
3687 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3688                                        void __user *arg)
3689 {
3690         struct btrfs_ioctl_scrub_args *sa;
3691         int ret;
3692
3693         if (!capable(CAP_SYS_ADMIN))
3694                 return -EPERM;
3695
3696         sa = memdup_user(arg, sizeof(*sa));
3697         if (IS_ERR(sa))
3698                 return PTR_ERR(sa);
3699
3700         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3701
3702         if (copy_to_user(arg, sa, sizeof(*sa)))
3703                 ret = -EFAULT;
3704
3705         kfree(sa);
3706         return ret;
3707 }
3708
3709 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3710                                       void __user *arg)
3711 {
3712         struct btrfs_ioctl_get_dev_stats *sa;
3713         int ret;
3714
3715         sa = memdup_user(arg, sizeof(*sa));
3716         if (IS_ERR(sa))
3717                 return PTR_ERR(sa);
3718
3719         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3720                 kfree(sa);
3721                 return -EPERM;
3722         }
3723
3724         ret = btrfs_get_dev_stats(root, sa);
3725
3726         if (copy_to_user(arg, sa, sizeof(*sa)))
3727                 ret = -EFAULT;
3728
3729         kfree(sa);
3730         return ret;
3731 }
3732
3733 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3734 {
3735         struct btrfs_ioctl_dev_replace_args *p;
3736         int ret;
3737
3738         if (!capable(CAP_SYS_ADMIN))
3739                 return -EPERM;
3740
3741         p = memdup_user(arg, sizeof(*p));
3742         if (IS_ERR(p))
3743                 return PTR_ERR(p);
3744
3745         switch (p->cmd) {
3746         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3747                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
3748                         ret = -EROFS;
3749                         goto out;
3750                 }
3751                 if (atomic_xchg(
3752                         &root->fs_info->mutually_exclusive_operation_running,
3753                         1)) {
3754                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3755                 } else {
3756                         ret = btrfs_dev_replace_start(root, p);
3757                         atomic_set(
3758                          &root->fs_info->mutually_exclusive_operation_running,
3759                          0);
3760                 }
3761                 break;
3762         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3763                 btrfs_dev_replace_status(root->fs_info, p);
3764                 ret = 0;
3765                 break;
3766         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3767                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3768                 break;
3769         default:
3770                 ret = -EINVAL;
3771                 break;
3772         }
3773
3774         if (copy_to_user(arg, p, sizeof(*p)))
3775                 ret = -EFAULT;
3776 out:
3777         kfree(p);
3778         return ret;
3779 }
3780
3781 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3782 {
3783         int ret = 0;
3784         int i;
3785         u64 rel_ptr;
3786         int size;
3787         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3788         struct inode_fs_paths *ipath = NULL;
3789         struct btrfs_path *path;
3790
3791         if (!capable(CAP_DAC_READ_SEARCH))
3792                 return -EPERM;
3793
3794         path = btrfs_alloc_path();
3795         if (!path) {
3796                 ret = -ENOMEM;
3797                 goto out;
3798         }
3799
3800         ipa = memdup_user(arg, sizeof(*ipa));
3801         if (IS_ERR(ipa)) {
3802                 ret = PTR_ERR(ipa);
3803                 ipa = NULL;
3804                 goto out;
3805         }
3806
3807         size = min_t(u32, ipa->size, 4096);
3808         ipath = init_ipath(size, root, path);
3809         if (IS_ERR(ipath)) {
3810                 ret = PTR_ERR(ipath);
3811                 ipath = NULL;
3812                 goto out;
3813         }
3814
3815         ret = paths_from_inode(ipa->inum, ipath);
3816         if (ret < 0)
3817                 goto out;
3818
3819         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3820                 rel_ptr = ipath->fspath->val[i] -
3821                           (u64)(unsigned long)ipath->fspath->val;
3822                 ipath->fspath->val[i] = rel_ptr;
3823         }
3824
3825         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3826                            (void *)(unsigned long)ipath->fspath, size);
3827         if (ret) {
3828                 ret = -EFAULT;
3829                 goto out;
3830         }
3831
3832 out:
3833         btrfs_free_path(path);
3834         free_ipath(ipath);
3835         kfree(ipa);
3836
3837         return ret;
3838 }
3839
3840 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3841 {
3842         struct btrfs_data_container *inodes = ctx;
3843         const size_t c = 3 * sizeof(u64);
3844
3845         if (inodes->bytes_left >= c) {
3846                 inodes->bytes_left -= c;
3847                 inodes->val[inodes->elem_cnt] = inum;
3848                 inodes->val[inodes->elem_cnt + 1] = offset;
3849                 inodes->val[inodes->elem_cnt + 2] = root;
