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
 140         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 141
 142         if (ip->flags & BTRFS_INODE_SYNC)
 143                 inode->i_flags |= S_SYNC;
 144         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 145                 inode->i_flags |= S_IMMUTABLE;
 146         if (ip->flags & BTRFS_INODE_APPEND)
 147                 inode->i_flags |= S_APPEND;
 148         if (ip->flags & BTRFS_INODE_NOATIME)
 149                 inode->i_flags |= S_NOATIME;
 150         if (ip->flags & BTRFS_INODE_DIRSYNC)
 151                 inode->i_flags |= S_DIRSYNC;
 152 }
 153
 154 /*
 155  * Inherit flags from the parent inode.
 156  *
 157  * Currently only the compression flags and the cow flags are inherited.
 158  */
 159 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 160 {
 161         unsigned int flags;
 162
 163         if (!dir)
 164                 return;
 165
 166         flags = BTRFS_I(dir)->flags;
 167
 168         if (flags & BTRFS_INODE_NOCOMPRESS) {
 169                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 170                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 171         } else if (flags & BTRFS_INODE_COMPRESS) {
 172                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 173                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 174         }
 175
 176         if (flags & BTRFS_INODE_NODATACOW) {
 177                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 178                 if (S_ISREG(inode->i_mode))
 179                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 180         }
 181
 182         btrfs_update_iflags(inode);
 183 }
 184
 185 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 186 {
 187         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 188         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 189
 190         if (copy_to_user(arg, &flags, sizeof(flags)))
 191                 return -EFAULT;
 192         return 0;
 193 }
 194
 195 static int check_flags(unsigned int flags)
 196 {
 197         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 198                       FS_NOATIME_FL | FS_NODUMP_FL | \
 199                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 200                       FS_NOCOMP_FL | FS_COMPR_FL |
 201                       FS_NOCOW_FL))
 202                 return -EOPNOTSUPP;
 203
 204         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 205                 return -EINVAL;
 206
 207         return 0;
 208 }
 209
 210 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 211 {
 212         struct inode *inode = file_inode(file);
 213         struct btrfs_inode *ip = BTRFS_I(inode);
 214         struct btrfs_root *root = ip->root;
 215         struct btrfs_trans_handle *trans;
 216         unsigned int flags, oldflags;
 217         int ret;
 218         u64 ip_oldflags;
 219         unsigned int i_oldflags;
 220         umode_t mode;
 221
 222         if (!inode_owner_or_capable(inode))
 223                 return -EPERM;
 224
 225         if (btrfs_root_readonly(root))
 226                 return -EROFS;
 227
 228         if (copy_from_user(&flags, arg, sizeof(flags)))
 229                 return -EFAULT;
 230
 231         ret = check_flags(flags);
 232         if (ret)
 233                 return ret;
 234
 235         ret = mnt_want_write_file(file);
 236         if (ret)
 237                 return ret;
 238
 239         mutex_lock(&inode->i_mutex);
 240
 241         ip_oldflags = ip->flags;
 242         i_oldflags = inode->i_flags;
 243         mode = inode->i_mode;
 244
 245         flags = btrfs_mask_flags(inode->i_mode, flags);
 246         oldflags = btrfs_flags_to_ioctl(ip->flags);
 247         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 248                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 249                         ret = -EPERM;
 250                         goto out_unlock;
 251                 }
 252         }
 253
 254         if (flags & FS_SYNC_FL)
 255                 ip->flags |= BTRFS_INODE_SYNC;
 256         else
 257                 ip->flags &= ~BTRFS_INODE_SYNC;
 258         if (flags & FS_IMMUTABLE_FL)
 259                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 260         else
 261                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 262         if (flags & FS_APPEND_FL)
 263                 ip->flags |= BTRFS_INODE_APPEND;
 264         else
 265                 ip->flags &= ~BTRFS_INODE_APPEND;
 266         if (flags & FS_NODUMP_FL)
 267                 ip->flags |= BTRFS_INODE_NODUMP;
 268         else
 269                 ip->flags &= ~BTRFS_INODE_NODUMP;
 270         if (flags & FS_NOATIME_FL)
 271                 ip->flags |= BTRFS_INODE_NOATIME;
 272         else
 273                 ip->flags &= ~BTRFS_INODE_NOATIME;
 274         if (flags & FS_DIRSYNC_FL)
 275                 ip->flags |= BTRFS_INODE_DIRSYNC;
 276         else
 277                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 278         if (flags & FS_NOCOW_FL) {
 279                 if (S_ISREG(mode)) {
 280                         /*
 281                          * It's safe to turn csums off here, no extents exist.
 282                          * Otherwise we want the flag to reflect the real COW
 283                          * status of the file and will not set it.
 284                          */
 285                         if (inode->i_size == 0)
 286                                 ip->flags |= BTRFS_INODE_NODATACOW
 287                                            | BTRFS_INODE_NODATASUM;
 288                 } else {
 289                         ip->flags |= BTRFS_INODE_NODATACOW;
 290                 }
 291         } else {
 292                 /*
 293                  * Revert back under same assuptions as above
 294                  */
 295                 if (S_ISREG(mode)) {
 296                         if (inode->i_size == 0)
 297                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 298                                              | BTRFS_INODE_NODATASUM);
 299                 } else {
 300                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 301                 }
 302         }
 303
 304         /*
 305          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 306          * flag may be changed automatically if compression code won't make
 307          * things smaller.
 308          */
 309         if (flags & FS_NOCOMP_FL) {
 310                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 311                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 312
 313                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 314                 if (ret && ret != -ENODATA)
 315                         goto out_drop;
 316         } else if (flags & FS_COMPR_FL) {
 317                 const char *comp;
 318
 319                 ip->flags |= BTRFS_INODE_COMPRESS;
 320                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 321
 322                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 323                         comp = "lzo";
 324                 else
 325                         comp = "zlib";
 326                 ret = btrfs_set_prop(inode, "btrfs.compression",
 327                                      comp, strlen(comp), 0);
 328                 if (ret)
 329                         goto out_drop;
 330
 331         } else {
 332                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 333         }
 334
 335         trans = btrfs_start_transaction(root, 1);
 336         if (IS_ERR(trans)) {
 337                 ret = PTR_ERR(trans);
 338                 goto out_drop;
 339         }
 340
 341         btrfs_update_iflags(inode);
 342         inode_inc_iversion(inode);
 343         inode->i_ctime = CURRENT_TIME;
 344         ret = btrfs_update_inode(trans, root, inode);
 345
 346         btrfs_end_transaction(trans, root);
 347  out_drop:
 348         if (ret) {
 349                 ip->flags = ip_oldflags;
 350                 inode->i_flags = i_oldflags;
 351         }
 352
 353  out_unlock:
 354         mutex_unlock(&inode->i_mutex);
 355         mnt_drop_write_file(file);
 356         return ret;
 357 }
 358
 359 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 360 {
 361         struct inode *inode = file_inode(file);
 362
 363         return put_user(inode->i_generation, arg);
 364 }
 365
 366 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 367 {
 368         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 369         struct btrfs_device *device;
 370         struct request_queue *q;
 371         struct fstrim_range range;
 372         u64 minlen = ULLONG_MAX;
 373         u64 num_devices = 0;
 374         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 375         int ret;
 376
 377         if (!capable(CAP_SYS_ADMIN))
 378                 return -EPERM;
 379
 380         rcu_read_lock();
 381         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 382                                 dev_list) {
 383                 if (!device->bdev)
 384                         continue;
 385                 q = bdev_get_queue(device->bdev);
 386                 if (blk_queue_discard(q)) {
 387                         num_devices++;
 388                         minlen = min((u64)q->limits.discard_granularity,
 389                                      minlen);
 390                 }
 391         }
 392         rcu_read_unlock();
 393
 394         if (!num_devices)
 395                 return -EOPNOTSUPP;
 396         if (copy_from_user(&range, arg, sizeof(range)))
 397                 return -EFAULT;
 398         if (range.start > total_bytes ||
 399             range.len < fs_info->sb->s_blocksize)
 400                 return -EINVAL;
 401
 402         range.len = min(range.len, total_bytes - range.start);
 403         range.minlen = max(range.minlen, minlen);
 404         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 405         if (ret < 0)
 406                 return ret;
 407
 408         if (copy_to_user(arg, &range, sizeof(range)))
 409                 return -EFAULT;
 410
 411         return 0;
 412 }
 413
 414 int btrfs_is_empty_uuid(u8 *uuid)
 415 {
 416         int i;
 417
 418         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 419                 if (uuid[i])
 420                         return 0;
 421         }
 422         return 1;
 423 }
 424
 425 static noinline int create_subvol(struct inode *dir,
 426                                   struct dentry *dentry,
 427                                   char *name, int namelen,
 428                                   u64 *async_transid,
 429                                   struct btrfs_qgroup_inherit *inherit)
 430 {
 431         struct btrfs_trans_handle *trans;
 432         struct btrfs_key key;
 433         struct btrfs_root_item root_item;
 434         struct btrfs_inode_item *inode_item;
 435         struct extent_buffer *leaf;
 436         struct btrfs_root *root = BTRFS_I(dir)->root;
 437         struct btrfs_root *new_root;
 438         struct btrfs_block_rsv block_rsv;
 439         struct timespec cur_time = CURRENT_TIME;
 440         struct inode *inode;
 441         int ret;
 442         int err;
 443         u64 objectid;
 444         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 445         u64 index = 0;
 446         u64 qgroup_reserved;
 447         uuid_le new_uuid;
 448
 449         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 450         if (ret)
 451                 return ret;
 452
 453         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 454         /*
 455          * The same as the snapshot creation, please see the comment
 456          * of create_snapshot().
 457          */
 458         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 459                                                8, &qgroup_reserved, false);
 460         if (ret)
 461                 return ret;
 462
 463         trans = btrfs_start_transaction(root, 0);
 464         if (IS_ERR(trans)) {
 465                 ret = PTR_ERR(trans);
 466                 btrfs_subvolume_release_metadata(root, &block_rsv,
 467                                                  qgroup_reserved);
 468                 return ret;
 469         }
 470         trans->block_rsv = &block_rsv;
 471         trans->bytes_reserved = block_rsv.size;
 472
 473         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 474         if (ret)
 475                 goto fail;
 476
 477         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 478                                       0, objectid, NULL, 0, 0, 0);
 479         if (IS_ERR(leaf)) {
 480                 ret = PTR_ERR(leaf);
 481                 goto fail;
 482         }
 483
 484         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 485         btrfs_set_header_bytenr(leaf, leaf->start);
 486         btrfs_set_header_generation(leaf, trans->transid);
 487         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 488         btrfs_set_header_owner(leaf, objectid);
 489
 490         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 491                             BTRFS_FSID_SIZE);
 492         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 493                             btrfs_header_chunk_tree_uuid(leaf),
 494                             BTRFS_UUID_SIZE);
 495         btrfs_mark_buffer_dirty(leaf);
 496
 497         memset(&root_item, 0, sizeof(root_item));
 498
 499         inode_item = &root_item.inode;
 500         btrfs_set_stack_inode_generation(inode_item, 1);
 501         btrfs_set_stack_inode_size(inode_item, 3);
 502         btrfs_set_stack_inode_nlink(inode_item, 1);
 503         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
 504         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 505
 506         btrfs_set_root_flags(&root_item, 0);
 507         btrfs_set_root_limit(&root_item, 0);
 508         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 509
 510         btrfs_set_root_bytenr(&root_item, leaf->start);
 511         btrfs_set_root_generation(&root_item, trans->transid);
 512         btrfs_set_root_level(&root_item, 0);
 513         btrfs_set_root_refs(&root_item, 1);
 514         btrfs_set_root_used(&root_item, leaf->len);
 515         btrfs_set_root_last_snapshot(&root_item, 0);
 516
 517         btrfs_set_root_generation_v2(&root_item,
 518                         btrfs_root_generation(&root_item));
 519         uuid_le_gen(&new_uuid);
 520         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 521         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 522         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 523         root_item.ctime = root_item.otime;
 524         btrfs_set_root_ctransid(&root_item, trans->transid);
 525         btrfs_set_root_otransid(&root_item, trans->transid);
 526
 527         btrfs_tree_unlock(leaf);
 528         free_extent_buffer(leaf);
 529         leaf = NULL;
 530
 531         btrfs_set_root_dirid(&root_item, new_dirid);
 532
 533         key.objectid = objectid;
 534         key.offset = 0;
 535         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 536         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 537                                 &root_item);
 538         if (ret)
 539                 goto fail;
 540
 541         key.offset = (u64)-1;
 542         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 543         if (IS_ERR(new_root)) {
 544                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 545                 ret = PTR_ERR(new_root);
 546                 goto fail;
 547         }
 548
 549         btrfs_record_root_in_trans(trans, new_root);
 550
 551         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 552         if (ret) {
 553                 /* We potentially lose an unused inode item here */
 554                 btrfs_abort_transaction(trans, root, ret);
 555                 goto fail;
 556         }
 557
 558         /*
 559          * insert the directory item
 560          */
 561         ret = btrfs_set_inode_index(dir, &index);
 562         if (ret) {
 563                 btrfs_abort_transaction(trans, root, ret);
 564                 goto fail;
 565         }
 566
 567         ret = btrfs_insert_dir_item(trans, root,
 568                                     name, namelen, dir, &key,
 569                                     BTRFS_FT_DIR, index);
 570         if (ret) {
 571                 btrfs_abort_transaction(trans, root, ret);
 572                 goto fail;
 573         }
 574
 575         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 576         ret = btrfs_update_inode(trans, root, dir);
 577         BUG_ON(ret);
 578
 579         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 580                                  objectid, root->root_key.objectid,
 581                                  btrfs_ino(dir), index, name, namelen);
 582         BUG_ON(ret);
 583
 584         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 585                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 586                                   objectid);
 587         if (ret)
 588                 btrfs_abort_transaction(trans, root, ret);
 589
 590 fail:
 591         trans->block_rsv = NULL;
 592         trans->bytes_reserved = 0;
 593         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 594
 595         if (async_transid) {
 596                 *async_transid = trans->transid;
 597                 err = btrfs_commit_transaction_async(trans, root, 1);
 598                 if (err)
 599                         err = btrfs_commit_transaction(trans, root);
 600         } else {
 601                 err = btrfs_commit_transaction(trans, root);
 602         }
 603         if (err && !ret)
 604                 ret = err;
 605
 606         if (!ret) {
 607                 inode = btrfs_lookup_dentry(dir, dentry);
 608                 if (IS_ERR(inode))
 609                         return PTR_ERR(inode);
 610                 d_instantiate(dentry, inode);
 611         }
 612         return ret;
 613 }
 614
 615 static void btrfs_wait_nocow_write(struct btrfs_root *root)
 616 {
 617         s64 writers;
 618         DEFINE_WAIT(wait);
 619
 620         do {
 621                 prepare_to_wait(&root->subv_writers->wait, &wait,
 622                                 TASK_UNINTERRUPTIBLE);
 623
 624                 writers = percpu_counter_sum(&root->subv_writers->counter);
 625                 if (writers)
 626                         schedule();
 627
 628                 finish_wait(&root->subv_writers->wait, &wait);
 629         } while (writers);
 630 }
 631
 632 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 633                            struct dentry *dentry, char *name, int namelen,
 634                            u64 *async_transid, bool readonly,
 635                            struct btrfs_qgroup_inherit *inherit)
 636 {
 637         struct inode *inode;
 638         struct btrfs_pending_snapshot *pending_snapshot;
 639         struct btrfs_trans_handle *trans;
 640         int ret;
 641
 642         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 643                 return -EINVAL;
 644
 645         atomic_inc(&root->will_be_snapshoted);
 646         smp_mb__after_atomic();
 647         btrfs_wait_nocow_write(root);
 648
 649         ret = btrfs_start_delalloc_inodes(root, 0);
 650         if (ret)
 651                 goto out;
 652
 653         btrfs_wait_ordered_extents(root, -1);
 654
 655         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 656         if (!pending_snapshot) {
 657                 ret = -ENOMEM;
 658                 goto out;
 659         }
 660
 661         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 662                              BTRFS_BLOCK_RSV_TEMP);
 663         /*
 664          * 1 - parent dir inode
 665          * 2 - dir entries
 666          * 1 - root item
 667          * 2 - root ref/backref
 668          * 1 - root of snapshot
 669          * 1 - UUID item
 670          */
 671         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 672                                         &pending_snapshot->block_rsv, 8,
 673                                         &pending_snapshot->qgroup_reserved,
 674                                         false);
 675         if (ret)
 676                 goto free;
 677
 678         pending_snapshot->dentry = dentry;
 679         pending_snapshot->root = root;
 680         pending_snapshot->readonly = readonly;
 681         pending_snapshot->dir = dir;
 682         pending_snapshot->inherit = inherit;
 683
 684         trans = btrfs_start_transaction(root, 0);
 685         if (IS_ERR(trans)) {
 686                 ret = PTR_ERR(trans);
 687                 goto fail;
 688         }
 689
 690         spin_lock(&root->fs_info->trans_lock);
 691         list_add(&pending_snapshot->list,
 692                  &trans->transaction->pending_snapshots);
 693         spin_unlock(&root->fs_info->trans_lock);
 694         if (async_transid) {
 695                 *async_transid = trans->transid;
 696                 ret = btrfs_commit_transaction_async(trans,
 697                                      root->fs_info->extent_root, 1);
 698                 if (ret)
 699                         ret = btrfs_commit_transaction(trans, root);
 700         } else {
 701                 ret = btrfs_commit_transaction(trans,
 702                                                root->fs_info->extent_root);
 703         }
 704         if (ret)
 705                 goto fail;
 706
 707         ret = pending_snapshot->error;
 708         if (ret)
 709                 goto fail;
 710
 711         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 712         if (ret)
 713                 goto fail;
 714
 715         /*
 716          * If orphan cleanup did remove any orphans, it means the tree was
 717          * modified and therefore the commit root is not the same as the
 718          * current root anymore. This is a problem, because send uses the
 719          * commit root and therefore can see inode items that don't exist
 720          * in the current root anymore, and for example make calls to
 721          * btrfs_iget, which will do tree lookups based on the current root
 722          * and not on the commit root. Those lookups will fail, returning a
 723          * -ESTALE error, and making send fail with that error. So make sure
 724          * a send does not see any orphans we have just removed, and that it
 725          * will see the same inodes regardless of whether a transaction
 726          * commit happened before it started (meaning that the commit root
 727          * will be the same as the current root) or not.
 728          */
 729         if (readonly && pending_snapshot->snap->node !=
 730             pending_snapshot->snap->commit_root) {
 731                 trans = btrfs_join_transaction(pending_snapshot->snap);
 732                 if (IS_ERR(trans) && PTR_ERR(trans) != -ENOENT) {
 733                         ret = PTR_ERR(trans);
 734                         goto fail;
 735                 }
 736                 if (!IS_ERR(trans)) {
 737                         ret = btrfs_commit_transaction(trans,
 738                                                        pending_snapshot->snap);
 739                         if (ret)
 740                                 goto fail;
 741                 }
 742         }
 743
 744         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 745         if (IS_ERR(inode)) {
 746                 ret = PTR_ERR(inode);
 747                 goto fail;
 748         }
 749
 750         d_instantiate(dentry, inode);
 751         ret = 0;
 752 fail:
 753         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 754                                          &pending_snapshot->block_rsv,
 755                                          pending_snapshot->qgroup_reserved);
 756 free:
 757         kfree(pending_snapshot);
 758 out:
 759         atomic_dec(&root->will_be_snapshoted);
 760         return ret;
 761 }
 762
 763 /*  copy of check_sticky in fs/namei.c()
 764 * It's inline, so penalty for filesystems that don't use sticky bit is
 765 * minimal.
