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 #include "qgroup.h"
  62
  63 #ifdef CONFIG_64BIT
  64 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  65  * structures are incorrect, as the timespec structure from userspace
  66  * is 4 bytes too small. We define these alternatives here to teach
  67  * the kernel about the 32-bit struct packing.
  68  */
  69 struct btrfs_ioctl_timespec_32 {
  70         __u64 sec;
  71         __u32 nsec;
  72 } __attribute__ ((__packed__));
  73
  74 struct btrfs_ioctl_received_subvol_args_32 {
  75         char    uuid[BTRFS_UUID_SIZE];  /* in */
  76         __u64   stransid;               /* in */
  77         __u64   rtransid;               /* out */
  78         struct btrfs_ioctl_timespec_32 stime; /* in */
  79         struct btrfs_ioctl_timespec_32 rtime; /* out */
  80         __u64   flags;                  /* in */
  81         __u64   reserved[16];           /* in */
  82 } __attribute__ ((__packed__));
  83
  84 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  85                                 struct btrfs_ioctl_received_subvol_args_32)
  86 #endif
  87
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  91
  92 /* Mask out flags that are inappropriate for the given type of inode. */
  93 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  94 {
  95         if (S_ISDIR(mode))
  96                 return flags;
  97         else if (S_ISREG(mode))
  98                 return flags & ~FS_DIRSYNC_FL;
  99         else
 100                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 101 }
 102
 103 /*
 104  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 105  */
 106 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 126                 iflags |= FS_COMPR_FL;
 127         else if (flags & BTRFS_INODE_NOCOMPRESS)
 128                 iflags |= FS_NOCOMP_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_update_iflags(struct inode *inode)
 137 {
 138         struct btrfs_inode *ip = BTRFS_I(inode);
 139         unsigned int new_fl = 0;
 140
 141         if (ip->flags & BTRFS_INODE_SYNC)
 142                 new_fl |= S_SYNC;
 143         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 144                 new_fl |= S_IMMUTABLE;
 145         if (ip->flags & BTRFS_INODE_APPEND)
 146                 new_fl |= S_APPEND;
 147         if (ip->flags & BTRFS_INODE_NOATIME)
 148                 new_fl |= S_NOATIME;
 149         if (ip->flags & BTRFS_INODE_DIRSYNC)
 150                 new_fl |= S_DIRSYNC;
 151
 152         set_mask_bits(&inode->i_flags,
 153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 154                       new_fl);
 155 }
 156
 157 /*
 158  * Inherit flags from the parent inode.
 159  *
 160  * Currently only the compression flags and the cow flags are inherited.
 161  */
 162 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 163 {
 164         unsigned int flags;
 165
 166         if (!dir)
 167                 return;
 168
 169         flags = BTRFS_I(dir)->flags;
 170
 171         if (flags & BTRFS_INODE_NOCOMPRESS) {
 172                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 173                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 174         } else if (flags & BTRFS_INODE_COMPRESS) {
 175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 177         }
 178
 179         if (flags & BTRFS_INODE_NODATACOW) {
 180                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 181                 if (S_ISREG(inode->i_mode))
 182                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 183         }
 184
 185         btrfs_update_iflags(inode);
 186 }
 187
 188 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 189 {
 190         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 191         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 192
 193         if (copy_to_user(arg, &flags, sizeof(flags)))
 194                 return -EFAULT;
 195         return 0;
 196 }
 197
 198 static int check_flags(unsigned int flags)
 199 {
 200         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 201                       FS_NOATIME_FL | FS_NODUMP_FL | \
 202                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 203                       FS_NOCOMP_FL | FS_COMPR_FL |
 204                       FS_NOCOW_FL))
 205                 return -EOPNOTSUPP;
 206
 207         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 208                 return -EINVAL;
 209
 210         return 0;
 211 }
 212
 213 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 214 {
 215         struct inode *inode = file_inode(file);
 216         struct btrfs_inode *ip = BTRFS_I(inode);
 217         struct btrfs_root *root = ip->root;
 218         struct btrfs_trans_handle *trans;
 219         unsigned int flags, oldflags;
 220         int ret;
 221         u64 ip_oldflags;
 222         unsigned int i_oldflags;
 223         umode_t mode;
 224
 225         if (!inode_owner_or_capable(inode))
 226                 return -EPERM;
 227
 228         if (btrfs_root_readonly(root))
 229                 return -EROFS;
 230
 231         if (copy_from_user(&flags, arg, sizeof(flags)))
 232                 return -EFAULT;
 233
 234         ret = check_flags(flags);
 235         if (ret)
 236                 return ret;
 237
 238         ret = mnt_want_write_file(file);
 239         if (ret)
 240                 return ret;
 241
 242         mutex_lock(&inode->i_mutex);
 243
 244         ip_oldflags = ip->flags;
 245         i_oldflags = inode->i_flags;
 246         mode = inode->i_mode;
 247
 248         flags = btrfs_mask_flags(inode->i_mode, flags);
 249         oldflags = btrfs_flags_to_ioctl(ip->flags);
 250         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 251                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 252                         ret = -EPERM;
 253                         goto out_unlock;
 254                 }
 255         }
 256
 257         if (flags & FS_SYNC_FL)
 258                 ip->flags |= BTRFS_INODE_SYNC;
 259         else
 260                 ip->flags &= ~BTRFS_INODE_SYNC;
 261         if (flags & FS_IMMUTABLE_FL)
 262                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 263         else
 264                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 265         if (flags & FS_APPEND_FL)
 266                 ip->flags |= BTRFS_INODE_APPEND;
 267         else
 268                 ip->flags &= ~BTRFS_INODE_APPEND;
 269         if (flags & FS_NODUMP_FL)
 270                 ip->flags |= BTRFS_INODE_NODUMP;
 271         else
 272                 ip->flags &= ~BTRFS_INODE_NODUMP;
 273         if (flags & FS_NOATIME_FL)
 274                 ip->flags |= BTRFS_INODE_NOATIME;
 275         else
 276                 ip->flags &= ~BTRFS_INODE_NOATIME;
 277         if (flags & FS_DIRSYNC_FL)
 278                 ip->flags |= BTRFS_INODE_DIRSYNC;
 279         else
 280                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 281         if (flags & FS_NOCOW_FL) {
 282                 if (S_ISREG(mode)) {
 283                         /*
 284                          * It's safe to turn csums off here, no extents exist.
 285                          * Otherwise we want the flag to reflect the real COW
 286                          * status of the file and will not set it.
 287                          */
 288                         if (inode->i_size == 0)
 289                                 ip->flags |= BTRFS_INODE_NODATACOW
 290                                            | BTRFS_INODE_NODATASUM;
 291                 } else {
 292                         ip->flags |= BTRFS_INODE_NODATACOW;
 293                 }
 294         } else {
 295                 /*
 296                  * Revert back under same assuptions as above
 297                  */
 298                 if (S_ISREG(mode)) {
 299                         if (inode->i_size == 0)
 300                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 301                                              | BTRFS_INODE_NODATASUM);
 302                 } else {
 303                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 304                 }
 305         }
 306
 307         /*
 308          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 309          * flag may be changed automatically if compression code won't make
 310          * things smaller.
 311          */
 312         if (flags & FS_NOCOMP_FL) {
 313                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 314                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 315
 316                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 317                 if (ret && ret != -ENODATA)
 318                         goto out_drop;
 319         } else if (flags & FS_COMPR_FL) {
 320                 const char *comp;
 321
 322                 ip->flags |= BTRFS_INODE_COMPRESS;
 323                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 324
 325                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 326                         comp = "lzo";
 327                 else
 328                         comp = "zlib";
 329                 ret = btrfs_set_prop(inode, "btrfs.compression",
 330                                      comp, strlen(comp), 0);
 331                 if (ret)
 332                         goto out_drop;
 333
 334         } else {
 335                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 336         }
 337
 338         trans = btrfs_start_transaction(root, 1);
 339         if (IS_ERR(trans)) {
 340                 ret = PTR_ERR(trans);
 341                 goto out_drop;
 342         }
 343
 344         btrfs_update_iflags(inode);
 345         inode_inc_iversion(inode);
 346         inode->i_ctime = CURRENT_TIME;
 347         ret = btrfs_update_inode(trans, root, inode);
 348
 349         btrfs_end_transaction(trans, root);
 350  out_drop:
 351         if (ret) {
 352                 ip->flags = ip_oldflags;
 353                 inode->i_flags = i_oldflags;
 354         }
 355
 356  out_unlock:
 357         mutex_unlock(&inode->i_mutex);
 358         mnt_drop_write_file(file);
 359         return ret;
 360 }
 361
 362 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 363 {
 364         struct inode *inode = file_inode(file);
 365
 366         return put_user(inode->i_generation, arg);
 367 }
 368
 369 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 370 {
 371         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 372         struct btrfs_device *device;
 373         struct request_queue *q;
 374         struct fstrim_range range;
 375         u64 minlen = ULLONG_MAX;
 376         u64 num_devices = 0;
 377         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 378         int ret;
 379
 380         if (!capable(CAP_SYS_ADMIN))
 381                 return -EPERM;
 382
 383         rcu_read_lock();
 384         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 385                                 dev_list) {
 386                 if (!device->bdev)
 387                         continue;
 388                 q = bdev_get_queue(device->bdev);
 389                 if (blk_queue_discard(q)) {
 390                         num_devices++;
 391                         minlen = min((u64)q->limits.discard_granularity,
 392                                      minlen);
 393                 }
 394         }
 395         rcu_read_unlock();
 396
 397         if (!num_devices)
 398                 return -EOPNOTSUPP;
 399         if (copy_from_user(&range, arg, sizeof(range)))
 400                 return -EFAULT;
 401         if (range.start > total_bytes ||
 402             range.len < fs_info->sb->s_blocksize)
 403                 return -EINVAL;
 404
 405         range.len = min(range.len, total_bytes - range.start);
 406         range.minlen = max(range.minlen, minlen);
 407         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 408         if (ret < 0)
 409                 return ret;
 410
 411         if (copy_to_user(arg, &range, sizeof(range)))
 412                 return -EFAULT;
 413
 414         return 0;
 415 }
 416
 417 int btrfs_is_empty_uuid(u8 *uuid)
 418 {
 419         int i;
 420
 421         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 422                 if (uuid[i])
 423                         return 0;
 424         }
 425         return 1;
 426 }
 427
 428 static noinline int create_subvol(struct inode *dir,
 429                                   struct dentry *dentry,
 430                                   char *name, int namelen,
 431                                   u64 *async_transid,
 432                                   struct btrfs_qgroup_inherit *inherit)
 433 {
 434         struct btrfs_trans_handle *trans;
 435         struct btrfs_key key;
 436         struct btrfs_root_item root_item;
 437         struct btrfs_inode_item *inode_item;
 438         struct extent_buffer *leaf;
 439         struct btrfs_root *root = BTRFS_I(dir)->root;
 440         struct btrfs_root *new_root;
 441         struct btrfs_block_rsv block_rsv;
 442         struct timespec cur_time = CURRENT_TIME;
 443         struct inode *inode;
 444         int ret;
 445         int err;
 446         u64 objectid;
 447         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 448         u64 index = 0;
 449         u64 qgroup_reserved;
 450         uuid_le new_uuid;
 451
 452         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 453         if (ret)
 454                 return ret;
 455
 456         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 457         /*
 458          * The same as the snapshot creation, please see the comment
 459          * of create_snapshot().
 460          */
 461         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 462                                                8, &qgroup_reserved, false);
 463         if (ret)
 464                 return ret;
 465
 466         trans = btrfs_start_transaction(root, 0);
 467         if (IS_ERR(trans)) {
 468                 ret = PTR_ERR(trans);
 469                 btrfs_subvolume_release_metadata(root, &block_rsv,
 470                                                  qgroup_reserved);
 471                 return ret;
 472         }
 473         trans->block_rsv = &block_rsv;
 474         trans->bytes_reserved = block_rsv.size;
 475
 476         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 477         if (ret)
 478                 goto fail;
 479
 480         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 481                                       0, objectid, NULL, 0, 0, 0);
 482         if (IS_ERR(leaf)) {
 483                 ret = PTR_ERR(leaf);
 484                 goto fail;
 485         }
 486
 487         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 488         btrfs_set_header_bytenr(leaf, leaf->start);
 489         btrfs_set_header_generation(leaf, trans->transid);
 490         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 491         btrfs_set_header_owner(leaf, objectid);
 492
 493         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 494                             BTRFS_FSID_SIZE);
 495         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 496                             btrfs_header_chunk_tree_uuid(leaf),
 497                             BTRFS_UUID_SIZE);
 498         btrfs_mark_buffer_dirty(leaf);
 499
 500         memset(&root_item, 0, sizeof(root_item));
 501
 502         inode_item = &root_item.inode;
 503         btrfs_set_stack_inode_generation(inode_item, 1);
 504         btrfs_set_stack_inode_size(inode_item, 3);
 505         btrfs_set_stack_inode_nlink(inode_item, 1);
 506         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
 507         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 508
 509         btrfs_set_root_flags(&root_item, 0);
 510         btrfs_set_root_limit(&root_item, 0);
 511         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 512
 513         btrfs_set_root_bytenr(&root_item, leaf->start);
 514         btrfs_set_root_generation(&root_item, trans->transid);
 515         btrfs_set_root_level(&root_item, 0);
 516         btrfs_set_root_refs(&root_item, 1);
 517         btrfs_set_root_used(&root_item, leaf->len);
 518         btrfs_set_root_last_snapshot(&root_item, 0);
 519
 520         btrfs_set_root_generation_v2(&root_item,
 521                         btrfs_root_generation(&root_item));
 522         uuid_le_gen(&new_uuid);
 523         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 524         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 525         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 526         root_item.ctime = root_item.otime;
 527         btrfs_set_root_ctransid(&root_item, trans->transid);
 528         btrfs_set_root_otransid(&root_item, trans->transid);
 529
 530         btrfs_tree_unlock(leaf);
 531         free_extent_buffer(leaf);
 532         leaf = NULL;
 533
 534         btrfs_set_root_dirid(&root_item, new_dirid);
 535
 536         key.objectid = objectid;
 537         key.offset = 0;
 538         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 539         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 540                                 &root_item);
 541         if (ret)
 542                 goto fail;
 543
 544         key.offset = (u64)-1;
 545         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 546         if (IS_ERR(new_root)) {
 547                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 548                 ret = PTR_ERR(new_root);
 549                 goto fail;
 550         }
 551
 552         btrfs_record_root_in_trans(trans, new_root);
 553
 554         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 555         if (ret) {
 556                 /* We potentially lose an unused inode item here */
 557                 btrfs_abort_transaction(trans, root, ret);
 558                 goto fail;
 559         }
 560
 561         /*
 562          * insert the directory item
 563          */
 564         ret = btrfs_set_inode_index(dir, &index);
 565         if (ret) {
 566                 btrfs_abort_transaction(trans, root, ret);
 567                 goto fail;
 568         }
 569
 570         ret = btrfs_insert_dir_item(trans, root,
 571                                     name, namelen, dir, &key,
 572                                     BTRFS_FT_DIR, index);
 573         if (ret) {
 574                 btrfs_abort_transaction(trans, root, ret);
 575                 goto fail;
 576         }
 577
 578         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 579         ret = btrfs_update_inode(trans, root, dir);
 580         BUG_ON(ret);
 581
 582         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 583                                  objectid, root->root_key.objectid,
 584                                  btrfs_ino(dir), index, name, namelen);
 585         BUG_ON(ret);
 586
 587         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 588                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 589                                   objectid);
 590         if (ret)
 591                 btrfs_abort_transaction(trans, root, ret);
 592
 593 fail:
 594         trans->block_rsv = NULL;
 595         trans->bytes_reserved = 0;
 596         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 597
 598         if (async_transid) {
 599                 *async_transid = trans->transid;
 600                 err = btrfs_commit_transaction_async(trans, root, 1);
 601                 if (err)
 602                         err = btrfs_commit_transaction(trans, root);
 603         } else {
 604                 err = btrfs_commit_transaction(trans, root);
 605         }
 606         if (err && !ret)
 607                 ret = err;
 608
 609         if (!ret) {
 610                 inode = btrfs_lookup_dentry(dir, dentry);
 611                 if (IS_ERR(inode))
 612                         return PTR_ERR(inode);
 613                 d_instantiate(dentry, inode);
 614         }
 615         return ret;
 616 }
 617
 618 static void btrfs_wait_nocow_write(struct btrfs_root *root)
 619 {
 620         s64 writers;
 621         DEFINE_WAIT(wait);
 622
 623         do {
 624                 prepare_to_wait(&root->subv_writers->wait, &wait,
 625                                 TASK_UNINTERRUPTIBLE);
 626
 627                 writers = percpu_counter_sum(&root->subv_writers->counter);
 628                 if (writers)
 629                         schedule();
 630
 631                 finish_wait(&root->subv_writers->wait, &wait);
 632         } while (writers);
 633 }
 634
 635 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 636                            struct dentry *dentry, char *name, int namelen,
 637                            u64 *async_transid, bool readonly,
 638                            struct btrfs_qgroup_inherit *inherit)
 639 {
 640         struct inode *inode;
 641         struct btrfs_pending_snapshot *pending_snapshot;
 642         struct btrfs_trans_handle *trans;
 643         int ret;
 644
 645         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 646                 return -EINVAL;
 647
 648         atomic_inc(&root->will_be_snapshoted);
 649         smp_mb__after_atomic();
 650         btrfs_wait_nocow_write(root);
 651
 652         ret = btrfs_start_delalloc_inodes(root, 0);
 653         if (ret)
 654                 goto out;
 655
 656         btrfs_wait_ordered_extents(root, -1);
 657
 658         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 659         if (!pending_snapshot) {
 660                 ret = -ENOMEM;
 661                 goto out;
 662         }
 663
 664         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 665                              BTRFS_BLOCK_RSV_TEMP);
 666         /*
 667          * 1 - parent dir inode
 668          * 2 - dir entries
 669          * 1 - root item
 670          * 2 - root ref/backref
 671          * 1 - root of snapshot
 672          * 1 - UUID item
 673          */
 674         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 675                                         &pending_snapshot->block_rsv, 8,
 676                                         &pending_snapshot->qgroup_reserved,
 677                                         false);
 678         if (ret)
 679                 goto free;
 680
 681         pending_snapshot->dentry = dentry;
 682         pending_snapshot->root = root;
 683         pending_snapshot->readonly = readonly;
 684         pending_snapshot->dir = dir;
 685         pending_snapshot->inherit = inherit;
 686
 687         trans = btrfs_start_transaction(root, 0);
 688         if (IS_ERR(trans)) {
 689                 ret = PTR_ERR(trans);
 690                 goto fail;
 691         }
 692
 693         spin_lock(&root->fs_info->trans_lock);
 694         list_add(&pending_snapshot->list,
 695                  &trans->transaction->pending_snapshots);
 696         spin_unlock(&root->fs_info->trans_lock);
 697         if (async_transid) {
 698                 *async_transid = trans->transid;
 699                 ret = btrfs_commit_transaction_async(trans,
 700                                      root->fs_info->extent_root, 1);
 701                 if (ret)
 702                         ret = btrfs_commit_transaction(trans, root);
 703         } else {
 704                 ret = btrfs_commit_transaction(trans,
 705                                                root->fs_info->extent_root);
 706         }
 707         if (ret)
 708                 goto fail;
 709
 710         ret = pending_snapshot->error;
 711         if (ret)
 712                 goto fail;
 713
 714         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 715         if (IS_ERR(inode)) {
 716                 ret = PTR_ERR(inode);
 717                 goto fail;
 718         }
 719
 720         d_instantiate(dentry, inode);
 721         ret = 0;
 722 fail:
 723         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 724                                          &pending_snapshot->block_rsv,
 725                                          pending_snapshot->qgroup_reserved);
 726 free:
 727         kfree(pending_snapshot);
 728 out:
 729         atomic_dec(&root->will_be_snapshoted);
 730         return ret;
 731 }
 732
 733 /*  copy of check_sticky in fs/namei.c()
 734 * It's inline, so penalty for filesystems that don't use sticky bit is
 735 * minimal.
