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_inc();
 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 *buf,
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                         item_len = 0;
1995
1996                 if (sizeof(sh) + item_len + *sk_offset > buf_size) {
1997                         ret = 1;
1998                         goto out;
1999                 }
2000
2001                 sh.objectid = key->objectid;
2002                 sh.offset = key->offset;
2003                 sh.type = key->type;
2004                 sh.len = item_len;
2005                 sh.transid = found_transid;
2006
2007                 /* copy search result header */
2008                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
2009                 *sk_offset += sizeof(sh);
2010
2011                 if (item_len) {
2012                         char *p = buf + *sk_offset;
2013                         /* copy the item */
2014                         read_extent_buffer(leaf, p,
2015                                            item_off, item_len);
2016                         *sk_offset += item_len;
2017                 }
2018                 (*num_found)++;
2019
2020                 if (*num_found >= sk->nr_items) {
2021                         ret = 1;
2022                         goto out;
2023                 }
2024         }
2025 advance_key:
2026         ret = 0;
2027         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2028                 key->offset++;
2029         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2030                 key->offset = 0;
2031                 key->type++;
2032         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2033                 key->offset = 0;
2034                 key->type = 0;
2035                 key->objectid++;
2036         } else
2037                 ret = 1;
2038 out:
2039         return ret;
2040 }
2041
2042 static noinline int search_ioctl(struct inode *inode,
2043                                  struct btrfs_ioctl_search_key *sk,
2044                                  size_t buf_size,
2045                                  char *buf)
2046 {
2047         struct btrfs_root *root;
2048         struct btrfs_key key;
2049         struct btrfs_path *path;
2050         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2051         int ret;
2052         int num_found = 0;
2053         unsigned long sk_offset = 0;
2054
2055         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2056                 return -EOVERFLOW;
2057
2058         path = btrfs_alloc_path();
2059         if (!path)
2060                 return -ENOMEM;
2061
2062         if (sk->tree_id == 0) {
2063                 /* search the root of the inode that was passed */
2064                 root = BTRFS_I(inode)->root;
2065         } else {
2066                 key.objectid = sk->tree_id;
2067                 key.type = BTRFS_ROOT_ITEM_KEY;
2068                 key.offset = (u64)-1;
2069                 root = btrfs_read_fs_root_no_name(info, &key);
2070                 if (IS_ERR(root)) {
2071                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2072                                sk->tree_id);
2073                         btrfs_free_path(path);
2074                         return -ENOENT;
2075                 }
2076         }
2077
2078         key.objectid = sk->min_objectid;
2079         key.type = sk->min_type;
2080         key.offset = sk->min_offset;
2081
2082         path->keep_locks = 1;
2083
2084         while (1) {
2085                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2086                 if (ret != 0) {
2087                         if (ret > 0)
2088                                 ret = 0;
2089                         goto err;
2090                 }
2091                 ret = copy_to_sk(root, path, &key, sk, buf_size, buf,
2092                                  &sk_offset, &num_found);
2093                 btrfs_release_path(path);
2094                 if (ret)
2095                         break;
2096
2097         }
2098         ret = 0;
2099 err:
2100         sk->nr_items = num_found;
2101         btrfs_free_path(path);
2102         return ret;
2103 }
2104
2105 static noinline int btrfs_ioctl_tree_search(struct file *file,
2106                                            void __user *argp)
2107 {
2108          struct btrfs_ioctl_search_args *args;
2109          struct inode *inode;
2110          int ret;
2111
2112         if (!capable(CAP_SYS_ADMIN))
2113                 return -EPERM;
2114
2115         args = memdup_user(argp, sizeof(*args));
2116         if (IS_ERR(args))
2117                 return PTR_ERR(args);
2118
2119         inode = file_inode(file);
2120         ret = search_ioctl(inode, &args->key, sizeof(args->buf), args->buf);
2121         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2122                 ret = -EFAULT;
2123         kfree(args);
2124         return ret;
2125 }
2126
2127 /*
2128  * Search INODE_REFs to identify path name of 'dirid' directory
2129  * in a 'tree_id' tree. and sets path name to 'name'.