 766 */
 767 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 768 {
 769         kuid_t fsuid = current_fsuid();
 770
 771         if (!(dir->i_mode & S_ISVTX))
 772                 return 0;
 773         if (uid_eq(inode->i_uid, fsuid))
 774                 return 0;
 775         if (uid_eq(dir->i_uid, fsuid))
 776                 return 0;
 777         return !capable(CAP_FOWNER);
 778 }
 779
 780 /*  copy of may_delete in fs/namei.c()
 781  *      Check whether we can remove a link victim from directory dir, check
 782  *  whether the type of victim is right.
 783  *  1. We can't do it if dir is read-only (done in permission())
 784  *  2. We should have write and exec permissions on dir
 785  *  3. We can't remove anything from append-only dir
 786  *  4. We can't do anything with immutable dir (done in permission())
 787  *  5. If the sticky bit on dir is set we should either
 788  *      a. be owner of dir, or
 789  *      b. be owner of victim, or
 790  *      c. have CAP_FOWNER capability
 791  *  6. If the victim is append-only or immutable we can't do antyhing with
 792  *     links pointing to it.
 793  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 794  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 795  *  9. We can't remove a root or mountpoint.
 796  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 797  *     nfs_async_unlink().
 798  */
 799
 800 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 801 {
 802         int error;
 803
 804         if (!victim->d_inode)
 805                 return -ENOENT;
 806
 807         BUG_ON(victim->d_parent->d_inode != dir);
 808         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 809
 810         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 811         if (error)
 812                 return error;
 813         if (IS_APPEND(dir))
 814                 return -EPERM;
 815         if (btrfs_check_sticky(dir, victim->d_inode)||
 816                 IS_APPEND(victim->d_inode)||
 817             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 818                 return -EPERM;
 819         if (isdir) {
 820                 if (!S_ISDIR(victim->d_inode->i_mode))
 821                         return -ENOTDIR;
 822                 if (IS_ROOT(victim))
 823                         return -EBUSY;
 824         } else if (S_ISDIR(victim->d_inode->i_mode))
 825                 return -EISDIR;
 826         if (IS_DEADDIR(dir))
 827                 return -ENOENT;
 828         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 829                 return -EBUSY;
 830         return 0;
 831 }
 832
 833 /* copy of may_create in fs/namei.c() */
 834 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 835 {
 836         if (child->d_inode)
 837                 return -EEXIST;
 838         if (IS_DEADDIR(dir))
 839                 return -ENOENT;
 840         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 841 }
 842
 843 /*
 844  * Create a new subvolume below @parent.  This is largely modeled after
 845  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 846  * inside this filesystem so it's quite a bit simpler.
 847  */
 848 static noinline int btrfs_mksubvol(struct path *parent,
 849                                    char *name, int namelen,
 850                                    struct btrfs_root *snap_src,
 851                                    u64 *async_transid, bool readonly,
 852                                    struct btrfs_qgroup_inherit *inherit)
 853 {
 854         struct inode *dir  = parent->dentry->d_inode;
 855         struct dentry *dentry;
 856         int error;
 857
 858         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 859         if (error == -EINTR)
 860                 return error;
 861
 862         dentry = lookup_one_len(name, parent->dentry, namelen);
 863         error = PTR_ERR(dentry);
 864         if (IS_ERR(dentry))
 865                 goto out_unlock;
 866
 867         error = -EEXIST;
 868         if (dentry->d_inode)
 869                 goto out_dput;
 870
 871         error = btrfs_may_create(dir, dentry);
 872         if (error)
 873                 goto out_dput;
 874
 875         /*
 876          * even if this name doesn't exist, we may get hash collisions.
 877          * check for them now when we can safely fail
 878          */
 879         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 880                                                dir->i_ino, name,
 881                                                namelen);
 882         if (error)
 883                 goto out_dput;
 884
 885         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 886
 887         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 888                 goto out_up_read;
 889
 890         if (snap_src) {
 891                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 892                                         async_transid, readonly, inherit);
 893         } else {
 894                 error = create_subvol(dir, dentry, name, namelen,
 895                                       async_transid, inherit);
 896         }
 897         if (!error)
 898                 fsnotify_mkdir(dir, dentry);
 899 out_up_read:
 900         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 901 out_dput:
 902         dput(dentry);
 903 out_unlock:
 904         mutex_unlock(&dir->i_mutex);
 905         return error;
 906 }
 907
 908 /*
 909  * When we're defragging a range, we don't want to kick it off again
 910  * if it is really just waiting for delalloc to send it down.
 911  * If we find a nice big extent or delalloc range for the bytes in the
 912  * file you want to defrag, we return 0 to let you know to skip this
 913  * part of the file
 914  */
 915 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 916 {
 917         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 918         struct extent_map *em = NULL;
 919         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 920         u64 end;
 921
 922         read_lock(&em_tree->lock);
 923         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 924         read_unlock(&em_tree->lock);
 925
 926         if (em) {
 927                 end = extent_map_end(em);
 928                 free_extent_map(em);
 929                 if (end - offset > thresh)
 930                         return 0;
 931         }
 932         /* if we already have a nice delalloc here, just stop */
 933         thresh /= 2;
 934         end = count_range_bits(io_tree, &offset, offset + thresh,
 935                                thresh, EXTENT_DELALLOC, 1);
 936         if (end >= thresh)
 937                 return 0;
 938         return 1;
 939 }
 940
 941 /*
 942  * helper function to walk through a file and find extents
 943  * newer than a specific transid, and smaller than thresh.
 944  *
 945  * This is used by the defragging code to find new and small
 946  * extents
 947  */
 948 static int find_new_extents(struct btrfs_root *root,
 949                             struct inode *inode, u64 newer_than,
 950                             u64 *off, int thresh)
 951 {
 952         struct btrfs_path *path;
 953         struct btrfs_key min_key;
 954         struct extent_buffer *leaf;
 955         struct btrfs_file_extent_item *extent;
 956         int type;
 957         int ret;
 958         u64 ino = btrfs_ino(inode);
 959
 960         path = btrfs_alloc_path();
 961         if (!path)
 962                 return -ENOMEM;
 963
 964         min_key.objectid = ino;
 965         min_key.type = BTRFS_EXTENT_DATA_KEY;
 966         min_key.offset = *off;
 967
 968         while (1) {
 969                 path->keep_locks = 1;
 970                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 971                 if (ret != 0)
 972                         goto none;
 973                 path->keep_locks = 0;
 974                 btrfs_unlock_up_safe(path, 1);
 975 process_slot:
 976                 if (min_key.objectid != ino)
 977                         goto none;
 978                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 979                         goto none;
 980
 981                 leaf = path->nodes[0];
 982                 extent = btrfs_item_ptr(leaf, path->slots[0],
 983                                         struct btrfs_file_extent_item);
 984
 985                 type = btrfs_file_extent_type(leaf, extent);
 986                 if (type == BTRFS_FILE_EXTENT_REG &&
 987                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 988                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 989                         *off = min_key.offset;
 990                         btrfs_free_path(path);
 991                         return 0;
 992                 }
 993
 994                 path->slots[0]++;
 995                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 996                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 997                         goto process_slot;
 998                 }
 999
1000                 if (min_key.offset == (u64)-1)
1001                         goto none;
1002
1003                 min_key.offset++;
1004                 btrfs_release_path(path);
1005         }
1006 none:
1007         btrfs_free_path(path);
1008         return -ENOENT;
1009 }
1010
1011 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1012 {
1013         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1014         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1015         struct extent_map *em;
1016         u64 len = PAGE_CACHE_SIZE;
1017
1018         /*
1019          * hopefully we have this extent in the tree already, try without
1020          * the full extent lock
1021          */
1022         read_lock(&em_tree->lock);
1023         em = lookup_extent_mapping(em_tree, start, len);
1024         read_unlock(&em_tree->lock);
1025
1026         if (!em) {
1027                 struct extent_state *cached = NULL;
1028                 u64 end = start + len - 1;
1029
1030                 /* get the big lock and read metadata off disk */
1031                 lock_extent_bits(io_tree, start, end, 0, &cached);
1032                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1033                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1034
1035                 if (IS_ERR(em))
1036                         return NULL;
1037         }
1038
1039         return em;
1040 }
1041
1042 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1043 {
1044         struct extent_map *next;
1045         bool ret = true;
1046
1047         /* this is the last extent */
1048         if (em->start + em->len >= i_size_read(inode))
1049                 return false;
1050
1051         next = defrag_lookup_extent(inode, em->start + em->len);
1052         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE ||
1053             (em->block_start + em->block_len == next->block_start))
1054                 ret = false;
1055
1056         free_extent_map(next);
1057         return ret;
1058 }
1059
1060 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
1061                                u64 *last_len, u64 *skip, u64 *defrag_end,
1062                                int compress)
1063 {
1064         struct extent_map *em;
1065         int ret = 1;
1066         bool next_mergeable = true;
1067
1068         /*
1069          * make sure that once we start defragging an extent, we keep on
1070          * defragging it
1071          */
1072         if (start < *defrag_end)
1073                 return 1;
1074
1075         *skip = 0;
1076
1077         em = defrag_lookup_extent(inode, start);
1078         if (!em)
1079                 return 0;
1080
1081         /* this will cover holes, and inline extents */
1082         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1083                 ret = 0;
1084                 goto out;
1085         }
1086
1087         next_mergeable = defrag_check_next_extent(inode, em);
1088
1089         /*
1090          * we hit a real extent, if it is big or the next extent is not a
1091          * real extent, don't bother defragging it
1092          */
1093         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1094             (em->len >= thresh || !next_mergeable))
1095                 ret = 0;
1096 out:
1097         /*
1098          * last_len ends up being a counter of how many bytes we've defragged.
1099          * every time we choose not to defrag an extent, we reset *last_len
1100          * so that the next tiny extent will force a defrag.
1101          *
1102          * The end result of this is that tiny extents before a single big
1103          * extent will force at least part of that big extent to be defragged.
1104          */
1105         if (ret) {
1106                 *defrag_end = extent_map_end(em);
1107         } else {
1108                 *last_len = 0;
1109                 *skip = extent_map_end(em);
1110                 *defrag_end = 0;
1111         }
1112
1113         free_extent_map(em);
1114         return ret;
1115 }
1116
1117 /*
1118  * it doesn't do much good to defrag one or two pages
1119  * at a time.  This pulls in a nice chunk of pages
1120  * to COW and defrag.
1121  *
1122  * It also makes sure the delalloc code has enough
1123  * dirty data to avoid making new small extents as part
1124  * of the defrag
1125  *
1126  * It's a good idea to start RA on this range
1127  * before calling this.
1128  */
1129 static int cluster_pages_for_defrag(struct inode *inode,
1130                                     struct page **pages,
1131                                     unsigned long start_index,
1132                                     unsigned long num_pages)
1133 {
1134         unsigned long file_end;
1135         u64 isize = i_size_read(inode);
1136         u64 page_start;
1137         u64 page_end;
1138         u64 page_cnt;
1139         int ret;
1140         int i;
1141         int i_done;
1142         struct btrfs_ordered_extent *ordered;
1143         struct extent_state *cached_state = NULL;
1144         struct extent_io_tree *tree;
1145         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1146
1147         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1148         if (!isize || start_index > file_end)
1149                 return 0;
1150
1151         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1152
1153         ret = btrfs_delalloc_reserve_space(inode,
1154                                            page_cnt << PAGE_CACHE_SHIFT);
1155         if (ret)
1156                 return ret;
1157         i_done = 0;
1158         tree = &BTRFS_I(inode)->io_tree;
1159
1160         /* step one, lock all the pages */
1161         for (i = 0; i < page_cnt; i++) {
1162                 struct page *page;
1163 again:
1164                 page = find_or_create_page(inode->i_mapping,
1165                                            start_index + i, mask);
1166                 if (!page)
1167                         break;
1168
1169                 page_start = page_offset(page);
1170                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1171                 while (1) {
1172                         lock_extent_bits(tree, page_start, page_end,
1173                                          0, &cached_state);
1174                         ordered = btrfs_lookup_ordered_extent(inode,
1175                                                               page_start);
1176                         unlock_extent_cached(tree, page_start, page_end,
1177                                              &cached_state, GFP_NOFS);
1178                         if (!ordered)
1179                                 break;
1180
1181                         unlock_page(page);
1182                         btrfs_start_ordered_extent(inode, ordered, 1);
1183                         btrfs_put_ordered_extent(ordered);
1184                         lock_page(page);
1185                         /*
1186                          * we unlocked the page above, so we need check if
1187                          * it was released or not.
1188                          */
1189                         if (page->mapping != inode->i_mapping) {
1190                                 unlock_page(page);
1191                                 page_cache_release(page);
1192                                 goto again;
1193                         }
1194                 }
1195
1196                 if (!PageUptodate(page)) {
1197                         btrfs_readpage(NULL, page);
1198                         lock_page(page);
1199                         if (!PageUptodate(page)) {
1200                                 unlock_page(page);
1201                                 page_cache_release(page);
1202                                 ret = -EIO;
1203                                 break;
1204                         }
1205                 }
1206
1207                 if (page->mapping != inode->i_mapping) {
1208                         unlock_page(page);
1209                         page_cache_release(page);
1210                         goto again;
1211                 }
1212
1213                 pages[i] = page;
1214                 i_done++;
1215         }
1216         if (!i_done || ret)
1217                 goto out;
1218
1219         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1220                 goto out;
1221
1222         /*
1223          * so now we have a nice long stream of locked
1224          * and up to date pages, lets wait on them
1225          */
1226         for (i = 0; i < i_done; i++)
1227                 wait_on_page_writeback(pages[i]);
1228
1229         page_start = page_offset(pages[0]);
1230         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1231
1232         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1233                          page_start, page_end - 1, 0, &cached_state);
1234         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1235                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1236                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1237                           &cached_state, GFP_NOFS);
1238
1239         if (i_done != page_cnt) {
1240                 spin_lock(&BTRFS_I(inode)->lock);
1241                 BTRFS_I(inode)->outstanding_extents++;
1242                 spin_unlock(&BTRFS_I(inode)->lock);
1243                 btrfs_delalloc_release_space(inode,
1244                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1245         }
1246
1247
1248         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1249                           &cached_state, GFP_NOFS);
1250
1251         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1252                              page_start, page_end - 1, &cached_state,
1253                              GFP_NOFS);
1254
1255         for (i = 0; i < i_done; i++) {
1256                 clear_page_dirty_for_io(pages[i]);
1257                 ClearPageChecked(pages[i]);
1258                 set_page_extent_mapped(pages[i]);
1259                 set_page_dirty(pages[i]);
1260                 unlock_page(pages[i]);
1261                 page_cache_release(pages[i]);
1262         }
1263         return i_done;
1264 out:
1265         for (i = 0; i < i_done; i++) {
1266                 unlock_page(pages[i]);
1267                 page_cache_release(pages[i]);
1268         }
1269         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1270         return ret;
1271
1272 }
1273
1274 int btrfs_defrag_file(struct inode *inode, struct file *file,
1275                       struct btrfs_ioctl_defrag_range_args *range,
1276                       u64 newer_than, unsigned long max_to_defrag)
1277 {
1278         struct btrfs_root *root = BTRFS_I(inode)->root;
1279         struct file_ra_state *ra = NULL;
1280         unsigned long last_index;
1281         u64 isize = i_size_read(inode);
1282         u64 last_len = 0;
1283         u64 skip = 0;
1284         u64 defrag_end = 0;
1285         u64 newer_off = range->start;
1286         unsigned long i;
1287         unsigned long ra_index = 0;
1288         int ret;
1289         int defrag_count = 0;
1290         int compress_type = BTRFS_COMPRESS_ZLIB;
1291         int extent_thresh = range->extent_thresh;
1292         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1293         unsigned long cluster = max_cluster;
1294         u64 new_align = ~((u64)128 * 1024 - 1);
1295         struct page **pages = NULL;
1296
1297         if (isize == 0)
1298                 return 0;
1299
1300         if (range->start >= isize)
1301                 return -EINVAL;
1302
1303         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1304                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1305                         return -EINVAL;
1306                 if (range->compress_type)
1307                         compress_type = range->compress_type;
1308         }
1309
1310         if (extent_thresh == 0)
1311                 extent_thresh = 256 * 1024;
1312
1313         /*
1314          * if we were not given a file, allocate a readahead
1315          * context
1316          */
1317         if (!file) {
1318                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1319                 if (!ra)
1320                         return -ENOMEM;
1321                 file_ra_state_init(ra, inode->i_mapping);
1322         } else {
1323                 ra = &file->f_ra;
1324         }
1325
1326         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1327                         GFP_NOFS);
1328         if (!pages) {
1329                 ret = -ENOMEM;
1330                 goto out_ra;
1331         }
1332
1333         /* find the last page to defrag */
1334         if (range->start + range->len > range->start) {
1335                 last_index = min_t(u64, isize - 1,
1336                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1337         } else {
1338                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1339         }
1340
1341         if (newer_than) {
1342                 ret = find_new_extents(root, inode, newer_than,
1343                                        &newer_off, 64 * 1024);
1344                 if (!ret) {
1345                         range->start = newer_off;
1346                         /*
1347                          * we always align our defrag to help keep
1348                          * the extents in the file evenly spaced
1349                          */
1350                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1351                 } else
1352                         goto out_ra;
1353         } else {
1354                 i = range->start >> PAGE_CACHE_SHIFT;
1355         }
1356         if (!max_to_defrag)
1357                 max_to_defrag = last_index + 1;
1358
1359         /*
1360          * make writeback starts from i, so the defrag range can be
1361          * written sequentially.