 736 */
 737 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 738 {
 739         kuid_t fsuid = current_fsuid();
 740
 741         if (!(dir->i_mode & S_ISVTX))
 742                 return 0;
 743         if (uid_eq(inode->i_uid, fsuid))
 744                 return 0;
 745         if (uid_eq(dir->i_uid, fsuid))
 746                 return 0;
 747         return !capable(CAP_FOWNER);
 748 }
 749
 750 /*  copy of may_delete in fs/namei.c()
 751  *      Check whether we can remove a link victim from directory dir, check
 752  *  whether the type of victim is right.
 753  *  1. We can't do it if dir is read-only (done in permission())
 754  *  2. We should have write and exec permissions on dir
 755  *  3. We can't remove anything from append-only dir
 756  *  4. We can't do anything with immutable dir (done in permission())
 757  *  5. If the sticky bit on dir is set we should either
 758  *      a. be owner of dir, or
 759  *      b. be owner of victim, or
 760  *      c. have CAP_FOWNER capability
 761  *  6. If the victim is append-only or immutable we can't do antyhing with
 762  *     links pointing to it.
 763  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 764  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 765  *  9. We can't remove a root or mountpoint.
 766  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 767  *     nfs_async_unlink().
 768  */
 769
 770 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 771 {
 772         int error;
 773
 774         if (!victim->d_inode)
 775                 return -ENOENT;
 776
 777         BUG_ON(victim->d_parent->d_inode != dir);
 778         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 779
 780         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 781         if (error)
 782                 return error;
 783         if (IS_APPEND(dir))
 784                 return -EPERM;
 785         if (btrfs_check_sticky(dir, victim->d_inode)||
 786                 IS_APPEND(victim->d_inode)||
 787             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 788                 return -EPERM;
 789         if (isdir) {
 790                 if (!S_ISDIR(victim->d_inode->i_mode))
 791                         return -ENOTDIR;
 792                 if (IS_ROOT(victim))
 793                         return -EBUSY;
 794         } else if (S_ISDIR(victim->d_inode->i_mode))
 795                 return -EISDIR;
 796         if (IS_DEADDIR(dir))
 797                 return -ENOENT;
 798         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 799                 return -EBUSY;
 800         return 0;
 801 }
 802
 803 /* copy of may_create in fs/namei.c() */
 804 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 805 {
 806         if (child->d_inode)
 807                 return -EEXIST;
 808         if (IS_DEADDIR(dir))
 809                 return -ENOENT;
 810         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 811 }
 812
 813 /*
 814  * Create a new subvolume below @parent.  This is largely modeled after
 815  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 816  * inside this filesystem so it's quite a bit simpler.
 817  */
 818 static noinline int btrfs_mksubvol(struct path *parent,
 819                                    char *name, int namelen,
 820                                    struct btrfs_root *snap_src,
 821                                    u64 *async_transid, bool readonly,
 822                                    struct btrfs_qgroup_inherit *inherit)
 823 {
 824         struct inode *dir  = parent->dentry->d_inode;
 825         struct dentry *dentry;
 826         int error;
 827
 828         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 829         if (error == -EINTR)
 830                 return error;
 831
 832         dentry = lookup_one_len(name, parent->dentry, namelen);
 833         error = PTR_ERR(dentry);
 834         if (IS_ERR(dentry))
 835                 goto out_unlock;
 836
 837         error = -EEXIST;
 838         if (dentry->d_inode)
 839                 goto out_dput;
 840
 841         error = btrfs_may_create(dir, dentry);
 842         if (error)
 843                 goto out_dput;
 844
 845         /*
 846          * even if this name doesn't exist, we may get hash collisions.
 847          * check for them now when we can safely fail
 848          */
 849         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 850                                                dir->i_ino, name,
 851                                                namelen);
 852         if (error)
 853                 goto out_dput;
 854
 855         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 856
 857         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 858                 goto out_up_read;
 859
 860         if (snap_src) {
 861                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 862                                         async_transid, readonly, inherit);
 863         } else {
 864                 error = create_subvol(dir, dentry, name, namelen,
 865                                       async_transid, inherit);
 866         }
 867         if (!error)
 868                 fsnotify_mkdir(dir, dentry);
 869 out_up_read:
 870         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 871 out_dput:
 872         dput(dentry);
 873 out_unlock:
 874         mutex_unlock(&dir->i_mutex);
 875         return error;
 876 }
 877
 878 /*
 879  * When we're defragging a range, we don't want to kick it off again
 880  * if it is really just waiting for delalloc to send it down.
 881  * If we find a nice big extent or delalloc range for the bytes in the
 882  * file you want to defrag, we return 0 to let you know to skip this
 883  * part of the file
 884  */
 885 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 886 {
 887         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 888         struct extent_map *em = NULL;
 889         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 890         u64 end;
 891
 892         read_lock(&em_tree->lock);
 893         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 894         read_unlock(&em_tree->lock);
 895
 896         if (em) {
 897                 end = extent_map_end(em);
 898                 free_extent_map(em);
 899                 if (end - offset > thresh)
 900                         return 0;
 901         }
 902         /* if we already have a nice delalloc here, just stop */
 903         thresh /= 2;
 904         end = count_range_bits(io_tree, &offset, offset + thresh,
 905                                thresh, EXTENT_DELALLOC, 1);
 906         if (end >= thresh)
 907                 return 0;
 908         return 1;
 909 }
 910
 911 /*
 912  * helper function to walk through a file and find extents
 913  * newer than a specific transid, and smaller than thresh.
 914  *
 915  * This is used by the defragging code to find new and small
 916  * extents
 917  */
 918 static int find_new_extents(struct btrfs_root *root,
 919                             struct inode *inode, u64 newer_than,
 920                             u64 *off, int thresh)
 921 {
 922         struct btrfs_path *path;
 923         struct btrfs_key min_key;
 924         struct extent_buffer *leaf;
 925         struct btrfs_file_extent_item *extent;
 926         int type;
 927         int ret;
 928         u64 ino = btrfs_ino(inode);
 929
 930         path = btrfs_alloc_path();
 931         if (!path)
 932                 return -ENOMEM;
 933
 934         min_key.objectid = ino;
 935         min_key.type = BTRFS_EXTENT_DATA_KEY;
 936         min_key.offset = *off;
 937
 938         while (1) {
 939                 path->keep_locks = 1;
 940                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 941                 if (ret != 0)
 942                         goto none;
 943                 path->keep_locks = 0;
 944                 btrfs_unlock_up_safe(path, 1);
 945 process_slot:
 946                 if (min_key.objectid != ino)
 947                         goto none;
 948                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 949                         goto none;
 950
 951                 leaf = path->nodes[0];
 952                 extent = btrfs_item_ptr(leaf, path->slots[0],
 953                                         struct btrfs_file_extent_item);
 954
 955                 type = btrfs_file_extent_type(leaf, extent);
 956                 if (type == BTRFS_FILE_EXTENT_REG &&
 957                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 958                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 959                         *off = min_key.offset;
 960                         btrfs_free_path(path);
 961                         return 0;
 962                 }
 963
 964                 path->slots[0]++;
 965                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 966                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 967                         goto process_slot;
 968                 }
 969
 970                 if (min_key.offset == (u64)-1)
 971                         goto none;
 972
 973                 min_key.offset++;
 974                 btrfs_release_path(path);
 975         }
 976 none:
 977         btrfs_free_path(path);
 978         return -ENOENT;
 979 }
 980
 981 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 982 {
 983         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 984         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 985         struct extent_map *em;
 986         u64 len = PAGE_CACHE_SIZE;
 987
 988         /*
 989          * hopefully we have this extent in the tree already, try without
 990          * the full extent lock
 991          */
 992         read_lock(&em_tree->lock);
 993         em = lookup_extent_mapping(em_tree, start, len);
 994         read_unlock(&em_tree->lock);
 995
 996         if (!em) {
 997                 struct extent_state *cached = NULL;
 998                 u64 end = start + len - 1;
 999
1000                 /* get the big lock and read metadata off disk */
1001                 lock_extent_bits(io_tree, start, end, 0, &cached);
1002                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1003                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1004
1005                 if (IS_ERR(em))
1006                         return NULL;
1007         }
1008
1009         return em;
1010 }
1011
1012 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1013 {
1014         struct extent_map *next;
1015         bool ret = true;
1016
1017         /* this is the last extent */
1018         if (em->start + em->len >= i_size_read(inode))
1019                 return false;
1020
1021         next = defrag_lookup_extent(inode, em->start + em->len);
1022         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE ||
1023             (em->block_start + em->block_len == next->block_start))
1024                 ret = false;
1025
1026         free_extent_map(next);
1027         return ret;
1028 }
1029
1030 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
1031                                u64 *last_len, u64 *skip, u64 *defrag_end,
1032                                int compress)
1033 {
1034         struct extent_map *em;
1035         int ret = 1;
1036         bool next_mergeable = true;
1037
1038         /*
1039          * make sure that once we start defragging an extent, we keep on
1040          * defragging it
1041          */
1042         if (start < *defrag_end)
1043                 return 1;
1044
1045         *skip = 0;
1046
1047         em = defrag_lookup_extent(inode, start);
1048         if (!em)
1049                 return 0;
1050
1051         /* this will cover holes, and inline extents */
1052         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1053                 ret = 0;
1054                 goto out;
1055         }
1056
1057         next_mergeable = defrag_check_next_extent(inode, em);
1058
1059         /*
1060          * we hit a real extent, if it is big or the next extent is not a
1061          * real extent, don't bother defragging it
1062          */
1063         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1064             (em->len >= thresh || !next_mergeable))
1065                 ret = 0;
1066 out:
1067         /*
1068          * last_len ends up being a counter of how many bytes we've defragged.
1069          * every time we choose not to defrag an extent, we reset *last_len
1070          * so that the next tiny extent will force a defrag.
1071          *
1072          * The end result of this is that tiny extents before a single big
1073          * extent will force at least part of that big extent to be defragged.
1074          */
1075         if (ret) {
1076                 *defrag_end = extent_map_end(em);
1077         } else {
1078                 *last_len = 0;
1079                 *skip = extent_map_end(em);
1080                 *defrag_end = 0;
1081         }
1082
1083         free_extent_map(em);
1084         return ret;
1085 }
1086
1087 /*
1088  * it doesn't do much good to defrag one or two pages
1089  * at a time.  This pulls in a nice chunk of pages
1090  * to COW and defrag.
1091  *
1092  * It also makes sure the delalloc code has enough
1093  * dirty data to avoid making new small extents as part
1094  * of the defrag
1095  *
1096  * It's a good idea to start RA on this range
1097  * before calling this.
1098  */
1099 static int cluster_pages_for_defrag(struct inode *inode,
1100                                     struct page **pages,
1101                                     unsigned long start_index,
1102                                     unsigned long num_pages)
1103 {
1104         unsigned long file_end;
1105         u64 isize = i_size_read(inode);
1106         u64 page_start;
1107         u64 page_end;
1108         u64 page_cnt;
1109         int ret;
1110         int i;
1111         int i_done;
1112         struct btrfs_ordered_extent *ordered;
1113         struct extent_state *cached_state = NULL;
1114         struct extent_io_tree *tree;
1115         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1116
1117         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1118         if (!isize || start_index > file_end)
1119                 return 0;
1120
1121         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1122
1123         ret = btrfs_delalloc_reserve_space(inode,
1124                                            page_cnt << PAGE_CACHE_SHIFT);
1125         if (ret)
1126                 return ret;
1127         i_done = 0;
1128         tree = &BTRFS_I(inode)->io_tree;
1129
1130         /* step one, lock all the pages */
1131         for (i = 0; i < page_cnt; i++) {
1132                 struct page *page;
1133 again:
1134                 page = find_or_create_page(inode->i_mapping,
1135                                            start_index + i, mask);
1136                 if (!page)
1137                         break;
1138
1139                 page_start = page_offset(page);
1140                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1141                 while (1) {
1142                         lock_extent_bits(tree, page_start, page_end,
1143                                          0, &cached_state);
1144                         ordered = btrfs_lookup_ordered_extent(inode,
1145                                                               page_start);
1146                         unlock_extent_cached(tree, page_start, page_end,
1147                                              &cached_state, GFP_NOFS);
1148                         if (!ordered)
1149                                 break;
1150
1151                         unlock_page(page);
1152                         btrfs_start_ordered_extent(inode, ordered, 1);
1153                         btrfs_put_ordered_extent(ordered);
1154                         lock_page(page);
1155                         /*
1156                          * we unlocked the page above, so we need check if
1157                          * it was released or not.
1158                          */
1159                         if (page->mapping != inode->i_mapping) {
1160                                 unlock_page(page);
1161                                 page_cache_release(page);
1162                                 goto again;
1163                         }
1164                 }
1165
1166                 if (!PageUptodate(page)) {
1167                         btrfs_readpage(NULL, page);
1168                         lock_page(page);
1169                         if (!PageUptodate(page)) {
1170                                 unlock_page(page);
1171                                 page_cache_release(page);
1172                                 ret = -EIO;
1173                                 break;
1174                         }
1175                 }
1176
1177                 if (page->mapping != inode->i_mapping) {
1178                         unlock_page(page);
1179                         page_cache_release(page);
1180                         goto again;
1181                 }
1182
1183                 pages[i] = page;
1184                 i_done++;
1185         }
1186         if (!i_done || ret)
1187                 goto out;
1188
1189         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1190                 goto out;
1191
1192         /*
1193          * so now we have a nice long stream of locked
1194          * and up to date pages, lets wait on them
1195          */
1196         for (i = 0; i < i_done; i++)
1197                 wait_on_page_writeback(pages[i]);
1198
1199         page_start = page_offset(pages[0]);
1200         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1201
1202         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1203                          page_start, page_end - 1, 0, &cached_state);
1204         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1205                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1206                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1207                           &cached_state, GFP_NOFS);
1208
1209         if (i_done != page_cnt) {
1210                 spin_lock(&BTRFS_I(inode)->lock);
1211                 BTRFS_I(inode)->outstanding_extents++;
1212                 spin_unlock(&BTRFS_I(inode)->lock);
1213                 btrfs_delalloc_release_space(inode,
1214                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1215         }
1216
1217
1218         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1219                           &cached_state, GFP_NOFS);
1220
1221         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1222                              page_start, page_end - 1, &cached_state,
1223                              GFP_NOFS);
1224
1225         for (i = 0; i < i_done; i++) {
1226                 clear_page_dirty_for_io(pages[i]);
1227                 ClearPageChecked(pages[i]);
1228                 set_page_extent_mapped(pages[i]);
1229                 set_page_dirty(pages[i]);
1230                 unlock_page(pages[i]);
1231                 page_cache_release(pages[i]);
1232         }
1233         return i_done;
1234 out:
1235         for (i = 0; i < i_done; i++) {
1236                 unlock_page(pages[i]);
1237                 page_cache_release(pages[i]);
1238         }
1239         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1240         return ret;
1241
1242 }
1243
1244 int btrfs_defrag_file(struct inode *inode, struct file *file,
1245                       struct btrfs_ioctl_defrag_range_args *range,
1246                       u64 newer_than, unsigned long max_to_defrag)
1247 {
1248         struct btrfs_root *root = BTRFS_I(inode)->root;
1249         struct file_ra_state *ra = NULL;
1250         unsigned long last_index;
1251         u64 isize = i_size_read(inode);
1252         u64 last_len = 0;
1253         u64 skip = 0;
1254         u64 defrag_end = 0;
1255         u64 newer_off = range->start;
1256         unsigned long i;
1257         unsigned long ra_index = 0;
1258         int ret;
1259         int defrag_count = 0;
1260         int compress_type = BTRFS_COMPRESS_ZLIB;
1261         int extent_thresh = range->extent_thresh;
1262         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1263         unsigned long cluster = max_cluster;
1264         u64 new_align = ~((u64)128 * 1024 - 1);
1265         struct page **pages = NULL;
1266
1267         if (isize == 0)
1268                 return 0;
1269
1270         if (range->start >= isize)
1271                 return -EINVAL;
1272
1273         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1274                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1275                         return -EINVAL;
1276                 if (range->compress_type)
1277                         compress_type = range->compress_type;
1278         }
1279
1280         if (extent_thresh == 0)
1281                 extent_thresh = 256 * 1024;
1282
1283         /*
1284          * if we were not given a file, allocate a readahead
1285          * context
1286          */
1287         if (!file) {
1288                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1289                 if (!ra)
1290                         return -ENOMEM;
1291                 file_ra_state_init(ra, inode->i_mapping);
1292         } else {
1293                 ra = &file->f_ra;
1294         }
1295
1296         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1297                         GFP_NOFS);
1298         if (!pages) {
1299                 ret = -ENOMEM;
1300                 goto out_ra;
1301         }
1302
1303         /* find the last page to defrag */
1304         if (range->start + range->len > range->start) {
1305                 last_index = min_t(u64, isize - 1,
1306                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1307         } else {
1308                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1309         }
1310
1311         if (newer_than) {
1312                 ret = find_new_extents(root, inode, newer_than,
1313                                        &newer_off, 64 * 1024);
1314                 if (!ret) {
1315                         range->start = newer_off;
1316                         /*
1317                          * we always align our defrag to help keep
1318                          * the extents in the file evenly spaced
1319                          */
1320                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1321                 } else
1322                         goto out_ra;
1323         } else {
1324                 i = range->start >> PAGE_CACHE_SHIFT;
1325         }
1326         if (!max_to_defrag)
1327                 max_to_defrag = last_index + 1;
1328
1329         /*
1330          * make writeback starts from i, so the defrag range can be
1331          * written sequentially.