2130  */
2131 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2132                                 u64 tree_id, u64 dirid, char *name)
2133 {
2134         struct btrfs_root *root;
2135         struct btrfs_key key;
2136         char *ptr;
2137         int ret = -1;
2138         int slot;
2139         int len;
2140         int total_len = 0;
2141         struct btrfs_inode_ref *iref;
2142         struct extent_buffer *l;
2143         struct btrfs_path *path;
2144
2145         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2146                 name[0]='\0';
2147                 return 0;
2148         }
2149
2150         path = btrfs_alloc_path();
2151         if (!path)
2152                 return -ENOMEM;
2153
2154         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2155
2156         key.objectid = tree_id;
2157         key.type = BTRFS_ROOT_ITEM_KEY;
2158         key.offset = (u64)-1;
2159         root = btrfs_read_fs_root_no_name(info, &key);
2160         if (IS_ERR(root)) {
2161                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2162                 ret = -ENOENT;
2163                 goto out;
2164         }
2165
2166         key.objectid = dirid;
2167         key.type = BTRFS_INODE_REF_KEY;
2168         key.offset = (u64)-1;
2169
2170         while (1) {
2171                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2172                 if (ret < 0)
2173                         goto out;
2174                 else if (ret > 0) {
2175                         ret = btrfs_previous_item(root, path, dirid,
2176                                                   BTRFS_INODE_REF_KEY);
2177                         if (ret < 0)
2178                                 goto out;
2179                         else if (ret > 0) {
2180                                 ret = -ENOENT;
2181                                 goto out;
2182                         }
2183                 }
2184
2185                 l = path->nodes[0];
2186                 slot = path->slots[0];
2187                 btrfs_item_key_to_cpu(l, &key, slot);
2188
2189                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2190                 len = btrfs_inode_ref_name_len(l, iref);
2191                 ptr -= len + 1;
2192                 total_len += len + 1;
2193                 if (ptr < name) {
2194                         ret = -ENAMETOOLONG;
2195                         goto out;
2196                 }
2197
2198                 *(ptr + len) = '/';
2199                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2200
2201                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2202                         break;
2203
2204                 btrfs_release_path(path);
2205                 key.objectid = key.offset;
2206                 key.offset = (u64)-1;
2207                 dirid = key.objectid;
2208         }
2209         memmove(name, ptr, total_len);
2210         name[total_len] = '\0';
2211         ret = 0;
2212 out:
2213         btrfs_free_path(path);
2214         return ret;
2215 }
2216
2217 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2218                                            void __user *argp)
2219 {
2220          struct btrfs_ioctl_ino_lookup_args *args;
2221          struct inode *inode;
2222          int ret;
2223
2224         if (!capable(CAP_SYS_ADMIN))
2225                 return -EPERM;
2226
2227         args = memdup_user(argp, sizeof(*args));
2228         if (IS_ERR(args))
2229                 return PTR_ERR(args);
2230
2231         inode = file_inode(file);
2232
2233         if (args->treeid == 0)
2234                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2235
2236         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2237                                         args->treeid, args->objectid,
2238                                         args->name);
2239
2240         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2241                 ret = -EFAULT;
2242
2243         kfree(args);
2244         return ret;
2245 }
2246
2247 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2248                                              void __user *arg)
2249 {
2250         struct dentry *parent = file->f_path.dentry;
2251         struct dentry *dentry;
2252         struct inode *dir = parent->d_inode;
2253         struct inode *inode;
2254         struct btrfs_root *root = BTRFS_I(dir)->root;
2255         struct btrfs_root *dest = NULL;
2256         struct btrfs_ioctl_vol_args *vol_args;
2257         struct btrfs_trans_handle *trans;
2258         struct btrfs_block_rsv block_rsv;
2259         u64 root_flags;
2260         u64 qgroup_reserved;
2261         int namelen;
2262         int ret;
2263         int err = 0;
2264
2265         vol_args = memdup_user(arg, sizeof(*vol_args));
2266         if (IS_ERR(vol_args))
2267                 return PTR_ERR(vol_args);
2268
2269         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2270         namelen = strlen(vol_args->name);
2271         if (strchr(vol_args->name, '/') ||
2272             strncmp(vol_args->name, "..", namelen) == 0) {
2273                 err = -EINVAL;
2274                 goto out;
2275         }
2276
2277         err = mnt_want_write_file(file);
2278         if (err)
2279                 goto out;
2280
2281
2282         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2283         if (err == -EINTR)
2284                 goto out_drop_write;
2285         dentry = lookup_one_len(vol_args->name, parent, namelen);
2286         if (IS_ERR(dentry)) {
2287                 err = PTR_ERR(dentry);
2288                 goto out_unlock_dir;
2289         }
2290
2291         if (!dentry->d_inode) {
2292                 err = -ENOENT;
2293                 goto out_dput;
2294         }
2295
2296         inode = dentry->d_inode;
2297         dest = BTRFS_I(inode)->root;
2298         if (!capable(CAP_SYS_ADMIN)) {
2299                 /*
2300                  * Regular user.  Only allow this with a special mount
2301                  * option, when the user has write+exec access to the
2302                  * subvol root, and when rmdir(2) would have been
2303                  * allowed.
2304                  *
2305                  * Note that this is _not_ check that the subvol is
2306                  * empty or doesn't contain data that we wouldn't
2307                  * otherwise be able to delete.
2308                  *
2309                  * Users who want to delete empty subvols should try
2310                  * rmdir(2).