1362          */
1363         if (i < inode->i_mapping->writeback_index)
1364                 inode->i_mapping->writeback_index = i;
1365
1366         while (i <= last_index && defrag_count < max_to_defrag &&
1367                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1368                 PAGE_CACHE_SHIFT)) {
1369                 /*
1370                  * make sure we stop running if someone unmounts
1371                  * the FS
1372                  */
1373                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1374                         break;
1375
1376                 if (btrfs_defrag_cancelled(root->fs_info)) {
1377                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1378                         ret = -EAGAIN;
1379                         break;
1380                 }
1381
1382                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1383                                          extent_thresh, &last_len, &skip,
1384                                          &defrag_end, range->flags &
1385                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1386                         unsigned long next;
1387                         /*
1388                          * the should_defrag function tells us how much to skip
1389                          * bump our counter by the suggested amount
1390                          */
1391                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1392                         i = max(i + 1, next);
1393                         continue;
1394                 }
1395
1396                 if (!newer_than) {
1397                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1398                                    PAGE_CACHE_SHIFT) - i;
1399                         cluster = min(cluster, max_cluster);
1400                 } else {
1401                         cluster = max_cluster;
1402                 }
1403
1404                 if (i + cluster > ra_index) {
1405                         ra_index = max(i, ra_index);
1406                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1407                                        cluster);
1408                         ra_index += max_cluster;
1409                 }
1410
1411                 mutex_lock(&inode->i_mutex);
1412                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1413                         BTRFS_I(inode)->force_compress = compress_type;
1414                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1415                 if (ret < 0) {
1416                         mutex_unlock(&inode->i_mutex);
1417                         goto out_ra;
1418                 }
1419
1420                 defrag_count += ret;
1421                 balance_dirty_pages_ratelimited(inode->i_mapping);
1422                 mutex_unlock(&inode->i_mutex);
1423
1424                 if (newer_than) {
1425                         if (newer_off == (u64)-1)
1426                                 break;
1427
1428                         if (ret > 0)
1429                                 i += ret;
1430
1431                         newer_off = max(newer_off + 1,
1432                                         (u64)i << PAGE_CACHE_SHIFT);
1433
1434                         ret = find_new_extents(root, inode,
1435                                                newer_than, &newer_off,
1436                                                64 * 1024);
1437                         if (!ret) {
1438                                 range->start = newer_off;
1439                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1440                         } else {
1441                                 break;
1442                         }
1443                 } else {
1444                         if (ret > 0) {
1445                                 i += ret;
1446                                 last_len += ret << PAGE_CACHE_SHIFT;
1447                         } else {
1448                                 i++;
1449                                 last_len = 0;
1450                         }
1451                 }
1452         }
1453
1454         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1455                 filemap_flush(inode->i_mapping);
1456                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1457                              &BTRFS_I(inode)->runtime_flags))
1458                         filemap_flush(inode->i_mapping);
1459         }
1460
1461         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1462                 /* the filemap_flush will queue IO into the worker threads, but
1463                  * we have to make sure the IO is actually started and that
1464                  * ordered extents get created before we return
1465                  */
1466                 atomic_inc(&root->fs_info->async_submit_draining);
1467                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1468                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1469                         wait_event(root->fs_info->async_submit_wait,
1470                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1471                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1472                 }
1473                 atomic_dec(&root->fs_info->async_submit_draining);
1474         }
1475
1476         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1477                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1478         }
1479
1480         ret = defrag_count;
1481
1482 out_ra:
1483         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1484                 mutex_lock(&inode->i_mutex);
1485                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1486                 mutex_unlock(&inode->i_mutex);
1487         }
1488         if (!file)
1489                 kfree(ra);
1490         kfree(pages);
1491         return ret;
1492 }
1493
1494 static noinline int btrfs_ioctl_resize(struct file *file,
1495                                         void __user *arg)
1496 {
1497         u64 new_size;
1498         u64 old_size;
1499         u64 devid = 1;
1500         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1501         struct btrfs_ioctl_vol_args *vol_args;
1502         struct btrfs_trans_handle *trans;
1503         struct btrfs_device *device = NULL;
1504         char *sizestr;
1505         char *retptr;
1506         char *devstr = NULL;
1507         int ret = 0;
1508         int mod = 0;
1509
1510         if (!capable(CAP_SYS_ADMIN))
1511                 return -EPERM;
1512
1513         ret = mnt_want_write_file(file);
1514         if (ret)
1515                 return ret;
1516
1517         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1518                         1)) {
1519                 mnt_drop_write_file(file);
1520                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1521         }
1522
1523         mutex_lock(&root->fs_info->volume_mutex);
1524         vol_args = memdup_user(arg, sizeof(*vol_args));
1525         if (IS_ERR(vol_args)) {
1526                 ret = PTR_ERR(vol_args);
1527                 goto out;
1528         }
1529
1530         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1531
1532         sizestr = vol_args->name;
1533         devstr = strchr(sizestr, ':');
1534         if (devstr) {
1535                 sizestr = devstr + 1;
1536                 *devstr = '\0';
1537                 devstr = vol_args->name;
1538                 ret = kstrtoull(devstr, 10, &devid);
1539                 if (ret)
1540                         goto out_free;
1541                 if (!devid) {
1542                         ret = -EINVAL;
1543                         goto out_free;
1544                 }
1545                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1546         }
1547
1548         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1549         if (!device) {
1550                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1551                        devid);
1552                 ret = -ENODEV;
1553                 goto out_free;
1554         }
1555
1556         if (!device->writeable) {
1557                 btrfs_info(root->fs_info,
1558                            "resizer unable to apply on readonly device %llu",
1559                        devid);
1560                 ret = -EPERM;
1561                 goto out_free;
1562         }
1563
1564         if (!strcmp(sizestr, "max"))
1565                 new_size = device->bdev->bd_inode->i_size;
1566         else {
1567                 if (sizestr[0] == '-') {
1568                         mod = -1;
1569                         sizestr++;
1570                 } else if (sizestr[0] == '+') {
1571                         mod = 1;
1572                         sizestr++;
1573                 }
1574                 new_size = memparse(sizestr, &retptr);
1575                 if (*retptr != '\0' || new_size == 0) {
1576                         ret = -EINVAL;
1577                         goto out_free;
1578                 }
1579         }
1580
1581         if (device->is_tgtdev_for_dev_replace) {
1582                 ret = -EPERM;
1583                 goto out_free;
1584         }
1585
1586         old_size = device->total_bytes;
1587
1588         if (mod < 0) {
1589                 if (new_size > old_size) {
1590                         ret = -EINVAL;
1591                         goto out_free;
1592                 }
1593                 new_size = old_size - new_size;
1594         } else if (mod > 0) {
1595                 if (new_size > ULLONG_MAX - old_size) {
1596                         ret = -ERANGE;
1597                         goto out_free;
1598                 }
1599                 new_size = old_size + new_size;
1600         }
1601
1602         if (new_size < 256 * 1024 * 1024) {
1603                 ret = -EINVAL;
1604                 goto out_free;
1605         }
1606         if (new_size > device->bdev->bd_inode->i_size) {
1607                 ret = -EFBIG;
1608                 goto out_free;
1609         }
1610
1611         do_div(new_size, root->sectorsize);
1612         new_size *= root->sectorsize;
1613
1614         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1615                       rcu_str_deref(device->name), new_size);
1616
1617         if (new_size > old_size) {
1618                 trans = btrfs_start_transaction(root, 0);
1619                 if (IS_ERR(trans)) {
1620                         ret = PTR_ERR(trans);
1621                         goto out_free;
1622                 }
1623                 ret = btrfs_grow_device(trans, device, new_size);
1624                 btrfs_commit_transaction(trans, root);
1625         } else if (new_size < old_size) {
1626                 ret = btrfs_shrink_device(device, new_size);
1627         } /* equal, nothing need to do */
1628
1629 out_free:
1630         kfree(vol_args);
1631 out:
1632         mutex_unlock(&root->fs_info->volume_mutex);
1633         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1634         mnt_drop_write_file(file);
1635         return ret;
1636 }
1637
1638 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1639                                 char *name, unsigned long fd, int subvol,
1640                                 u64 *transid, bool readonly,
1641                                 struct btrfs_qgroup_inherit *inherit)
1642 {
1643         int namelen;
1644         int ret = 0;
1645
1646         ret = mnt_want_write_file(file);
1647         if (ret)
1648                 goto out;
1649
1650         namelen = strlen(name);
1651         if (strchr(name, '/')) {
1652                 ret = -EINVAL;
1653                 goto out_drop_write;
1654         }
1655
1656         if (name[0] == '.' &&
1657            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1658                 ret = -EEXIST;
1659                 goto out_drop_write;
1660         }
1661
1662         if (subvol) {
1663                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1664                                      NULL, transid, readonly, inherit);
1665         } else {
1666                 struct fd src = fdget(fd);
1667                 struct inode *src_inode;
1668                 if (!src.file) {
1669                         ret = -EINVAL;
1670                         goto out_drop_write;
1671                 }
1672
1673                 src_inode = file_inode(src.file);
1674                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1675                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1676                                    "Snapshot src from another FS");
1677                         ret = -EXDEV;
1678                 } else if (!inode_owner_or_capable(src_inode)) {
1679                         /*
1680                          * Subvolume creation is not restricted, but snapshots
1681                          * are limited to own subvolumes only
1682                          */
1683                         ret = -EPERM;
1684                 } else {
1685                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1686                                              BTRFS_I(src_inode)->root,
1687                                              transid, readonly, inherit);
1688                 }
1689                 fdput(src);
1690         }
1691 out_drop_write:
1692         mnt_drop_write_file(file);
1693 out:
1694         return ret;
1695 }
1696
1697 static noinline int btrfs_ioctl_snap_create(struct file *file,
1698                                             void __user *arg, int subvol)
1699 {
1700         struct btrfs_ioctl_vol_args *vol_args;
1701         int ret;
1702
1703         vol_args = memdup_user(arg, sizeof(*vol_args));
1704         if (IS_ERR(vol_args))
1705                 return PTR_ERR(vol_args);
1706         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1707
1708         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1709                                               vol_args->fd, subvol,
1710                                               NULL, false, NULL);
1711
1712         kfree(vol_args);
1713         return ret;
1714 }
1715
1716 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1717                                                void __user *arg, int subvol)
1718 {
1719         struct btrfs_ioctl_vol_args_v2 *vol_args;
1720         int ret;
1721         u64 transid = 0;
1722         u64 *ptr = NULL;
1723         bool readonly = false;
1724         struct btrfs_qgroup_inherit *inherit = NULL;
1725
1726         vol_args = memdup_user(arg, sizeof(*vol_args));
1727         if (IS_ERR(vol_args))
1728                 return PTR_ERR(vol_args);
1729         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1730
1731         if (vol_args->flags &
1732             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1733               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1734                 ret = -EOPNOTSUPP;
1735                 goto out;
1736         }
1737
1738         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1739                 ptr = &transid;
1740         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1741                 readonly = true;
1742         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1743                 if (vol_args->size > PAGE_CACHE_SIZE) {
1744                         ret = -EINVAL;
1745                         goto out;
1746                 }
1747                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1748                 if (IS_ERR(inherit)) {
1749                         ret = PTR_ERR(inherit);
1750                         goto out;
1751                 }
1752         }
1753
1754         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1755                                               vol_args->fd, subvol, ptr,
1756                                               readonly, inherit);
1757
1758         if (ret == 0 && ptr &&
1759             copy_to_user(arg +
1760                          offsetof(struct btrfs_ioctl_vol_args_v2,
1761                                   transid), ptr, sizeof(*ptr)))
1762                 ret = -EFAULT;
1763 out:
1764         kfree(vol_args);
1765         kfree(inherit);
1766         return ret;
1767 }
1768
1769 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1770                                                 void __user *arg)
1771 {
1772         struct inode *inode = file_inode(file);
1773         struct btrfs_root *root = BTRFS_I(inode)->root;
1774         int ret = 0;
1775         u64 flags = 0;
1776
1777         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1778                 return -EINVAL;
1779
1780         down_read(&root->fs_info->subvol_sem);
1781         if (btrfs_root_readonly(root))
1782                 flags |= BTRFS_SUBVOL_RDONLY;
1783         up_read(&root->fs_info->subvol_sem);
1784
1785         if (copy_to_user(arg, &flags, sizeof(flags)))
1786                 ret = -EFAULT;
1787
1788         return ret;
1789 }
1790
1791 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1792                                               void __user *arg)
1793 {
1794         struct inode *inode = file_inode(file);
1795         struct btrfs_root *root = BTRFS_I(inode)->root;
1796         struct btrfs_trans_handle *trans;
1797         u64 root_flags;
1798         u64 flags;
1799         int ret = 0;
1800
1801         if (!inode_owner_or_capable(inode))
1802                 return -EPERM;
1803
1804         ret = mnt_want_write_file(file);
1805         if (ret)
1806                 goto out;
1807
1808         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1809                 ret = -EINVAL;
1810                 goto out_drop_write;
1811         }
1812
1813         if (copy_from_user(&flags, arg, sizeof(flags))) {
1814                 ret = -EFAULT;
1815                 goto out_drop_write;
1816         }
1817
1818         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1819                 ret = -EINVAL;
1820                 goto out_drop_write;
1821         }
1822
1823         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1824                 ret = -EOPNOTSUPP;
1825                 goto out_drop_write;
1826         }
1827
1828         down_write(&root->fs_info->subvol_sem);
1829
1830         /* nothing to do */
1831         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1832                 goto out_drop_sem;
1833
1834         root_flags = btrfs_root_flags(&root->root_item);
1835         if (flags & BTRFS_SUBVOL_RDONLY) {
1836                 btrfs_set_root_flags(&root->root_item,
1837                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1838         } else {
1839                 /*
1840                  * Block RO -> RW transition if this subvolume is involved in
1841                  * send
1842                  */
1843                 spin_lock(&root->root_item_lock);
1844                 if (root->send_in_progress == 0) {
1845                         btrfs_set_root_flags(&root->root_item,
1846                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1847                         spin_unlock(&root->root_item_lock);
1848                 } else {
1849                         spin_unlock(&root->root_item_lock);
1850                         btrfs_warn(root->fs_info,
1851                         "Attempt to set subvolume %llu read-write during send",
1852                                         root->root_key.objectid);
1853                         ret = -EPERM;
1854                         goto out_drop_sem;
1855                 }
1856         }
1857
1858         trans = btrfs_start_transaction(root, 1);
1859         if (IS_ERR(trans)) {
1860                 ret = PTR_ERR(trans);
1861                 goto out_reset;
1862         }
1863
1864         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1865                                 &root->root_key, &root->root_item);
1866
1867         btrfs_commit_transaction(trans, root);
1868 out_reset:
1869         if (ret)
1870                 btrfs_set_root_flags(&root->root_item, root_flags);
1871 out_drop_sem:
1872         up_write(&root->fs_info->subvol_sem);
1873 out_drop_write:
1874         mnt_drop_write_file(file);
1875 out:
1876         return ret;
1877 }
1878
1879 /*
1880  * helper to check if the subvolume references other subvolumes
1881  */
1882 static noinline int may_destroy_subvol(struct btrfs_root *root)
1883 {
1884         struct btrfs_path *path;
1885         struct btrfs_dir_item *di;
1886         struct btrfs_key key;
1887         u64 dir_id;
1888         int ret;
1889
1890         path = btrfs_alloc_path();
1891         if (!path)
1892                 return -ENOMEM;
1893
1894         /* Make sure this root isn't set as the default subvol */
1895         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1896         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1897                                    dir_id, "default", 7, 0);
1898         if (di && !IS_ERR(di)) {
1899                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1900                 if (key.objectid == root->root_key.objectid) {
1901                         ret = -EPERM;
1902                         btrfs_err(root->fs_info, "deleting default subvolume "
1903                                   "%llu is not allowed", key.objectid);
1904                         goto out;
1905                 }
1906                 btrfs_release_path(path);
1907         }
1908
1909         key.objectid = root->root_key.objectid;
1910         key.type = BTRFS_ROOT_REF_KEY;
1911         key.offset = (u64)-1;
1912
1913         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1914                                 &key, path, 0, 0);
1915         if (ret < 0)
1916                 goto out;
1917         BUG_ON(ret == 0);
1918
1919         ret = 0;
1920         if (path->slots[0] > 0) {
1921                 path->slots[0]--;
1922                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1923                 if (key.objectid == root->root_key.objectid &&
1924                     key.type == BTRFS_ROOT_REF_KEY)
1925                         ret = -ENOTEMPTY;
1926         }
1927 out:
1928         btrfs_free_path(path);
1929         return ret;
1930 }
1931
1932 static noinline int key_in_sk(struct btrfs_key *key,
1933                               struct btrfs_ioctl_search_key *sk)
1934 {
1935         struct btrfs_key test;
1936         int ret;
1937
1938         test.objectid = sk->min_objectid;
1939         test.type = sk->min_type;
1940         test.offset = sk->min_offset;
1941
1942         ret = btrfs_comp_cpu_keys(key, &test);
1943         if (ret < 0)
1944                 return 0;
1945
1946         test.objectid = sk->max_objectid;
1947         test.type = sk->max_type;
1948         test.offset = sk->max_offset;
1949
1950         ret = btrfs_comp_cpu_keys(key, &test);
1951         if (ret > 0)
1952                 return 0;
1953         return 1;
1954 }
1955
1956 static noinline int copy_to_sk(struct btrfs_root *root,
1957                                struct btrfs_path *path,
1958                                struct btrfs_key *key,
1959                                struct btrfs_ioctl_search_key *sk,
1960                                size_t *buf_size,
1961                                char __user *ubuf,
1962                                unsigned long *sk_offset,
1963                                int *num_found)
1964 {
1965         u64 found_transid;
1966         struct extent_buffer *leaf;
1967         struct btrfs_ioctl_search_header sh;
1968         unsigned long item_off;
1969         unsigned long item_len;
1970         int nritems;
1971         int i;
1972         int slot;
1973         int ret = 0;
1974
1975         leaf = path->nodes[0];
1976         slot = path->slots[0];
1977         nritems = btrfs_header_nritems(leaf);
1978
1979         if (btrfs_header_generation(leaf) > sk->max_transid) {
1980                 i = nritems;
1981                 goto advance_key;
1982         }
1983         found_transid = btrfs_header_generation(leaf);
1984
1985         for (i = slot; i < nritems; i++) {
1986                 item_off = btrfs_item_ptr_offset(leaf, i);
1987                 item_len = btrfs_item_size_nr(leaf, i);
1988
1989                 btrfs_item_key_to_cpu(leaf, key, i);
1990                 if (!key_in_sk(key, sk))
1991                         continue;
1992
1993                 if (sizeof(sh) + item_len > *buf_size) {
1994                         if (*num_found) {
1995                                 ret = 1;
1996                                 goto out;
1997                         }
1998
1999                         /*
2000                          * return one empty item back for v1, which does not
2001                          * handle -EOVERFLOW
2002                          */
2003
2004                         *buf_size = sizeof(sh) + item_len;
2005                         item_len = 0;
2006                         ret = -EOVERFLOW;
2007                 }
2008
2009                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2010                         ret = 1;
2011                         goto out;
2012                 }
2013
2014                 sh.objectid = key->objectid;
2015                 sh.offset = key->offset;
2016                 sh.type = key->type;
2017                 sh.len = item_len;
2018                 sh.transid = found_transid;
2019
2020                 /* copy search result header */
2021                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2022                         ret = -EFAULT;
2023                         goto out;
2024                 }
2025
2026                 *sk_offset += sizeof(sh);
2027
2028                 if (item_len) {
2029                         char __user *up = ubuf + *sk_offset;
2030                         /* copy the item */
2031                         if (read_extent_buffer_to_user(leaf, up,
2032                                                        item_off, item_len)) {
2033                                 ret = -EFAULT;
2034                                 goto out;
2035                         }
2036
2037                         *sk_offset += item_len;
2038                 }
2039                 (*num_found)++;
2040
2041                 if (ret) /* -EOVERFLOW from above */
2042                         goto out;
2043
2044                 if (*num_found >= sk->nr_items) {
2045                         ret = 1;
2046                         goto out;
2047                 }
2048         }
2049 advance_key:
2050         ret = 0;
2051         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2052                 key->offset++;
2053         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2054                 key->offset = 0;
2055                 key->type++;
2056         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2057                 key->offset = 0;
2058                 key->type = 0;
2059                 key->objectid++;
2060         } else
2061                 ret = 1;
2062 out:
2063         /*
2064          *  0: all items from this leaf copied, continue with next
2065          *  1: * more items can be copied, but unused buffer is too small
2066          *     * all items were found
2067          *     Either way, it will stops the loop which iterates to the next
2068          *     leaf
2069          *  -EOVERFLOW: item was to large for buffer
2070          *  -EFAULT: could not copy extent buffer back to userspace
2071          */
2072         return ret;
2073 }
2074
2075 static noinline int search_ioctl(struct inode *inode,
2076                                  struct btrfs_ioctl_search_key *sk,
2077                                  size_t *buf_size,
2078                                  char __user *ubuf)
2079 {
2080         struct btrfs_root *root;
2081         struct btrfs_key key;
2082         struct btrfs_path *path;
2083         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2084         int ret;
2085         int num_found = 0;
2086         unsigned long sk_offset = 0;
2087
2088         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2089                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2090                 return -EOVERFLOW;
2091         }
2092
2093         path = btrfs_alloc_path();
2094         if (!path)
2095                 return -ENOMEM;
2096
2097         if (sk->tree_id == 0) {
2098                 /* search the root of the inode that was passed */
2099                 root = BTRFS_I(inode)->root;
2100         } else {
2101                 key.objectid = sk->tree_id;
2102                 key.type = BTRFS_ROOT_ITEM_KEY;
2103                 key.offset = (u64)-1;
2104                 root = btrfs_read_fs_root_no_name(info, &key);
2105                 if (IS_ERR(root)) {
2106                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2107                                sk->tree_id);
2108                         btrfs_free_path(path);
2109                         return -ENOENT;
2110                 }
2111         }
2112
2113         key.objectid = sk->min_objectid;
2114         key.type = sk->min_type;
2115         key.offset = sk->min_offset;
2116
2117         path->keep_locks = 1;
2118
2119         while (1) {
2120                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2121                 if (ret != 0) {
2122                         if (ret > 0)
2123                                 ret = 0;
2124                         goto err;
2125                 }
2126                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2127                                  &sk_offset, &num_found);
2128                 btrfs_release_path(path);
2129                 if (ret)
2130                         break;
2131
2132         }
2133         if (ret > 0)
2134                 ret = 0;
2135 err:
2136         sk->nr_items = num_found;
2137         btrfs_free_path(path);
2138         return ret;
2139 }
2140
2141 static noinline int btrfs_ioctl_tree_search(struct file *file,
2142                                            void __user *argp)
2143 {
2144         struct btrfs_ioctl_search_args __user *uargs;
2145         struct btrfs_ioctl_search_key sk;
2146         struct inode *inode;
2147         int ret;
2148         size_t buf_size;
2149
2150         if (!capable(CAP_SYS_ADMIN))
2151                 return -EPERM;
2152
2153         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2154
2155         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2156                 return -EFAULT;
2157
2158         buf_size = sizeof(uargs->buf);
2159
2160         inode = file_inode(file);
2161         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2162
2163         /*
2164          * In the origin implementation an overflow is handled by returning a
2165          * search header with a len of zero, so reset ret.