1332          */
1333         if (i < inode->i_mapping->writeback_index)
1334                 inode->i_mapping->writeback_index = i;
1335
1336         while (i <= last_index && defrag_count < max_to_defrag &&
1337                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1338                 PAGE_CACHE_SHIFT)) {
1339                 /*
1340                  * make sure we stop running if someone unmounts
1341                  * the FS
1342                  */
1343                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1344                         break;
1345
1346                 if (btrfs_defrag_cancelled(root->fs_info)) {
1347                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1348                         ret = -EAGAIN;
1349                         break;
1350                 }
1351
1352                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1353                                          extent_thresh, &last_len, &skip,
1354                                          &defrag_end, range->flags &
1355                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1356                         unsigned long next;
1357                         /*
1358                          * the should_defrag function tells us how much to skip
1359                          * bump our counter by the suggested amount
1360                          */
1361                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1362                         i = max(i + 1, next);
1363                         continue;
1364                 }
1365
1366                 if (!newer_than) {
1367                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1368                                    PAGE_CACHE_SHIFT) - i;
1369                         cluster = min(cluster, max_cluster);
1370                 } else {
1371                         cluster = max_cluster;
1372                 }
1373
1374                 if (i + cluster > ra_index) {
1375                         ra_index = max(i, ra_index);
1376                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1377                                        cluster);
1378                         ra_index += max_cluster;
1379                 }
1380
1381                 mutex_lock(&inode->i_mutex);
1382                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1383                         BTRFS_I(inode)->force_compress = compress_type;
1384                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1385                 if (ret < 0) {
1386                         mutex_unlock(&inode->i_mutex);
1387                         goto out_ra;
1388                 }
1389
1390                 defrag_count += ret;
1391                 balance_dirty_pages_ratelimited(inode->i_mapping);
1392                 mutex_unlock(&inode->i_mutex);
1393
1394                 if (newer_than) {
1395                         if (newer_off == (u64)-1)
1396                                 break;
1397
1398                         if (ret > 0)
1399                                 i += ret;
1400
1401                         newer_off = max(newer_off + 1,
1402                                         (u64)i << PAGE_CACHE_SHIFT);
1403
1404                         ret = find_new_extents(root, inode,
1405                                                newer_than, &newer_off,
1406                                                64 * 1024);
1407                         if (!ret) {
1408                                 range->start = newer_off;
1409                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1410                         } else {
1411                                 break;
1412                         }
1413                 } else {
1414                         if (ret > 0) {
1415                                 i += ret;
1416                                 last_len += ret << PAGE_CACHE_SHIFT;
1417                         } else {
1418                                 i++;
1419                                 last_len = 0;
1420                         }
1421                 }
1422         }
1423
1424         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1425                 filemap_flush(inode->i_mapping);
1426                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1427                              &BTRFS_I(inode)->runtime_flags))
1428                         filemap_flush(inode->i_mapping);
1429         }
1430
1431         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1432                 /* the filemap_flush will queue IO into the worker threads, but
1433                  * we have to make sure the IO is actually started and that
1434                  * ordered extents get created before we return
1435                  */
1436                 atomic_inc(&root->fs_info->async_submit_draining);
1437                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1438                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1439                         wait_event(root->fs_info->async_submit_wait,
1440                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1441                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1442                 }
1443                 atomic_dec(&root->fs_info->async_submit_draining);
1444         }
1445
1446         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1447                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1448         }
1449
1450         ret = defrag_count;
1451
1452 out_ra:
1453         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1454                 mutex_lock(&inode->i_mutex);
1455                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1456                 mutex_unlock(&inode->i_mutex);
1457         }
1458         if (!file)
1459                 kfree(ra);
1460         kfree(pages);
1461         return ret;
1462 }
1463
1464 static noinline int btrfs_ioctl_resize(struct file *file,
1465                                         void __user *arg)
1466 {
1467         u64 new_size;
1468         u64 old_size;
1469         u64 devid = 1;
1470         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1471         struct btrfs_ioctl_vol_args *vol_args;
1472         struct btrfs_trans_handle *trans;
1473         struct btrfs_device *device = NULL;
1474         char *sizestr;
1475         char *retptr;
1476         char *devstr = NULL;
1477         int ret = 0;
1478         int mod = 0;
1479
1480         if (!capable(CAP_SYS_ADMIN))
1481                 return -EPERM;
1482
1483         ret = mnt_want_write_file(file);
1484         if (ret)
1485                 return ret;
1486
1487         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1488                         1)) {
1489                 mnt_drop_write_file(file);
1490                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1491         }
1492
1493         mutex_lock(&root->fs_info->volume_mutex);
1494         vol_args = memdup_user(arg, sizeof(*vol_args));
1495         if (IS_ERR(vol_args)) {
1496                 ret = PTR_ERR(vol_args);
1497                 goto out;
1498         }
1499
1500         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1501
1502         sizestr = vol_args->name;
1503         devstr = strchr(sizestr, ':');
1504         if (devstr) {
1505                 sizestr = devstr + 1;
1506                 *devstr = '\0';
1507                 devstr = vol_args->name;
1508                 ret = kstrtoull(devstr, 10, &devid);
1509                 if (ret)
1510                         goto out_free;
1511                 if (!devid) {
1512                         ret = -EINVAL;
1513                         goto out_free;
1514                 }
1515                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1516         }
1517
1518         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1519         if (!device) {
1520                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1521                        devid);
1522                 ret = -ENODEV;
1523                 goto out_free;
1524         }
1525
1526         if (!device->writeable) {
1527                 btrfs_info(root->fs_info,
1528                            "resizer unable to apply on readonly device %llu",
1529                        devid);
1530                 ret = -EPERM;
1531                 goto out_free;
1532         }
1533
1534         if (!strcmp(sizestr, "max"))
1535                 new_size = device->bdev->bd_inode->i_size;
1536         else {
1537                 if (sizestr[0] == '-') {
1538                         mod = -1;
1539                         sizestr++;
1540                 } else if (sizestr[0] == '+') {
1541                         mod = 1;
1542                         sizestr++;
1543                 }
1544                 new_size = memparse(sizestr, &retptr);
1545                 if (*retptr != '\0' || new_size == 0) {
1546                         ret = -EINVAL;
1547                         goto out_free;
1548                 }
1549         }
1550
1551         if (device->is_tgtdev_for_dev_replace) {
1552                 ret = -EPERM;
1553                 goto out_free;
1554         }
1555
1556         old_size = device->total_bytes;
1557
1558         if (mod < 0) {
1559                 if (new_size > old_size) {
1560                         ret = -EINVAL;
1561                         goto out_free;
1562                 }
1563                 new_size = old_size - new_size;
1564         } else if (mod > 0) {
1565                 if (new_size > ULLONG_MAX - old_size) {
1566                         ret = -ERANGE;
1567                         goto out_free;
1568                 }
1569                 new_size = old_size + new_size;
1570         }
1571
1572         if (new_size < 256 * 1024 * 1024) {
1573                 ret = -EINVAL;
1574                 goto out_free;
1575         }
1576         if (new_size > device->bdev->bd_inode->i_size) {
1577                 ret = -EFBIG;
1578                 goto out_free;
1579         }
1580
1581         do_div(new_size, root->sectorsize);
1582         new_size *= root->sectorsize;
1583
1584         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1585                       rcu_str_deref(device->name), new_size);
1586
1587         if (new_size > old_size) {
1588                 trans = btrfs_start_transaction(root, 0);
1589                 if (IS_ERR(trans)) {
1590                         ret = PTR_ERR(trans);
1591                         goto out_free;
1592                 }
1593                 ret = btrfs_grow_device(trans, device, new_size);
1594                 btrfs_commit_transaction(trans, root);
1595         } else if (new_size < old_size) {
1596                 ret = btrfs_shrink_device(device, new_size);
1597         } /* equal, nothing need to do */
1598
1599 out_free:
1600         kfree(vol_args);
1601 out:
1602         mutex_unlock(&root->fs_info->volume_mutex);
1603         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1604         mnt_drop_write_file(file);
1605         return ret;
1606 }
1607
1608 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1609                                 char *name, unsigned long fd, int subvol,
1610                                 u64 *transid, bool readonly,
1611                                 struct btrfs_qgroup_inherit *inherit)
1612 {
1613         int namelen;
1614         int ret = 0;
1615
1616         ret = mnt_want_write_file(file);
1617         if (ret)
1618                 goto out;
1619
1620         namelen = strlen(name);
1621         if (strchr(name, '/')) {
1622                 ret = -EINVAL;
1623                 goto out_drop_write;
1624         }
1625
1626         if (name[0] == '.' &&
1627            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1628                 ret = -EEXIST;
1629                 goto out_drop_write;
1630         }
1631
1632         if (subvol) {
1633                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1634                                      NULL, transid, readonly, inherit);
1635         } else {
1636                 struct fd src = fdget(fd);
1637                 struct inode *src_inode;
1638                 if (!src.file) {
1639                         ret = -EINVAL;
1640                         goto out_drop_write;
1641                 }
1642
1643                 src_inode = file_inode(src.file);
1644                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1645                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1646                                    "Snapshot src from another FS");
1647                         ret = -EXDEV;
1648                 } else if (!inode_owner_or_capable(src_inode)) {
1649                         /*
1650                          * Subvolume creation is not restricted, but snapshots
1651                          * are limited to own subvolumes only
1652                          */
1653                         ret = -EPERM;
1654                 } else {
1655                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1656                                              BTRFS_I(src_inode)->root,
1657                                              transid, readonly, inherit);
1658                 }
1659                 fdput(src);
1660         }
1661 out_drop_write:
1662         mnt_drop_write_file(file);
1663 out:
1664         return ret;
1665 }
1666
1667 static noinline int btrfs_ioctl_snap_create(struct file *file,
1668                                             void __user *arg, int subvol)
1669 {
1670         struct btrfs_ioctl_vol_args *vol_args;
1671         int ret;
1672
1673         vol_args = memdup_user(arg, sizeof(*vol_args));
1674         if (IS_ERR(vol_args))
1675                 return PTR_ERR(vol_args);
1676         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1677
1678         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1679                                               vol_args->fd, subvol,
1680                                               NULL, false, NULL);
1681
1682         kfree(vol_args);
1683         return ret;
1684 }
1685
1686 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1687                                                void __user *arg, int subvol)
1688 {
1689         struct btrfs_ioctl_vol_args_v2 *vol_args;
1690         int ret;
1691         u64 transid = 0;
1692         u64 *ptr = NULL;
1693         bool readonly = false;
1694         struct btrfs_qgroup_inherit *inherit = NULL;
1695
1696         vol_args = memdup_user(arg, sizeof(*vol_args));
1697         if (IS_ERR(vol_args))
1698                 return PTR_ERR(vol_args);
1699         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1700
1701         if (vol_args->flags &
1702             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1703               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1704                 ret = -EOPNOTSUPP;
1705                 goto out;
1706         }
1707
1708         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1709                 ptr = &transid;
1710         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1711                 readonly = true;
1712         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1713                 if (vol_args->size > PAGE_CACHE_SIZE) {
1714                         ret = -EINVAL;
1715                         goto out;
1716                 }
1717                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1718                 if (IS_ERR(inherit)) {
1719                         ret = PTR_ERR(inherit);
1720                         goto out;
1721                 }
1722         }
1723
1724         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1725                                               vol_args->fd, subvol, ptr,
1726                                               readonly, inherit);
1727
1728         if (ret == 0 && ptr &&
1729             copy_to_user(arg +
1730                          offsetof(struct btrfs_ioctl_vol_args_v2,
1731                                   transid), ptr, sizeof(*ptr)))
1732                 ret = -EFAULT;
1733 out:
1734         kfree(vol_args);
1735         kfree(inherit);
1736         return ret;
1737 }
1738
1739 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1740                                                 void __user *arg)
1741 {
1742         struct inode *inode = file_inode(file);
1743         struct btrfs_root *root = BTRFS_I(inode)->root;
1744         int ret = 0;
1745         u64 flags = 0;
1746
1747         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1748                 return -EINVAL;
1749
1750         down_read(&root->fs_info->subvol_sem);
1751         if (btrfs_root_readonly(root))
1752                 flags |= BTRFS_SUBVOL_RDONLY;
1753         up_read(&root->fs_info->subvol_sem);
1754
1755         if (copy_to_user(arg, &flags, sizeof(flags)))
1756                 ret = -EFAULT;
1757
1758         return ret;
1759 }
1760
1761 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1762                                               void __user *arg)
1763 {
1764         struct inode *inode = file_inode(file);
1765         struct btrfs_root *root = BTRFS_I(inode)->root;
1766         struct btrfs_trans_handle *trans;
1767         u64 root_flags;
1768         u64 flags;
1769         int ret = 0;
1770
1771         if (!inode_owner_or_capable(inode))
1772                 return -EPERM;
1773
1774         ret = mnt_want_write_file(file);
1775         if (ret)
1776                 goto out;
1777
1778         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1779                 ret = -EINVAL;
1780                 goto out_drop_write;
1781         }
1782
1783         if (copy_from_user(&flags, arg, sizeof(flags))) {
1784                 ret = -EFAULT;
1785                 goto out_drop_write;
1786         }
1787
1788         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1789                 ret = -EINVAL;
1790                 goto out_drop_write;
1791         }
1792
1793         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1794                 ret = -EOPNOTSUPP;
1795                 goto out_drop_write;
1796         }
1797
1798         down_write(&root->fs_info->subvol_sem);
1799
1800         /* nothing to do */
1801         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1802                 goto out_drop_sem;
1803
1804         root_flags = btrfs_root_flags(&root->root_item);
1805         if (flags & BTRFS_SUBVOL_RDONLY) {
1806                 btrfs_set_root_flags(&root->root_item,
1807                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1808         } else {
1809                 /*
1810                  * Block RO -> RW transition if this subvolume is involved in
1811                  * send
1812                  */
1813                 spin_lock(&root->root_item_lock);
1814                 if (root->send_in_progress == 0) {
1815                         btrfs_set_root_flags(&root->root_item,
1816                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1817                         spin_unlock(&root->root_item_lock);
1818                 } else {
1819                         spin_unlock(&root->root_item_lock);
1820                         btrfs_warn(root->fs_info,
1821                         "Attempt to set subvolume %llu read-write during send",
1822                                         root->root_key.objectid);
1823                         ret = -EPERM;
1824                         goto out_drop_sem;
1825                 }
1826         }
1827
1828         trans = btrfs_start_transaction(root, 1);
1829         if (IS_ERR(trans)) {
1830                 ret = PTR_ERR(trans);
1831                 goto out_reset;
1832         }
1833
1834         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1835                                 &root->root_key, &root->root_item);
1836
1837         btrfs_commit_transaction(trans, root);
1838 out_reset:
1839         if (ret)
1840                 btrfs_set_root_flags(&root->root_item, root_flags);
1841 out_drop_sem:
1842         up_write(&root->fs_info->subvol_sem);
1843 out_drop_write:
1844         mnt_drop_write_file(file);
1845 out:
1846         return ret;
1847 }
1848
1849 /*
1850  * helper to check if the subvolume references other subvolumes
1851  */
1852 static noinline int may_destroy_subvol(struct btrfs_root *root)
1853 {
1854         struct btrfs_path *path;
1855         struct btrfs_dir_item *di;
1856         struct btrfs_key key;
1857         u64 dir_id;
1858         int ret;
1859
1860         path = btrfs_alloc_path();
1861         if (!path)
1862                 return -ENOMEM;
1863
1864         /* Make sure this root isn't set as the default subvol */
1865         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1866         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1867                                    dir_id, "default", 7, 0);
1868         if (di && !IS_ERR(di)) {
1869                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1870                 if (key.objectid == root->root_key.objectid) {
1871                         ret = -EPERM;
1872                         btrfs_err(root->fs_info, "deleting default subvolume "
1873                                   "%llu is not allowed", key.objectid);
1874                         goto out;
1875                 }
1876                 btrfs_release_path(path);
1877         }
1878
1879         key.objectid = root->root_key.objectid;
1880         key.type = BTRFS_ROOT_REF_KEY;
1881         key.offset = (u64)-1;
1882
1883         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1884                                 &key, path, 0, 0);
1885         if (ret < 0)
1886                 goto out;
1887         BUG_ON(ret == 0);
1888
1889         ret = 0;
1890         if (path->slots[0] > 0) {
1891                 path->slots[0]--;
1892                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1893                 if (key.objectid == root->root_key.objectid &&
1894                     key.type == BTRFS_ROOT_REF_KEY)
1895                         ret = -ENOTEMPTY;
1896         }
1897 out:
1898         btrfs_free_path(path);
1899         return ret;
1900 }
1901
1902 static noinline int key_in_sk(struct btrfs_key *key,
1903                               struct btrfs_ioctl_search_key *sk)
1904 {
1905         struct btrfs_key test;
1906         int ret;
1907
1908         test.objectid = sk->min_objectid;
1909         test.type = sk->min_type;
1910         test.offset = sk->min_offset;
1911
1912         ret = btrfs_comp_cpu_keys(key, &test);
1913         if (ret < 0)
1914                 return 0;
1915
1916         test.objectid = sk->max_objectid;
1917         test.type = sk->max_type;
1918         test.offset = sk->max_offset;
1919
1920         ret = btrfs_comp_cpu_keys(key, &test);
1921         if (ret > 0)
1922                 return 0;
1923         return 1;
1924 }
1925
1926 static noinline int copy_to_sk(struct btrfs_root *root,
1927                                struct btrfs_path *path,
1928                                struct btrfs_key *key,
1929                                struct btrfs_ioctl_search_key *sk,
1930                                size_t *buf_size,
1931                                char __user *ubuf,
1932                                unsigned long *sk_offset,
1933                                int *num_found)
1934 {
1935         u64 found_transid;
1936         struct extent_buffer *leaf;
1937         struct btrfs_ioctl_search_header sh;
1938         unsigned long item_off;
1939         unsigned long item_len;
1940         int nritems;
1941         int i;
1942         int slot;
1943         int ret = 0;
1944
1945         leaf = path->nodes[0];
1946         slot = path->slots[0];
1947         nritems = btrfs_header_nritems(leaf);
1948
1949         if (btrfs_header_generation(leaf) > sk->max_transid) {
1950                 i = nritems;
1951                 goto advance_key;
1952         }
1953         found_transid = btrfs_header_generation(leaf);
1954
1955         for (i = slot; i < nritems; i++) {
1956                 item_off = btrfs_item_ptr_offset(leaf, i);
1957                 item_len = btrfs_item_size_nr(leaf, i);
1958
1959                 btrfs_item_key_to_cpu(leaf, key, i);
1960                 if (!key_in_sk(key, sk))
1961                         continue;
1962
1963                 if (sizeof(sh) + item_len > *buf_size) {
1964                         if (*num_found) {
1965                                 ret = 1;
1966                                 goto out;
1967                         }
1968
1969                         /*
1970                          * return one empty item back for v1, which does not
1971                          * handle -EOVERFLOW
1972                          */
1973
1974                         *buf_size = sizeof(sh) + item_len;
1975                         item_len = 0;
1976                         ret = -EOVERFLOW;
1977                 }
1978
1979                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1980                         ret = 1;
1981                         goto out;
1982                 }
1983
1984                 sh.objectid = key->objectid;
1985                 sh.offset = key->offset;
1986                 sh.type = key->type;
1987                 sh.len = item_len;
1988                 sh.transid = found_transid;
1989
1990                 /* copy search result header */
1991                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1992                         ret = -EFAULT;
1993                         goto out;
1994                 }
1995
1996                 *sk_offset += sizeof(sh);
1997
1998                 if (item_len) {
1999                         char __user *up = ubuf + *sk_offset;
2000                         /* copy the item */
2001                         if (read_extent_buffer_to_user(leaf, up,
2002                                                        item_off, item_len)) {
2003                                 ret = -EFAULT;
2004                                 goto out;
2005                         }
2006
2007                         *sk_offset += item_len;
2008                 }
2009                 (*num_found)++;
2010
2011                 if (ret) /* -EOVERFLOW from above */
2012                         goto out;
2013
2014                 if (*num_found >= sk->nr_items) {
2015                         ret = 1;
2016                         goto out;
2017                 }
2018         }
2019 advance_key:
2020         ret = 0;
2021         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2022                 key->offset++;
2023         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2024                 key->offset = 0;
2025                 key->type++;
2026         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2027                 key->offset = 0;
2028                 key->type = 0;
2029                 key->objectid++;
2030         } else
2031                 ret = 1;
2032 out:
2033         /*
2034          *  0: all items from this leaf copied, continue with next
2035          *  1: * more items can be copied, but unused buffer is too small
2036          *     * all items were found
2037          *     Either way, it will stops the loop which iterates to the next
2038          *     leaf
2039          *  -EOVERFLOW: item was to large for buffer
2040          *  -EFAULT: could not copy extent buffer back to userspace
2041          */
2042         return ret;
2043 }
2044
2045 static noinline int search_ioctl(struct inode *inode,
2046                                  struct btrfs_ioctl_search_key *sk,
2047                                  size_t *buf_size,
2048                                  char __user *ubuf)
2049 {
2050         struct btrfs_root *root;
2051         struct btrfs_key key;
2052         struct btrfs_path *path;
2053         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2054         int ret;
2055         int num_found = 0;
2056         unsigned long sk_offset = 0;
2057
2058         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2059                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2060                 return -EOVERFLOW;
2061         }
2062
2063         path = btrfs_alloc_path();
2064         if (!path)
2065                 return -ENOMEM;
2066
2067         if (sk->tree_id == 0) {
2068                 /* search the root of the inode that was passed */
2069                 root = BTRFS_I(inode)->root;
2070         } else {
2071                 key.objectid = sk->tree_id;
2072                 key.type = BTRFS_ROOT_ITEM_KEY;
2073                 key.offset = (u64)-1;
2074                 root = btrfs_read_fs_root_no_name(info, &key);
2075                 if (IS_ERR(root)) {
2076                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2077                                sk->tree_id);
2078                         btrfs_free_path(path);
2079                         return -ENOENT;
2080                 }
2081         }
2082
2083         key.objectid = sk->min_objectid;
2084         key.type = sk->min_type;
2085         key.offset = sk->min_offset;
2086
2087         path->keep_locks = 1;
2088
2089         while (1) {
2090                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2091                 if (ret != 0) {
2092                         if (ret > 0)
2093                                 ret = 0;
2094                         goto err;
2095                 }
2096                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2097                                  &sk_offset, &num_found);
2098                 btrfs_release_path(path);
2099                 if (ret)
2100                         break;
2101
2102         }
2103         if (ret > 0)
2104                 ret = 0;
2105 err:
2106         sk->nr_items = num_found;
2107         btrfs_free_path(path);
2108         return ret;
2109 }
2110
2111 static noinline int btrfs_ioctl_tree_search(struct file *file,
2112                                            void __user *argp)
2113 {
2114         struct btrfs_ioctl_search_args __user *uargs;
2115         struct btrfs_ioctl_search_key sk;
2116         struct inode *inode;
2117         int ret;
2118         size_t buf_size;
2119
2120         if (!capable(CAP_SYS_ADMIN))
2121                 return -EPERM;
2122
2123         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2124
2125         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2126                 return -EFAULT;
2127
2128         buf_size = sizeof(uargs->buf);
2129
2130         inode = file_inode(file);
2131         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2132
2133         /*
2134          * In the origin implementation an overflow is handled by returning a
2135          * search header with a len of zero, so reset ret.