3850                 inodes->elem_cnt += 3;
3851         } else {
3852                 inodes->bytes_missing += c - inodes->bytes_left;
3853                 inodes->bytes_left = 0;
3854                 inodes->elem_missed += 3;
3855         }
3856
3857         return 0;
3858 }
3859
3860 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3861                                         void __user *arg)
3862 {
3863         int ret = 0;
3864         int size;
3865         struct btrfs_ioctl_logical_ino_args *loi;
3866         struct btrfs_data_container *inodes = NULL;
3867         struct btrfs_path *path = NULL;
3868
3869         if (!capable(CAP_SYS_ADMIN))
3870                 return -EPERM;
3871
3872         loi = memdup_user(arg, sizeof(*loi));
3873         if (IS_ERR(loi)) {
3874                 ret = PTR_ERR(loi);
3875                 loi = NULL;
3876                 goto out;
3877         }
3878
3879         path = btrfs_alloc_path();
3880         if (!path) {
3881                 ret = -ENOMEM;
3882                 goto out;
3883         }
3884
3885         size = min_t(u32, loi->size, 64 * 1024);
3886         inodes = init_data_container(size);
3887         if (IS_ERR(inodes)) {
3888                 ret = PTR_ERR(inodes);
3889                 inodes = NULL;
3890                 goto out;
3891         }
3892
3893         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3894                                           build_ino_list, inodes);
3895         if (ret == -EINVAL)
3896                 ret = -ENOENT;
3897         if (ret < 0)
3898                 goto out;
3899
3900         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3901                            (void *)(unsigned long)inodes, size);
3902         if (ret)
3903                 ret = -EFAULT;
3904
3905 out:
3906         btrfs_free_path(path);
3907         vfree(inodes);
3908         kfree(loi);
3909
3910         return ret;
3911 }
3912
3913 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3914                                struct btrfs_ioctl_balance_args *bargs)
3915 {
3916         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3917
3918         bargs->flags = bctl->flags;
3919
3920         if (atomic_read(&fs_info->balance_running))
3921                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3922         if (atomic_read(&fs_info->balance_pause_req))
3923                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3924         if (atomic_read(&fs_info->balance_cancel_req))
3925                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3926
3927         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3928         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3929         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3930
3931         if (lock) {
3932                 spin_lock(&fs_info->balance_lock);
3933                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3934                 spin_unlock(&fs_info->balance_lock);
3935         } else {
3936                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3937         }
3938 }
3939
3940 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3941 {
3942         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3943         struct btrfs_fs_info *fs_info = root->fs_info;
3944         struct btrfs_ioctl_balance_args *bargs;
3945         struct btrfs_balance_control *bctl;
3946         bool need_unlock; /* for mut. excl. ops lock */
3947         int ret;
3948
3949         if (!capable(CAP_SYS_ADMIN))
3950                 return -EPERM;
3951
3952         ret = mnt_want_write_file(file);
3953         if (ret)
3954                 return ret;
3955
3956 again:
3957         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
3958                 mutex_lock(&fs_info->volume_mutex);
3959                 mutex_lock(&fs_info->balance_mutex);
3960                 need_unlock = true;
3961                 goto locked;
3962         }
3963
3964         /*
3965          * mut. excl. ops lock is locked.  Three possibilites:
3966          *   (1) some other op is running
3967          *   (2) balance is running
3968          *   (3) balance is paused -- special case (think resume)
3969          */
3970         mutex_lock(&fs_info->balance_mutex);
3971         if (fs_info->balance_ctl) {
3972                 /* this is either (2) or (3) */
3973                 if (!atomic_read(&fs_info->balance_running)) {
3974                         mutex_unlock(&fs_info->balance_mutex);
3975                         if (!mutex_trylock(&fs_info->volume_mutex))
3976                                 goto again;
3977                         mutex_lock(&fs_info->balance_mutex);
3978
3979                         if (fs_info->balance_ctl &&
3980                             !atomic_read(&fs_info->balance_running)) {
3981                                 /* this is (3) */
3982                                 need_unlock = false;
3983                                 goto locked;
3984                         }
3985
3986                         mutex_unlock(&fs_info->balance_mutex);
3987                         mutex_unlock(&fs_info->volume_mutex);
3988                         goto again;
3989                 } else {
3990                         /* this is (2) */
3991                         mutex_unlock(&fs_info->balance_mutex);