2166          */
2167         if (ret == -EOVERFLOW)
2168                 ret = 0;
2169
2170         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2171                 ret = -EFAULT;
2172         return ret;
2173 }
2174
2175 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2176                                                void __user *argp)
2177 {
2178         struct btrfs_ioctl_search_args_v2 __user *uarg;
2179         struct btrfs_ioctl_search_args_v2 args;
2180         struct inode *inode;
2181         int ret;
2182         size_t buf_size;
2183         const size_t buf_limit = 16 * 1024 * 1024;
2184
2185         if (!capable(CAP_SYS_ADMIN))
2186                 return -EPERM;
2187
2188         /* copy search header and buffer size */
2189         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2190         if (copy_from_user(&args, uarg, sizeof(args)))
2191                 return -EFAULT;
2192
2193         buf_size = args.buf_size;
2194
2195         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2196                 return -EOVERFLOW;
2197
2198         /* limit result size to 16MB */
2199         if (buf_size > buf_limit)
2200                 buf_size = buf_limit;
2201
2202         inode = file_inode(file);
2203         ret = search_ioctl(inode, &args.key, &buf_size,
2204                            (char *)(&uarg->buf[0]));
2205         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2206                 ret = -EFAULT;
2207         else if (ret == -EOVERFLOW &&
2208                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2209                 ret = -EFAULT;
2210
2211         return ret;
2212 }
2213
2214 /*
2215  * Search INODE_REFs to identify path name of 'dirid' directory
2216  * in a 'tree_id' tree. and sets path name to 'name'.
2217  */
2218 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2219                                 u64 tree_id, u64 dirid, char *name)
2220 {
2221         struct btrfs_root *root;
2222         struct btrfs_key key;
2223         char *ptr;
2224         int ret = -1;
2225         int slot;
2226         int len;
2227         int total_len = 0;
2228         struct btrfs_inode_ref *iref;
2229         struct extent_buffer *l;
2230         struct btrfs_path *path;
2231
2232         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2233                 name[0]='\0';
2234                 return 0;
2235         }
2236
2237         path = btrfs_alloc_path();
2238         if (!path)
2239                 return -ENOMEM;
2240
2241         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2242
2243         key.objectid = tree_id;
2244         key.type = BTRFS_ROOT_ITEM_KEY;
2245         key.offset = (u64)-1;
2246         root = btrfs_read_fs_root_no_name(info, &key);
2247         if (IS_ERR(root)) {
2248                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2249                 ret = -ENOENT;
2250                 goto out;
2251         }
2252
2253         key.objectid = dirid;
2254         key.type = BTRFS_INODE_REF_KEY;
2255         key.offset = (u64)-1;
2256
2257         while (1) {
2258                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2259                 if (ret < 0)
2260                         goto out;
2261                 else if (ret > 0) {
2262                         ret = btrfs_previous_item(root, path, dirid,
2263                                                   BTRFS_INODE_REF_KEY);
2264                         if (ret < 0)
2265                                 goto out;
2266                         else if (ret > 0) {
2267                                 ret = -ENOENT;
2268                                 goto out;
2269                         }
2270                 }
2271
2272                 l = path->nodes[0];
2273                 slot = path->slots[0];
2274                 btrfs_item_key_to_cpu(l, &key, slot);
2275
2276                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2277                 len = btrfs_inode_ref_name_len(l, iref);
2278                 ptr -= len + 1;
2279                 total_len += len + 1;
2280                 if (ptr < name) {
2281                         ret = -ENAMETOOLONG;
2282                         goto out;
2283                 }
2284
2285                 *(ptr + len) = '/';
2286                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2287
2288                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2289                         break;
2290
2291                 btrfs_release_path(path);
2292                 key.objectid = key.offset;
2293                 key.offset = (u64)-1;
2294                 dirid = key.objectid;
2295         }
2296         memmove(name, ptr, total_len);
2297         name[total_len] = '\0';
2298         ret = 0;
2299 out:
2300         btrfs_free_path(path);
2301         return ret;
2302 }
2303
2304 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2305                                            void __user *argp)
2306 {
2307          struct btrfs_ioctl_ino_lookup_args *args;
2308          struct inode *inode;
2309          int ret;
2310
2311         if (!capable(CAP_SYS_ADMIN))
2312                 return -EPERM;
2313
2314         args = memdup_user(argp, sizeof(*args));
2315         if (IS_ERR(args))
2316                 return PTR_ERR(args);
2317
2318         inode = file_inode(file);
2319
2320         if (args->treeid == 0)
2321                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2322
2323         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2324                                         args->treeid, args->objectid,
2325                                         args->name);
2326
2327         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2328                 ret = -EFAULT;
2329
2330         kfree(args);
2331         return ret;
2332 }
2333
2334 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2335                                              void __user *arg)
2336 {
2337         struct dentry *parent = file->f_path.dentry;
2338         struct dentry *dentry;
2339         struct inode *dir = parent->d_inode;
2340         struct inode *inode;
2341         struct btrfs_root *root = BTRFS_I(dir)->root;
2342         struct btrfs_root *dest = NULL;
2343         struct btrfs_ioctl_vol_args *vol_args;
2344         struct btrfs_trans_handle *trans;
2345         struct btrfs_block_rsv block_rsv;
2346         u64 root_flags;
2347         u64 qgroup_reserved;
2348         int namelen;
2349         int ret;
2350         int err = 0;
2351
2352         vol_args = memdup_user(arg, sizeof(*vol_args));
2353         if (IS_ERR(vol_args))
2354                 return PTR_ERR(vol_args);
2355
2356         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2357         namelen = strlen(vol_args->name);
2358         if (strchr(vol_args->name, '/') ||
2359             strncmp(vol_args->name, "..", namelen) == 0) {
2360                 err = -EINVAL;
2361                 goto out;
2362         }
2363
2364         err = mnt_want_write_file(file);
2365         if (err)
2366                 goto out;
2367
2368
2369         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2370         if (err == -EINTR)
2371                 goto out_drop_write;
2372         dentry = lookup_one_len(vol_args->name, parent, namelen);
2373         if (IS_ERR(dentry)) {
2374                 err = PTR_ERR(dentry);
2375                 goto out_unlock_dir;
2376         }
2377
2378         if (!dentry->d_inode) {
2379                 err = -ENOENT;
2380                 goto out_dput;
2381         }
2382
2383         inode = dentry->d_inode;
2384         dest = BTRFS_I(inode)->root;
2385         if (!capable(CAP_SYS_ADMIN)) {
2386                 /*
2387                  * Regular user.  Only allow this with a special mount
2388                  * option, when the user has write+exec access to the
2389                  * subvol root, and when rmdir(2) would have been
2390                  * allowed.
2391                  *
2392                  * Note that this is _not_ check that the subvol is
2393                  * empty or doesn't contain data that we wouldn't
2394                  * otherwise be able to delete.
2395                  *
2396                  * Users who want to delete empty subvols should try
2397                  * rmdir(2).
2398                  */
2399                 err = -EPERM;
2400                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2401                         goto out_dput;
2402
2403                 /*
2404                  * Do not allow deletion if the parent dir is the same
2405                  * as the dir to be deleted.  That means the ioctl
2406                  * must be called on the dentry referencing the root
2407                  * of the subvol, not a random directory contained
2408                  * within it.
2409                  */
2410                 err = -EINVAL;
2411                 if (root == dest)
2412                         goto out_dput;
2413
2414                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2415                 if (err)
2416                         goto out_dput;
2417         }
2418
2419         /* check if subvolume may be deleted by a user */
2420         err = btrfs_may_delete(dir, dentry, 1);
2421         if (err)
2422                 goto out_dput;
2423
2424         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2425                 err = -EINVAL;
2426                 goto out_dput;
2427         }
2428
2429         mutex_lock(&inode->i_mutex);
2430
2431         /*
2432          * Don't allow to delete a subvolume with send in progress. This is
2433          * inside the i_mutex so the error handling that has to drop the bit
2434          * again is not run concurrently.
2435          */
2436         spin_lock(&dest->root_item_lock);
2437         root_flags = btrfs_root_flags(&dest->root_item);
2438         if (dest->send_in_progress == 0) {
2439                 btrfs_set_root_flags(&dest->root_item,
2440                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2441                 spin_unlock(&dest->root_item_lock);
2442         } else {
2443                 spin_unlock(&dest->root_item_lock);
2444                 btrfs_warn(root->fs_info,
2445                         "Attempt to delete subvolume %llu during send",
2446                         dest->root_key.objectid);
2447                 err = -EPERM;
2448                 goto out_dput;
2449         }
2450
2451         err = d_invalidate(dentry);
2452         if (err)
2453                 goto out_unlock;
2454
2455         down_write(&root->fs_info->subvol_sem);
2456
2457         err = may_destroy_subvol(dest);
2458         if (err)
2459                 goto out_up_write;
2460
2461         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2462         /*
2463          * One for dir inode, two for dir entries, two for root
2464          * ref/backref.
2465          */
2466         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2467                                                5, &qgroup_reserved, true);
2468         if (err)
2469                 goto out_up_write;
2470
2471         trans = btrfs_start_transaction(root, 0);
2472         if (IS_ERR(trans)) {
2473                 err = PTR_ERR(trans);
2474                 goto out_release;
2475         }
2476         trans->block_rsv = &block_rsv;
2477         trans->bytes_reserved = block_rsv.size;
2478
2479         ret = btrfs_unlink_subvol(trans, root, dir,
2480                                 dest->root_key.objectid,
2481                                 dentry->d_name.name,
2482                                 dentry->d_name.len);
2483         if (ret) {
2484                 err = ret;
2485                 btrfs_abort_transaction(trans, root, ret);
2486                 goto out_end_trans;
2487         }
2488
2489         btrfs_record_root_in_trans(trans, dest);
2490
2491         memset(&dest->root_item.drop_progress, 0,
2492                 sizeof(dest->root_item.drop_progress));
2493         dest->root_item.drop_level = 0;
2494         btrfs_set_root_refs(&dest->root_item, 0);
2495
2496         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2497                 ret = btrfs_insert_orphan_item(trans,
2498                                         root->fs_info->tree_root,
2499                                         dest->root_key.objectid);
2500                 if (ret) {
2501                         btrfs_abort_transaction(trans, root, ret);
2502                         err = ret;
2503                         goto out_end_trans;
2504                 }
2505         }
2506
2507         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2508                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2509                                   dest->root_key.objectid);
2510         if (ret && ret != -ENOENT) {
2511                 btrfs_abort_transaction(trans, root, ret);
2512                 err = ret;
2513                 goto out_end_trans;
2514         }
2515         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2516                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2517                                           dest->root_item.received_uuid,
2518                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2519                                           dest->root_key.objectid);
2520                 if (ret && ret != -ENOENT) {
2521                         btrfs_abort_transaction(trans, root, ret);
2522                         err = ret;
2523                         goto out_end_trans;
2524                 }
2525         }
2526
2527 out_end_trans:
2528         trans->block_rsv = NULL;
2529         trans->bytes_reserved = 0;
2530         ret = btrfs_end_transaction(trans, root);
2531         if (ret && !err)
2532                 err = ret;
2533         inode->i_flags |= S_DEAD;
2534 out_release:
2535         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2536 out_up_write:
2537         up_write(&root->fs_info->subvol_sem);
2538 out_unlock:
2539         if (err) {
2540                 spin_lock(&dest->root_item_lock);
2541                 root_flags = btrfs_root_flags(&dest->root_item);
2542                 btrfs_set_root_flags(&dest->root_item,
2543                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2544                 spin_unlock(&dest->root_item_lock);
2545         }
2546         mutex_unlock(&inode->i_mutex);
2547         if (!err) {
2548                 shrink_dcache_sb(root->fs_info->sb);
2549                 btrfs_invalidate_inodes(dest);
2550                 d_delete(dentry);
2551                 ASSERT(dest->send_in_progress == 0);
2552
2553                 /* the last ref */
2554                 if (dest->cache_inode) {
2555                         iput(dest->cache_inode);
2556                         dest->cache_inode = NULL;
2557                 }
2558         }
2559 out_dput:
2560         dput(dentry);
2561 out_unlock_dir:
2562         mutex_unlock(&dir->i_mutex);
2563 out_drop_write:
2564         mnt_drop_write_file(file);
2565 out:
2566         kfree(vol_args);
2567         return err;
2568 }
2569
2570 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2571 {
2572         struct inode *inode = file_inode(file);
2573         struct btrfs_root *root = BTRFS_I(inode)->root;
2574         struct btrfs_ioctl_defrag_range_args *range;
2575         int ret;
2576
2577         ret = mnt_want_write_file(file);
2578         if (ret)
2579                 return ret;
2580
2581         if (btrfs_root_readonly(root)) {
2582                 ret = -EROFS;
2583                 goto out;
2584         }
2585
2586         switch (inode->i_mode & S_IFMT) {
2587         case S_IFDIR:
2588                 if (!capable(CAP_SYS_ADMIN)) {
2589                         ret = -EPERM;
2590                         goto out;
2591                 }
2592                 ret = btrfs_defrag_root(root);
2593                 if (ret)
2594                         goto out;
2595                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2596                 break;
2597         case S_IFREG:
2598                 if (!(file->f_mode & FMODE_WRITE)) {
2599                         ret = -EINVAL;
2600                         goto out;
2601                 }
2602
2603                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2604                 if (!range) {
2605                         ret = -ENOMEM;
2606                         goto out;
2607                 }
2608
2609                 if (argp) {
2610                         if (copy_from_user(range, argp,
2611                                            sizeof(*range))) {
2612                                 ret = -EFAULT;
2613                                 kfree(range);
2614                                 goto out;
2615                         }
2616                         /* compression requires us to start the IO */
2617                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2618                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2619                                 range->extent_thresh = (u32)-1;
2620                         }
2621                 } else {
2622                         /* the rest are all set to zero by kzalloc */
2623                         range->len = (u64)-1;
2624                 }
2625                 ret = btrfs_defrag_file(file_inode(file), file,
2626                                         range, 0, 0);
2627                 if (ret > 0)
2628                         ret = 0;
2629                 kfree(range);
2630                 break;
2631         default:
2632                 ret = -EINVAL;
2633         }
2634 out:
2635         mnt_drop_write_file(file);
2636         return ret;
2637 }
2638
2639 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2640 {
2641         struct btrfs_ioctl_vol_args *vol_args;
2642         int ret;
2643
2644         if (!capable(CAP_SYS_ADMIN))
2645                 return -EPERM;
2646
2647         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2648                         1)) {
2649                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2650         }
2651
2652         mutex_lock(&root->fs_info->volume_mutex);
2653         vol_args = memdup_user(arg, sizeof(*vol_args));
2654         if (IS_ERR(vol_args)) {
2655                 ret = PTR_ERR(vol_args);
2656                 goto out;
2657         }
2658
2659         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2660         ret = btrfs_init_new_device(root, vol_args->name);
2661
2662         kfree(vol_args);
2663 out:
2664         mutex_unlock(&root->fs_info->volume_mutex);
2665         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2666         return ret;
2667 }
2668
2669 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2670 {
2671         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2672         struct btrfs_ioctl_vol_args *vol_args;
2673         int ret;
2674
2675         if (!capable(CAP_SYS_ADMIN))
2676                 return -EPERM;
2677
2678         ret = mnt_want_write_file(file);
2679         if (ret)
2680                 return ret;
2681
2682         vol_args = memdup_user(arg, sizeof(*vol_args));
2683         if (IS_ERR(vol_args)) {
2684                 ret = PTR_ERR(vol_args);
2685                 goto out;
2686         }
2687
2688         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2689
2690         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2691                         1)) {
2692                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2693                 goto out;
2694         }
2695
2696         mutex_lock(&root->fs_info->volume_mutex);
2697         ret = btrfs_rm_device(root, vol_args->name);
2698         mutex_unlock(&root->fs_info->volume_mutex);
2699         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2700
2701 out:
2702         kfree(vol_args);
2703         mnt_drop_write_file(file);
2704         return ret;
2705 }
2706
2707 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2708 {
2709         struct btrfs_ioctl_fs_info_args *fi_args;
2710         struct btrfs_device *device;
2711         struct btrfs_device *next;
2712         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2713         int ret = 0;
2714
2715         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2716         if (!fi_args)
2717                 return -ENOMEM;
2718
2719         mutex_lock(&fs_devices->device_list_mutex);
2720         fi_args->num_devices = fs_devices->num_devices;
2721         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2722
2723         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2724                 if (device->devid > fi_args->max_id)
2725                         fi_args->max_id = device->devid;
2726         }
2727         mutex_unlock(&fs_devices->device_list_mutex);
2728
2729         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2730         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2731         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2732
2733         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2734                 ret = -EFAULT;
2735
2736         kfree(fi_args);
2737         return ret;
2738 }
2739
2740 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2741 {
2742         struct btrfs_ioctl_dev_info_args *di_args;
2743         struct btrfs_device *dev;
2744         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2745         int ret = 0;
2746         char *s_uuid = NULL;
2747
2748         di_args = memdup_user(arg, sizeof(*di_args));
2749         if (IS_ERR(di_args))
2750                 return PTR_ERR(di_args);
2751
2752         if (!btrfs_is_empty_uuid(di_args->uuid))
2753                 s_uuid = di_args->uuid;
2754
2755         mutex_lock(&fs_devices->device_list_mutex);
2756         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2757
2758         if (!dev) {
2759                 ret = -ENODEV;
2760                 goto out;
2761         }
2762
2763         di_args->devid = dev->devid;
2764         di_args->bytes_used = dev->bytes_used;
2765         di_args->total_bytes = dev->total_bytes;
2766         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2767         if (dev->name) {
2768                 struct rcu_string *name;
2769
2770                 rcu_read_lock();
2771                 name = rcu_dereference(dev->name);
2772                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2773                 rcu_read_unlock();
2774                 di_args->path[sizeof(di_args->path) - 1] = 0;
2775         } else {
2776                 di_args->path[0] = '\0';
2777         }
2778
2779 out:
2780         mutex_unlock(&fs_devices->device_list_mutex);
2781         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2782                 ret = -EFAULT;
2783
2784         kfree(di_args);
2785         return ret;
2786 }
2787
2788 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2789 {
2790         struct page *page;
2791         pgoff_t index;
2792         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2793
2794         index = off >> PAGE_CACHE_SHIFT;
2795
2796         page = grab_cache_page(inode->i_mapping, index);
2797         if (!page)
2798                 return NULL;
2799
2800         if (!PageUptodate(page)) {
2801                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2802                                                  0))
2803                         return NULL;
2804                 lock_page(page);
2805                 if (!PageUptodate(page)) {
2806                         unlock_page(page);
2807                         page_cache_release(page);
2808                         return NULL;
2809                 }
2810         }
2811         unlock_page(page);
2812
2813         return page;
2814 }
2815
2816 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2817 {
2818         /* do any pending delalloc/csum calc on src, one way or
2819            another, and lock file content */
2820         while (1) {
2821                 struct btrfs_ordered_extent *ordered;
2822                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2823                 ordered = btrfs_lookup_first_ordered_extent(inode,
2824                                                             off + len - 1);
2825                 if ((!ordered ||
2826                      ordered->file_offset + ordered->len <= off ||
2827                      ordered->file_offset >= off + len) &&
2828                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2829                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2830                         if (ordered)
2831                                 btrfs_put_ordered_extent(ordered);
2832                         break;
2833                 }
2834                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2835                 if (ordered)
2836                         btrfs_put_ordered_extent(ordered);
2837                 btrfs_wait_ordered_range(inode, off, len);
2838         }
2839 }
2840
2841 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2842                                 struct inode *inode2, u64 loff2, u64 len)
2843 {
2844         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2845         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2846
2847         mutex_unlock(&inode1->i_mutex);
2848         mutex_unlock(&inode2->i_mutex);
2849 }
2850
2851 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2852                               struct inode *inode2, u64 loff2, u64 len)
2853 {
2854         if (inode1 < inode2) {
2855                 swap(inode1, inode2);
2856                 swap(loff1, loff2);
2857         }
2858
2859         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2860         lock_extent_range(inode1, loff1, len);
2861         if (inode1 != inode2) {
2862                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2863                 lock_extent_range(inode2, loff2, len);
2864         }
2865 }
2866
2867 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2868                           u64 dst_loff, u64 len)
2869 {
2870         int ret = 0;
2871         struct page *src_page, *dst_page;
2872         unsigned int cmp_len = PAGE_CACHE_SIZE;
2873         void *addr, *dst_addr;
2874
2875         while (len) {
2876                 if (len < PAGE_CACHE_SIZE)
2877                         cmp_len = len;
2878
2879                 src_page = extent_same_get_page(src, loff);
2880                 if (!src_page)
2881                         return -EINVAL;
2882                 dst_page = extent_same_get_page(dst, dst_loff);
2883                 if (!dst_page) {
2884                         page_cache_release(src_page);
2885                         return -EINVAL;
2886                 }
2887                 addr = kmap_atomic(src_page);
2888                 dst_addr = kmap_atomic(dst_page);
2889
2890                 flush_dcache_page(src_page);
2891                 flush_dcache_page(dst_page);
2892
2893                 if (memcmp(addr, dst_addr, cmp_len))
2894                         ret = BTRFS_SAME_DATA_DIFFERS;
2895
2896                 kunmap_atomic(addr);
2897                 kunmap_atomic(dst_addr);
2898                 page_cache_release(src_page);
2899                 page_cache_release(dst_page);
2900
2901                 if (ret)
2902                         break;
2903
2904                 loff += cmp_len;
2905                 dst_loff += cmp_len;
2906                 len -= cmp_len;
2907         }
2908
2909         return ret;
2910 }
2911
2912 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2913 {
2914         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2915
2916         if (off + len > inode->i_size || off + len < off)
2917                 return -EINVAL;
2918         /* Check that we are block aligned - btrfs_clone() requires this */
2919         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2920                 return -EINVAL;
2921
2922         return 0;
2923 }
2924
2925 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2926                              struct inode *dst, u64 dst_loff)
2927 {
2928         int ret;
2929
2930         /*
2931          * btrfs_clone() can't handle extents in the same file
2932          * yet. Once that works, we can drop this check and replace it
2933          * with a check for the same inode, but overlapping extents.