2136          */
2137         if (ret == -EOVERFLOW)
2138                 ret = 0;
2139
2140         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2141                 ret = -EFAULT;
2142         return ret;
2143 }
2144
2145 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2146                                                void __user *argp)
2147 {
2148         struct btrfs_ioctl_search_args_v2 __user *uarg;
2149         struct btrfs_ioctl_search_args_v2 args;
2150         struct inode *inode;
2151         int ret;
2152         size_t buf_size;
2153         const size_t buf_limit = 16 * 1024 * 1024;
2154
2155         if (!capable(CAP_SYS_ADMIN))
2156                 return -EPERM;
2157
2158         /* copy search header and buffer size */
2159         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2160         if (copy_from_user(&args, uarg, sizeof(args)))
2161                 return -EFAULT;
2162
2163         buf_size = args.buf_size;
2164
2165         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2166                 return -EOVERFLOW;
2167
2168         /* limit result size to 16MB */
2169         if (buf_size > buf_limit)
2170                 buf_size = buf_limit;
2171
2172         inode = file_inode(file);
2173         ret = search_ioctl(inode, &args.key, &buf_size,
2174                            (char *)(&uarg->buf[0]));
2175         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2176                 ret = -EFAULT;
2177         else if (ret == -EOVERFLOW &&
2178                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2179                 ret = -EFAULT;
2180
2181         return ret;
2182 }
2183
2184 /*
2185  * Search INODE_REFs to identify path name of 'dirid' directory
2186  * in a 'tree_id' tree. and sets path name to 'name'.
2187  */
2188 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2189                                 u64 tree_id, u64 dirid, char *name)
2190 {
2191         struct btrfs_root *root;
2192         struct btrfs_key key;
2193         char *ptr;
2194         int ret = -1;
2195         int slot;
2196         int len;
2197         int total_len = 0;
2198         struct btrfs_inode_ref *iref;
2199         struct extent_buffer *l;
2200         struct btrfs_path *path;
2201
2202         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2203                 name[0]='\0';
2204                 return 0;
2205         }
2206
2207         path = btrfs_alloc_path();
2208         if (!path)
2209                 return -ENOMEM;
2210
2211         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2212
2213         key.objectid = tree_id;
2214         key.type = BTRFS_ROOT_ITEM_KEY;
2215         key.offset = (u64)-1;
2216         root = btrfs_read_fs_root_no_name(info, &key);
2217         if (IS_ERR(root)) {
2218                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2219                 ret = -ENOENT;
2220                 goto out;
2221         }
2222
2223         key.objectid = dirid;
2224         key.type = BTRFS_INODE_REF_KEY;
2225         key.offset = (u64)-1;
2226
2227         while (1) {
2228                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2229                 if (ret < 0)
2230                         goto out;
2231                 else if (ret > 0) {
2232                         ret = btrfs_previous_item(root, path, dirid,
2233                                                   BTRFS_INODE_REF_KEY);
2234                         if (ret < 0)
2235                                 goto out;
2236                         else if (ret > 0) {
2237                                 ret = -ENOENT;
2238                                 goto out;
2239                         }
2240                 }
2241
2242                 l = path->nodes[0];
2243                 slot = path->slots[0];
2244                 btrfs_item_key_to_cpu(l, &key, slot);
2245
2246                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2247                 len = btrfs_inode_ref_name_len(l, iref);
2248                 ptr -= len + 1;
2249                 total_len += len + 1;
2250                 if (ptr < name) {
2251                         ret = -ENAMETOOLONG;
2252                         goto out;
2253                 }
2254
2255                 *(ptr + len) = '/';
2256                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2257
2258                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2259                         break;
2260
2261                 btrfs_release_path(path);
2262                 key.objectid = key.offset;
2263                 key.offset = (u64)-1;
2264                 dirid = key.objectid;
2265         }
2266         memmove(name, ptr, total_len);
2267         name[total_len] = '\0';
2268         ret = 0;
2269 out:
2270         btrfs_free_path(path);
2271         return ret;
2272 }
2273
2274 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2275                                            void __user *argp)
2276 {
2277          struct btrfs_ioctl_ino_lookup_args *args;
2278          struct inode *inode;
2279          int ret;
2280
2281         if (!capable(CAP_SYS_ADMIN))
2282                 return -EPERM;
2283
2284         args = memdup_user(argp, sizeof(*args));
2285         if (IS_ERR(args))
2286                 return PTR_ERR(args);
2287
2288         inode = file_inode(file);
2289
2290         if (args->treeid == 0)
2291                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2292
2293         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2294                                         args->treeid, args->objectid,
2295                                         args->name);
2296
2297         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2298                 ret = -EFAULT;
2299
2300         kfree(args);
2301         return ret;
2302 }
2303
2304 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2305                                              void __user *arg)
2306 {
2307         struct dentry *parent = file->f_path.dentry;
2308         struct dentry *dentry;
2309         struct inode *dir = parent->d_inode;
2310         struct inode *inode;
2311         struct btrfs_root *root = BTRFS_I(dir)->root;
2312         struct btrfs_root *dest = NULL;
2313         struct btrfs_ioctl_vol_args *vol_args;
2314         struct btrfs_trans_handle *trans;
2315         struct btrfs_block_rsv block_rsv;
2316         u64 root_flags;
2317         u64 qgroup_reserved;
2318         int namelen;
2319         int ret;
2320         int err = 0;
2321
2322         vol_args = memdup_user(arg, sizeof(*vol_args));
2323         if (IS_ERR(vol_args))
2324                 return PTR_ERR(vol_args);
2325
2326         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2327         namelen = strlen(vol_args->name);
2328         if (strchr(vol_args->name, '/') ||
2329             strncmp(vol_args->name, "..", namelen) == 0) {
2330                 err = -EINVAL;
2331                 goto out;
2332         }
2333
2334         err = mnt_want_write_file(file);
2335         if (err)
2336                 goto out;
2337
2338
2339         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2340         if (err == -EINTR)
2341                 goto out_drop_write;
2342         dentry = lookup_one_len(vol_args->name, parent, namelen);
2343         if (IS_ERR(dentry)) {
2344                 err = PTR_ERR(dentry);
2345                 goto out_unlock_dir;
2346         }
2347
2348         if (!dentry->d_inode) {
2349                 err = -ENOENT;
2350                 goto out_dput;
2351         }
2352
2353         inode = dentry->d_inode;
2354         dest = BTRFS_I(inode)->root;
2355         if (!capable(CAP_SYS_ADMIN)) {
2356                 /*
2357                  * Regular user.  Only allow this with a special mount
2358                  * option, when the user has write+exec access to the
2359                  * subvol root, and when rmdir(2) would have been
2360                  * allowed.
2361                  *
2362                  * Note that this is _not_ check that the subvol is
2363                  * empty or doesn't contain data that we wouldn't
2364                  * otherwise be able to delete.
2365                  *
2366                  * Users who want to delete empty subvols should try
2367                  * rmdir(2).
2368                  */
2369                 err = -EPERM;
2370                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2371                         goto out_dput;
2372
2373                 /*
2374                  * Do not allow deletion if the parent dir is the same
2375                  * as the dir to be deleted.  That means the ioctl
2376                  * must be called on the dentry referencing the root
2377                  * of the subvol, not a random directory contained
2378                  * within it.
2379                  */
2380                 err = -EINVAL;
2381                 if (root == dest)
2382                         goto out_dput;
2383
2384                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2385                 if (err)
2386                         goto out_dput;
2387         }
2388
2389         /* check if subvolume may be deleted by a user */
2390         err = btrfs_may_delete(dir, dentry, 1);
2391         if (err)
2392                 goto out_dput;
2393
2394         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2395                 err = -EINVAL;
2396                 goto out_dput;
2397         }
2398
2399         mutex_lock(&inode->i_mutex);
2400
2401         /*
2402          * Don't allow to delete a subvolume with send in progress. This is
2403          * inside the i_mutex so the error handling that has to drop the bit
2404          * again is not run concurrently.
2405          */
2406         spin_lock(&dest->root_item_lock);
2407         root_flags = btrfs_root_flags(&dest->root_item);
2408         if (dest->send_in_progress == 0) {
2409                 btrfs_set_root_flags(&dest->root_item,
2410                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2411                 spin_unlock(&dest->root_item_lock);
2412         } else {
2413                 spin_unlock(&dest->root_item_lock);
2414                 btrfs_warn(root->fs_info,
2415                         "Attempt to delete subvolume %llu during send",
2416                         dest->root_key.objectid);
2417                 err = -EPERM;
2418                 goto out_dput;
2419         }
2420
2421         err = d_invalidate(dentry);
2422         if (err)
2423                 goto out_unlock;
2424
2425         down_write(&root->fs_info->subvol_sem);
2426
2427         err = may_destroy_subvol(dest);
2428         if (err)
2429                 goto out_up_write;
2430
2431         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2432         /*
2433          * One for dir inode, two for dir entries, two for root
2434          * ref/backref.
2435          */
2436         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2437                                                5, &qgroup_reserved, true);
2438         if (err)
2439                 goto out_up_write;
2440
2441         trans = btrfs_start_transaction(root, 0);
2442         if (IS_ERR(trans)) {
2443                 err = PTR_ERR(trans);
2444                 goto out_release;
2445         }
2446         trans->block_rsv = &block_rsv;
2447         trans->bytes_reserved = block_rsv.size;
2448
2449         ret = btrfs_unlink_subvol(trans, root, dir,
2450                                 dest->root_key.objectid,
2451                                 dentry->d_name.name,
2452                                 dentry->d_name.len);
2453         if (ret) {
2454                 err = ret;
2455                 btrfs_abort_transaction(trans, root, ret);
2456                 goto out_end_trans;
2457         }
2458
2459         btrfs_record_root_in_trans(trans, dest);
2460
2461         memset(&dest->root_item.drop_progress, 0,
2462                 sizeof(dest->root_item.drop_progress));
2463         dest->root_item.drop_level = 0;
2464         btrfs_set_root_refs(&dest->root_item, 0);
2465
2466         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2467                 ret = btrfs_insert_orphan_item(trans,
2468                                         root->fs_info->tree_root,
2469                                         dest->root_key.objectid);
2470                 if (ret) {
2471                         btrfs_abort_transaction(trans, root, ret);
2472                         err = ret;
2473                         goto out_end_trans;
2474                 }
2475         }
2476
2477         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2478                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2479                                   dest->root_key.objectid);
2480         if (ret && ret != -ENOENT) {
2481                 btrfs_abort_transaction(trans, root, ret);
2482                 err = ret;
2483                 goto out_end_trans;
2484         }
2485         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2486                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2487                                           dest->root_item.received_uuid,
2488                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2489                                           dest->root_key.objectid);
2490                 if (ret && ret != -ENOENT) {
2491                         btrfs_abort_transaction(trans, root, ret);
2492                         err = ret;
2493                         goto out_end_trans;
2494                 }
2495         }
2496
2497 out_end_trans:
2498         trans->block_rsv = NULL;
2499         trans->bytes_reserved = 0;
2500         ret = btrfs_end_transaction(trans, root);
2501         if (ret && !err)
2502                 err = ret;
2503         inode->i_flags |= S_DEAD;
2504 out_release:
2505         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2506 out_up_write:
2507         up_write(&root->fs_info->subvol_sem);
2508 out_unlock:
2509         if (err) {
2510                 spin_lock(&dest->root_item_lock);
2511                 root_flags = btrfs_root_flags(&dest->root_item);
2512                 btrfs_set_root_flags(&dest->root_item,
2513                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2514                 spin_unlock(&dest->root_item_lock);
2515         }
2516         mutex_unlock(&inode->i_mutex);
2517         if (!err) {
2518                 shrink_dcache_sb(root->fs_info->sb);
2519                 btrfs_invalidate_inodes(dest);
2520                 d_delete(dentry);
2521                 ASSERT(dest->send_in_progress == 0);
2522
2523                 /* the last ref */
2524                 if (dest->cache_inode) {
2525                         iput(dest->cache_inode);
2526                         dest->cache_inode = NULL;
2527                 }
2528         }
2529 out_dput:
2530         dput(dentry);
2531 out_unlock_dir:
2532         mutex_unlock(&dir->i_mutex);
2533 out_drop_write:
2534         mnt_drop_write_file(file);
2535 out:
2536         kfree(vol_args);
2537         return err;
2538 }
2539
2540 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2541 {
2542         struct inode *inode = file_inode(file);
2543         struct btrfs_root *root = BTRFS_I(inode)->root;
2544         struct btrfs_ioctl_defrag_range_args *range;
2545         int ret;
2546
2547         ret = mnt_want_write_file(file);
2548         if (ret)
2549                 return ret;
2550
2551         if (btrfs_root_readonly(root)) {
2552                 ret = -EROFS;
2553                 goto out;
2554         }
2555
2556         switch (inode->i_mode & S_IFMT) {
2557         case S_IFDIR:
2558                 if (!capable(CAP_SYS_ADMIN)) {
2559                         ret = -EPERM;
2560                         goto out;
2561                 }
2562                 ret = btrfs_defrag_root(root);
2563                 if (ret)
2564                         goto out;
2565                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2566                 break;
2567         case S_IFREG:
2568                 if (!(file->f_mode & FMODE_WRITE)) {
2569                         ret = -EINVAL;
2570                         goto out;
2571                 }
2572
2573                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2574                 if (!range) {
2575                         ret = -ENOMEM;
2576                         goto out;
2577                 }
2578
2579                 if (argp) {
2580                         if (copy_from_user(range, argp,
2581                                            sizeof(*range))) {
2582                                 ret = -EFAULT;
2583                                 kfree(range);
2584                                 goto out;
2585                         }
2586                         /* compression requires us to start the IO */
2587                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2588                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2589                                 range->extent_thresh = (u32)-1;
2590                         }
2591                 } else {
2592                         /* the rest are all set to zero by kzalloc */
2593                         range->len = (u64)-1;
2594                 }
2595                 ret = btrfs_defrag_file(file_inode(file), file,
2596                                         range, 0, 0);
2597                 if (ret > 0)
2598                         ret = 0;
2599                 kfree(range);
2600                 break;
2601         default:
2602                 ret = -EINVAL;
2603         }
2604 out:
2605         mnt_drop_write_file(file);
2606         return ret;
2607 }
2608
2609 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2610 {
2611         struct btrfs_ioctl_vol_args *vol_args;
2612         int ret;
2613
2614         if (!capable(CAP_SYS_ADMIN))
2615                 return -EPERM;
2616
2617         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2618                         1)) {
2619                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2620         }
2621
2622         mutex_lock(&root->fs_info->volume_mutex);
2623         vol_args = memdup_user(arg, sizeof(*vol_args));
2624         if (IS_ERR(vol_args)) {
2625                 ret = PTR_ERR(vol_args);
2626                 goto out;
2627         }
2628
2629         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2630         ret = btrfs_init_new_device(root, vol_args->name);
2631
2632         kfree(vol_args);
2633 out:
2634         mutex_unlock(&root->fs_info->volume_mutex);
2635         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2636         return ret;
2637 }
2638
2639 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2640 {
2641         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2642         struct btrfs_ioctl_vol_args *vol_args;
2643         int ret;
2644
2645         if (!capable(CAP_SYS_ADMIN))
2646                 return -EPERM;
2647
2648         ret = mnt_want_write_file(file);
2649         if (ret)
2650                 return ret;
2651
2652         vol_args = memdup_user(arg, sizeof(*vol_args));
2653         if (IS_ERR(vol_args)) {
2654                 ret = PTR_ERR(vol_args);
2655                 goto out;
2656         }
2657
2658         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2659
2660         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2661                         1)) {
2662                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2663                 goto out;
2664         }
2665
2666         mutex_lock(&root->fs_info->volume_mutex);
2667         ret = btrfs_rm_device(root, vol_args->name);
2668         mutex_unlock(&root->fs_info->volume_mutex);
2669         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2670
2671 out:
2672         kfree(vol_args);
2673         mnt_drop_write_file(file);
2674         return ret;
2675 }
2676
2677 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2678 {
2679         struct btrfs_ioctl_fs_info_args *fi_args;
2680         struct btrfs_device *device;
2681         struct btrfs_device *next;
2682         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2683         int ret = 0;
2684
2685         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2686         if (!fi_args)
2687                 return -ENOMEM;
2688
2689         mutex_lock(&fs_devices->device_list_mutex);
2690         fi_args->num_devices = fs_devices->num_devices;
2691         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2692
2693         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2694                 if (device->devid > fi_args->max_id)
2695                         fi_args->max_id = device->devid;
2696         }
2697         mutex_unlock(&fs_devices->device_list_mutex);
2698
2699         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2700         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2701         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2702
2703         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2704                 ret = -EFAULT;
2705
2706         kfree(fi_args);
2707         return ret;
2708 }
2709
2710 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2711 {
2712         struct btrfs_ioctl_dev_info_args *di_args;
2713         struct btrfs_device *dev;
2714         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2715         int ret = 0;
2716         char *s_uuid = NULL;
2717
2718         di_args = memdup_user(arg, sizeof(*di_args));
2719         if (IS_ERR(di_args))
2720                 return PTR_ERR(di_args);
2721
2722         if (!btrfs_is_empty_uuid(di_args->uuid))
2723                 s_uuid = di_args->uuid;
2724
2725         mutex_lock(&fs_devices->device_list_mutex);
2726         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2727
2728         if (!dev) {
2729                 ret = -ENODEV;
2730                 goto out;
2731         }
2732
2733         di_args->devid = dev->devid;
2734         di_args->bytes_used = dev->bytes_used;
2735         di_args->total_bytes = dev->total_bytes;
2736         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2737         if (dev->name) {
2738                 struct rcu_string *name;
2739
2740                 rcu_read_lock();
2741                 name = rcu_dereference(dev->name);
2742                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2743                 rcu_read_unlock();
2744                 di_args->path[sizeof(di_args->path) - 1] = 0;
2745         } else {
2746                 di_args->path[0] = '\0';
2747         }
2748
2749 out:
2750         mutex_unlock(&fs_devices->device_list_mutex);
2751         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2752                 ret = -EFAULT;
2753
2754         kfree(di_args);
2755         return ret;
2756 }
2757
2758 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2759 {
2760         struct page *page;
2761         pgoff_t index;
2762         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2763
2764         index = off >> PAGE_CACHE_SHIFT;
2765
2766         page = grab_cache_page(inode->i_mapping, index);
2767         if (!page)
2768                 return NULL;
2769
2770         if (!PageUptodate(page)) {
2771                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2772                                                  0))
2773                         return NULL;
2774                 lock_page(page);
2775                 if (!PageUptodate(page)) {
2776                         unlock_page(page);
2777                         page_cache_release(page);
2778                         return NULL;
2779                 }
2780         }
2781         unlock_page(page);
2782
2783         return page;
2784 }
2785
2786 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2787 {
2788         /* do any pending delalloc/csum calc on src, one way or
2789            another, and lock file content */
2790         while (1) {
2791                 struct btrfs_ordered_extent *ordered;
2792                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2793                 ordered = btrfs_lookup_first_ordered_extent(inode,
2794                                                             off + len - 1);
2795                 if ((!ordered ||
2796                      ordered->file_offset + ordered->len <= off ||
2797                      ordered->file_offset >= off + len) &&
2798                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2799                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2800                         if (ordered)
2801                                 btrfs_put_ordered_extent(ordered);
2802                         break;
2803                 }
2804                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2805                 if (ordered)
2806                         btrfs_put_ordered_extent(ordered);
2807                 btrfs_wait_ordered_range(inode, off, len);
2808         }
2809 }
2810
2811 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2812                                 struct inode *inode2, u64 loff2, u64 len)
2813 {
2814         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2815         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2816
2817         mutex_unlock(&inode1->i_mutex);
2818         mutex_unlock(&inode2->i_mutex);
2819 }
2820
2821 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2822                               struct inode *inode2, u64 loff2, u64 len)
2823 {
2824         if (inode1 < inode2) {
2825                 swap(inode1, inode2);
2826                 swap(loff1, loff2);
2827         }
2828
2829         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2830         lock_extent_range(inode1, loff1, len);
2831         if (inode1 != inode2) {
2832                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2833                 lock_extent_range(inode2, loff2, len);
2834         }
2835 }
2836
2837 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2838                           u64 dst_loff, u64 len)
2839 {
2840         int ret = 0;
2841         struct page *src_page, *dst_page;
2842         unsigned int cmp_len = PAGE_CACHE_SIZE;
2843         void *addr, *dst_addr;
2844
2845         while (len) {
2846                 if (len < PAGE_CACHE_SIZE)
2847                         cmp_len = len;
2848
2849                 src_page = extent_same_get_page(src, loff);
2850                 if (!src_page)
2851                         return -EINVAL;
2852                 dst_page = extent_same_get_page(dst, dst_loff);
2853                 if (!dst_page) {
2854                         page_cache_release(src_page);
2855                         return -EINVAL;
2856                 }
2857                 addr = kmap_atomic(src_page);
2858                 dst_addr = kmap_atomic(dst_page);
2859
2860                 flush_dcache_page(src_page);
2861                 flush_dcache_page(dst_page);
2862
2863                 if (memcmp(addr, dst_addr, cmp_len))
2864                         ret = BTRFS_SAME_DATA_DIFFERS;
2865
2866                 kunmap_atomic(addr);
2867                 kunmap_atomic(dst_addr);
2868                 page_cache_release(src_page);
2869                 page_cache_release(dst_page);
2870
2871                 if (ret)
2872                         break;
2873
2874                 loff += cmp_len;
2875                 dst_loff += cmp_len;
2876                 len -= cmp_len;
2877         }
2878
2879         return ret;
2880 }
2881
2882 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2883 {
2884         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2885
2886         if (off + len > inode->i_size || off + len < off)
2887                 return -EINVAL;
2888         /* Check that we are block aligned - btrfs_clone() requires this */
2889         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2890                 return -EINVAL;
2891
2892         return 0;
2893 }
2894
2895 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2896                              struct inode *dst, u64 dst_loff)
2897 {
2898         int ret;
2899
2900         /*
2901          * btrfs_clone() can't handle extents in the same file
2902          * yet. Once that works, we can drop this check and replace it
2903          * with a check for the same inode, but overlapping extents.