3992                         ret = -EINPROGRESS;
3993                         goto out;
3994                 }
3995         } else {
3996                 /* this is (1) */
3997                 mutex_unlock(&fs_info->balance_mutex);
3998                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3999                 goto out;
4000         }
4001
4002 locked:
4003         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4004
4005         if (arg) {
4006                 bargs = memdup_user(arg, sizeof(*bargs));
4007                 if (IS_ERR(bargs)) {
4008                         ret = PTR_ERR(bargs);
4009                         goto out_unlock;
4010                 }
4011
4012                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4013                         if (!fs_info->balance_ctl) {
4014                                 ret = -ENOTCONN;
4015                                 goto out_bargs;
4016                         }
4017
4018                         bctl = fs_info->balance_ctl;
4019                         spin_lock(&fs_info->balance_lock);
4020                         bctl->flags |= BTRFS_BALANCE_RESUME;
4021                         spin_unlock(&fs_info->balance_lock);
4022
4023                         goto do_balance;
4024                 }
4025         } else {
4026                 bargs = NULL;
4027         }
4028
4029         if (fs_info->balance_ctl) {
4030                 ret = -EINPROGRESS;
4031                 goto out_bargs;
4032         }
4033
4034         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4035         if (!bctl) {
4036                 ret = -ENOMEM;
4037                 goto out_bargs;
4038         }
4039
4040         bctl->fs_info = fs_info;
4041         if (arg) {
4042                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4043                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4044                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4045
4046                 bctl->flags = bargs->flags;
4047         } else {
4048                 /* balance everything - no filters */
4049                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4050         }
4051
4052 do_balance:
4053         /*
4054          * Ownership of bctl and mutually_exclusive_operation_running
4055          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4056          * or, if restriper was paused all the way until unmount, in
4057          * free_fs_info.  mutually_exclusive_operation_running is
4058          * cleared in __cancel_balance.
4059          */
4060         need_unlock = false;
4061
4062         ret = btrfs_balance(bctl, bargs);
4063
4064         if (arg) {
4065                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4066                         ret = -EFAULT;
4067         }
4068
4069 out_bargs:
4070         kfree(bargs);
4071 out_unlock:
4072         mutex_unlock(&fs_info->balance_mutex);
4073         mutex_unlock(&fs_info->volume_mutex);
4074         if (need_unlock)
4075                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4076 out:
4077         mnt_drop_write_file(file);
4078         return ret;
4079 }
4080
4081 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4082 {
4083         if (!capable(CAP_SYS_ADMIN))
4084                 return -EPERM;
4085
4086         switch (cmd) {
4087         case BTRFS_BALANCE_CTL_PAUSE:
4088                 return btrfs_pause_balance(root->fs_info);
4089         case BTRFS_BALANCE_CTL_CANCEL:
4090                 return btrfs_cancel_balance(root->fs_info);
4091         }
4092
4093         return -EINVAL;
4094 }
4095
4096 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4097                                          void __user *arg)
4098 {
4099         struct btrfs_fs_info *fs_info = root->fs_info;
4100         struct btrfs_ioctl_balance_args *bargs;
4101         int ret = 0;
4102
4103         if (!capable(CAP_SYS_ADMIN))
4104                 return -EPERM;
4105
4106         mutex_lock(&fs_info->balance_mutex);
4107         if (!fs_info->balance_ctl) {
4108                 ret = -ENOTCONN;
4109                 goto out;
4110         }
4111
4112         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4113         if (!bargs) {
4114                 ret = -ENOMEM;
4115                 goto out;
4116         }
4117
4118         update_ioctl_balance_args(fs_info, 1, bargs);
4119
4120         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4121                 ret = -EFAULT;
4122
4123         kfree(bargs);
4124 out:
4125         mutex_unlock(&fs_info->balance_mutex);
4126         return ret;
4127 }
4128
4129 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4130 {
4131         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4132         struct btrfs_ioctl_quota_ctl_args *sa;
4133         struct btrfs_trans_handle *trans = NULL;
4134         int ret;
4135         int err;
4136
4137         if (!capable(CAP_SYS_ADMIN))
4138                 return -EPERM;
4139
4140         ret = mnt_want_write_file(file);
4141         if (ret)
4142                 return ret;
4143
4144         sa = memdup_user(arg, sizeof(*sa));
4145         if (IS_ERR(sa)) {
4146                 ret = PTR_ERR(sa);
4147                 goto drop_write;
4148         }
4149
4150         down_write(&root->fs_info->subvol_sem);