2934          */
2935         if (src == dst)
2936                 return -EINVAL;
2937
2938         btrfs_double_lock(src, loff, dst, dst_loff, len);
2939
2940         ret = extent_same_check_offsets(src, loff, len);
2941         if (ret)
2942                 goto out_unlock;
2943
2944         ret = extent_same_check_offsets(dst, dst_loff, len);
2945         if (ret)
2946                 goto out_unlock;
2947
2948         /* don't make the dst file partly checksummed */
2949         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2950             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2951                 ret = -EINVAL;
2952                 goto out_unlock;
2953         }
2954
2955         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2956         if (ret == 0)
2957                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2958
2959 out_unlock:
2960         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2961
2962         return ret;
2963 }
2964
2965 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2966
2967 static long btrfs_ioctl_file_extent_same(struct file *file,
2968                         struct btrfs_ioctl_same_args __user *argp)
2969 {
2970         struct btrfs_ioctl_same_args *same;
2971         struct btrfs_ioctl_same_extent_info *info;
2972         struct inode *src = file_inode(file);
2973         u64 off;
2974         u64 len;
2975         int i;
2976         int ret;
2977         unsigned long size;
2978         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2979         bool is_admin = capable(CAP_SYS_ADMIN);
2980         u16 count;
2981
2982         if (!(file->f_mode & FMODE_READ))
2983                 return -EINVAL;
2984
2985         ret = mnt_want_write_file(file);
2986         if (ret)
2987                 return ret;
2988
2989         if (get_user(count, &argp->dest_count)) {
2990                 ret = -EFAULT;
2991                 goto out;
2992         }
2993
2994         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2995
2996         same = memdup_user(argp, size);
2997
2998         if (IS_ERR(same)) {
2999                 ret = PTR_ERR(same);
3000                 goto out;
3001         }
3002
3003         off = same->logical_offset;
3004         len = same->length;
3005
3006         /*
3007          * Limit the total length we will dedupe for each operation.
3008          * This is intended to bound the total time spent in this
3009          * ioctl to something sane.
3010          */
3011         if (len > BTRFS_MAX_DEDUPE_LEN)
3012                 len = BTRFS_MAX_DEDUPE_LEN;
3013
3014         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3015                 /*
3016                  * Btrfs does not support blocksize < page_size. As a
3017                  * result, btrfs_cmp_data() won't correctly handle
3018                  * this situation without an update.
3019                  */
3020                 ret = -EINVAL;
3021                 goto out;
3022         }
3023
3024         ret = -EISDIR;
3025         if (S_ISDIR(src->i_mode))
3026                 goto out;
3027
3028         ret = -EACCES;
3029         if (!S_ISREG(src->i_mode))
3030                 goto out;
3031
3032         /* pre-format output fields to sane values */
3033         for (i = 0; i < count; i++) {
3034                 same->info[i].bytes_deduped = 0ULL;
3035                 same->info[i].status = 0;
3036         }
3037
3038         for (i = 0, info = same->info; i < count; i++, info++) {
3039                 struct inode *dst;
3040                 struct fd dst_file = fdget(info->fd);
3041                 if (!dst_file.file) {
3042                         info->status = -EBADF;
3043                         continue;
3044                 }
3045                 dst = file_inode(dst_file.file);
3046
3047                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3048                         info->status = -EINVAL;
3049                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3050                         info->status = -EXDEV;
3051                 } else if (S_ISDIR(dst->i_mode)) {
3052                         info->status = -EISDIR;
3053                 } else if (!S_ISREG(dst->i_mode)) {
3054                         info->status = -EACCES;
3055                 } else {
3056                         info->status = btrfs_extent_same(src, off, len, dst,
3057                                                         info->logical_offset);
3058                         if (info->status == 0)
3059                                 info->bytes_deduped += len;
3060                 }
3061                 fdput(dst_file);
3062         }
3063
3064         ret = copy_to_user(argp, same, size);
3065         if (ret)
3066                 ret = -EFAULT;
3067
3068 out:
3069         mnt_drop_write_file(file);
3070         return ret;
3071 }
3072
3073 /* Helper to check and see if this root currently has a ref on the given disk
3074  * bytenr.  If it does then we need to update the quota for this root.  This
3075  * doesn't do anything if quotas aren't enabled.
3076  */
3077 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3078                      u64 disko)
3079 {
3080         struct seq_list tree_mod_seq_elem = {};
3081         struct ulist *roots;
3082         struct ulist_iterator uiter;
3083         struct ulist_node *root_node = NULL;
3084         int ret;
3085
3086         if (!root->fs_info->quota_enabled)
3087                 return 1;
3088
3089         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3090         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3091                                    tree_mod_seq_elem.seq, &roots);
3092         if (ret < 0)
3093                 goto out;
3094         ret = 0;
3095         ULIST_ITER_INIT(&uiter);
3096         while ((root_node = ulist_next(roots, &uiter))) {
3097                 if (root_node->val == root->objectid) {
3098                         ret = 1;
3099                         break;
3100                 }
3101         }
3102         ulist_free(roots);
3103 out:
3104         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3105         return ret;
3106 }
3107
3108 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3109                                      struct inode *inode,
3110                                      u64 endoff,
3111                                      const u64 destoff,
3112                                      const u64 olen)
3113 {
3114         struct btrfs_root *root = BTRFS_I(inode)->root;
3115         int ret;
3116
3117         inode_inc_iversion(inode);
3118         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3119         /*
3120          * We round up to the block size at eof when determining which
3121          * extents to clone above, but shouldn't round up the file size.
3122          */
3123         if (endoff > destoff + olen)
3124                 endoff = destoff + olen;
3125         if (endoff > inode->i_size)
3126                 btrfs_i_size_write(inode, endoff);
3127
3128         ret = btrfs_update_inode(trans, root, inode);
3129         if (ret) {
3130                 btrfs_abort_transaction(trans, root, ret);
3131                 btrfs_end_transaction(trans, root);
3132                 goto out;
3133         }
3134         ret = btrfs_end_transaction(trans, root);
3135 out:
3136         return ret;
3137 }
3138
3139 static void clone_update_extent_map(struct inode *inode,
3140                                     const struct btrfs_trans_handle *trans,
3141                                     const struct btrfs_path *path,
3142                                     struct btrfs_file_extent_item *fi,
3143                                     const u64 hole_offset,
3144                                     const u64 hole_len)
3145 {
3146         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3147         struct extent_map *em;
3148         int ret;
3149
3150         em = alloc_extent_map();
3151         if (!em) {
3152                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3153                         &BTRFS_I(inode)->runtime_flags);
3154                 return;
3155         }
3156
3157         if (fi) {
3158                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3159                 em->generation = -1;
3160                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3161                     BTRFS_FILE_EXTENT_INLINE)
3162                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3163                                 &BTRFS_I(inode)->runtime_flags);
3164         } else {
3165                 em->start = hole_offset;
3166                 em->len = hole_len;
3167                 em->ram_bytes = em->len;
3168                 em->orig_start = hole_offset;
3169                 em->block_start = EXTENT_MAP_HOLE;
3170                 em->block_len = 0;
3171                 em->orig_block_len = 0;
3172                 em->compress_type = BTRFS_COMPRESS_NONE;
3173                 em->generation = trans->transid;
3174         }
3175
3176         while (1) {
3177                 write_lock(&em_tree->lock);
3178                 ret = add_extent_mapping(em_tree, em, 1);
3179                 write_unlock(&em_tree->lock);
3180                 if (ret != -EEXIST) {
3181                         free_extent_map(em);
3182                         break;
3183                 }
3184                 btrfs_drop_extent_cache(inode, em->start,
3185                                         em->start + em->len - 1, 0);
3186         }
3187
3188         if (unlikely(ret))
3189                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3190                         &BTRFS_I(inode)->runtime_flags);
3191 }
3192
3193 /**
3194  * btrfs_clone() - clone a range from inode file to another
3195  *
3196  * @src: Inode to clone from
3197  * @inode: Inode to clone to
3198  * @off: Offset within source to start clone from
3199  * @olen: Original length, passed by user, of range to clone
3200  * @olen_aligned: Block-aligned value of olen, extent_same uses
3201  *               identical values here
3202  * @destoff: Offset within @inode to start clone
3203  */
3204 static int btrfs_clone(struct inode *src, struct inode *inode,
3205                        const u64 off, const u64 olen, const u64 olen_aligned,
3206                        const u64 destoff)
3207 {
3208         struct btrfs_root *root = BTRFS_I(inode)->root;
3209         struct btrfs_path *path = NULL;
3210         struct extent_buffer *leaf;
3211         struct btrfs_trans_handle *trans;
3212         char *buf = NULL;
3213         struct btrfs_key key;
3214         u32 nritems;
3215         int slot;
3216         int ret;
3217         int no_quota;
3218         const u64 len = olen_aligned;
3219         u64 last_disko = 0;
3220         u64 last_dest_end = destoff;
3221
3222         ret = -ENOMEM;
3223         buf = vmalloc(btrfs_level_size(root, 0));
3224         if (!buf)
3225                 return ret;
3226
3227         path = btrfs_alloc_path();
3228         if (!path) {
3229                 vfree(buf);
3230                 return ret;
3231         }
3232
3233         path->reada = 2;
3234         /* clone data */
3235         key.objectid = btrfs_ino(src);
3236         key.type = BTRFS_EXTENT_DATA_KEY;
3237         key.offset = off;
3238
3239         while (1) {
3240                 /*
3241                  * note the key will change type as we walk through the
3242                  * tree.
3243                  */
3244                 path->leave_spinning = 1;
3245                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3246                                 0, 0);
3247                 if (ret < 0)
3248                         goto out;
3249                 /*
3250                  * First search, if no extent item that starts at offset off was
3251                  * found but the previous item is an extent item, it's possible
3252                  * it might overlap our target range, therefore process it.
3253                  */
3254                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3255                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3256                                               path->slots[0] - 1);
3257                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3258                                 path->slots[0]--;
3259                 }
3260
3261                 nritems = btrfs_header_nritems(path->nodes[0]);
3262 process_slot:
3263                 no_quota = 1;
3264                 if (path->slots[0] >= nritems) {
3265                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3266                         if (ret < 0)
3267                                 goto out;
3268                         if (ret > 0)
3269                                 break;
3270                         nritems = btrfs_header_nritems(path->nodes[0]);
3271                 }
3272                 leaf = path->nodes[0];
3273                 slot = path->slots[0];
3274
3275                 btrfs_item_key_to_cpu(leaf, &key, slot);
3276                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
3277                     key.objectid != btrfs_ino(src))
3278                         break;
3279
3280                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
3281                         struct btrfs_file_extent_item *extent;
3282                         int type;
3283                         u32 size;
3284                         struct btrfs_key new_key;
3285                         u64 disko = 0, diskl = 0;
3286                         u64 datao = 0, datal = 0;
3287                         u8 comp;
3288                         u64 drop_start;
3289
3290                         extent = btrfs_item_ptr(leaf, slot,
3291                                                 struct btrfs_file_extent_item);
3292                         comp = btrfs_file_extent_compression(leaf, extent);
3293                         type = btrfs_file_extent_type(leaf, extent);
3294                         if (type == BTRFS_FILE_EXTENT_REG ||
3295                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3296                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3297                                                                       extent);
3298                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3299                                                                  extent);
3300                                 datao = btrfs_file_extent_offset(leaf, extent);
3301                                 datal = btrfs_file_extent_num_bytes(leaf,
3302                                                                     extent);
3303                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3304                                 /* take upper bound, may be compressed */
3305                                 datal = btrfs_file_extent_ram_bytes(leaf,
3306                                                                     extent);
3307                         }
3308
3309                         /*
3310                          * The first search might have left us at an extent
3311                          * item that ends before our target range's start, can
3312                          * happen if we have holes and NO_HOLES feature enabled.
3313                          */
3314                         if (key.offset + datal <= off) {
3315                                 path->slots[0]++;
3316                                 goto process_slot;
3317                         } else if (key.offset >= off + len) {
3318                                 break;
3319                         }
3320
3321                         size = btrfs_item_size_nr(leaf, slot);
3322                         read_extent_buffer(leaf, buf,
3323                                            btrfs_item_ptr_offset(leaf, slot),
3324                                            size);
3325
3326                         btrfs_release_path(path);
3327                         path->leave_spinning = 0;
3328
3329                         memcpy(&new_key, &key, sizeof(new_key));
3330                         new_key.objectid = btrfs_ino(inode);
3331                         if (off <= key.offset)
3332                                 new_key.offset = key.offset + destoff - off;
3333                         else
3334                                 new_key.offset = destoff;
3335
3336                         /*
3337                          * Deal with a hole that doesn't have an extent item
3338                          * that represents it (NO_HOLES feature enabled).
3339                          * This hole is either in the middle of the cloning
3340                          * range or at the beginning (fully overlaps it or
3341                          * partially overlaps it).
3342                          */
3343                         if (new_key.offset != last_dest_end)
3344                                 drop_start = last_dest_end;
3345                         else
3346                                 drop_start = new_key.offset;
3347
3348                         /*
3349                          * 1 - adjusting old extent (we may have to split it)
3350                          * 1 - add new extent
3351                          * 1 - inode update
3352                          */
3353                         trans = btrfs_start_transaction(root, 3);
3354                         if (IS_ERR(trans)) {
3355                                 ret = PTR_ERR(trans);
3356                                 goto out;
3357                         }
3358
3359                         if (type == BTRFS_FILE_EXTENT_REG ||
3360                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3361                                 /*
3362                                  *    a  | --- range to clone ---|  b
3363                                  * | ------------- extent ------------- |
3364                                  */
3365
3366                                 /* subtract range b */
3367                                 if (key.offset + datal > off + len)
3368                                         datal = off + len - key.offset;
3369
3370                                 /* subtract range a */
3371                                 if (off > key.offset) {
3372                                         datao += off - key.offset;
3373                                         datal -= off - key.offset;
3374                                 }
3375
3376                                 ret = btrfs_drop_extents(trans, root, inode,
3377                                                          drop_start,
3378                                                          new_key.offset + datal,
3379                                                          1);
3380                                 if (ret) {
3381                                         if (ret != -EOPNOTSUPP)
3382                                                 btrfs_abort_transaction(trans,
3383                                                                 root, ret);
3384                                         btrfs_end_transaction(trans, root);
3385                                         goto out;
3386                                 }
3387
3388                                 ret = btrfs_insert_empty_item(trans, root, path,
3389                                                               &new_key, size);
3390                                 if (ret) {
3391                                         btrfs_abort_transaction(trans, root,
3392                                                                 ret);
3393                                         btrfs_end_transaction(trans, root);
3394                                         goto out;
3395                                 }
3396
3397                                 leaf = path->nodes[0];
3398                                 slot = path->slots[0];
3399                                 write_extent_buffer(leaf, buf,
3400                                             btrfs_item_ptr_offset(leaf, slot),
3401                                             size);
3402
3403                                 extent = btrfs_item_ptr(leaf, slot,
3404                                                 struct btrfs_file_extent_item);
3405
3406                                 /* disko == 0 means it's a hole */
3407                                 if (!disko)
3408                                         datao = 0;
3409
3410                                 btrfs_set_file_extent_offset(leaf, extent,
3411                                                              datao);
3412                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3413                                                                 datal);
3414
3415                                 /*
3416                                  * We need to look up the roots that point at
3417                                  * this bytenr and see if the new root does.  If
3418                                  * it does not we need to make sure we update
3419                                  * quotas appropriately.