2904          */
2905         if (src == dst)
2906                 return -EINVAL;
2907
2908         btrfs_double_lock(src, loff, dst, dst_loff, len);
2909
2910         ret = extent_same_check_offsets(src, loff, len);
2911         if (ret)
2912                 goto out_unlock;
2913
2914         ret = extent_same_check_offsets(dst, dst_loff, len);
2915         if (ret)
2916                 goto out_unlock;
2917
2918         /* don't make the dst file partly checksummed */
2919         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2920             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2921                 ret = -EINVAL;
2922                 goto out_unlock;
2923         }
2924
2925         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2926         if (ret == 0)
2927                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2928
2929 out_unlock:
2930         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2931
2932         return ret;
2933 }
2934
2935 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2936
2937 static long btrfs_ioctl_file_extent_same(struct file *file,
2938                         struct btrfs_ioctl_same_args __user *argp)
2939 {
2940         struct btrfs_ioctl_same_args *same;
2941         struct btrfs_ioctl_same_extent_info *info;
2942         struct inode *src = file_inode(file);
2943         u64 off;
2944         u64 len;
2945         int i;
2946         int ret;
2947         unsigned long size;
2948         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2949         bool is_admin = capable(CAP_SYS_ADMIN);
2950         u16 count;
2951
2952         if (!(file->f_mode & FMODE_READ))
2953                 return -EINVAL;
2954
2955         ret = mnt_want_write_file(file);
2956         if (ret)
2957                 return ret;
2958
2959         if (get_user(count, &argp->dest_count)) {
2960                 ret = -EFAULT;
2961                 goto out;
2962         }
2963
2964         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2965
2966         same = memdup_user(argp, size);
2967
2968         if (IS_ERR(same)) {
2969                 ret = PTR_ERR(same);
2970                 goto out;
2971         }
2972
2973         off = same->logical_offset;
2974         len = same->length;
2975
2976         /*
2977          * Limit the total length we will dedupe for each operation.
2978          * This is intended to bound the total time spent in this
2979          * ioctl to something sane.
2980          */
2981         if (len > BTRFS_MAX_DEDUPE_LEN)
2982                 len = BTRFS_MAX_DEDUPE_LEN;
2983
2984         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2985                 /*
2986                  * Btrfs does not support blocksize < page_size. As a
2987                  * result, btrfs_cmp_data() won't correctly handle
2988                  * this situation without an update.
2989                  */
2990                 ret = -EINVAL;
2991                 goto out;
2992         }
2993
2994         ret = -EISDIR;
2995         if (S_ISDIR(src->i_mode))
2996                 goto out;
2997
2998         ret = -EACCES;
2999         if (!S_ISREG(src->i_mode))
3000                 goto out;
3001
3002         /* pre-format output fields to sane values */
3003         for (i = 0; i < count; i++) {
3004                 same->info[i].bytes_deduped = 0ULL;
3005                 same->info[i].status = 0;
3006         }
3007
3008         for (i = 0, info = same->info; i < count; i++, info++) {
3009                 struct inode *dst;
3010                 struct fd dst_file = fdget(info->fd);
3011                 if (!dst_file.file) {
3012                         info->status = -EBADF;
3013                         continue;
3014                 }
3015                 dst = file_inode(dst_file.file);
3016
3017                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3018                         info->status = -EINVAL;
3019                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3020                         info->status = -EXDEV;
3021                 } else if (S_ISDIR(dst->i_mode)) {
3022                         info->status = -EISDIR;
3023                 } else if (!S_ISREG(dst->i_mode)) {
3024                         info->status = -EACCES;
3025                 } else {
3026                         info->status = btrfs_extent_same(src, off, len, dst,
3027                                                         info->logical_offset);
3028                         if (info->status == 0)
3029                                 info->bytes_deduped += len;
3030                 }
3031                 fdput(dst_file);
3032         }
3033
3034         ret = copy_to_user(argp, same, size);
3035         if (ret)
3036                 ret = -EFAULT;
3037
3038 out:
3039         mnt_drop_write_file(file);
3040         return ret;
3041 }
3042
3043 /* Helper to check and see if this root currently has a ref on the given disk
3044  * bytenr.  If it does then we need to update the quota for this root.  This
3045  * doesn't do anything if quotas aren't enabled.
3046  */
3047 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3048                      u64 disko)
3049 {
3050         struct seq_list tree_mod_seq_elem = {};
3051         struct ulist *roots;
3052         struct ulist_iterator uiter;
3053         struct ulist_node *root_node = NULL;
3054         int ret;
3055
3056         if (!root->fs_info->quota_enabled)
3057                 return 1;
3058
3059         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3060         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3061                                    tree_mod_seq_elem.seq, &roots);
3062         if (ret < 0)
3063                 goto out;
3064         ret = 0;
3065         ULIST_ITER_INIT(&uiter);
3066         while ((root_node = ulist_next(roots, &uiter))) {
3067                 if (root_node->val == root->objectid) {
3068                         ret = 1;
3069                         break;
3070                 }
3071         }
3072         ulist_free(roots);
3073 out:
3074         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3075         return ret;
3076 }
3077
3078 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3079                                      struct inode *inode,
3080                                      u64 endoff,
3081                                      const u64 destoff,
3082                                      const u64 olen)
3083 {
3084         struct btrfs_root *root = BTRFS_I(inode)->root;
3085         int ret;
3086
3087         inode_inc_iversion(inode);
3088         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3089         /*
3090          * We round up to the block size at eof when determining which
3091          * extents to clone above, but shouldn't round up the file size.
3092          */
3093         if (endoff > destoff + olen)
3094                 endoff = destoff + olen;
3095         if (endoff > inode->i_size)
3096                 btrfs_i_size_write(inode, endoff);
3097
3098         ret = btrfs_update_inode(trans, root, inode);
3099         if (ret) {
3100                 btrfs_abort_transaction(trans, root, ret);
3101                 btrfs_end_transaction(trans, root);
3102                 goto out;
3103         }
3104         ret = btrfs_end_transaction(trans, root);
3105 out:
3106         return ret;
3107 }
3108
3109 static void clone_update_extent_map(struct inode *inode,
3110                                     const struct btrfs_trans_handle *trans,
3111                                     const struct btrfs_path *path,
3112                                     const u64 hole_offset,
3113                                     const u64 hole_len)
3114 {
3115         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3116         struct extent_map *em;
3117         int ret;
3118
3119         em = alloc_extent_map();
3120         if (!em) {
3121                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3122                         &BTRFS_I(inode)->runtime_flags);
3123                 return;
3124         }
3125
3126         if (path) {
3127                 struct btrfs_file_extent_item *fi;
3128
3129                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3130                                     struct btrfs_file_extent_item);
3131                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3132                 em->generation = -1;
3133                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3134                     BTRFS_FILE_EXTENT_INLINE)
3135                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3136                                 &BTRFS_I(inode)->runtime_flags);
3137         } else {
3138                 em->start = hole_offset;
3139                 em->len = hole_len;
3140                 em->ram_bytes = em->len;
3141                 em->orig_start = hole_offset;
3142                 em->block_start = EXTENT_MAP_HOLE;
3143                 em->block_len = 0;
3144                 em->orig_block_len = 0;
3145                 em->compress_type = BTRFS_COMPRESS_NONE;
3146                 em->generation = trans->transid;
3147         }
3148
3149         while (1) {
3150                 write_lock(&em_tree->lock);
3151                 ret = add_extent_mapping(em_tree, em, 1);
3152                 write_unlock(&em_tree->lock);
3153                 if (ret != -EEXIST) {
3154                         free_extent_map(em);
3155                         break;
3156                 }
3157                 btrfs_drop_extent_cache(inode, em->start,
3158                                         em->start + em->len - 1, 0);
3159         }
3160
3161         if (unlikely(ret))
3162                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3163                         &BTRFS_I(inode)->runtime_flags);
3164 }
3165
3166 /**
3167  * btrfs_clone() - clone a range from inode file to another
3168  *
3169  * @src: Inode to clone from
3170  * @inode: Inode to clone to
3171  * @off: Offset within source to start clone from
3172  * @olen: Original length, passed by user, of range to clone
3173  * @olen_aligned: Block-aligned value of olen, extent_same uses
3174  *               identical values here
3175  * @destoff: Offset within @inode to start clone
3176  */
3177 static int btrfs_clone(struct inode *src, struct inode *inode,
3178                        const u64 off, const u64 olen, const u64 olen_aligned,
3179                        const u64 destoff)
3180 {
3181         struct btrfs_root *root = BTRFS_I(inode)->root;
3182         struct btrfs_path *path = NULL;
3183         struct extent_buffer *leaf;
3184         struct btrfs_trans_handle *trans;
3185         char *buf = NULL;
3186         struct btrfs_key key;
3187         u32 nritems;
3188         int slot;
3189         int ret;
3190         int no_quota;
3191         const u64 len = olen_aligned;
3192         u64 last_disko = 0;
3193         u64 last_dest_end = destoff;
3194
3195         ret = -ENOMEM;
3196         buf = vmalloc(btrfs_level_size(root, 0));
3197         if (!buf)
3198                 return ret;
3199
3200         path = btrfs_alloc_path();
3201         if (!path) {
3202                 vfree(buf);
3203                 return ret;
3204         }
3205
3206         path->reada = 2;
3207         /* clone data */
3208         key.objectid = btrfs_ino(src);
3209         key.type = BTRFS_EXTENT_DATA_KEY;
3210         key.offset = off;
3211
3212         while (1) {
3213                 /*
3214                  * note the key will change type as we walk through the
3215                  * tree.
3216                  */
3217                 path->leave_spinning = 1;
3218                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3219                                 0, 0);
3220                 if (ret < 0)
3221                         goto out;
3222                 /*
3223                  * First search, if no extent item that starts at offset off was
3224                  * found but the previous item is an extent item, it's possible
3225                  * it might overlap our target range, therefore process it.
3226                  */
3227                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3228                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3229                                               path->slots[0] - 1);
3230                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3231                                 path->slots[0]--;
3232                 }
3233
3234                 nritems = btrfs_header_nritems(path->nodes[0]);
3235 process_slot:
3236                 no_quota = 1;
3237                 if (path->slots[0] >= nritems) {
3238                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3239                         if (ret < 0)
3240                                 goto out;
3241                         if (ret > 0)
3242                                 break;
3243                         nritems = btrfs_header_nritems(path->nodes[0]);
3244                 }
3245                 leaf = path->nodes[0];
3246                 slot = path->slots[0];
3247
3248                 btrfs_item_key_to_cpu(leaf, &key, slot);
3249                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
3250                     key.objectid != btrfs_ino(src))
3251                         break;
3252
3253                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
3254                         struct btrfs_file_extent_item *extent;
3255                         int type;
3256                         u32 size;
3257                         struct btrfs_key new_key;
3258                         u64 disko = 0, diskl = 0;
3259                         u64 datao = 0, datal = 0;
3260                         u8 comp;
3261                         u64 drop_start;
3262
3263                         extent = btrfs_item_ptr(leaf, slot,
3264                                                 struct btrfs_file_extent_item);
3265                         comp = btrfs_file_extent_compression(leaf, extent);
3266                         type = btrfs_file_extent_type(leaf, extent);
3267                         if (type == BTRFS_FILE_EXTENT_REG ||
3268                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3269                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3270                                                                       extent);
3271                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3272                                                                  extent);
3273                                 datao = btrfs_file_extent_offset(leaf, extent);
3274                                 datal = btrfs_file_extent_num_bytes(leaf,
3275                                                                     extent);
3276                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3277                                 /* take upper bound, may be compressed */
3278                                 datal = btrfs_file_extent_ram_bytes(leaf,
3279                                                                     extent);
3280                         }
3281
3282                         /*
3283                          * The first search might have left us at an extent
3284                          * item that ends before our target range's start, can
3285                          * happen if we have holes and NO_HOLES feature enabled.
3286                          */
3287                         if (key.offset + datal <= off) {
3288                                 path->slots[0]++;
3289                                 goto process_slot;
3290                         } else if (key.offset >= off + len) {
3291                                 break;
3292                         }
3293
3294                         size = btrfs_item_size_nr(leaf, slot);
3295                         read_extent_buffer(leaf, buf,
3296                                            btrfs_item_ptr_offset(leaf, slot),
3297                                            size);
3298
3299                         btrfs_release_path(path);
3300                         path->leave_spinning = 0;
3301
3302                         memcpy(&new_key, &key, sizeof(new_key));
3303                         new_key.objectid = btrfs_ino(inode);
3304                         if (off <= key.offset)
3305                                 new_key.offset = key.offset + destoff - off;
3306                         else
3307                                 new_key.offset = destoff;
3308
3309                         /*
3310                          * Deal with a hole that doesn't have an extent item
3311                          * that represents it (NO_HOLES feature enabled).
3312                          * This hole is either in the middle of the cloning
3313                          * range or at the beginning (fully overlaps it or
3314                          * partially overlaps it).
3315                          */
3316                         if (new_key.offset != last_dest_end)
3317                                 drop_start = last_dest_end;
3318                         else
3319                                 drop_start = new_key.offset;
3320
3321                         /*
3322                          * 1 - adjusting old extent (we may have to split it)
3323                          * 1 - add new extent
3324                          * 1 - inode update
3325                          */
3326                         trans = btrfs_start_transaction(root, 3);
3327                         if (IS_ERR(trans)) {
3328                                 ret = PTR_ERR(trans);
3329                                 goto out;
3330                         }
3331
3332                         if (type == BTRFS_FILE_EXTENT_REG ||
3333                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3334                                 /*
3335                                  *    a  | --- range to clone ---|  b
3336                                  * | ------------- extent ------------- |
3337                                  */
3338
3339                                 /* subtract range b */
3340                                 if (key.offset + datal > off + len)
3341                                         datal = off + len - key.offset;
3342
3343                                 /* subtract range a */
3344                                 if (off > key.offset) {
3345                                         datao += off - key.offset;
3346                                         datal -= off - key.offset;
3347                                 }
3348
3349                                 ret = btrfs_drop_extents(trans, root, inode,
3350                                                          drop_start,
3351                                                          new_key.offset + datal,
3352                                                          1);
3353                                 if (ret) {
3354                                         if (ret != -EOPNOTSUPP)
3355                                                 btrfs_abort_transaction(trans,
3356                                                                 root, ret);
3357                                         btrfs_end_transaction(trans, root);
3358                                         goto out;
3359                                 }
3360
3361                                 ret = btrfs_insert_empty_item(trans, root, path,
3362                                                               &new_key, size);
3363                                 if (ret) {
3364                                         btrfs_abort_transaction(trans, root,
3365                                                                 ret);
3366                                         btrfs_end_transaction(trans, root);
3367                                         goto out;
3368                                 }
3369
3370                                 leaf = path->nodes[0];
3371                                 slot = path->slots[0];
3372                                 write_extent_buffer(leaf, buf,
3373                                             btrfs_item_ptr_offset(leaf, slot),
3374                                             size);
3375
3376                                 extent = btrfs_item_ptr(leaf, slot,
3377                                                 struct btrfs_file_extent_item);
3378
3379                                 /* disko == 0 means it's a hole */
3380                                 if (!disko)
3381                                         datao = 0;
3382
3383                                 btrfs_set_file_extent_offset(leaf, extent,
3384                                                              datao);
3385                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3386                                                                 datal);
3387
3388                                 /*
3389                                  * We need to look up the roots that point at
3390                                  * this bytenr and see if the new root does.  If
3391                                  * it does not we need to make sure we update
3392                                  * quotas appropriately.