4151         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4152         if (IS_ERR(trans)) {
4153                 ret = PTR_ERR(trans);
4154                 goto out;
4155         }
4156
4157         switch (sa->cmd) {
4158         case BTRFS_QUOTA_CTL_ENABLE:
4159                 ret = btrfs_quota_enable(trans, root->fs_info);
4160                 break;
4161         case BTRFS_QUOTA_CTL_DISABLE:
4162                 ret = btrfs_quota_disable(trans, root->fs_info);
4163                 break;
4164         default:
4165                 ret = -EINVAL;
4166                 break;
4167         }
4168
4169         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4170         if (err && !ret)
4171                 ret = err;
4172 out:
4173         kfree(sa);
4174         up_write(&root->fs_info->subvol_sem);
4175 drop_write:
4176         mnt_drop_write_file(file);
4177         return ret;
4178 }
4179
4180 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4181 {
4182         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4183         struct btrfs_ioctl_qgroup_assign_args *sa;
4184         struct btrfs_trans_handle *trans;
4185         int ret;
4186         int err;
4187
4188         if (!capable(CAP_SYS_ADMIN))
4189                 return -EPERM;
4190
4191         ret = mnt_want_write_file(file);
4192         if (ret)
4193                 return ret;
4194
4195         sa = memdup_user(arg, sizeof(*sa));
4196         if (IS_ERR(sa)) {
4197                 ret = PTR_ERR(sa);
4198                 goto drop_write;
4199         }
4200
4201         trans = btrfs_join_transaction(root);
4202         if (IS_ERR(trans)) {
4203                 ret = PTR_ERR(trans);
4204                 goto out;
4205         }
4206
4207         /* FIXME: check if the IDs really exist */
4208         if (sa->assign) {
4209                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4210                                                 sa->src, sa->dst);
4211         } else {
4212                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4213                                                 sa->src, sa->dst);
4214         }
4215
4216         err = btrfs_end_transaction(trans, root);
4217         if (err && !ret)
4218                 ret = err;
4219
4220 out:
4221         kfree(sa);
4222 drop_write:
4223         mnt_drop_write_file(file);
4224         return ret;
4225 }
4226
4227 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4228 {
4229         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4230         struct btrfs_ioctl_qgroup_create_args *sa;
4231         struct btrfs_trans_handle *trans;
4232         int ret;
4233         int err;
4234
4235         if (!capable(CAP_SYS_ADMIN))
4236                 return -EPERM;
4237
4238         ret = mnt_want_write_file(file);
4239         if (ret)
4240                 return ret;
4241
4242         sa = memdup_user(arg, sizeof(*sa));
4243         if (IS_ERR(sa)) {
4244                 ret = PTR_ERR(sa);
4245                 goto drop_write;
4246         }
4247
4248         if (!sa->qgroupid) {
4249                 ret = -EINVAL;
4250                 goto out;
4251         }
4252
4253         trans = btrfs_join_transaction(root);
4254         if (IS_ERR(trans)) {
4255                 ret = PTR_ERR(trans);
4256                 goto out;
4257         }
4258
4259         /* FIXME: check if the IDs really exist */
4260         if (sa->create) {
4261                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4262                                           NULL);
4263         } else {
4264                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4265         }
4266
4267         err = btrfs_end_transaction(trans, root);
4268         if (err && !ret)
4269                 ret = err;
4270
4271 out:
4272         kfree(sa);
4273 drop_write:
4274         mnt_drop_write_file(file);
4275         return ret;
4276 }
4277
4278 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4279 {
4280         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4281         struct btrfs_ioctl_qgroup_limit_args *sa;
4282         struct btrfs_trans_handle *trans;
4283         int ret;
4284         int err;
4285         u64 qgroupid;
4286
4287         if (!capable(CAP_SYS_ADMIN))
4288                 return -EPERM;
4289
4290         ret = mnt_want_write_file(file);
4291         if (ret)
4292                 return ret;
4293
4294         sa = memdup_user(arg, sizeof(*sa));
4295         if (IS_ERR(sa)) {
4296                 ret = PTR_ERR(sa);
4297                 goto drop_write;
4298         }
4299
4300         trans = btrfs_join_transaction(root);
4301         if (IS_ERR(trans)) {
4302                 ret = PTR_ERR(trans);
4303                 goto out;
4304         }
4305
4306         qgroupid = sa->qgroupid;
4307         if (!qgroupid) {
4308                 /* take the current subvol as qgroup */
4309                 qgroupid = root->root_key.objectid;
4310         }
4311
4312         /* FIXME: check if the IDs really exist */
4313         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4314
4315         err = btrfs_end_transaction(trans, root);
4316         if (err && !ret)
4317                 ret = err;
4318
4319 out:
4320         kfree(sa);
4321 drop_write:
4322         mnt_drop_write_file(file);
4323         return ret;
4324 }
4325
4326 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4327 {