3420                                  */
3421                                 if (disko && root != BTRFS_I(src)->root &&
3422                                     disko != last_disko) {
3423                                         no_quota = check_ref(trans, root,
3424                                                              disko);
3425                                         if (no_quota < 0) {
3426                                                 btrfs_abort_transaction(trans,
3427                                                                         root,
3428                                                                         ret);
3429                                                 btrfs_end_transaction(trans,
3430                                                                       root);
3431                                                 ret = no_quota;
3432                                                 goto out;
3433                                         }
3434                                 }
3435
3436                                 if (disko) {
3437                                         inode_add_bytes(inode, datal);
3438                                         ret = btrfs_inc_extent_ref(trans, root,
3439                                                         disko, diskl, 0,
3440                                                         root->root_key.objectid,
3441                                                         btrfs_ino(inode),
3442                                                         new_key.offset - datao,
3443                                                         no_quota);
3444                                         if (ret) {
3445                                                 btrfs_abort_transaction(trans,
3446                                                                         root,
3447                                                                         ret);
3448                                                 btrfs_end_transaction(trans,
3449                                                                       root);
3450                                                 goto out;
3451
3452                                         }
3453                                 }
3454                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3455                                 u64 skip = 0;
3456                                 u64 trim = 0;
3457                                 u64 aligned_end = 0;
3458
3459                                 if (off > key.offset) {
3460                                         skip = off - key.offset;
3461                                         new_key.offset += skip;
3462                                 }
3463
3464                                 if (key.offset + datal > off + len)
3465                                         trim = key.offset + datal - (off + len);
3466
3467                                 if (comp && (skip || trim)) {
3468                                         ret = -EINVAL;
3469                                         btrfs_end_transaction(trans, root);
3470                                         goto out;
3471                                 }
3472                                 size -= skip + trim;
3473                                 datal -= skip + trim;
3474
3475                                 aligned_end = ALIGN(new_key.offset + datal,
3476                                                     root->sectorsize);
3477                                 ret = btrfs_drop_extents(trans, root, inode,
3478                                                          drop_start,
3479                                                          aligned_end,
3480                                                          1);
3481                                 if (ret) {
3482                                         if (ret != -EOPNOTSUPP)
3483                                                 btrfs_abort_transaction(trans,
3484                                                         root, ret);
3485                                         btrfs_end_transaction(trans, root);
3486                                         goto out;
3487                                 }
3488
3489                                 ret = btrfs_insert_empty_item(trans, root, path,
3490                                                               &new_key, size);
3491                                 if (ret) {
3492                                         btrfs_abort_transaction(trans, root,
3493                                                                 ret);
3494                                         btrfs_end_transaction(trans, root);
3495                                         goto out;
3496                                 }
3497
3498                                 if (skip) {
3499                                         u32 start =
3500                                           btrfs_file_extent_calc_inline_size(0);
3501                                         memmove(buf+start, buf+start+skip,
3502                                                 datal);
3503                                 }
3504
3505                                 leaf = path->nodes[0];
3506                                 slot = path->slots[0];
3507                                 write_extent_buffer(leaf, buf,
3508                                             btrfs_item_ptr_offset(leaf, slot),
3509                                             size);
3510                                 inode_add_bytes(inode, datal);
3511                                 extent = btrfs_item_ptr(leaf, slot,
3512                                                 struct btrfs_file_extent_item);
3513                         }
3514
3515                         /* If we have an implicit hole (NO_HOLES feature). */
3516                         if (drop_start < new_key.offset)
3517                                 clone_update_extent_map(inode, trans,
3518                                                 path, NULL, drop_start,
3519                                                 new_key.offset - drop_start);
3520
3521                         clone_update_extent_map(inode, trans, path,
3522                                                 extent, 0, 0);
3523
3524                         btrfs_mark_buffer_dirty(leaf);
3525                         btrfs_release_path(path);
3526
3527                         last_dest_end = new_key.offset + datal;
3528                         ret = clone_finish_inode_update(trans, inode,
3529                                                         last_dest_end,
3530                                                         destoff, olen);
3531                         if (ret)
3532                                 goto out;
3533                         if (new_key.offset + datal >= destoff + len)
3534                                 break;
3535                 }
3536                 btrfs_release_path(path);
3537                 key.offset++;
3538         }
3539         ret = 0;
3540
3541         if (last_dest_end < destoff + len) {
3542                 /*
3543                  * We have an implicit hole (NO_HOLES feature is enabled) that
3544                  * fully or partially overlaps our cloning range at its end.
3545                  */
3546                 btrfs_release_path(path);
3547
3548                 /*
3549                  * 1 - remove extent(s)
3550                  * 1 - inode update
3551                  */
3552                 trans = btrfs_start_transaction(root, 2);
3553                 if (IS_ERR(trans)) {
3554                         ret = PTR_ERR(trans);
3555                         goto out;
3556                 }
3557                 ret = btrfs_drop_extents(trans, root, inode,
3558                                          last_dest_end, destoff + len, 1);
3559                 if (ret) {
3560                         if (ret != -EOPNOTSUPP)
3561                                 btrfs_abort_transaction(trans, root, ret);
3562                         btrfs_end_transaction(trans, root);
3563                         goto out;
3564                 }
3565                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3566                                                 destoff, olen);
3567                 if (ret)
3568                         goto out;
3569                 clone_update_extent_map(inode, trans, path, NULL, last_dest_end,
3570                                         destoff + len - last_dest_end);
3571         }
3572
3573 out:
3574         btrfs_free_path(path);
3575         vfree(buf);
3576         return ret;
3577 }
3578
3579 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3580                                        u64 off, u64 olen, u64 destoff)
3581 {
3582         struct inode *inode = file_inode(file);
3583         struct btrfs_root *root = BTRFS_I(inode)->root;
3584         struct fd src_file;
3585         struct inode *src;
3586         int ret;
3587         u64 len = olen;
3588         u64 bs = root->fs_info->sb->s_blocksize;
3589         int same_inode = 0;
3590
3591         /*
3592          * TODO:
3593          * - split compressed inline extents.  annoying: we need to
3594          *   decompress into destination's address_space (the file offset
3595          *   may change, so source mapping won't do), then recompress (or
3596          *   otherwise reinsert) a subrange.
3597          *
3598          * - split destination inode's inline extents.  The inline extents can
3599          *   be either compressed or non-compressed.
3600          */
3601
3602         /* the destination must be opened for writing */
3603         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3604                 return -EINVAL;
3605
3606         if (btrfs_root_readonly(root))
3607                 return -EROFS;
3608
3609         ret = mnt_want_write_file(file);
3610         if (ret)
3611                 return ret;
3612
3613         src_file = fdget(srcfd);
3614         if (!src_file.file) {
3615                 ret = -EBADF;
3616                 goto out_drop_write;
3617         }
3618
3619         ret = -EXDEV;
3620         if (src_file.file->f_path.mnt != file->f_path.mnt)
3621                 goto out_fput;
3622
3623         src = file_inode(src_file.file);
3624
3625         ret = -EINVAL;
3626         if (src == inode)
3627                 same_inode = 1;
3628
3629         /* the src must be open for reading */
3630         if (!(src_file.file->f_mode & FMODE_READ))
3631                 goto out_fput;
3632
3633         /* don't make the dst file partly checksummed */
3634         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3635             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3636                 goto out_fput;
3637
3638         ret = -EISDIR;
3639         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3640                 goto out_fput;
3641
3642         ret = -EXDEV;
3643         if (src->i_sb != inode->i_sb)
3644                 goto out_fput;
3645
3646         if (!same_inode) {
3647                 if (inode < src) {
3648                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3649                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3650                 } else {
3651                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3652                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3653                 }
3654         } else {
3655                 mutex_lock(&src->i_mutex);
3656         }
3657
3658         /* determine range to clone */
3659         ret = -EINVAL;
3660         if (off + len > src->i_size || off + len < off)
3661                 goto out_unlock;
3662         if (len == 0)
3663                 olen = len = src->i_size - off;
3664         /* if we extend to eof, continue to block boundary */
3665         if (off + len == src->i_size)
3666                 len = ALIGN(src->i_size, bs) - off;
3667
3668         /* verify the end result is block aligned */
3669         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3670             !IS_ALIGNED(destoff, bs))
3671                 goto out_unlock;
3672
3673         /* verify if ranges are overlapped within the same file */
3674         if (same_inode) {
3675                 if (destoff + len > off && destoff < off + len)
3676                         goto out_unlock;
3677         }
3678
3679         if (destoff > inode->i_size) {
3680                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3681                 if (ret)
3682                         goto out_unlock;
3683         }
3684
3685         /*
3686          * Lock the target range too. Right after we replace the file extent
3687          * items in the fs tree (which now point to the cloned data), we might
3688          * have a worker replace them with extent items relative to a write
3689          * operation that was issued before this clone operation (i.e. confront
3690          * with inode.c:btrfs_finish_ordered_io).
3691          */
3692         if (same_inode) {
3693                 u64 lock_start = min_t(u64, off, destoff);
3694                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3695
3696                 lock_extent_range(src, lock_start, lock_len);
3697         } else {
3698                 lock_extent_range(src, off, len);
3699                 lock_extent_range(inode, destoff, len);
3700         }
3701
3702         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3703
3704         if (same_inode) {
3705                 u64 lock_start = min_t(u64, off, destoff);
3706                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3707
3708                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3709         } else {
3710                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3711                 unlock_extent(&BTRFS_I(inode)->io_tree, destoff,
3712                               destoff + len - 1);
3713         }
3714         /*
3715          * Truncate page cache pages so that future reads will see the cloned
3716          * data immediately and not the previous data.
3717          */
3718         truncate_inode_pages_range(&inode->i_data, destoff,
3719                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3720 out_unlock:
3721         if (!same_inode) {
3722                 if (inode < src) {
3723                         mutex_unlock(&src->i_mutex);
3724                         mutex_unlock(&inode->i_mutex);
3725                 } else {
3726                         mutex_unlock(&inode->i_mutex);
3727                         mutex_unlock(&src->i_mutex);
3728                 }
3729         } else {
3730                 mutex_unlock(&src->i_mutex);
3731         }
3732 out_fput:
3733         fdput(src_file);
3734 out_drop_write:
3735         mnt_drop_write_file(file);
3736         return ret;
3737 }
3738
3739 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3740 {
3741         struct btrfs_ioctl_clone_range_args args;
3742
3743         if (copy_from_user(&args, argp, sizeof(args)))
3744                 return -EFAULT;
3745         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3746                                  args.src_length, args.dest_offset);
3747 }
3748
3749 /*
3750  * there are many ways the trans_start and trans_end ioctls can lead
3751  * to deadlocks.  They should only be used by applications that
3752  * basically own the machine, and have a very in depth understanding
3753  * of all the possible deadlocks and enospc problems.
3754  */
3755 static long btrfs_ioctl_trans_start(struct file *file)
3756 {
3757         struct inode *inode = file_inode(file);
3758         struct btrfs_root *root = BTRFS_I(inode)->root;
3759         struct btrfs_trans_handle *trans;
3760         int ret;
3761
3762         ret = -EPERM;
3763         if (!capable(CAP_SYS_ADMIN))
3764                 goto out;
3765
3766         ret = -EINPROGRESS;
3767         if (file->private_data)
3768                 goto out;
3769
3770         ret = -EROFS;
3771         if (btrfs_root_readonly(root))
3772                 goto out;
3773
3774         ret = mnt_want_write_file(file);
3775         if (ret)
3776                 goto out;
3777
3778         atomic_inc(&root->fs_info->open_ioctl_trans);
3779
3780         ret = -ENOMEM;
3781         trans = btrfs_start_ioctl_transaction(root);
3782         if (IS_ERR(trans))
3783                 goto out_drop;
3784
3785         file->private_data = trans;
3786         return 0;
3787
3788 out_drop:
3789         atomic_dec(&root->fs_info->open_ioctl_trans);
3790         mnt_drop_write_file(file);
3791 out:
3792         return ret;
3793 }
3794
3795 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3796 {
3797         struct inode *inode = file_inode(file);
3798         struct btrfs_root *root = BTRFS_I(inode)->root;
3799         struct btrfs_root *new_root;
3800         struct btrfs_dir_item *di;
3801         struct btrfs_trans_handle *trans;
3802         struct btrfs_path *path;
3803         struct btrfs_key location;
3804         struct btrfs_disk_key disk_key;
3805         u64 objectid = 0;
3806         u64 dir_id;
3807         int ret;
3808
3809         if (!capable(CAP_SYS_ADMIN))
3810                 return -EPERM;
3811
3812         ret = mnt_want_write_file(file);
3813         if (ret)
3814                 return ret;
3815
3816         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3817                 ret = -EFAULT;
3818                 goto out;
3819         }
3820
3821         if (!objectid)
3822                 objectid = BTRFS_FS_TREE_OBJECTID;
3823
3824         location.objectid = objectid;
3825         location.type = BTRFS_ROOT_ITEM_KEY;
3826         location.offset = (u64)-1;
3827
3828         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3829         if (IS_ERR(new_root)) {
3830                 ret = PTR_ERR(new_root);
3831                 goto out;
3832         }
3833
3834         path = btrfs_alloc_path();
3835         if (!path) {
3836                 ret = -ENOMEM;
3837                 goto out;
3838         }
3839         path->leave_spinning = 1;
3840
3841         trans = btrfs_start_transaction(root, 1);
3842         if (IS_ERR(trans)) {
3843                 btrfs_free_path(path);
3844                 ret = PTR_ERR(trans);
3845                 goto out;
3846         }
3847
3848         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3849         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3850                                    dir_id, "default", 7, 1);
3851         if (IS_ERR_OR_NULL(di)) {
3852                 btrfs_free_path(path);
3853                 btrfs_end_transaction(trans, root);
3854                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3855                            "item, this isn't going to work");
3856                 ret = -ENOENT;
3857                 goto out;
3858         }
3859
3860         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3861         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3862         btrfs_mark_buffer_dirty(path->nodes[0]);
3863         btrfs_free_path(path);
3864
3865         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3866         btrfs_end_transaction(trans, root);
3867 out:
3868         mnt_drop_write_file(file);
3869         return ret;
3870 }
3871
3872 void btrfs_get_block_group_info(struct list_head *groups_list,
3873                                 struct btrfs_ioctl_space_info *space)
3874 {
3875         struct btrfs_block_group_cache *block_group;
3876
3877         space->total_bytes = 0;
3878         space->used_bytes = 0;
3879         space->flags = 0;
3880         list_for_each_entry(block_group, groups_list, list) {
3881                 space->flags = block_group->flags;
3882                 space->total_bytes += block_group->key.offset;
3883                 space->used_bytes +=
3884                         btrfs_block_group_used(&block_group->item);
3885         }
3886 }
3887
3888 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3889 {
3890         struct btrfs_ioctl_space_args space_args;
3891         struct btrfs_ioctl_space_info space;
3892         struct btrfs_ioctl_space_info *dest;
3893         struct btrfs_ioctl_space_info *dest_orig;
3894         struct btrfs_ioctl_space_info __user *user_dest;
3895         struct btrfs_space_info *info;
3896         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3897                        BTRFS_BLOCK_GROUP_SYSTEM,
3898                        BTRFS_BLOCK_GROUP_METADATA,
3899                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3900         int num_types = 4;
3901         int alloc_size;
3902         int ret = 0;
3903         u64 slot_count = 0;
3904         int i, c;
3905
3906         if (copy_from_user(&space_args,
3907                            (struct btrfs_ioctl_space_args __user *)arg,
3908                            sizeof(space_args)))
3909                 return -EFAULT;
3910
3911         for (i = 0; i < num_types; i++) {
3912                 struct btrfs_space_info *tmp;
3913
3914                 info = NULL;
3915                 rcu_read_lock();
3916                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3917                                         list) {
3918                         if (tmp->flags == types[i]) {
3919                                 info = tmp;
3920                                 break;
3921                         }
3922                 }
3923                 rcu_read_unlock();
3924
3925                 if (!info)
3926                         continue;
3927
3928                 down_read(&info->groups_sem);
3929                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3930                         if (!list_empty(&info->block_groups[c]))
3931                                 slot_count++;
3932                 }
3933                 up_read(&info->groups_sem);
3934         }
3935
3936         /*
3937          * Global block reserve, exported as a space_info
3938          */
3939         slot_count++;
3940
3941         /* space_slots == 0 means they are asking for a count */
3942         if (space_args.space_slots == 0) {
3943                 space_args.total_spaces = slot_count;
3944                 goto out;
3945         }
3946
3947         slot_count = min_t(u64, space_args.space_slots, slot_count);
3948
3949         alloc_size = sizeof(*dest) * slot_count;
3950
3951         /* we generally have at most 6 or so space infos, one for each raid
3952          * level.  So, a whole page should be more than enough for everyone
3953          */
3954         if (alloc_size > PAGE_CACHE_SIZE)
3955                 return -ENOMEM;
3956
3957         space_args.total_spaces = 0;
3958         dest = kmalloc(alloc_size, GFP_NOFS);
3959         if (!dest)
3960                 return -ENOMEM;
3961         dest_orig = dest;
3962
3963         /* now we have a buffer to copy into */
3964         for (i = 0; i < num_types; i++) {
3965                 struct btrfs_space_info *tmp;
3966
3967                 if (!slot_count)
3968                         break;
3969
3970                 info = NULL;
3971                 rcu_read_lock();
3972                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3973                                         list) {
3974                         if (tmp->flags == types[i]) {
3975                                 info = tmp;
3976                                 break;
3977                         }
3978                 }
3979                 rcu_read_unlock();
3980
3981                 if (!info)
3982                         continue;
3983                 down_read(&info->groups_sem);
3984                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3985                         if (!list_empty(&info->block_groups[c])) {
3986                                 btrfs_get_block_group_info(
3987                                         &info->block_groups[c], &space);
3988                                 memcpy(dest, &space, sizeof(space));
3989                                 dest++;
3990                                 space_args.total_spaces++;
3991                                 slot_count--;
3992                         }
3993                         if (!slot_count)
3994                                 break;
3995                 }
3996                 up_read(&info->groups_sem);
3997         }
3998
3999         /*
4000          * Add global block reserve
4001          */
4002         if (slot_count) {
4003                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4004
4005                 spin_lock(&block_rsv->lock);
4006                 space.total_bytes = block_rsv->size;
4007                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4008                 spin_unlock(&block_rsv->lock);
4009                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4010                 memcpy(dest, &space, sizeof(space));
4011                 space_args.total_spaces++;
4012         }
4013
4014         user_dest = (struct btrfs_ioctl_space_info __user *)
4015                 (arg + sizeof(struct btrfs_ioctl_space_args));
4016
4017         if (copy_to_user(user_dest, dest_orig, alloc_size))
4018                 ret = -EFAULT;
4019
4020         kfree(dest_orig);
4021 out:
4022         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4023                 ret = -EFAULT;
4024
4025         return ret;
4026 }
4027
4028 /*
4029  * there are many ways the trans_start and trans_end ioctls can lead
4030  * to deadlocks.  They should only be used by applications that
4031  * basically own the machine, and have a very in depth understanding
4032  * of all the possible deadlocks and enospc problems.