3393                                  */
3394                                 if (disko && root != BTRFS_I(src)->root &&
3395                                     disko != last_disko) {
3396                                         no_quota = check_ref(trans, root,
3397                                                              disko);
3398                                         if (no_quota < 0) {
3399                                                 btrfs_abort_transaction(trans,
3400                                                                         root,
3401                                                                         ret);
3402                                                 btrfs_end_transaction(trans,
3403                                                                       root);
3404                                                 ret = no_quota;
3405                                                 goto out;
3406                                         }
3407                                 }
3408
3409                                 if (disko) {
3410                                         inode_add_bytes(inode, datal);
3411                                         ret = btrfs_inc_extent_ref(trans, root,
3412                                                         disko, diskl, 0,
3413                                                         root->root_key.objectid,
3414                                                         btrfs_ino(inode),
3415                                                         new_key.offset - datao,
3416                                                         no_quota);
3417                                         if (ret) {
3418                                                 btrfs_abort_transaction(trans,
3419                                                                         root,
3420                                                                         ret);
3421                                                 btrfs_end_transaction(trans,
3422                                                                       root);
3423                                                 goto out;
3424
3425                                         }
3426                                 }
3427                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3428                                 u64 skip = 0;
3429                                 u64 trim = 0;
3430                                 u64 aligned_end = 0;
3431
3432                                 if (off > key.offset) {
3433                                         skip = off - key.offset;
3434                                         new_key.offset += skip;
3435                                 }
3436
3437                                 if (key.offset + datal > off + len)
3438                                         trim = key.offset + datal - (off + len);
3439
3440                                 if (comp && (skip || trim)) {
3441                                         ret = -EINVAL;
3442                                         btrfs_end_transaction(trans, root);
3443                                         goto out;
3444                                 }
3445                                 size -= skip + trim;
3446                                 datal -= skip + trim;
3447
3448                                 aligned_end = ALIGN(new_key.offset + datal,
3449                                                     root->sectorsize);
3450                                 ret = btrfs_drop_extents(trans, root, inode,
3451                                                          drop_start,
3452                                                          aligned_end,
3453                                                          1);
3454                                 if (ret) {
3455                                         if (ret != -EOPNOTSUPP)
3456                                                 btrfs_abort_transaction(trans,
3457                                                         root, ret);
3458                                         btrfs_end_transaction(trans, root);
3459                                         goto out;
3460                                 }
3461
3462                                 ret = btrfs_insert_empty_item(trans, root, path,
3463                                                               &new_key, size);
3464                                 if (ret) {
3465                                         btrfs_abort_transaction(trans, root,
3466                                                                 ret);
3467                                         btrfs_end_transaction(trans, root);
3468                                         goto out;
3469                                 }
3470
3471                                 if (skip) {
3472                                         u32 start =
3473                                           btrfs_file_extent_calc_inline_size(0);
3474                                         memmove(buf+start, buf+start+skip,
3475                                                 datal);
3476                                 }
3477
3478                                 leaf = path->nodes[0];
3479                                 slot = path->slots[0];
3480                                 write_extent_buffer(leaf, buf,
3481                                             btrfs_item_ptr_offset(leaf, slot),
3482                                             size);
3483                                 inode_add_bytes(inode, datal);
3484                         }
3485
3486                         /* If we have an implicit hole (NO_HOLES feature). */
3487                         if (drop_start < new_key.offset)
3488                                 clone_update_extent_map(inode, trans,
3489                                                 NULL, drop_start,
3490                                                 new_key.offset - drop_start);
3491
3492                         clone_update_extent_map(inode, trans, path, 0, 0);
3493
3494                         btrfs_mark_buffer_dirty(leaf);
3495                         btrfs_release_path(path);
3496
3497                         last_dest_end = ALIGN(new_key.offset + datal,
3498                                               root->sectorsize);
3499                         ret = clone_finish_inode_update(trans, inode,
3500                                                         last_dest_end,
3501                                                         destoff, olen);
3502                         if (ret)
3503                                 goto out;
3504                         if (new_key.offset + datal >= destoff + len)
3505                                 break;
3506                 }
3507                 btrfs_release_path(path);
3508                 key.offset++;
3509         }
3510         ret = 0;
3511
3512         if (last_dest_end < destoff + len) {
3513                 /*
3514                  * We have an implicit hole (NO_HOLES feature is enabled) that
3515                  * fully or partially overlaps our cloning range at its end.
3516                  */
3517                 btrfs_release_path(path);
3518
3519                 /*
3520                  * 1 - remove extent(s)
3521                  * 1 - inode update
3522                  */
3523                 trans = btrfs_start_transaction(root, 2);
3524                 if (IS_ERR(trans)) {
3525                         ret = PTR_ERR(trans);
3526                         goto out;
3527                 }
3528                 ret = btrfs_drop_extents(trans, root, inode,
3529                                          last_dest_end, destoff + len, 1);
3530                 if (ret) {
3531                         if (ret != -EOPNOTSUPP)
3532                                 btrfs_abort_transaction(trans, root, ret);
3533                         btrfs_end_transaction(trans, root);
3534                         goto out;
3535                 }
3536                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3537                                         destoff + len - last_dest_end);
3538                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3539                                                 destoff, olen);
3540         }
3541
3542 out:
3543         btrfs_free_path(path);
3544         vfree(buf);
3545         return ret;
3546 }
3547
3548 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3549                                        u64 off, u64 olen, u64 destoff)
3550 {
3551         struct inode *inode = file_inode(file);
3552         struct btrfs_root *root = BTRFS_I(inode)->root;
3553         struct fd src_file;
3554         struct inode *src;
3555         int ret;
3556         u64 len = olen;
3557         u64 bs = root->fs_info->sb->s_blocksize;
3558         int same_inode = 0;
3559
3560         /*
3561          * TODO:
3562          * - split compressed inline extents.  annoying: we need to
3563          *   decompress into destination's address_space (the file offset
3564          *   may change, so source mapping won't do), then recompress (or
3565          *   otherwise reinsert) a subrange.
3566          *
3567          * - split destination inode's inline extents.  The inline extents can
3568          *   be either compressed or non-compressed.
3569          */
3570
3571         /* the destination must be opened for writing */
3572         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3573                 return -EINVAL;
3574
3575         if (btrfs_root_readonly(root))
3576                 return -EROFS;
3577
3578         ret = mnt_want_write_file(file);
3579         if (ret)
3580                 return ret;
3581
3582         src_file = fdget(srcfd);
3583         if (!src_file.file) {
3584                 ret = -EBADF;
3585                 goto out_drop_write;
3586         }
3587
3588         ret = -EXDEV;
3589         if (src_file.file->f_path.mnt != file->f_path.mnt)
3590                 goto out_fput;
3591
3592         src = file_inode(src_file.file);
3593
3594         ret = -EINVAL;
3595         if (src == inode)
3596                 same_inode = 1;
3597
3598         /* the src must be open for reading */
3599         if (!(src_file.file->f_mode & FMODE_READ))
3600                 goto out_fput;
3601
3602         /* don't make the dst file partly checksummed */
3603         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3604             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3605                 goto out_fput;
3606
3607         ret = -EISDIR;
3608         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3609                 goto out_fput;
3610
3611         ret = -EXDEV;
3612         if (src->i_sb != inode->i_sb)
3613                 goto out_fput;
3614
3615         if (!same_inode) {
3616                 if (inode < src) {
3617                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3618                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3619                 } else {
3620                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3621                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3622                 }
3623         } else {
3624                 mutex_lock(&src->i_mutex);
3625         }
3626
3627         /* determine range to clone */
3628         ret = -EINVAL;
3629         if (off + len > src->i_size || off + len < off)
3630                 goto out_unlock;
3631         if (len == 0)
3632                 olen = len = src->i_size - off;
3633         /* if we extend to eof, continue to block boundary */
3634         if (off + len == src->i_size)
3635                 len = ALIGN(src->i_size, bs) - off;
3636
3637         /* verify the end result is block aligned */
3638         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3639             !IS_ALIGNED(destoff, bs))
3640                 goto out_unlock;
3641
3642         /* verify if ranges are overlapped within the same file */
3643         if (same_inode) {
3644                 if (destoff + len > off && destoff < off + len)
3645                         goto out_unlock;
3646         }
3647
3648         if (destoff > inode->i_size) {
3649                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3650                 if (ret)
3651                         goto out_unlock;
3652         }
3653
3654         /*
3655          * Lock the target range too. Right after we replace the file extent
3656          * items in the fs tree (which now point to the cloned data), we might
3657          * have a worker replace them with extent items relative to a write
3658          * operation that was issued before this clone operation (i.e. confront
3659          * with inode.c:btrfs_finish_ordered_io).
3660          */
3661         if (same_inode) {
3662                 u64 lock_start = min_t(u64, off, destoff);
3663                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3664
3665                 lock_extent_range(src, lock_start, lock_len);
3666         } else {
3667                 lock_extent_range(src, off, len);
3668                 lock_extent_range(inode, destoff, len);
3669         }
3670
3671         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3672
3673         if (same_inode) {
3674                 u64 lock_start = min_t(u64, off, destoff);
3675                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3676
3677                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3678         } else {
3679                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3680                 unlock_extent(&BTRFS_I(inode)->io_tree, destoff,
3681                               destoff + len - 1);
3682         }
3683         /*
3684          * Truncate page cache pages so that future reads will see the cloned
3685          * data immediately and not the previous data.
3686          */
3687         truncate_inode_pages_range(&inode->i_data, destoff,
3688                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3689 out_unlock:
3690         if (!same_inode) {
3691                 if (inode < src) {
3692                         mutex_unlock(&src->i_mutex);
3693                         mutex_unlock(&inode->i_mutex);
3694                 } else {
3695                         mutex_unlock(&inode->i_mutex);
3696                         mutex_unlock(&src->i_mutex);
3697                 }
3698         } else {
3699                 mutex_unlock(&src->i_mutex);
3700         }
3701 out_fput:
3702         fdput(src_file);
3703 out_drop_write:
3704         mnt_drop_write_file(file);
3705         return ret;
3706 }
3707
3708 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3709 {
3710         struct btrfs_ioctl_clone_range_args args;
3711
3712         if (copy_from_user(&args, argp, sizeof(args)))
3713                 return -EFAULT;
3714         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3715                                  args.src_length, args.dest_offset);
3716 }
3717
3718 /*
3719  * there are many ways the trans_start and trans_end ioctls can lead
3720  * to deadlocks.  They should only be used by applications that
3721  * basically own the machine, and have a very in depth understanding
3722  * of all the possible deadlocks and enospc problems.
3723  */
3724 static long btrfs_ioctl_trans_start(struct file *file)
3725 {
3726         struct inode *inode = file_inode(file);
3727         struct btrfs_root *root = BTRFS_I(inode)->root;
3728         struct btrfs_trans_handle *trans;
3729         int ret;
3730
3731         ret = -EPERM;
3732         if (!capable(CAP_SYS_ADMIN))
3733                 goto out;
3734
3735         ret = -EINPROGRESS;
3736         if (file->private_data)
3737                 goto out;
3738
3739         ret = -EROFS;
3740         if (btrfs_root_readonly(root))
3741                 goto out;
3742
3743         ret = mnt_want_write_file(file);
3744         if (ret)
3745                 goto out;
3746
3747         atomic_inc(&root->fs_info->open_ioctl_trans);
3748
3749         ret = -ENOMEM;
3750         trans = btrfs_start_ioctl_transaction(root);
3751         if (IS_ERR(trans))
3752                 goto out_drop;
3753
3754         file->private_data = trans;
3755         return 0;
3756
3757 out_drop:
3758         atomic_dec(&root->fs_info->open_ioctl_trans);
3759         mnt_drop_write_file(file);
3760 out:
3761         return ret;
3762 }
3763
3764 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3765 {
3766         struct inode *inode = file_inode(file);
3767         struct btrfs_root *root = BTRFS_I(inode)->root;
3768         struct btrfs_root *new_root;
3769         struct btrfs_dir_item *di;
3770         struct btrfs_trans_handle *trans;
3771         struct btrfs_path *path;
3772         struct btrfs_key location;
3773         struct btrfs_disk_key disk_key;
3774         u64 objectid = 0;
3775         u64 dir_id;
3776         int ret;
3777
3778         if (!capable(CAP_SYS_ADMIN))
3779                 return -EPERM;
3780
3781         ret = mnt_want_write_file(file);
3782         if (ret)
3783                 return ret;
3784
3785         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3786                 ret = -EFAULT;
3787                 goto out;
3788         }
3789
3790         if (!objectid)
3791                 objectid = BTRFS_FS_TREE_OBJECTID;
3792
3793         location.objectid = objectid;
3794         location.type = BTRFS_ROOT_ITEM_KEY;
3795         location.offset = (u64)-1;
3796
3797         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3798         if (IS_ERR(new_root)) {
3799                 ret = PTR_ERR(new_root);
3800                 goto out;
3801         }
3802
3803         path = btrfs_alloc_path();
3804         if (!path) {
3805                 ret = -ENOMEM;
3806                 goto out;
3807         }
3808         path->leave_spinning = 1;
3809
3810         trans = btrfs_start_transaction(root, 1);
3811         if (IS_ERR(trans)) {
3812                 btrfs_free_path(path);
3813                 ret = PTR_ERR(trans);
3814                 goto out;
3815         }
3816
3817         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3818         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3819                                    dir_id, "default", 7, 1);
3820         if (IS_ERR_OR_NULL(di)) {
3821                 btrfs_free_path(path);
3822                 btrfs_end_transaction(trans, root);
3823                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3824                            "item, this isn't going to work");
3825                 ret = -ENOENT;
3826                 goto out;
3827         }
3828
3829         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3830         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3831         btrfs_mark_buffer_dirty(path->nodes[0]);
3832         btrfs_free_path(path);
3833
3834         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3835         btrfs_end_transaction(trans, root);
3836 out:
3837         mnt_drop_write_file(file);
3838         return ret;
3839 }
3840
3841 void btrfs_get_block_group_info(struct list_head *groups_list,
3842                                 struct btrfs_ioctl_space_info *space)
3843 {
3844         struct btrfs_block_group_cache *block_group;
3845
3846         space->total_bytes = 0;
3847         space->used_bytes = 0;
3848         space->flags = 0;
3849         list_for_each_entry(block_group, groups_list, list) {
3850                 space->flags = block_group->flags;
3851                 space->total_bytes += block_group->key.offset;
3852                 space->used_bytes +=
3853                         btrfs_block_group_used(&block_group->item);
3854         }
3855 }
3856
3857 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3858 {
3859         struct btrfs_ioctl_space_args space_args;
3860         struct btrfs_ioctl_space_info space;
3861         struct btrfs_ioctl_space_info *dest;
3862         struct btrfs_ioctl_space_info *dest_orig;
3863         struct btrfs_ioctl_space_info __user *user_dest;
3864         struct btrfs_space_info *info;
3865         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3866                        BTRFS_BLOCK_GROUP_SYSTEM,
3867                        BTRFS_BLOCK_GROUP_METADATA,
3868                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3869         int num_types = 4;
3870         int alloc_size;
3871         int ret = 0;
3872         u64 slot_count = 0;
3873         int i, c;
3874
3875         if (copy_from_user(&space_args,
3876                            (struct btrfs_ioctl_space_args __user *)arg,
3877                            sizeof(space_args)))
3878                 return -EFAULT;
3879
3880         for (i = 0; i < num_types; i++) {
3881                 struct btrfs_space_info *tmp;
3882
3883                 info = NULL;
3884                 rcu_read_lock();
3885                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3886                                         list) {
3887                         if (tmp->flags == types[i]) {
3888                                 info = tmp;
3889                                 break;
3890                         }
3891                 }
3892                 rcu_read_unlock();
3893
3894                 if (!info)
3895                         continue;
3896
3897                 down_read(&info->groups_sem);
3898                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3899                         if (!list_empty(&info->block_groups[c]))
3900                                 slot_count++;
3901                 }
3902                 up_read(&info->groups_sem);
3903         }
3904
3905         /*
3906          * Global block reserve, exported as a space_info
3907          */
3908         slot_count++;
3909
3910         /* space_slots == 0 means they are asking for a count */
3911         if (space_args.space_slots == 0) {
3912                 space_args.total_spaces = slot_count;
3913                 goto out;
3914         }
3915
3916         slot_count = min_t(u64, space_args.space_slots, slot_count);
3917
3918         alloc_size = sizeof(*dest) * slot_count;
3919
3920         /* we generally have at most 6 or so space infos, one for each raid
3921          * level.  So, a whole page should be more than enough for everyone
3922          */
3923         if (alloc_size > PAGE_CACHE_SIZE)
3924                 return -ENOMEM;
3925
3926         space_args.total_spaces = 0;
3927         dest = kmalloc(alloc_size, GFP_NOFS);
3928         if (!dest)
3929                 return -ENOMEM;
3930         dest_orig = dest;
3931
3932         /* now we have a buffer to copy into */
3933         for (i = 0; i < num_types; i++) {
3934                 struct btrfs_space_info *tmp;
3935
3936                 if (!slot_count)
3937                         break;
3938
3939                 info = NULL;
3940                 rcu_read_lock();
3941                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3942                                         list) {
3943                         if (tmp->flags == types[i]) {
3944                                 info = tmp;
3945                                 break;
3946                         }
3947                 }
3948                 rcu_read_unlock();
3949
3950                 if (!info)
3951                         continue;
3952                 down_read(&info->groups_sem);
3953                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3954                         if (!list_empty(&info->block_groups[c])) {
3955                                 btrfs_get_block_group_info(
3956                                         &info->block_groups[c], &space);
3957                                 memcpy(dest, &space, sizeof(space));
3958                                 dest++;
3959                                 space_args.total_spaces++;
3960                                 slot_count--;
3961                         }
3962                         if (!slot_count)
3963                                 break;
3964                 }
3965                 up_read(&info->groups_sem);
3966         }
3967
3968         /*
3969          * Add global block reserve
3970          */
3971         if (slot_count) {
3972                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3973
3974                 spin_lock(&block_rsv->lock);
3975                 space.total_bytes = block_rsv->size;
3976                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3977                 spin_unlock(&block_rsv->lock);
3978                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3979                 memcpy(dest, &space, sizeof(space));
3980                 space_args.total_spaces++;
3981         }
3982
3983         user_dest = (struct btrfs_ioctl_space_info __user *)
3984                 (arg + sizeof(struct btrfs_ioctl_space_args));
3985
3986         if (copy_to_user(user_dest, dest_orig, alloc_size))
3987                 ret = -EFAULT;
3988
3989         kfree(dest_orig);
3990 out:
3991         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3992                 ret = -EFAULT;
3993
3994         return ret;
3995 }
3996
3997 /*
3998  * there are many ways the trans_start and trans_end ioctls can lead
3999  * to deadlocks.  They should only be used by applications that
4000  * basically own the machine, and have a very in depth understanding
4001  * of all the possible deadlocks and enospc problems.