4328         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4329         struct btrfs_ioctl_quota_rescan_args *qsa;
4330         int ret;
4331
4332         if (!capable(CAP_SYS_ADMIN))
4333                 return -EPERM;
4334
4335         ret = mnt_want_write_file(file);
4336         if (ret)
4337                 return ret;
4338
4339         qsa = memdup_user(arg, sizeof(*qsa));
4340         if (IS_ERR(qsa)) {
4341                 ret = PTR_ERR(qsa);
4342                 goto drop_write;
4343         }
4344
4345         if (qsa->flags) {
4346                 ret = -EINVAL;
4347                 goto out;
4348         }
4349
4350         ret = btrfs_qgroup_rescan(root->fs_info);
4351
4352 out:
4353         kfree(qsa);
4354 drop_write:
4355         mnt_drop_write_file(file);
4356         return ret;
4357 }
4358
4359 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4360 {
4361         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4362         struct btrfs_ioctl_quota_rescan_args *qsa;
4363         int ret = 0;
4364
4365         if (!capable(CAP_SYS_ADMIN))
4366                 return -EPERM;
4367
4368         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4369         if (!qsa)
4370                 return -ENOMEM;
4371
4372         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4373                 qsa->flags = 1;
4374                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4375         }
4376
4377         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4378                 ret = -EFAULT;
4379
4380         kfree(qsa);
4381         return ret;
4382 }
4383
4384 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4385 {
4386         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4387
4388         if (!capable(CAP_SYS_ADMIN))
4389                 return -EPERM;
4390
4391         return btrfs_qgroup_wait_for_completion(root->fs_info);
4392 }
4393
4394 static long btrfs_ioctl_set_received_subvol(struct file *file,
4395                                             void __user *arg)
4396 {
4397         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4398         struct inode *inode = file_inode(file);
4399         struct btrfs_root *root = BTRFS_I(inode)->root;
4400         struct btrfs_root_item *root_item = &root->root_item;
4401         struct btrfs_trans_handle *trans;
4402         struct timespec ct = CURRENT_TIME;
4403         int ret = 0;
4404         int received_uuid_changed;
4405
4406         ret = mnt_want_write_file(file);
4407         if (ret < 0)
4408                 return ret;
4409
4410         down_write(&root->fs_info->subvol_sem);
4411
4412         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4413                 ret = -EINVAL;
4414                 goto out;
4415         }
4416
4417         if (btrfs_root_readonly(root)) {
4418                 ret = -EROFS;
4419                 goto out;
4420         }
4421
4422         if (!inode_owner_or_capable(inode)) {
4423                 ret = -EACCES;
4424                 goto out;
4425         }
4426
4427         sa = memdup_user(arg, sizeof(*sa));
4428         if (IS_ERR(sa)) {
4429                 ret = PTR_ERR(sa);
4430                 sa = NULL;
4431                 goto out;
4432         }
4433
4434         /*
4435          * 1 - root item
4436          * 2 - uuid items (received uuid + subvol uuid)
4437          */
4438         trans = btrfs_start_transaction(root, 3);
4439         if (IS_ERR(trans)) {
4440                 ret = PTR_ERR(trans);
4441                 trans = NULL;
4442                 goto out;
4443         }
4444
4445         sa->rtransid = trans->transid;
4446         sa->rtime.sec = ct.tv_sec;
4447         sa->rtime.nsec = ct.tv_nsec;
4448
4449         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4450                                        BTRFS_UUID_SIZE);
4451         if (received_uuid_changed &&
4452             !btrfs_is_empty_uuid(root_item->received_uuid))
4453                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4454                                     root_item->received_uuid,
4455                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4456                                     root->root_key.objectid);
4457         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4458         btrfs_set_root_stransid(root_item, sa->stransid);
4459         btrfs_set_root_rtransid(root_item, sa->rtransid);
4460         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4461         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4462         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4463         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4464
4465         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4466                                 &root->root_key, &root->root_item);
4467         if (ret < 0) {
4468                 btrfs_end_transaction(trans, root);
4469                 goto out;
4470         }
4471         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4472                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4473                                           sa->uuid,
4474                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4475                                           root->root_key.objectid);
4476                 if (ret < 0 && ret != -EEXIST) {