4033  */
4034 long btrfs_ioctl_trans_end(struct file *file)
4035 {
4036         struct inode *inode = file_inode(file);
4037         struct btrfs_root *root = BTRFS_I(inode)->root;
4038         struct btrfs_trans_handle *trans;
4039
4040         trans = file->private_data;
4041         if (!trans)
4042                 return -EINVAL;
4043         file->private_data = NULL;
4044
4045         btrfs_end_transaction(trans, root);
4046
4047         atomic_dec(&root->fs_info->open_ioctl_trans);
4048
4049         mnt_drop_write_file(file);
4050         return 0;
4051 }
4052
4053 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4054                                             void __user *argp)
4055 {
4056         struct btrfs_trans_handle *trans;
4057         u64 transid;
4058         int ret;
4059
4060         trans = btrfs_attach_transaction_barrier(root);
4061         if (IS_ERR(trans)) {
4062                 if (PTR_ERR(trans) != -ENOENT)
4063                         return PTR_ERR(trans);
4064
4065                 /* No running transaction, don't bother */
4066                 transid = root->fs_info->last_trans_committed;
4067                 goto out;
4068         }
4069         transid = trans->transid;
4070         ret = btrfs_commit_transaction_async(trans, root, 0);
4071         if (ret) {
4072                 btrfs_end_transaction(trans, root);
4073                 return ret;
4074         }
4075 out:
4076         if (argp)
4077                 if (copy_to_user(argp, &transid, sizeof(transid)))
4078                         return -EFAULT;
4079         return 0;
4080 }
4081
4082 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4083                                            void __user *argp)
4084 {
4085         u64 transid;
4086
4087         if (argp) {
4088                 if (copy_from_user(&transid, argp, sizeof(transid)))
4089                         return -EFAULT;
4090         } else {
4091                 transid = 0;  /* current trans */
4092         }
4093         return btrfs_wait_for_commit(root, transid);
4094 }
4095
4096 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4097 {
4098         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4099         struct btrfs_ioctl_scrub_args *sa;
4100         int ret;
4101
4102         if (!capable(CAP_SYS_ADMIN))
4103                 return -EPERM;
4104
4105         sa = memdup_user(arg, sizeof(*sa));
4106         if (IS_ERR(sa))
4107                 return PTR_ERR(sa);
4108
4109         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4110                 ret = mnt_want_write_file(file);
4111                 if (ret)
4112                         goto out;
4113         }
4114
4115         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4116                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4117                               0);
4118
4119         if (copy_to_user(arg, sa, sizeof(*sa)))
4120                 ret = -EFAULT;
4121
4122         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4123                 mnt_drop_write_file(file);
4124 out:
4125         kfree(sa);
4126         return ret;
4127 }
4128
4129 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4130 {
4131         if (!capable(CAP_SYS_ADMIN))
4132                 return -EPERM;
4133
4134         return btrfs_scrub_cancel(root->fs_info);
4135 }
4136
4137 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4138                                        void __user *arg)
4139 {
4140         struct btrfs_ioctl_scrub_args *sa;
4141         int ret;
4142
4143         if (!capable(CAP_SYS_ADMIN))
4144                 return -EPERM;
4145
4146         sa = memdup_user(arg, sizeof(*sa));
4147         if (IS_ERR(sa))
4148                 return PTR_ERR(sa);
4149
4150         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4151
4152         if (copy_to_user(arg, sa, sizeof(*sa)))
4153                 ret = -EFAULT;
4154
4155         kfree(sa);
4156         return ret;
4157 }
4158
4159 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4160                                       void __user *arg)
4161 {
4162         struct btrfs_ioctl_get_dev_stats *sa;
4163         int ret;
4164
4165         sa = memdup_user(arg, sizeof(*sa));
4166         if (IS_ERR(sa))
4167                 return PTR_ERR(sa);
4168
4169         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4170                 kfree(sa);
4171                 return -EPERM;
4172         }
4173
4174         ret = btrfs_get_dev_stats(root, sa);
4175
4176         if (copy_to_user(arg, sa, sizeof(*sa)))
4177                 ret = -EFAULT;
4178
4179         kfree(sa);
4180         return ret;
4181 }
4182
4183 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4184 {
4185         struct btrfs_ioctl_dev_replace_args *p;
4186         int ret;
4187
4188         if (!capable(CAP_SYS_ADMIN))
4189                 return -EPERM;
4190
4191         p = memdup_user(arg, sizeof(*p));
4192         if (IS_ERR(p))
4193                 return PTR_ERR(p);
4194
4195         switch (p->cmd) {
4196         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4197                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4198                         ret = -EROFS;
4199                         goto out;
4200                 }
4201                 if (atomic_xchg(
4202                         &root->fs_info->mutually_exclusive_operation_running,
4203                         1)) {
4204                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4205                 } else {
4206                         ret = btrfs_dev_replace_start(root, p);
4207                         atomic_set(
4208                          &root->fs_info->mutually_exclusive_operation_running,
4209                          0);
4210                 }
4211                 break;
4212         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4213                 btrfs_dev_replace_status(root->fs_info, p);
4214                 ret = 0;
4215                 break;
4216         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4217                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4218                 break;
4219         default:
4220                 ret = -EINVAL;
4221                 break;
4222         }
4223
4224         if (copy_to_user(arg, p, sizeof(*p)))
4225                 ret = -EFAULT;
4226 out:
4227         kfree(p);
4228         return ret;
4229 }
4230
4231 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4232 {
4233         int ret = 0;
4234         int i;
4235         u64 rel_ptr;
4236         int size;
4237         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4238         struct inode_fs_paths *ipath = NULL;
4239         struct btrfs_path *path;
4240
4241         if (!capable(CAP_DAC_READ_SEARCH))
4242                 return -EPERM;
4243
4244         path = btrfs_alloc_path();
4245         if (!path) {
4246                 ret = -ENOMEM;
4247                 goto out;
4248         }
4249
4250         ipa = memdup_user(arg, sizeof(*ipa));
4251         if (IS_ERR(ipa)) {
4252                 ret = PTR_ERR(ipa);
4253                 ipa = NULL;
4254                 goto out;
4255         }
4256
4257         size = min_t(u32, ipa->size, 4096);
4258         ipath = init_ipath(size, root, path);
4259         if (IS_ERR(ipath)) {
4260                 ret = PTR_ERR(ipath);
4261                 ipath = NULL;
4262                 goto out;
4263         }
4264
4265         ret = paths_from_inode(ipa->inum, ipath);
4266         if (ret < 0)
4267                 goto out;
4268
4269         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4270                 rel_ptr = ipath->fspath->val[i] -
4271                           (u64)(unsigned long)ipath->fspath->val;
4272                 ipath->fspath->val[i] = rel_ptr;
4273         }
4274
4275         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4276                            (void *)(unsigned long)ipath->fspath, size);
4277         if (ret) {
4278                 ret = -EFAULT;
4279                 goto out;
4280         }
4281
4282 out:
4283         btrfs_free_path(path);
4284         free_ipath(ipath);
4285         kfree(ipa);
4286
4287         return ret;
4288 }
4289
4290 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4291 {
4292         struct btrfs_data_container *inodes = ctx;
4293         const size_t c = 3 * sizeof(u64);
4294
4295         if (inodes->bytes_left >= c) {
4296                 inodes->bytes_left -= c;
4297                 inodes->val[inodes->elem_cnt] = inum;
4298                 inodes->val[inodes->elem_cnt + 1] = offset;
4299                 inodes->val[inodes->elem_cnt + 2] = root;
4300                 inodes->elem_cnt += 3;
4301         } else {
4302                 inodes->bytes_missing += c - inodes->bytes_left;
4303                 inodes->bytes_left = 0;
4304                 inodes->elem_missed += 3;
4305         }
4306
4307         return 0;
4308 }
4309
4310 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4311                                         void __user *arg)
4312 {
4313         int ret = 0;
4314         int size;
4315         struct btrfs_ioctl_logical_ino_args *loi;
4316         struct btrfs_data_container *inodes = NULL;
4317         struct btrfs_path *path = NULL;
4318
4319         if (!capable(CAP_SYS_ADMIN))
4320                 return -EPERM;
4321
4322         loi = memdup_user(arg, sizeof(*loi));
4323         if (IS_ERR(loi)) {
4324                 ret = PTR_ERR(loi);
4325                 loi = NULL;
4326                 goto out;
4327         }
4328
4329         path = btrfs_alloc_path();
4330         if (!path) {
4331                 ret = -ENOMEM;
4332                 goto out;
4333         }
4334
4335         size = min_t(u32, loi->size, 64 * 1024);
4336         inodes = init_data_container(size);
4337         if (IS_ERR(inodes)) {
4338                 ret = PTR_ERR(inodes);
4339                 inodes = NULL;
4340                 goto out;
4341         }
4342
4343         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4344                                           build_ino_list, inodes);
4345         if (ret == -EINVAL)
4346                 ret = -ENOENT;
4347         if (ret < 0)
4348                 goto out;
4349
4350         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4351                            (void *)(unsigned long)inodes, size);
4352         if (ret)
4353                 ret = -EFAULT;
4354
4355 out:
4356         btrfs_free_path(path);
4357         vfree(inodes);
4358         kfree(loi);
4359
4360         return ret;
4361 }
4362
4363 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4364                                struct btrfs_ioctl_balance_args *bargs)
4365 {
4366         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4367
4368         bargs->flags = bctl->flags;
4369
4370         if (atomic_read(&fs_info->balance_running))
4371                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4372         if (atomic_read(&fs_info->balance_pause_req))
4373                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4374         if (atomic_read(&fs_info->balance_cancel_req))
4375                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4376
4377         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4378         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4379         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4380
4381         if (lock) {
4382                 spin_lock(&fs_info->balance_lock);
4383                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4384                 spin_unlock(&fs_info->balance_lock);
4385         } else {
4386                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4387         }
4388 }
4389
4390 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4391 {
4392         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4393         struct btrfs_fs_info *fs_info = root->fs_info;
4394         struct btrfs_ioctl_balance_args *bargs;
4395         struct btrfs_balance_control *bctl;
4396         bool need_unlock; /* for mut. excl. ops lock */
4397         int ret;
4398
4399         if (!capable(CAP_SYS_ADMIN))
4400                 return -EPERM;
4401
4402         ret = mnt_want_write_file(file);
4403         if (ret)
4404                 return ret;
4405
4406 again:
4407         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4408                 mutex_lock(&fs_info->volume_mutex);
4409                 mutex_lock(&fs_info->balance_mutex);
4410                 need_unlock = true;
4411                 goto locked;
4412         }
4413
4414         /*
4415          * mut. excl. ops lock is locked.  Three possibilites:
4416          *   (1) some other op is running
4417          *   (2) balance is running
4418          *   (3) balance is paused -- special case (think resume)
4419          */
4420         mutex_lock(&fs_info->balance_mutex);
4421         if (fs_info->balance_ctl) {
4422                 /* this is either (2) or (3) */
4423                 if (!atomic_read(&fs_info->balance_running)) {
4424                         mutex_unlock(&fs_info->balance_mutex);
4425                         if (!mutex_trylock(&fs_info->volume_mutex))
4426                                 goto again;
4427                         mutex_lock(&fs_info->balance_mutex);
4428
4429                         if (fs_info->balance_ctl &&
4430                             !atomic_read(&fs_info->balance_running)) {
4431                                 /* this is (3) */
4432                                 need_unlock = false;
4433                                 goto locked;
4434                         }
4435
4436                         mutex_unlock(&fs_info->balance_mutex);
4437                         mutex_unlock(&fs_info->volume_mutex);
4438                         goto again;
4439                 } else {
4440                         /* this is (2) */
4441                         mutex_unlock(&fs_info->balance_mutex);
4442                         ret = -EINPROGRESS;
4443                         goto out;
4444                 }
4445         } else {
4446                 /* this is (1) */
4447                 mutex_unlock(&fs_info->balance_mutex);
4448                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4449                 goto out;
4450         }
4451
4452 locked:
4453         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4454
4455         if (arg) {
4456                 bargs = memdup_user(arg, sizeof(*bargs));
4457                 if (IS_ERR(bargs)) {
4458                         ret = PTR_ERR(bargs);
4459                         goto out_unlock;
4460                 }
4461
4462                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4463                         if (!fs_info->balance_ctl) {
4464                                 ret = -ENOTCONN;
4465                                 goto out_bargs;
4466                         }
4467
4468                         bctl = fs_info->balance_ctl;
4469                         spin_lock(&fs_info->balance_lock);
4470                         bctl->flags |= BTRFS_BALANCE_RESUME;
4471                         spin_unlock(&fs_info->balance_lock);
4472
4473                         goto do_balance;
4474                 }
4475         } else {
4476                 bargs = NULL;
4477         }
4478
4479         if (fs_info->balance_ctl) {
4480                 ret = -EINPROGRESS;
4481                 goto out_bargs;
4482         }
4483
4484         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4485         if (!bctl) {
4486                 ret = -ENOMEM;
4487                 goto out_bargs;
4488         }
4489
4490         bctl->fs_info = fs_info;
4491         if (arg) {
4492                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4493                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4494                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4495
4496                 bctl->flags = bargs->flags;
4497         } else {
4498                 /* balance everything - no filters */
4499                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4500         }
4501
4502 do_balance:
4503         /*
4504          * Ownership of bctl and mutually_exclusive_operation_running
4505          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4506          * or, if restriper was paused all the way until unmount, in
4507          * free_fs_info.  mutually_exclusive_operation_running is
4508          * cleared in __cancel_balance.
4509          */
4510         need_unlock = false;
4511
4512         ret = btrfs_balance(bctl, bargs);
4513
4514         if (arg) {
4515                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4516                         ret = -EFAULT;
4517         }
4518
4519 out_bargs:
4520         kfree(bargs);
4521 out_unlock:
4522         mutex_unlock(&fs_info->balance_mutex);
4523         mutex_unlock(&fs_info->volume_mutex);
4524         if (need_unlock)
4525                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4526 out:
4527         mnt_drop_write_file(file);
4528         return ret;
4529 }
4530
4531 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4532 {
4533         if (!capable(CAP_SYS_ADMIN))
4534                 return -EPERM;
4535
4536         switch (cmd) {
4537         case BTRFS_BALANCE_CTL_PAUSE:
4538                 return btrfs_pause_balance(root->fs_info);
4539         case BTRFS_BALANCE_CTL_CANCEL:
4540                 return btrfs_cancel_balance(root->fs_info);
4541         }
4542
4543         return -EINVAL;
4544 }
4545
4546 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4547                                          void __user *arg)
4548 {
4549         struct btrfs_fs_info *fs_info = root->fs_info;
4550         struct btrfs_ioctl_balance_args *bargs;
4551         int ret = 0;
4552
4553         if (!capable(CAP_SYS_ADMIN))
4554                 return -EPERM;
4555
4556         mutex_lock(&fs_info->balance_mutex);
4557         if (!fs_info->balance_ctl) {
4558                 ret = -ENOTCONN;
4559                 goto out;
4560         }
4561
4562         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4563         if (!bargs) {
4564                 ret = -ENOMEM;
4565                 goto out;
4566         }
4567
4568         update_ioctl_balance_args(fs_info, 1, bargs);
4569
4570         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4571                 ret = -EFAULT;
4572
4573         kfree(bargs);
4574 out:
4575         mutex_unlock(&fs_info->balance_mutex);
4576         return ret;
4577 }
4578
4579 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4580 {
4581         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4582         struct btrfs_ioctl_quota_ctl_args *sa;
4583         struct btrfs_trans_handle *trans = NULL;
4584         int ret;
4585         int err;
4586
4587         if (!capable(CAP_SYS_ADMIN))
4588                 return -EPERM;
4589
4590         ret = mnt_want_write_file(file);
4591         if (ret)
4592                 return ret;
4593
4594         sa = memdup_user(arg, sizeof(*sa));
4595         if (IS_ERR(sa)) {
4596                 ret = PTR_ERR(sa);
4597                 goto drop_write;
4598         }
4599
4600         down_write(&root->fs_info->subvol_sem);
4601         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4602         if (IS_ERR(trans)) {
4603                 ret = PTR_ERR(trans);
4604                 goto out;
4605         }
4606
4607         switch (sa->cmd) {
4608         case BTRFS_QUOTA_CTL_ENABLE:
4609                 ret = btrfs_quota_enable(trans, root->fs_info);
4610                 break;
4611         case BTRFS_QUOTA_CTL_DISABLE:
4612                 ret = btrfs_quota_disable(trans, root->fs_info);
4613                 break;
4614         default:
4615                 ret = -EINVAL;
4616                 break;
4617         }
4618
4619         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4620         if (err && !ret)
4621                 ret = err;
4622 out:
4623         kfree(sa);
4624         up_write(&root->fs_info->subvol_sem);
4625 drop_write:
4626         mnt_drop_write_file(file);
4627         return ret;
4628 }
4629
4630 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4631 {
4632         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4633         struct btrfs_ioctl_qgroup_assign_args *sa;
4634         struct btrfs_trans_handle *trans;
4635         int ret;
4636         int err;
4637
4638         if (!capable(CAP_SYS_ADMIN))
4639                 return -EPERM;
4640
4641         ret = mnt_want_write_file(file);
4642         if (ret)
4643                 return ret;
4644
4645         sa = memdup_user(arg, sizeof(*sa));
4646         if (IS_ERR(sa)) {
4647                 ret = PTR_ERR(sa);
4648                 goto drop_write;
4649         }
4650
4651         trans = btrfs_join_transaction(root);
4652         if (IS_ERR(trans)) {
4653                 ret = PTR_ERR(trans);
4654                 goto out;
4655         }
4656
4657         /* FIXME: check if the IDs really exist */
4658         if (sa->assign) {
4659                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4660                                                 sa->src, sa->dst);
4661         } else {
4662                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4663                                                 sa->src, sa->dst);
4664         }
4665
4666         err = btrfs_end_transaction(trans, root);
4667         if (err && !ret)
4668                 ret = err;
4669
4670 out:
4671         kfree(sa);
4672 drop_write:
4673         mnt_drop_write_file(file);
4674         return ret;
4675 }
4676
4677 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4678 {
4679         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4680         struct btrfs_ioctl_qgroup_create_args *sa;
4681         struct btrfs_trans_handle *trans;
4682         int ret;
4683         int err;
4684
4685         if (!capable(CAP_SYS_ADMIN))
4686                 return -EPERM;
4687
4688         ret = mnt_want_write_file(file);
4689         if (ret)
4690                 return ret;
4691
4692         sa = memdup_user(arg, sizeof(*sa));
4693         if (IS_ERR(sa)) {
4694                 ret = PTR_ERR(sa);
4695                 goto drop_write;
4696         }
4697
4698         if (!sa->qgroupid) {
4699                 ret = -EINVAL;
4700                 goto out;
4701         }
4702
4703         trans = btrfs_join_transaction(root);
4704         if (IS_ERR(trans)) {
4705                 ret = PTR_ERR(trans);
4706                 goto out;
4707         }
4708
4709         /* FIXME: check if the IDs really exist */
4710         if (sa->create) {
4711                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4712                                           NULL);
4713         } else {
4714                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4715         }
4716
4717         err = btrfs_end_transaction(trans, root);
4718         if (err && !ret)
4719                 ret = err;
4720
4721 out:
4722         kfree(sa);
4723 drop_write:
4724         mnt_drop_write_file(file);
4725         return ret;
4726 }
4727
4728 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4729 {
4730         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4731         struct btrfs_ioctl_qgroup_limit_args *sa;
4732         struct btrfs_trans_handle *trans;
4733         int ret;
4734         int err;
4735         u64 qgroupid;
4736
4737         if (!capable(CAP_SYS_ADMIN))
4738                 return -EPERM;
4739
4740         ret = mnt_want_write_file(file);
4741         if (ret)
4742                 return ret;
4743
4744         sa = memdup_user(arg, sizeof(*sa));
4745         if (IS_ERR(sa)) {
4746                 ret = PTR_ERR(sa);
4747                 goto drop_write;
4748         }
4749
4750         trans = btrfs_join_transaction(root);
4751         if (IS_ERR(trans)) {
4752                 ret = PTR_ERR(trans);
4753                 goto out;
4754         }
4755
4756         qgroupid = sa->qgroupid;
4757         if (!qgroupid) {
4758                 /* take the current subvol as qgroup */
4759                 qgroupid = root->root_key.objectid;