4002  */
4003 long btrfs_ioctl_trans_end(struct file *file)
4004 {
4005         struct inode *inode = file_inode(file);
4006         struct btrfs_root *root = BTRFS_I(inode)->root;
4007         struct btrfs_trans_handle *trans;
4008
4009         trans = file->private_data;
4010         if (!trans)
4011                 return -EINVAL;
4012         file->private_data = NULL;
4013
4014         btrfs_end_transaction(trans, root);
4015
4016         atomic_dec(&root->fs_info->open_ioctl_trans);
4017
4018         mnt_drop_write_file(file);
4019         return 0;
4020 }
4021
4022 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4023                                             void __user *argp)
4024 {
4025         struct btrfs_trans_handle *trans;
4026         u64 transid;
4027         int ret;
4028
4029         trans = btrfs_attach_transaction_barrier(root);
4030         if (IS_ERR(trans)) {
4031                 if (PTR_ERR(trans) != -ENOENT)
4032                         return PTR_ERR(trans);
4033
4034                 /* No running transaction, don't bother */
4035                 transid = root->fs_info->last_trans_committed;
4036                 goto out;
4037         }
4038         transid = trans->transid;
4039         ret = btrfs_commit_transaction_async(trans, root, 0);
4040         if (ret) {
4041                 btrfs_end_transaction(trans, root);
4042                 return ret;
4043         }
4044 out:
4045         if (argp)
4046                 if (copy_to_user(argp, &transid, sizeof(transid)))
4047                         return -EFAULT;
4048         return 0;
4049 }
4050
4051 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4052                                            void __user *argp)
4053 {
4054         u64 transid;
4055
4056         if (argp) {
4057                 if (copy_from_user(&transid, argp, sizeof(transid)))
4058                         return -EFAULT;
4059         } else {
4060                 transid = 0;  /* current trans */
4061         }
4062         return btrfs_wait_for_commit(root, transid);
4063 }
4064
4065 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4066 {
4067         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4068         struct btrfs_ioctl_scrub_args *sa;
4069         int ret;
4070
4071         if (!capable(CAP_SYS_ADMIN))
4072                 return -EPERM;
4073
4074         sa = memdup_user(arg, sizeof(*sa));
4075         if (IS_ERR(sa))
4076                 return PTR_ERR(sa);
4077
4078         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4079                 ret = mnt_want_write_file(file);
4080                 if (ret)
4081                         goto out;
4082         }
4083
4084         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4085                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4086                               0);
4087
4088         if (copy_to_user(arg, sa, sizeof(*sa)))
4089                 ret = -EFAULT;
4090
4091         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4092                 mnt_drop_write_file(file);
4093 out:
4094         kfree(sa);
4095         return ret;
4096 }
4097
4098 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4099 {
4100         if (!capable(CAP_SYS_ADMIN))
4101                 return -EPERM;
4102
4103         return btrfs_scrub_cancel(root->fs_info);
4104 }
4105
4106 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4107                                        void __user *arg)
4108 {
4109         struct btrfs_ioctl_scrub_args *sa;
4110         int ret;
4111
4112         if (!capable(CAP_SYS_ADMIN))
4113                 return -EPERM;
4114
4115         sa = memdup_user(arg, sizeof(*sa));
4116         if (IS_ERR(sa))
4117                 return PTR_ERR(sa);
4118
4119         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4120
4121         if (copy_to_user(arg, sa, sizeof(*sa)))
4122                 ret = -EFAULT;
4123
4124         kfree(sa);
4125         return ret;
4126 }
4127
4128 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4129                                       void __user *arg)
4130 {
4131         struct btrfs_ioctl_get_dev_stats *sa;
4132         int ret;
4133
4134         sa = memdup_user(arg, sizeof(*sa));
4135         if (IS_ERR(sa))
4136                 return PTR_ERR(sa);
4137
4138         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4139                 kfree(sa);
4140                 return -EPERM;
4141         }
4142
4143         ret = btrfs_get_dev_stats(root, sa);
4144
4145         if (copy_to_user(arg, sa, sizeof(*sa)))
4146                 ret = -EFAULT;
4147
4148         kfree(sa);
4149         return ret;
4150 }
4151
4152 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4153 {
4154         struct btrfs_ioctl_dev_replace_args *p;
4155         int ret;
4156
4157         if (!capable(CAP_SYS_ADMIN))
4158                 return -EPERM;
4159
4160         p = memdup_user(arg, sizeof(*p));
4161         if (IS_ERR(p))
4162                 return PTR_ERR(p);
4163
4164         switch (p->cmd) {
4165         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4166                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4167                         ret = -EROFS;
4168                         goto out;
4169                 }
4170                 if (atomic_xchg(
4171                         &root->fs_info->mutually_exclusive_operation_running,
4172                         1)) {
4173                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4174                 } else {
4175                         ret = btrfs_dev_replace_start(root, p);
4176                         atomic_set(
4177                          &root->fs_info->mutually_exclusive_operation_running,
4178                          0);
4179                 }
4180                 break;
4181         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4182                 btrfs_dev_replace_status(root->fs_info, p);
4183                 ret = 0;
4184                 break;
4185         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4186                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4187                 break;
4188         default:
4189                 ret = -EINVAL;
4190                 break;
4191         }
4192
4193         if (copy_to_user(arg, p, sizeof(*p)))
4194                 ret = -EFAULT;
4195 out:
4196         kfree(p);
4197         return ret;
4198 }
4199
4200 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4201 {
4202         int ret = 0;
4203         int i;
4204         u64 rel_ptr;
4205         int size;
4206         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4207         struct inode_fs_paths *ipath = NULL;
4208         struct btrfs_path *path;
4209
4210         if (!capable(CAP_DAC_READ_SEARCH))
4211                 return -EPERM;
4212
4213         path = btrfs_alloc_path();
4214         if (!path) {
4215                 ret = -ENOMEM;
4216                 goto out;
4217         }
4218
4219         ipa = memdup_user(arg, sizeof(*ipa));
4220         if (IS_ERR(ipa)) {
4221                 ret = PTR_ERR(ipa);
4222                 ipa = NULL;
4223                 goto out;
4224         }
4225
4226         size = min_t(u32, ipa->size, 4096);
4227         ipath = init_ipath(size, root, path);
4228         if (IS_ERR(ipath)) {
4229                 ret = PTR_ERR(ipath);
4230                 ipath = NULL;
4231                 goto out;
4232         }
4233
4234         ret = paths_from_inode(ipa->inum, ipath);
4235         if (ret < 0)
4236                 goto out;
4237
4238         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4239                 rel_ptr = ipath->fspath->val[i] -
4240                           (u64)(unsigned long)ipath->fspath->val;
4241                 ipath->fspath->val[i] = rel_ptr;
4242         }
4243
4244         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4245                            (void *)(unsigned long)ipath->fspath, size);
4246         if (ret) {
4247                 ret = -EFAULT;
4248                 goto out;
4249         }
4250
4251 out:
4252         btrfs_free_path(path);
4253         free_ipath(ipath);
4254         kfree(ipa);
4255
4256         return ret;
4257 }
4258
4259 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4260 {
4261         struct btrfs_data_container *inodes = ctx;
4262         const size_t c = 3 * sizeof(u64);
4263
4264         if (inodes->bytes_left >= c) {
4265                 inodes->bytes_left -= c;
4266                 inodes->val[inodes->elem_cnt] = inum;
4267                 inodes->val[inodes->elem_cnt + 1] = offset;
4268                 inodes->val[inodes->elem_cnt + 2] = root;
4269                 inodes->elem_cnt += 3;
4270         } else {
4271                 inodes->bytes_missing += c - inodes->bytes_left;
4272                 inodes->bytes_left = 0;
4273                 inodes->elem_missed += 3;
4274         }
4275
4276         return 0;
4277 }
4278
4279 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4280                                         void __user *arg)
4281 {
4282         int ret = 0;
4283         int size;
4284         struct btrfs_ioctl_logical_ino_args *loi;
4285         struct btrfs_data_container *inodes = NULL;
4286         struct btrfs_path *path = NULL;
4287
4288         if (!capable(CAP_SYS_ADMIN))
4289                 return -EPERM;
4290
4291         loi = memdup_user(arg, sizeof(*loi));
4292         if (IS_ERR(loi)) {
4293                 ret = PTR_ERR(loi);
4294                 loi = NULL;
4295                 goto out;
4296         }
4297
4298         path = btrfs_alloc_path();
4299         if (!path) {
4300                 ret = -ENOMEM;
4301                 goto out;
4302         }
4303
4304         size = min_t(u32, loi->size, 64 * 1024);
4305         inodes = init_data_container(size);
4306         if (IS_ERR(inodes)) {
4307                 ret = PTR_ERR(inodes);
4308                 inodes = NULL;
4309                 goto out;
4310         }
4311
4312         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4313                                           build_ino_list, inodes);
4314         if (ret == -EINVAL)
4315                 ret = -ENOENT;
4316         if (ret < 0)
4317                 goto out;
4318
4319         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4320                            (void *)(unsigned long)inodes, size);
4321         if (ret)
4322                 ret = -EFAULT;
4323
4324 out:
4325         btrfs_free_path(path);
4326         vfree(inodes);
4327         kfree(loi);
4328
4329         return ret;
4330 }
4331
4332 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4333                                struct btrfs_ioctl_balance_args *bargs)
4334 {
4335         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4336
4337         bargs->flags = bctl->flags;
4338
4339         if (atomic_read(&fs_info->balance_running))
4340                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4341         if (atomic_read(&fs_info->balance_pause_req))
4342                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4343         if (atomic_read(&fs_info->balance_cancel_req))
4344                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4345
4346         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4347         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4348         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4349
4350         if (lock) {
4351                 spin_lock(&fs_info->balance_lock);
4352                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4353                 spin_unlock(&fs_info->balance_lock);
4354         } else {
4355                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4356         }
4357 }
4358
4359 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4360 {
4361         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4362         struct btrfs_fs_info *fs_info = root->fs_info;
4363         struct btrfs_ioctl_balance_args *bargs;
4364         struct btrfs_balance_control *bctl;
4365         bool need_unlock; /* for mut. excl. ops lock */
4366         int ret;
4367
4368         if (!capable(CAP_SYS_ADMIN))
4369                 return -EPERM;
4370
4371         ret = mnt_want_write_file(file);
4372         if (ret)
4373                 return ret;
4374
4375 again:
4376         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4377                 mutex_lock(&fs_info->volume_mutex);
4378                 mutex_lock(&fs_info->balance_mutex);
4379                 need_unlock = true;
4380                 goto locked;
4381         }
4382
4383         /*
4384          * mut. excl. ops lock is locked.  Three possibilites:
4385          *   (1) some other op is running
4386          *   (2) balance is running
4387          *   (3) balance is paused -- special case (think resume)
4388          */
4389         mutex_lock(&fs_info->balance_mutex);
4390         if (fs_info->balance_ctl) {
4391                 /* this is either (2) or (3) */
4392                 if (!atomic_read(&fs_info->balance_running)) {
4393                         mutex_unlock(&fs_info->balance_mutex);
4394                         if (!mutex_trylock(&fs_info->volume_mutex))
4395                                 goto again;
4396                         mutex_lock(&fs_info->balance_mutex);
4397
4398                         if (fs_info->balance_ctl &&
4399                             !atomic_read(&fs_info->balance_running)) {
4400                                 /* this is (3) */
4401                                 need_unlock = false;
4402                                 goto locked;
4403                         }
4404
4405                         mutex_unlock(&fs_info->balance_mutex);
4406                         mutex_unlock(&fs_info->volume_mutex);
4407                         goto again;
4408                 } else {
4409                         /* this is (2) */
4410                         mutex_unlock(&fs_info->balance_mutex);
4411                         ret = -EINPROGRESS;
4412                         goto out;
4413                 }
4414         } else {
4415                 /* this is (1) */
4416                 mutex_unlock(&fs_info->balance_mutex);
4417                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4418                 goto out;
4419         }
4420
4421 locked:
4422         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4423
4424         if (arg) {
4425                 bargs = memdup_user(arg, sizeof(*bargs));
4426                 if (IS_ERR(bargs)) {
4427                         ret = PTR_ERR(bargs);
4428                         goto out_unlock;
4429                 }
4430
4431                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4432                         if (!fs_info->balance_ctl) {
4433                                 ret = -ENOTCONN;
4434                                 goto out_bargs;
4435                         }
4436
4437                         bctl = fs_info->balance_ctl;
4438                         spin_lock(&fs_info->balance_lock);
4439                         bctl->flags |= BTRFS_BALANCE_RESUME;
4440                         spin_unlock(&fs_info->balance_lock);
4441
4442                         goto do_balance;
4443                 }
4444         } else {
4445                 bargs = NULL;
4446         }
4447
4448         if (fs_info->balance_ctl) {
4449                 ret = -EINPROGRESS;
4450                 goto out_bargs;
4451         }
4452
4453         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4454         if (!bctl) {
4455                 ret = -ENOMEM;
4456                 goto out_bargs;
4457         }
4458
4459         bctl->fs_info = fs_info;
4460         if (arg) {
4461                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4462                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4463                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4464
4465                 bctl->flags = bargs->flags;
4466         } else {
4467                 /* balance everything - no filters */
4468                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4469         }
4470
4471 do_balance:
4472         /*
4473          * Ownership of bctl and mutually_exclusive_operation_running
4474          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4475          * or, if restriper was paused all the way until unmount, in
4476          * free_fs_info.  mutually_exclusive_operation_running is
4477          * cleared in __cancel_balance.
4478          */
4479         need_unlock = false;
4480
4481         ret = btrfs_balance(bctl, bargs);
4482
4483         if (arg) {
4484                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4485                         ret = -EFAULT;
4486         }
4487
4488 out_bargs:
4489         kfree(bargs);
4490 out_unlock:
4491         mutex_unlock(&fs_info->balance_mutex);
4492         mutex_unlock(&fs_info->volume_mutex);
4493         if (need_unlock)
4494                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4495 out:
4496         mnt_drop_write_file(file);
4497         return ret;
4498 }
4499
4500 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4501 {
4502         if (!capable(CAP_SYS_ADMIN))
4503                 return -EPERM;
4504
4505         switch (cmd) {
4506         case BTRFS_BALANCE_CTL_PAUSE:
4507                 return btrfs_pause_balance(root->fs_info);
4508         case BTRFS_BALANCE_CTL_CANCEL:
4509                 return btrfs_cancel_balance(root->fs_info);
4510         }
4511
4512         return -EINVAL;
4513 }
4514
4515 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4516                                          void __user *arg)
4517 {
4518         struct btrfs_fs_info *fs_info = root->fs_info;
4519         struct btrfs_ioctl_balance_args *bargs;
4520         int ret = 0;
4521
4522         if (!capable(CAP_SYS_ADMIN))
4523                 return -EPERM;
4524
4525         mutex_lock(&fs_info->balance_mutex);
4526         if (!fs_info->balance_ctl) {
4527                 ret = -ENOTCONN;
4528                 goto out;
4529         }
4530
4531         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4532         if (!bargs) {
4533                 ret = -ENOMEM;
4534                 goto out;
4535         }
4536
4537         update_ioctl_balance_args(fs_info, 1, bargs);
4538
4539         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4540                 ret = -EFAULT;
4541
4542         kfree(bargs);
4543 out:
4544         mutex_unlock(&fs_info->balance_mutex);
4545         return ret;
4546 }
4547
4548 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4549 {
4550         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4551         struct btrfs_ioctl_quota_ctl_args *sa;
4552         struct btrfs_trans_handle *trans = NULL;
4553         int ret;
4554         int err;
4555
4556         if (!capable(CAP_SYS_ADMIN))
4557                 return -EPERM;
4558
4559         ret = mnt_want_write_file(file);
4560         if (ret)
4561                 return ret;
4562
4563         sa = memdup_user(arg, sizeof(*sa));
4564         if (IS_ERR(sa)) {
4565                 ret = PTR_ERR(sa);
4566                 goto drop_write;
4567         }
4568
4569         down_write(&root->fs_info->subvol_sem);
4570         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4571         if (IS_ERR(trans)) {
4572                 ret = PTR_ERR(trans);
4573                 goto out;
4574         }
4575
4576         switch (sa->cmd) {
4577         case BTRFS_QUOTA_CTL_ENABLE:
4578                 ret = btrfs_quota_enable(trans, root->fs_info);
4579                 break;
4580         case BTRFS_QUOTA_CTL_DISABLE:
4581                 ret = btrfs_quota_disable(trans, root->fs_info);
4582                 break;
4583         default:
4584                 ret = -EINVAL;
4585                 break;
4586         }
4587
4588         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4589         if (err && !ret)
4590                 ret = err;
4591 out:
4592         kfree(sa);
4593         up_write(&root->fs_info->subvol_sem);
4594 drop_write:
4595         mnt_drop_write_file(file);
4596         return ret;
4597 }
4598
4599 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4600 {
4601         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4602         struct btrfs_ioctl_qgroup_assign_args *sa;
4603         struct btrfs_trans_handle *trans;
4604         int ret;
4605         int err;
4606
4607         if (!capable(CAP_SYS_ADMIN))
4608                 return -EPERM;
4609
4610         ret = mnt_want_write_file(file);
4611         if (ret)
4612                 return ret;
4613
4614         sa = memdup_user(arg, sizeof(*sa));
4615         if (IS_ERR(sa)) {
4616                 ret = PTR_ERR(sa);
4617                 goto drop_write;
4618         }
4619
4620         trans = btrfs_join_transaction(root);
4621         if (IS_ERR(trans)) {
4622                 ret = PTR_ERR(trans);
4623                 goto out;
4624         }
4625
4626         /* FIXME: check if the IDs really exist */
4627         if (sa->assign) {
4628                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4629                                                 sa->src, sa->dst);
4630         } else {
4631                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4632                                                 sa->src, sa->dst);
4633         }
4634
4635         err = btrfs_end_transaction(trans, root);
4636         if (err && !ret)
4637                 ret = err;
4638
4639 out:
4640         kfree(sa);
4641 drop_write:
4642         mnt_drop_write_file(file);
4643         return ret;
4644 }
4645
4646 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4647 {
4648         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4649         struct btrfs_ioctl_qgroup_create_args *sa;
4650         struct btrfs_trans_handle *trans;
4651         int ret;
4652         int err;
4653
4654         if (!capable(CAP_SYS_ADMIN))
4655                 return -EPERM;
4656
4657         ret = mnt_want_write_file(file);
4658         if (ret)
4659                 return ret;
4660
4661         sa = memdup_user(arg, sizeof(*sa));
4662         if (IS_ERR(sa)) {
4663                 ret = PTR_ERR(sa);
4664                 goto drop_write;
4665         }
4666
4667         if (!sa->qgroupid) {
4668                 ret = -EINVAL;
4669                 goto out;
4670         }
4671
4672         trans = btrfs_join_transaction(root);
4673         if (IS_ERR(trans)) {
4674                 ret = PTR_ERR(trans);
4675                 goto out;
4676         }
4677
4678         /* FIXME: check if the IDs really exist */
4679         if (sa->create) {
4680                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4681                                           NULL);
4682         } else {
4683                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4684         }
4685
4686         err = btrfs_end_transaction(trans, root);
4687         if (err && !ret)
4688                 ret = err;
4689
4690 out:
4691         kfree(sa);
4692 drop_write:
4693         mnt_drop_write_file(file);
4694         return ret;
4695 }
4696
4697 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4698 {
4699         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4700         struct btrfs_ioctl_qgroup_limit_args *sa;
4701         struct btrfs_trans_handle *trans;
4702         int ret;
4703         int err;
4704         u64 qgroupid;
4705
4706         if (!capable(CAP_SYS_ADMIN))
4707                 return -EPERM;
4708
4709         ret = mnt_want_write_file(file);
4710         if (ret)
4711                 return ret;
4712
4713         sa = memdup_user(arg, sizeof(*sa));
4714         if (IS_ERR(sa)) {
4715                 ret = PTR_ERR(sa);
4716                 goto drop_write;
4717         }
4718
4719         trans = btrfs_join_transaction(root);
4720         if (IS_ERR(trans)) {
4721                 ret = PTR_ERR(trans);
4722                 goto out;
4723         }
4724
4725         qgroupid = sa->qgroupid;
4726         if (!qgroupid) {
4727                 /* take the current subvol as qgroup */
4728                 qgroupid = root->root_key.objectid;