4477                         btrfs_abort_transaction(trans, root, ret);
4478                         goto out;
4479                 }
4480         }
4481         ret = btrfs_commit_transaction(trans, root);
4482         if (ret < 0) {
4483                 btrfs_abort_transaction(trans, root, ret);
4484                 goto out;
4485         }
4486
4487         ret = copy_to_user(arg, sa, sizeof(*sa));
4488         if (ret)
4489                 ret = -EFAULT;
4490
4491 out:
4492         kfree(sa);
4493         up_write(&root->fs_info->subvol_sem);
4494         mnt_drop_write_file(file);
4495         return ret;
4496 }
4497
4498 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4499 {
4500         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4501         size_t len;
4502         int ret;
4503         char label[BTRFS_LABEL_SIZE];
4504
4505         spin_lock(&root->fs_info->super_lock);
4506         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4507         spin_unlock(&root->fs_info->super_lock);
4508
4509         len = strnlen(label, BTRFS_LABEL_SIZE);
4510
4511         if (len == BTRFS_LABEL_SIZE) {
4512                 btrfs_warn(root->fs_info,
4513                         "label is too long, return the first %zu bytes", --len);
4514         }
4515
4516         ret = copy_to_user(arg, label, len);
4517
4518         return ret ? -EFAULT : 0;
4519 }
4520
4521 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4522 {
4523         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4524         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4525         struct btrfs_trans_handle *trans;
4526         char label[BTRFS_LABEL_SIZE];
4527         int ret;
4528
4529         if (!capable(CAP_SYS_ADMIN))
4530                 return -EPERM;
4531
4532         if (copy_from_user(label, arg, sizeof(label)))
4533                 return -EFAULT;
4534
4535         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4536                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
4537                        BTRFS_LABEL_SIZE - 1);
4538                 return -EINVAL;
4539         }
4540
4541         ret = mnt_want_write_file(file);
4542         if (ret)
4543                 return ret;
4544
4545         trans = btrfs_start_transaction(root, 0);
4546         if (IS_ERR(trans)) {
4547                 ret = PTR_ERR(trans);
4548                 goto out_unlock;
4549         }
4550
4551         spin_lock(&root->fs_info->super_lock);
4552         strcpy(super_block->label, label);
4553         spin_unlock(&root->fs_info->super_lock);
4554         ret = btrfs_end_transaction(trans, root);
4555
4556 out_unlock:
4557         mnt_drop_write_file(file);
4558         return ret;
4559 }
4560
4561 #define INIT_FEATURE_FLAGS(suffix) \
4562         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4563           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4564           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4565
4566 static int btrfs_ioctl_get_supported_features(struct file *file,
4567                                               void __user *arg)
4568 {
4569         static struct btrfs_ioctl_feature_flags features[3] = {
4570                 INIT_FEATURE_FLAGS(SUPP),
4571                 INIT_FEATURE_FLAGS(SAFE_SET),
4572                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4573         };
4574
4575         if (copy_to_user(arg, &features, sizeof(features)))
4576                 return -EFAULT;
4577
4578         return 0;
4579 }
4580
4581 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
4582 {
4583         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4584         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4585         struct btrfs_ioctl_feature_flags features;
4586
4587         features.compat_flags = btrfs_super_compat_flags(super_block);
4588         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4589         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4590
4591         if (copy_to_user(arg, &features, sizeof(features)))
4592                 return -EFAULT;
4593
4594         return 0;
4595 }
4596
4597 static int check_feature_bits(struct btrfs_root *root,
4598                               enum btrfs_feature_set set,
4599                               u64 change_mask, u64 flags, u64 supported_flags,
4600                               u64 safe_set, u64 safe_clear)
4601 {
4602         const char *type = btrfs_feature_set_names[set];
4603         char *names;
4604         u64 disallowed, unsupported;
4605         u64 set_mask = flags & change_mask;
4606         u64 clear_mask = ~flags & change_mask;
4607
4608         unsupported = set_mask & ~supported_flags;
4609         if (unsupported) {
4610                 names = btrfs_printable_features(set, unsupported);
4611                 if (names) {
4612                         btrfs_warn(root->fs_info,
4613                            "this kernel does not support the %s feature bit%s",
4614                            names, strchr(names, ',') ? "s" : "");
4615                         kfree(names);
4616                 } else
4617                         btrfs_warn(root->fs_info,
4618                            "this kernel does not support %s bits 0x%llx",
4619                            type, unsupported);
4620                 return -EOPNOTSUPP;