4760         }
4761
4762         /* FIXME: check if the IDs really exist */
4763         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4764
4765         err = btrfs_end_transaction(trans, root);
4766         if (err && !ret)
4767                 ret = err;
4768
4769 out:
4770         kfree(sa);
4771 drop_write:
4772         mnt_drop_write_file(file);
4773         return ret;
4774 }
4775
4776 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4777 {
4778         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4779         struct btrfs_ioctl_quota_rescan_args *qsa;
4780         int ret;
4781
4782         if (!capable(CAP_SYS_ADMIN))
4783                 return -EPERM;
4784
4785         ret = mnt_want_write_file(file);
4786         if (ret)
4787                 return ret;
4788
4789         qsa = memdup_user(arg, sizeof(*qsa));
4790         if (IS_ERR(qsa)) {
4791                 ret = PTR_ERR(qsa);
4792                 goto drop_write;
4793         }
4794
4795         if (qsa->flags) {
4796                 ret = -EINVAL;
4797                 goto out;
4798         }
4799
4800         ret = btrfs_qgroup_rescan(root->fs_info);
4801
4802 out:
4803         kfree(qsa);
4804 drop_write:
4805         mnt_drop_write_file(file);
4806         return ret;
4807 }
4808
4809 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4810 {
4811         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4812         struct btrfs_ioctl_quota_rescan_args *qsa;
4813         int ret = 0;
4814
4815         if (!capable(CAP_SYS_ADMIN))
4816                 return -EPERM;
4817
4818         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4819         if (!qsa)
4820                 return -ENOMEM;
4821
4822         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4823                 qsa->flags = 1;
4824                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4825         }
4826
4827         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4828                 ret = -EFAULT;
4829
4830         kfree(qsa);
4831         return ret;
4832 }
4833
4834 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4835 {
4836         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4837
4838         if (!capable(CAP_SYS_ADMIN))
4839                 return -EPERM;
4840
4841         return btrfs_qgroup_wait_for_completion(root->fs_info);
4842 }
4843
4844 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4845                                             struct btrfs_ioctl_received_subvol_args *sa)
4846 {
4847         struct inode *inode = file_inode(file);
4848         struct btrfs_root *root = BTRFS_I(inode)->root;
4849         struct btrfs_root_item *root_item = &root->root_item;
4850         struct btrfs_trans_handle *trans;
4851         struct timespec ct = CURRENT_TIME;
4852         int ret = 0;
4853         int received_uuid_changed;
4854
4855         if (!inode_owner_or_capable(inode))
4856                 return -EPERM;
4857
4858         ret = mnt_want_write_file(file);
4859         if (ret < 0)
4860                 return ret;
4861
4862         down_write(&root->fs_info->subvol_sem);
4863
4864         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4865                 ret = -EINVAL;
4866                 goto out;
4867         }
4868
4869         if (btrfs_root_readonly(root)) {
4870                 ret = -EROFS;
4871                 goto out;
4872         }
4873
4874         /*
4875          * 1 - root item
4876          * 2 - uuid items (received uuid + subvol uuid)
4877          */
4878         trans = btrfs_start_transaction(root, 3);
4879         if (IS_ERR(trans)) {
4880                 ret = PTR_ERR(trans);
4881                 trans = NULL;
4882                 goto out;
4883         }
4884
4885         sa->rtransid = trans->transid;
4886         sa->rtime.sec = ct.tv_sec;
4887         sa->rtime.nsec = ct.tv_nsec;
4888
4889         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4890                                        BTRFS_UUID_SIZE);
4891         if (received_uuid_changed &&
4892             !btrfs_is_empty_uuid(root_item->received_uuid))
4893                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4894                                     root_item->received_uuid,
4895                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4896                                     root->root_key.objectid);
4897         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4898         btrfs_set_root_stransid(root_item, sa->stransid);
4899         btrfs_set_root_rtransid(root_item, sa->rtransid);
4900         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4901         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4902         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4903         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4904
4905         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4906                                 &root->root_key, &root->root_item);
4907         if (ret < 0) {
4908                 btrfs_end_transaction(trans, root);
4909                 goto out;
4910         }
4911         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4912                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4913                                           sa->uuid,
4914                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4915                                           root->root_key.objectid);
4916                 if (ret < 0 && ret != -EEXIST) {
4917                         btrfs_abort_transaction(trans, root, ret);
4918                         goto out;
4919                 }
4920         }
4921         ret = btrfs_commit_transaction(trans, root);
4922         if (ret < 0) {
4923                 btrfs_abort_transaction(trans, root, ret);
4924                 goto out;
4925         }
4926
4927 out:
4928         up_write(&root->fs_info->subvol_sem);
4929         mnt_drop_write_file(file);
4930         return ret;
4931 }
4932
4933 #ifdef CONFIG_64BIT
4934 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4935                                                 void __user *arg)
4936 {
4937         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4938         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4939         int ret = 0;
4940
4941         args32 = memdup_user(arg, sizeof(*args32));
4942         if (IS_ERR(args32)) {
4943                 ret = PTR_ERR(args32);
4944                 args32 = NULL;
4945                 goto out;
4946         }
4947
4948         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4949         if (!args64) {
4950                 ret = -ENOMEM;
4951                 goto out;
4952         }
4953
4954         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4955         args64->stransid = args32->stransid;
4956         args64->rtransid = args32->rtransid;
4957         args64->stime.sec = args32->stime.sec;
4958         args64->stime.nsec = args32->stime.nsec;
4959         args64->rtime.sec = args32->rtime.sec;
4960         args64->rtime.nsec = args32->rtime.nsec;
4961         args64->flags = args32->flags;
4962
4963         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4964         if (ret)
4965                 goto out;
4966
4967         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4968         args32->stransid = args64->stransid;
4969         args32->rtransid = args64->rtransid;
4970         args32->stime.sec = args64->stime.sec;
4971         args32->stime.nsec = args64->stime.nsec;
4972         args32->rtime.sec = args64->rtime.sec;
4973         args32->rtime.nsec = args64->rtime.nsec;
4974         args32->flags = args64->flags;
4975
4976         ret = copy_to_user(arg, args32, sizeof(*args32));
4977         if (ret)
4978                 ret = -EFAULT;
4979
4980 out:
4981         kfree(args32);
4982         kfree(args64);
4983         return ret;
4984 }
4985 #endif
4986
4987 static long btrfs_ioctl_set_received_subvol(struct file *file,
4988                                             void __user *arg)
4989 {
4990         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4991         int ret = 0;
4992
4993         sa = memdup_user(arg, sizeof(*sa));
4994         if (IS_ERR(sa)) {
4995                 ret = PTR_ERR(sa);
4996                 sa = NULL;
4997                 goto out;
4998         }
4999
5000         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5001
5002         if (ret)
5003                 goto out;
5004
5005         ret = copy_to_user(arg, sa, sizeof(*sa));
5006         if (ret)
5007                 ret = -EFAULT;
5008
5009 out:
5010         kfree(sa);
5011         return ret;
5012 }
5013
5014 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5015 {
5016         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5017         size_t len;
5018         int ret;
5019         char label[BTRFS_LABEL_SIZE];
5020
5021         spin_lock(&root->fs_info->super_lock);
5022         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5023         spin_unlock(&root->fs_info->super_lock);
5024
5025         len = strnlen(label, BTRFS_LABEL_SIZE);
5026
5027         if (len == BTRFS_LABEL_SIZE) {
5028                 btrfs_warn(root->fs_info,
5029                         "label is too long, return the first %zu bytes", --len);
5030         }
5031
5032         ret = copy_to_user(arg, label, len);
5033
5034         return ret ? -EFAULT : 0;
5035 }
5036
5037 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5038 {
5039         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5040         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5041         struct btrfs_trans_handle *trans;
5042         char label[BTRFS_LABEL_SIZE];
5043         int ret;
5044
5045         if (!capable(CAP_SYS_ADMIN))
5046                 return -EPERM;
5047
5048         if (copy_from_user(label, arg, sizeof(label)))
5049                 return -EFAULT;
5050
5051         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5052                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5053                        BTRFS_LABEL_SIZE - 1);
5054                 return -EINVAL;
5055         }
5056
5057         ret = mnt_want_write_file(file);
5058         if (ret)
5059                 return ret;
5060
5061         trans = btrfs_start_transaction(root, 0);
5062         if (IS_ERR(trans)) {
5063                 ret = PTR_ERR(trans);
5064                 goto out_unlock;
5065         }
5066
5067         spin_lock(&root->fs_info->super_lock);
5068         strcpy(super_block->label, label);
5069         spin_unlock(&root->fs_info->super_lock);
5070         ret = btrfs_commit_transaction(trans, root);
5071
5072 out_unlock:
5073         mnt_drop_write_file(file);
5074         return ret;
5075 }
5076
5077 #define INIT_FEATURE_FLAGS(suffix) \
5078         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5079           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5080           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5081
5082 static int btrfs_ioctl_get_supported_features(struct file *file,
5083                                               void __user *arg)
5084 {
5085         static struct btrfs_ioctl_feature_flags features[3] = {
5086                 INIT_FEATURE_FLAGS(SUPP),
5087                 INIT_FEATURE_FLAGS(SAFE_SET),
5088                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5089         };
5090
5091         if (copy_to_user(arg, &features, sizeof(features)))
5092                 return -EFAULT;
5093
5094         return 0;
5095 }
5096
5097 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5098 {
5099         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5100         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5101         struct btrfs_ioctl_feature_flags features;
5102
5103         features.compat_flags = btrfs_super_compat_flags(super_block);
5104         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5105         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5106
5107         if (copy_to_user(arg, &features, sizeof(features)))
5108                 return -EFAULT;
5109
5110         return 0;
5111 }
5112
5113 static int check_feature_bits(struct btrfs_root *root,
5114                               enum btrfs_feature_set set,
5115                               u64 change_mask, u64 flags, u64 supported_flags,
5116                               u64 safe_set, u64 safe_clear)
5117 {
5118         const char *type = btrfs_feature_set_names[set];
5119         char *names;
5120         u64 disallowed, unsupported;
5121         u64 set_mask = flags & change_mask;
5122         u64 clear_mask = ~flags & change_mask;
5123
5124         unsupported = set_mask & ~supported_flags;
5125         if (unsupported) {
5126                 names = btrfs_printable_features(set, unsupported);
5127                 if (names) {
5128                         btrfs_warn(root->fs_info,
5129                            "this kernel does not support the %s feature bit%s",
5130                            names, strchr(names, ',') ? "s" : "");
5131                         kfree(names);
5132                 } else
5133                         btrfs_warn(root->fs_info,
5134                            "this kernel does not support %s bits 0x%llx",
5135                            type, unsupported);
5136                 return -EOPNOTSUPP;
5137         }
5138
5139         disallowed = set_mask & ~safe_set;
5140         if (disallowed) {
5141                 names = btrfs_printable_features(set, disallowed);
5142                 if (names) {
5143                         btrfs_warn(root->fs_info,
5144                            "can't set the %s feature bit%s while mounted",
5145                            names, strchr(names, ',') ? "s" : "");
5146                         kfree(names);
5147                 } else
5148                         btrfs_warn(root->fs_info,
5149                            "can't set %s bits 0x%llx while mounted",
5150                            type, disallowed);
5151                 return -EPERM;
5152         }
5153
5154         disallowed = clear_mask & ~safe_clear;
5155         if (disallowed) {
5156                 names = btrfs_printable_features(set, disallowed);
5157                 if (names) {
5158                         btrfs_warn(root->fs_info,
5159                            "can't clear the %s feature bit%s while mounted",
5160                            names, strchr(names, ',') ? "s" : "");
5161                         kfree(names);
5162                 } else
5163                         btrfs_warn(root->fs_info,
5164                            "can't clear %s bits 0x%llx while mounted",
5165                            type, disallowed);
5166                 return -EPERM;
5167         }
5168
5169         return 0;
5170 }
5171
5172 #define check_feature(root, change_mask, flags, mask_base)      \
5173 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5174                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5175                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5176                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5177
5178 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5179 {
5180         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5181         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5182         struct btrfs_ioctl_feature_flags flags[2];
5183         struct btrfs_trans_handle *trans;
5184         u64 newflags;
5185         int ret;
5186
5187         if (!capable(CAP_SYS_ADMIN))
5188                 return -EPERM;
5189
5190         if (copy_from_user(flags, arg, sizeof(flags)))
5191                 return -EFAULT;
5192
5193         /* Nothing to do */
5194         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5195             !flags[0].incompat_flags)
5196                 return 0;
5197
5198         ret = check_feature(root, flags[0].compat_flags,
5199                             flags[1].compat_flags, COMPAT);
5200         if (ret)
5201                 return ret;
5202
5203         ret = check_feature(root, flags[0].compat_ro_flags,
5204                             flags[1].compat_ro_flags, COMPAT_RO);
5205         if (ret)
5206                 return ret;
5207
5208         ret = check_feature(root, flags[0].incompat_flags,
5209                             flags[1].incompat_flags, INCOMPAT);
5210         if (ret)
5211                 return ret;
5212
5213         trans = btrfs_start_transaction(root, 0);
5214         if (IS_ERR(trans))
5215                 return PTR_ERR(trans);
5216
5217         spin_lock(&root->fs_info->super_lock);
5218         newflags = btrfs_super_compat_flags(super_block);
5219         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5220         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5221         btrfs_set_super_compat_flags(super_block, newflags);
5222
5223         newflags = btrfs_super_compat_ro_flags(super_block);
5224         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5225         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5226         btrfs_set_super_compat_ro_flags(super_block, newflags);
5227
5228         newflags = btrfs_super_incompat_flags(super_block);
5229         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5230         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5231         btrfs_set_super_incompat_flags(super_block, newflags);
5232         spin_unlock(&root->fs_info->super_lock);
5233
5234         return btrfs_commit_transaction(trans, root);
5235 }
5236
5237 long btrfs_ioctl(struct file *file, unsigned int
5238                 cmd, unsigned long arg)
5239 {
5240         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5241         void __user *argp = (void __user *)arg;
5242
5243         switch (cmd) {
5244         case FS_IOC_GETFLAGS:
5245                 return btrfs_ioctl_getflags(file, argp);
5246         case FS_IOC_SETFLAGS:
5247                 return btrfs_ioctl_setflags(file, argp);
5248         case FS_IOC_GETVERSION:
5249                 return btrfs_ioctl_getversion(file, argp);
5250         case FITRIM:
5251                 return btrfs_ioctl_fitrim(file, argp);
5252         case BTRFS_IOC_SNAP_CREATE:
5253                 return btrfs_ioctl_snap_create(file, argp, 0);
5254         case BTRFS_IOC_SNAP_CREATE_V2:
5255                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5256         case BTRFS_IOC_SUBVOL_CREATE:
5257                 return btrfs_ioctl_snap_create(file, argp, 1);
5258         case BTRFS_IOC_SUBVOL_CREATE_V2:
5259                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5260         case BTRFS_IOC_SNAP_DESTROY:
5261                 return btrfs_ioctl_snap_destroy(file, argp);
5262         case BTRFS_IOC_SUBVOL_GETFLAGS:
5263                 return btrfs_ioctl_subvol_getflags(file, argp);
5264         case BTRFS_IOC_SUBVOL_SETFLAGS:
5265                 return btrfs_ioctl_subvol_setflags(file, argp);
5266         case BTRFS_IOC_DEFAULT_SUBVOL:
5267                 return btrfs_ioctl_default_subvol(file, argp);
5268         case BTRFS_IOC_DEFRAG:
5269                 return btrfs_ioctl_defrag(file, NULL);
5270         case BTRFS_IOC_DEFRAG_RANGE:
5271                 return btrfs_ioctl_defrag(file, argp);
5272         case BTRFS_IOC_RESIZE:
5273                 return btrfs_ioctl_resize(file, argp);
5274         case BTRFS_IOC_ADD_DEV:
5275                 return btrfs_ioctl_add_dev(root, argp);
5276         case BTRFS_IOC_RM_DEV:
5277                 return btrfs_ioctl_rm_dev(file, argp);
5278         case BTRFS_IOC_FS_INFO:
5279                 return btrfs_ioctl_fs_info(root, argp);
5280         case BTRFS_IOC_DEV_INFO:
5281                 return btrfs_ioctl_dev_info(root, argp);
5282         case BTRFS_IOC_BALANCE:
5283                 return btrfs_ioctl_balance(file, NULL);
5284         case BTRFS_IOC_CLONE:
5285                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5286         case BTRFS_IOC_CLONE_RANGE:
5287                 return btrfs_ioctl_clone_range(file, argp);
5288         case BTRFS_IOC_TRANS_START:
5289                 return btrfs_ioctl_trans_start(file);
5290         case BTRFS_IOC_TRANS_END:
5291                 return btrfs_ioctl_trans_end(file);
5292         case BTRFS_IOC_TREE_SEARCH:
5293                 return btrfs_ioctl_tree_search(file, argp);
5294         case BTRFS_IOC_TREE_SEARCH_V2:
5295                 return btrfs_ioctl_tree_search_v2(file, argp);
5296         case BTRFS_IOC_INO_LOOKUP:
5297                 return btrfs_ioctl_ino_lookup(file, argp);
5298         case BTRFS_IOC_INO_PATHS:
5299                 return btrfs_ioctl_ino_to_path(root, argp);
5300         case BTRFS_IOC_LOGICAL_INO:
5301                 return btrfs_ioctl_logical_to_ino(root, argp);
5302         case BTRFS_IOC_SPACE_INFO:
5303                 return btrfs_ioctl_space_info(root, argp);
5304         case BTRFS_IOC_SYNC: {
5305                 int ret;
5306
5307                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5308                 if (ret)
5309                         return ret;
5310                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
5311                 return ret;
5312         }
5313         case BTRFS_IOC_START_SYNC:
5314                 return btrfs_ioctl_start_sync(root, argp);
5315         case BTRFS_IOC_WAIT_SYNC:
5316                 return btrfs_ioctl_wait_sync(root, argp);
5317         case BTRFS_IOC_SCRUB:
5318                 return btrfs_ioctl_scrub(file, argp);
5319         case BTRFS_IOC_SCRUB_CANCEL:
5320                 return btrfs_ioctl_scrub_cancel(root, argp);
5321         case BTRFS_IOC_SCRUB_PROGRESS:
5322                 return btrfs_ioctl_scrub_progress(root, argp);
5323         case BTRFS_IOC_BALANCE_V2:
5324                 return btrfs_ioctl_balance(file, argp);
5325         case BTRFS_IOC_BALANCE_CTL:
5326                 return btrfs_ioctl_balance_ctl(root, arg);
5327         case BTRFS_IOC_BALANCE_PROGRESS:
5328                 return btrfs_ioctl_balance_progress(root, argp);
5329         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5330                 return btrfs_ioctl_set_received_subvol(file, argp);
5331 #ifdef CONFIG_64BIT
5332         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5333                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5334 #endif
5335         case BTRFS_IOC_SEND:
5336                 return btrfs_ioctl_send(file, argp);
5337         case BTRFS_IOC_GET_DEV_STATS:
5338                 return btrfs_ioctl_get_dev_stats(root, argp);
5339         case BTRFS_IOC_QUOTA_CTL:
5340                 return btrfs_ioctl_quota_ctl(file, argp);
5341         case BTRFS_IOC_QGROUP_ASSIGN:
5342                 return btrfs_ioctl_qgroup_assign(file, argp);
5343         case BTRFS_IOC_QGROUP_CREATE:
5344                 return btrfs_ioctl_qgroup_create(file, argp);
5345         case BTRFS_IOC_QGROUP_LIMIT:
5346                 return btrfs_ioctl_qgroup_limit(file, argp);
5347         case BTRFS_IOC_QUOTA_RESCAN:
5348                 return btrfs_ioctl_quota_rescan(file, argp);
5349         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5350                 return btrfs_ioctl_quota_rescan_status(file, argp);
5351         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5352                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5353         case BTRFS_IOC_DEV_REPLACE:
5354                 return btrfs_ioctl_dev_replace(root, argp);
5355         case BTRFS_IOC_GET_FSLABEL:
5356                 return btrfs_ioctl_get_fslabel(file, argp);
5357         case BTRFS_IOC_SET_FSLABEL:
5358                 return btrfs_ioctl_set_fslabel(file, argp);
5359         case BTRFS_IOC_FILE_EXTENT_SAME:
5360                 return btrfs_ioctl_file_extent_same(file, argp);
5361         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5362                 return btrfs_ioctl_get_supported_features(file, argp);
5363         case BTRFS_IOC_GET_FEATURES:
5364                 return btrfs_ioctl_get_features(file, argp);
5365         case BTRFS_IOC_SET_FEATURES:
5366                 return btrfs_ioctl_set_features(file, argp);
5367         }
5368
5369         return -ENOTTY;
5370 }