4729         }
4730
4731         /* FIXME: check if the IDs really exist */
4732         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4733
4734         err = btrfs_end_transaction(trans, root);
4735         if (err && !ret)
4736                 ret = err;
4737
4738 out:
4739         kfree(sa);
4740 drop_write:
4741         mnt_drop_write_file(file);
4742         return ret;
4743 }
4744
4745 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4746 {
4747         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4748         struct btrfs_ioctl_quota_rescan_args *qsa;
4749         int ret;
4750
4751         if (!capable(CAP_SYS_ADMIN))
4752                 return -EPERM;
4753
4754         ret = mnt_want_write_file(file);
4755         if (ret)
4756                 return ret;
4757
4758         qsa = memdup_user(arg, sizeof(*qsa));
4759         if (IS_ERR(qsa)) {
4760                 ret = PTR_ERR(qsa);
4761                 goto drop_write;
4762         }
4763
4764         if (qsa->flags) {
4765                 ret = -EINVAL;
4766                 goto out;
4767         }
4768
4769         ret = btrfs_qgroup_rescan(root->fs_info);
4770
4771 out:
4772         kfree(qsa);
4773 drop_write:
4774         mnt_drop_write_file(file);
4775         return ret;
4776 }
4777
4778 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4779 {
4780         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4781         struct btrfs_ioctl_quota_rescan_args *qsa;
4782         int ret = 0;
4783
4784         if (!capable(CAP_SYS_ADMIN))
4785                 return -EPERM;
4786
4787         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4788         if (!qsa)
4789                 return -ENOMEM;
4790
4791         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4792                 qsa->flags = 1;
4793                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4794         }
4795
4796         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4797                 ret = -EFAULT;
4798
4799         kfree(qsa);
4800         return ret;
4801 }
4802
4803 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4804 {
4805         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4806
4807         if (!capable(CAP_SYS_ADMIN))
4808                 return -EPERM;
4809
4810         return btrfs_qgroup_wait_for_completion(root->fs_info);
4811 }
4812
4813 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4814                                             struct btrfs_ioctl_received_subvol_args *sa)
4815 {
4816         struct inode *inode = file_inode(file);
4817         struct btrfs_root *root = BTRFS_I(inode)->root;
4818         struct btrfs_root_item *root_item = &root->root_item;
4819         struct btrfs_trans_handle *trans;
4820         struct timespec ct = CURRENT_TIME;
4821         int ret = 0;
4822         int received_uuid_changed;
4823
4824         if (!inode_owner_or_capable(inode))
4825                 return -EPERM;
4826
4827         ret = mnt_want_write_file(file);
4828         if (ret < 0)
4829                 return ret;
4830
4831         down_write(&root->fs_info->subvol_sem);
4832
4833         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4834                 ret = -EINVAL;
4835                 goto out;
4836         }
4837
4838         if (btrfs_root_readonly(root)) {
4839                 ret = -EROFS;
4840                 goto out;
4841         }
4842
4843         /*
4844          * 1 - root item
4845          * 2 - uuid items (received uuid + subvol uuid)
4846          */
4847         trans = btrfs_start_transaction(root, 3);
4848         if (IS_ERR(trans)) {
4849                 ret = PTR_ERR(trans);
4850                 trans = NULL;
4851                 goto out;
4852         }
4853
4854         sa->rtransid = trans->transid;
4855         sa->rtime.sec = ct.tv_sec;
4856         sa->rtime.nsec = ct.tv_nsec;
4857
4858         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4859                                        BTRFS_UUID_SIZE);
4860         if (received_uuid_changed &&
4861             !btrfs_is_empty_uuid(root_item->received_uuid))
4862                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4863                                     root_item->received_uuid,
4864                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4865                                     root->root_key.objectid);
4866         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4867         btrfs_set_root_stransid(root_item, sa->stransid);
4868         btrfs_set_root_rtransid(root_item, sa->rtransid);
4869         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4870         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4871         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4872         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4873
4874         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4875                                 &root->root_key, &root->root_item);
4876         if (ret < 0) {
4877                 btrfs_end_transaction(trans, root);
4878                 goto out;
4879         }
4880         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4881                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4882                                           sa->uuid,
4883                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4884                                           root->root_key.objectid);
4885                 if (ret < 0 && ret != -EEXIST) {
4886                         btrfs_abort_transaction(trans, root, ret);
4887                         goto out;
4888                 }
4889         }
4890         ret = btrfs_commit_transaction(trans, root);
4891         if (ret < 0) {
4892                 btrfs_abort_transaction(trans, root, ret);
4893                 goto out;
4894         }
4895
4896 out:
4897         up_write(&root->fs_info->subvol_sem);
4898         mnt_drop_write_file(file);
4899         return ret;
4900 }
4901
4902 #ifdef CONFIG_64BIT
4903 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4904                                                 void __user *arg)
4905 {
4906         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4907         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4908         int ret = 0;
4909
4910         args32 = memdup_user(arg, sizeof(*args32));
4911         if (IS_ERR(args32)) {
4912                 ret = PTR_ERR(args32);
4913                 args32 = NULL;
4914                 goto out;
4915         }
4916
4917         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4918         if (!args64) {
4919                 ret = -ENOMEM;
4920                 goto out;
4921         }
4922
4923         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4924         args64->stransid = args32->stransid;
4925         args64->rtransid = args32->rtransid;
4926         args64->stime.sec = args32->stime.sec;
4927         args64->stime.nsec = args32->stime.nsec;
4928         args64->rtime.sec = args32->rtime.sec;
4929         args64->rtime.nsec = args32->rtime.nsec;
4930         args64->flags = args32->flags;
4931
4932         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4933         if (ret)
4934                 goto out;
4935
4936         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4937         args32->stransid = args64->stransid;
4938         args32->rtransid = args64->rtransid;
4939         args32->stime.sec = args64->stime.sec;
4940         args32->stime.nsec = args64->stime.nsec;
4941         args32->rtime.sec = args64->rtime.sec;
4942         args32->rtime.nsec = args64->rtime.nsec;
4943         args32->flags = args64->flags;
4944
4945         ret = copy_to_user(arg, args32, sizeof(*args32));
4946         if (ret)
4947                 ret = -EFAULT;
4948
4949 out:
4950         kfree(args32);
4951         kfree(args64);
4952         return ret;
4953 }
4954 #endif
4955
4956 static long btrfs_ioctl_set_received_subvol(struct file *file,
4957                                             void __user *arg)
4958 {
4959         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4960         int ret = 0;
4961
4962         sa = memdup_user(arg, sizeof(*sa));
4963         if (IS_ERR(sa)) {
4964                 ret = PTR_ERR(sa);
4965                 sa = NULL;
4966                 goto out;
4967         }
4968
4969         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4970
4971         if (ret)
4972                 goto out;
4973
4974         ret = copy_to_user(arg, sa, sizeof(*sa));
4975         if (ret)
4976                 ret = -EFAULT;
4977
4978 out:
4979         kfree(sa);
4980         return ret;
4981 }
4982
4983 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4984 {
4985         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4986         size_t len;
4987         int ret;
4988         char label[BTRFS_LABEL_SIZE];
4989
4990         spin_lock(&root->fs_info->super_lock);
4991         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4992         spin_unlock(&root->fs_info->super_lock);
4993
4994         len = strnlen(label, BTRFS_LABEL_SIZE);
4995
4996         if (len == BTRFS_LABEL_SIZE) {
4997                 btrfs_warn(root->fs_info,
4998                         "label is too long, return the first %zu bytes", --len);
4999         }
5000
5001         ret = copy_to_user(arg, label, len);
5002
5003         return ret ? -EFAULT : 0;
5004 }
5005
5006 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5007 {
5008         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5009         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5010         struct btrfs_trans_handle *trans;
5011         char label[BTRFS_LABEL_SIZE];
5012         int ret;
5013
5014         if (!capable(CAP_SYS_ADMIN))
5015                 return -EPERM;
5016
5017         if (copy_from_user(label, arg, sizeof(label)))
5018                 return -EFAULT;
5019
5020         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5021                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5022                        BTRFS_LABEL_SIZE - 1);
5023                 return -EINVAL;
5024         }
5025
5026         ret = mnt_want_write_file(file);
5027         if (ret)
5028                 return ret;
5029
5030         trans = btrfs_start_transaction(root, 0);
5031         if (IS_ERR(trans)) {
5032                 ret = PTR_ERR(trans);
5033                 goto out_unlock;
5034         }
5035
5036         spin_lock(&root->fs_info->super_lock);
5037         strcpy(super_block->label, label);
5038         spin_unlock(&root->fs_info->super_lock);
5039         ret = btrfs_commit_transaction(trans, root);
5040
5041 out_unlock:
5042         mnt_drop_write_file(file);
5043         return ret;
5044 }
5045
5046 #define INIT_FEATURE_FLAGS(suffix) \
5047         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5048           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5049           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5050
5051 static int btrfs_ioctl_get_supported_features(struct file *file,
5052                                               void __user *arg)
5053 {
5054         static struct btrfs_ioctl_feature_flags features[3] = {
5055                 INIT_FEATURE_FLAGS(SUPP),
5056                 INIT_FEATURE_FLAGS(SAFE_SET),
5057                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5058         };
5059
5060         if (copy_to_user(arg, &features, sizeof(features)))
5061                 return -EFAULT;
5062
5063         return 0;
5064 }
5065
5066 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5067 {
5068         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5069         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5070         struct btrfs_ioctl_feature_flags features;
5071
5072         features.compat_flags = btrfs_super_compat_flags(super_block);
5073         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5074         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5075
5076         if (copy_to_user(arg, &features, sizeof(features)))
5077                 return -EFAULT;
5078
5079         return 0;
5080 }
5081
5082 static int check_feature_bits(struct btrfs_root *root,
5083                               enum btrfs_feature_set set,
5084                               u64 change_mask, u64 flags, u64 supported_flags,
5085                               u64 safe_set, u64 safe_clear)
5086 {
5087         const char *type = btrfs_feature_set_names[set];
5088         char *names;
5089         u64 disallowed, unsupported;
5090         u64 set_mask = flags & change_mask;
5091         u64 clear_mask = ~flags & change_mask;
5092
5093         unsupported = set_mask & ~supported_flags;
5094         if (unsupported) {
5095                 names = btrfs_printable_features(set, unsupported);
5096                 if (names) {
5097                         btrfs_warn(root->fs_info,
5098                            "this kernel does not support the %s feature bit%s",
5099                            names, strchr(names, ',') ? "s" : "");
5100                         kfree(names);
5101                 } else
5102                         btrfs_warn(root->fs_info,
5103                            "this kernel does not support %s bits 0x%llx",
5104                            type, unsupported);
5105                 return -EOPNOTSUPP;
5106         }
5107
5108         disallowed = set_mask & ~safe_set;
5109         if (disallowed) {
5110                 names = btrfs_printable_features(set, disallowed);
5111                 if (names) {
5112                         btrfs_warn(root->fs_info,
5113                            "can't set the %s feature bit%s while mounted",
5114                            names, strchr(names, ',') ? "s" : "");
5115                         kfree(names);
5116                 } else
5117                         btrfs_warn(root->fs_info,
5118                            "can't set %s bits 0x%llx while mounted",
5119                            type, disallowed);
5120                 return -EPERM;
5121         }
5122
5123         disallowed = clear_mask & ~safe_clear;
5124         if (disallowed) {
5125                 names = btrfs_printable_features(set, disallowed);
5126                 if (names) {
5127                         btrfs_warn(root->fs_info,
5128                            "can't clear the %s feature bit%s while mounted",
5129                            names, strchr(names, ',') ? "s" : "");
5130                         kfree(names);
5131                 } else
5132                         btrfs_warn(root->fs_info,
5133                            "can't clear %s bits 0x%llx while mounted",
5134                            type, disallowed);
5135                 return -EPERM;
5136         }
5137
5138         return 0;
5139 }
5140
5141 #define check_feature(root, change_mask, flags, mask_base)      \
5142 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5143                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5144                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5145                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5146
5147 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5148 {
5149         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5150         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5151         struct btrfs_ioctl_feature_flags flags[2];
5152         struct btrfs_trans_handle *trans;
5153         u64 newflags;
5154         int ret;
5155
5156         if (!capable(CAP_SYS_ADMIN))
5157                 return -EPERM;
5158
5159         if (copy_from_user(flags, arg, sizeof(flags)))
5160                 return -EFAULT;
5161
5162         /* Nothing to do */
5163         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5164             !flags[0].incompat_flags)
5165                 return 0;
5166
5167         ret = check_feature(root, flags[0].compat_flags,
5168                             flags[1].compat_flags, COMPAT);
5169         if (ret)
5170                 return ret;
5171
5172         ret = check_feature(root, flags[0].compat_ro_flags,
5173                             flags[1].compat_ro_flags, COMPAT_RO);
5174         if (ret)
5175                 return ret;
5176
5177         ret = check_feature(root, flags[0].incompat_flags,
5178                             flags[1].incompat_flags, INCOMPAT);
5179         if (ret)
5180                 return ret;
5181
5182         trans = btrfs_start_transaction(root, 0);
5183         if (IS_ERR(trans))
5184                 return PTR_ERR(trans);
5185
5186         spin_lock(&root->fs_info->super_lock);
5187         newflags = btrfs_super_compat_flags(super_block);
5188         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5189         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5190         btrfs_set_super_compat_flags(super_block, newflags);
5191
5192         newflags = btrfs_super_compat_ro_flags(super_block);
5193         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5194         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5195         btrfs_set_super_compat_ro_flags(super_block, newflags);
5196
5197         newflags = btrfs_super_incompat_flags(super_block);
5198         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5199         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5200         btrfs_set_super_incompat_flags(super_block, newflags);
5201         spin_unlock(&root->fs_info->super_lock);
5202
5203         return btrfs_commit_transaction(trans, root);
5204 }
5205
5206 long btrfs_ioctl(struct file *file, unsigned int
5207                 cmd, unsigned long arg)
5208 {
5209         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5210         void __user *argp = (void __user *)arg;
5211
5212         switch (cmd) {
5213         case FS_IOC_GETFLAGS:
5214                 return btrfs_ioctl_getflags(file, argp);
5215         case FS_IOC_SETFLAGS:
5216                 return btrfs_ioctl_setflags(file, argp);
5217         case FS_IOC_GETVERSION:
5218                 return btrfs_ioctl_getversion(file, argp);
5219         case FITRIM:
5220                 return btrfs_ioctl_fitrim(file, argp);
5221         case BTRFS_IOC_SNAP_CREATE:
5222                 return btrfs_ioctl_snap_create(file, argp, 0);
5223         case BTRFS_IOC_SNAP_CREATE_V2:
5224                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5225         case BTRFS_IOC_SUBVOL_CREATE:
5226                 return btrfs_ioctl_snap_create(file, argp, 1);
5227         case BTRFS_IOC_SUBVOL_CREATE_V2:
5228                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5229         case BTRFS_IOC_SNAP_DESTROY:
5230                 return btrfs_ioctl_snap_destroy(file, argp);
5231         case BTRFS_IOC_SUBVOL_GETFLAGS:
5232                 return btrfs_ioctl_subvol_getflags(file, argp);
5233         case BTRFS_IOC_SUBVOL_SETFLAGS:
5234                 return btrfs_ioctl_subvol_setflags(file, argp);
5235         case BTRFS_IOC_DEFAULT_SUBVOL:
5236                 return btrfs_ioctl_default_subvol(file, argp);
5237         case BTRFS_IOC_DEFRAG:
5238                 return btrfs_ioctl_defrag(file, NULL);
5239         case BTRFS_IOC_DEFRAG_RANGE:
5240                 return btrfs_ioctl_defrag(file, argp);
5241         case BTRFS_IOC_RESIZE:
5242                 return btrfs_ioctl_resize(file, argp);
5243         case BTRFS_IOC_ADD_DEV:
5244                 return btrfs_ioctl_add_dev(root, argp);
5245         case BTRFS_IOC_RM_DEV:
5246                 return btrfs_ioctl_rm_dev(file, argp);
5247         case BTRFS_IOC_FS_INFO:
5248                 return btrfs_ioctl_fs_info(root, argp);
5249         case BTRFS_IOC_DEV_INFO:
5250                 return btrfs_ioctl_dev_info(root, argp);
5251         case BTRFS_IOC_BALANCE:
5252                 return btrfs_ioctl_balance(file, NULL);
5253         case BTRFS_IOC_CLONE:
5254                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5255         case BTRFS_IOC_CLONE_RANGE:
5256                 return btrfs_ioctl_clone_range(file, argp);
5257         case BTRFS_IOC_TRANS_START:
5258                 return btrfs_ioctl_trans_start(file);
5259         case BTRFS_IOC_TRANS_END:
5260                 return btrfs_ioctl_trans_end(file);
5261         case BTRFS_IOC_TREE_SEARCH:
5262                 return btrfs_ioctl_tree_search(file, argp);
5263         case BTRFS_IOC_TREE_SEARCH_V2:
5264                 return btrfs_ioctl_tree_search_v2(file, argp);
5265         case BTRFS_IOC_INO_LOOKUP:
5266                 return btrfs_ioctl_ino_lookup(file, argp);
5267         case BTRFS_IOC_INO_PATHS:
5268                 return btrfs_ioctl_ino_to_path(root, argp);
5269         case BTRFS_IOC_LOGICAL_INO:
5270                 return btrfs_ioctl_logical_to_ino(root, argp);
5271         case BTRFS_IOC_SPACE_INFO:
5272                 return btrfs_ioctl_space_info(root, argp);
5273         case BTRFS_IOC_SYNC: {
5274                 int ret;
5275
5276                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5277                 if (ret)
5278                         return ret;
5279                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
5280                 return ret;
5281         }
5282         case BTRFS_IOC_START_SYNC:
5283                 return btrfs_ioctl_start_sync(root, argp);
5284         case BTRFS_IOC_WAIT_SYNC:
5285                 return btrfs_ioctl_wait_sync(root, argp);
5286         case BTRFS_IOC_SCRUB:
5287                 return btrfs_ioctl_scrub(file, argp);
5288         case BTRFS_IOC_SCRUB_CANCEL:
5289                 return btrfs_ioctl_scrub_cancel(root, argp);
5290         case BTRFS_IOC_SCRUB_PROGRESS:
5291                 return btrfs_ioctl_scrub_progress(root, argp);
5292         case BTRFS_IOC_BALANCE_V2:
5293                 return btrfs_ioctl_balance(file, argp);
5294         case BTRFS_IOC_BALANCE_CTL:
5295                 return btrfs_ioctl_balance_ctl(root, arg);
5296         case BTRFS_IOC_BALANCE_PROGRESS:
5297                 return btrfs_ioctl_balance_progress(root, argp);
5298         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5299                 return btrfs_ioctl_set_received_subvol(file, argp);
5300 #ifdef CONFIG_64BIT
5301         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5302                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5303 #endif
5304         case BTRFS_IOC_SEND:
5305                 return btrfs_ioctl_send(file, argp);
5306         case BTRFS_IOC_GET_DEV_STATS:
5307                 return btrfs_ioctl_get_dev_stats(root, argp);
5308         case BTRFS_IOC_QUOTA_CTL:
5309                 return btrfs_ioctl_quota_ctl(file, argp);
5310         case BTRFS_IOC_QGROUP_ASSIGN:
5311                 return btrfs_ioctl_qgroup_assign(file, argp);
5312         case BTRFS_IOC_QGROUP_CREATE:
5313                 return btrfs_ioctl_qgroup_create(file, argp);
5314         case BTRFS_IOC_QGROUP_LIMIT:
5315                 return btrfs_ioctl_qgroup_limit(file, argp);
5316         case BTRFS_IOC_QUOTA_RESCAN:
5317                 return btrfs_ioctl_quota_rescan(file, argp);
5318         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5319                 return btrfs_ioctl_quota_rescan_status(file, argp);
5320         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5321                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5322         case BTRFS_IOC_DEV_REPLACE:
5323                 return btrfs_ioctl_dev_replace(root, argp);
5324         case BTRFS_IOC_GET_FSLABEL:
5325                 return btrfs_ioctl_get_fslabel(file, argp);
5326         case BTRFS_IOC_SET_FSLABEL:
5327                 return btrfs_ioctl_set_fslabel(file, argp);
5328         case BTRFS_IOC_FILE_EXTENT_SAME:
5329                 return btrfs_ioctl_file_extent_same(file, argp);
5330         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5331                 return btrfs_ioctl_get_supported_features(file, argp);
5332         case BTRFS_IOC_GET_FEATURES:
5333                 return btrfs_ioctl_get_features(file, argp);
5334         case BTRFS_IOC_SET_FEATURES:
5335                 return btrfs_ioctl_set_features(file, argp);
5336         }
5337
5338         return -ENOTTY;
5339 }