4621         }
4622
4623         disallowed = set_mask & ~safe_set;
4624         if (disallowed) {
4625                 names = btrfs_printable_features(set, disallowed);
4626                 if (names) {
4627                         btrfs_warn(root->fs_info,
4628                            "can't set the %s feature bit%s while mounted",
4629                            names, strchr(names, ',') ? "s" : "");
4630                         kfree(names);
4631                 } else
4632                         btrfs_warn(root->fs_info,
4633                            "can't set %s bits 0x%llx while mounted",
4634                            type, disallowed);
4635                 return -EPERM;
4636         }
4637
4638         disallowed = clear_mask & ~safe_clear;
4639         if (disallowed) {
4640                 names = btrfs_printable_features(set, disallowed);
4641                 if (names) {
4642                         btrfs_warn(root->fs_info,
4643                            "can't clear the %s feature bit%s while mounted",
4644                            names, strchr(names, ',') ? "s" : "");
4645                         kfree(names);
4646                 } else
4647                         btrfs_warn(root->fs_info,
4648                            "can't clear %s bits 0x%llx while mounted",
4649                            type, disallowed);
4650                 return -EPERM;
4651         }
4652
4653         return 0;
4654 }
4655
4656 #define check_feature(root, change_mask, flags, mask_base)      \
4657 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
4658                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4659                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4660                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4661
4662 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4663 {
4664         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4665         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4666         struct btrfs_ioctl_feature_flags flags[2];
4667         struct btrfs_trans_handle *trans;
4668         u64 newflags;
4669         int ret;
4670
4671         if (!capable(CAP_SYS_ADMIN))
4672                 return -EPERM;
4673
4674         if (copy_from_user(flags, arg, sizeof(flags)))
4675                 return -EFAULT;
4676
4677         /* Nothing to do */
4678         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4679             !flags[0].incompat_flags)
4680                 return 0;
4681
4682         ret = check_feature(root, flags[0].compat_flags,
4683                             flags[1].compat_flags, COMPAT);
4684         if (ret)
4685                 return ret;
4686
4687         ret = check_feature(root, flags[0].compat_ro_flags,
4688                             flags[1].compat_ro_flags, COMPAT_RO);
4689         if (ret)
4690                 return ret;
4691
4692         ret = check_feature(root, flags[0].incompat_flags,
4693                             flags[1].incompat_flags, INCOMPAT);
4694         if (ret)
4695                 return ret;
4696
4697         trans = btrfs_start_transaction(root, 1);
4698         if (IS_ERR(trans))
4699                 return PTR_ERR(trans);
4700
4701         spin_lock(&root->fs_info->super_lock);
4702         newflags = btrfs_super_compat_flags(super_block);
4703         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4704         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4705         btrfs_set_super_compat_flags(super_block, newflags);
4706
4707         newflags = btrfs_super_compat_ro_flags(super_block);
4708         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4709         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4710         btrfs_set_super_compat_ro_flags(super_block, newflags);
4711
4712         newflags = btrfs_super_incompat_flags(super_block);
4713         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4714         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4715         btrfs_set_super_incompat_flags(super_block, newflags);
4716         spin_unlock(&root->fs_info->super_lock);
4717
4718         return btrfs_end_transaction(trans, root);
4719 }
4720
4721 long btrfs_ioctl(struct file *file, unsigned int
4722                 cmd, unsigned long arg)
4723 {
4724         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4725         void __user *argp = (void __user *)arg;
4726
4727         switch (cmd) {
4728         case FS_IOC_GETFLAGS:
4729                 return btrfs_ioctl_getflags(file, argp);
4730         case FS_IOC_SETFLAGS:
4731                 return btrfs_ioctl_setflags(file, argp);
4732         case FS_IOC_GETVERSION:
4733                 return btrfs_ioctl_getversion(file, argp);
4734         case FITRIM:
4735                 return btrfs_ioctl_fitrim(file, argp);
4736         case BTRFS_IOC_SNAP_CREATE:
4737                 return btrfs_ioctl_snap_create(file, argp, 0);
4738         case BTRFS_IOC_SNAP_CREATE_V2:
4739                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4740         case BTRFS_IOC_SUBVOL_CREATE:
4741                 return btrfs_ioctl_snap_create(file, argp, 1);
4742         case BTRFS_IOC_SUBVOL_CREATE_V2:
4743                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4744         case BTRFS_IOC_SNAP_DESTROY:
4745                 return btrfs_ioctl_snap_destroy(file, argp);
4746         case BTRFS_IOC_SUBVOL_GETFLAGS:
4747                 return btrfs_ioctl_subvol_getflags(file, argp);
4748         case BTRFS_IOC_SUBVOL_SETFLAGS:
4749                 return btrfs_ioctl_subvol_setflags(file, argp);
4750         case BTRFS_IOC_DEFAULT_SUBVOL:
4751                 return btrfs_ioctl_default_subvol(file, argp);
4752         case BTRFS_IOC_DEFRAG:
4753                 return btrfs_ioctl_defrag(file, NULL);
4754         case BTRFS_IOC_DEFRAG_RANGE:
4755                 return btrfs_ioctl_defrag(file, argp);
4756         case BTRFS_IOC_RESIZE:
4757                 return btrfs_ioctl_resize(file, argp);
4758         case BTRFS_IOC_ADD_DEV:
4759                 return btrfs_ioctl_add_dev(root, argp);
4760         case BTRFS_IOC_RM_DEV:
4761                 return btrfs_ioctl_rm_dev(file, argp);
4762         case BTRFS_IOC_FS_INFO:
4763                 return btrfs_ioctl_fs_info(root, argp);
4764         case BTRFS_IOC_DEV_INFO:
4765                 return btrfs_ioctl_dev_info(root, argp);
4766         case BTRFS_IOC_BALANCE:
4767                 return btrfs_ioctl_balance(file, NULL);
4768         case BTRFS_IOC_CLONE:
4769                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4770         case BTRFS_IOC_CLONE_RANGE:
4771                 return btrfs_ioctl_clone_range(file, argp);
4772         case BTRFS_IOC_TRANS_START:
4773                 return btrfs_ioctl_trans_start(file);
4774         case BTRFS_IOC_TRANS_END:
4775                 return btrfs_ioctl_trans_end(file);
4776         case BTRFS_IOC_TREE_SEARCH:
4777                 return btrfs_ioctl_tree_search(file, argp);
4778         case BTRFS_IOC_INO_LOOKUP:
4779                 return btrfs_ioctl_ino_lookup(file, argp);
4780         case BTRFS_IOC_INO_PATHS:
4781                 return btrfs_ioctl_ino_to_path(root, argp);
4782         case BTRFS_IOC_LOGICAL_INO:
4783                 return btrfs_ioctl_logical_to_ino(root, argp);
4784         case BTRFS_IOC_SPACE_INFO:
4785                 return btrfs_ioctl_space_info(root, argp);
4786         case BTRFS_IOC_GLOBAL_RSV:
4787                 return btrfs_ioctl_global_rsv(root, argp);
4788         case BTRFS_IOC_SYNC: {
4789                 int ret;
4790
4791                 ret = btrfs_start_delalloc_roots(root->fs_info, 0);
4792                 if (ret)
4793                         return ret;
4794                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
4795                 return ret;
4796         }
4797         case BTRFS_IOC_START_SYNC:
4798                 return btrfs_ioctl_start_sync(root, argp);
4799         case BTRFS_IOC_WAIT_SYNC:
4800                 return btrfs_ioctl_wait_sync(root, argp);
4801         case BTRFS_IOC_SCRUB:
4802                 return btrfs_ioctl_scrub(file, argp);
4803         case BTRFS_IOC_SCRUB_CANCEL:
4804                 return btrfs_ioctl_scrub_cancel(root, argp);
4805         case BTRFS_IOC_SCRUB_PROGRESS:
4806                 return btrfs_ioctl_scrub_progress(root, argp);
4807         case BTRFS_IOC_BALANCE_V2:
4808                 return btrfs_ioctl_balance(file, argp);
4809         case BTRFS_IOC_BALANCE_CTL:
4810                 return btrfs_ioctl_balance_ctl(root, arg);
4811         case BTRFS_IOC_BALANCE_PROGRESS:
4812                 return btrfs_ioctl_balance_progress(root, argp);
4813         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4814                 return btrfs_ioctl_set_received_subvol(file, argp);
4815         case BTRFS_IOC_SEND:
4816                 return btrfs_ioctl_send(file, argp);
4817         case BTRFS_IOC_GET_DEV_STATS:
4818                 return btrfs_ioctl_get_dev_stats(root, argp);
4819         case BTRFS_IOC_QUOTA_CTL:
4820                 return btrfs_ioctl_quota_ctl(file, argp);
4821         case BTRFS_IOC_QGROUP_ASSIGN:
4822                 return btrfs_ioctl_qgroup_assign(file, argp);
4823         case BTRFS_IOC_QGROUP_CREATE:
4824                 return btrfs_ioctl_qgroup_create(file, argp);
4825         case BTRFS_IOC_QGROUP_LIMIT:
4826                 return btrfs_ioctl_qgroup_limit(file, argp);
4827         case BTRFS_IOC_QUOTA_RESCAN:
4828                 return btrfs_ioctl_quota_rescan(file, argp);
4829         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4830                 return btrfs_ioctl_quota_rescan_status(file, argp);
4831         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4832                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4833         case BTRFS_IOC_DEV_REPLACE:
4834                 return btrfs_ioctl_dev_replace(root, argp);
4835         case BTRFS_IOC_GET_FSLABEL:
4836                 return btrfs_ioctl_get_fslabel(file, argp);
4837         case BTRFS_IOC_SET_FSLABEL:
4838                 return btrfs_ioctl_set_fslabel(file, argp);
4839         case BTRFS_IOC_FILE_EXTENT_SAME:
4840                 return btrfs_ioctl_file_extent_same(file, argp);
4841         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4842                 return btrfs_ioctl_get_supported_features(file, argp);
4843         case BTRFS_IOC_GET_FEATURES:
4844                 return btrfs_ioctl_get_features(file, argp);
4845         case BTRFS_IOC_SET_FEATURES:
4846                 return btrfs_ioctl_set_features(file, argp);
4847         }
4848
4849         return -ENOTTY;
4850 }