2311                  */
2312                 err = -EPERM;
2313                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2314                         goto out_dput;
2315
2316                 /*
2317                  * Do not allow deletion if the parent dir is the same
2318                  * as the dir to be deleted.  That means the ioctl
2319                  * must be called on the dentry referencing the root
2320                  * of the subvol, not a random directory contained
2321                  * within it.
2322                  */
2323                 err = -EINVAL;
2324                 if (root == dest)
2325                         goto out_dput;
2326
2327                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2328                 if (err)
2329                         goto out_dput;
2330         }
2331
2332         /* check if subvolume may be deleted by a user */
2333         err = btrfs_may_delete(dir, dentry, 1);
2334         if (err)
2335                 goto out_dput;
2336
2337         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2338                 err = -EINVAL;
2339                 goto out_dput;
2340         }
2341
2342         mutex_lock(&inode->i_mutex);
2343
2344         /*
2345          * Don't allow to delete a subvolume with send in progress. This is
2346          * inside the i_mutex so the error handling that has to drop the bit
2347          * again is not run concurrently.
2348          */
2349         spin_lock(&dest->root_item_lock);
2350         root_flags = btrfs_root_flags(&dest->root_item);
2351         if (dest->send_in_progress == 0) {
2352                 btrfs_set_root_flags(&dest->root_item,
2353                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2354                 spin_unlock(&dest->root_item_lock);
2355         } else {
2356                 spin_unlock(&dest->root_item_lock);
2357                 btrfs_warn(root->fs_info,
2358                         "Attempt to delete subvolume %llu during send",
2359                         dest->root_key.objectid);
2360                 err = -EPERM;
2361                 goto out_dput;
2362         }
2363
2364         err = d_invalidate(dentry);
2365         if (err)
2366                 goto out_unlock;
2367
2368         down_write(&root->fs_info->subvol_sem);
2369
2370         err = may_destroy_subvol(dest);
2371         if (err)
2372                 goto out_up_write;
2373
2374         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2375         /*
2376          * One for dir inode, two for dir entries, two for root
2377          * ref/backref.
2378          */
2379         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2380                                                5, &qgroup_reserved, true);
2381         if (err)
2382                 goto out_up_write;
2383
2384         trans = btrfs_start_transaction(root, 0);
2385         if (IS_ERR(trans)) {
2386                 err = PTR_ERR(trans);
2387                 goto out_release;
2388         }
2389         trans->block_rsv = &block_rsv;
2390         trans->bytes_reserved = block_rsv.size;
2391
2392         ret = btrfs_unlink_subvol(trans, root, dir,
2393                                 dest->root_key.objectid,
2394                                 dentry->d_name.name,
2395                                 dentry->d_name.len);
2396         if (ret) {
2397                 err = ret;
2398                 btrfs_abort_transaction(trans, root, ret);
2399                 goto out_end_trans;
2400         }
2401
2402         btrfs_record_root_in_trans(trans, dest);
2403
2404         memset(&dest->root_item.drop_progress, 0,
2405                 sizeof(dest->root_item.drop_progress));
2406         dest->root_item.drop_level = 0;
2407         btrfs_set_root_refs(&dest->root_item, 0);
2408
2409         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2410                 ret = btrfs_insert_orphan_item(trans,
2411                                         root->fs_info->tree_root,
2412                                         dest->root_key.objectid);
2413                 if (ret) {
2414                         btrfs_abort_transaction(trans, root, ret);
2415                         err = ret;
2416                         goto out_end_trans;
2417                 }
2418         }
2419
2420         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2421                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2422                                   dest->root_key.objectid);
2423         if (ret && ret != -ENOENT) {
2424                 btrfs_abort_transaction(trans, root, ret);
2425                 err = ret;
2426                 goto out_end_trans;
2427         }
2428         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2429                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2430                                           dest->root_item.received_uuid,
2431                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2432                                           dest->root_key.objectid);
2433                 if (ret && ret != -ENOENT) {
2434                         btrfs_abort_transaction(trans, root, ret);
2435                         err = ret;
2436                         goto out_end_trans;
2437                 }
2438         }
2439
2440 out_end_trans:
2441         trans->block_rsv = NULL;
2442         trans->bytes_reserved = 0;
2443         ret = btrfs_end_transaction(trans, root);
2444         if (ret && !err)
2445                 err = ret;
2446         inode->i_flags |= S_DEAD;
2447 out_release:
2448         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2449 out_up_write:
2450         up_write(&root->fs_info->subvol_sem);
2451 out_unlock:
2452         if (err) {
2453                 spin_lock(&dest->root_item_lock);
2454                 root_flags = btrfs_root_flags(&dest->root_item);
2455                 btrfs_set_root_flags(&dest->root_item,
2456                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2457                 spin_unlock(&dest->root_item_lock);
2458         }
2459         mutex_unlock(&inode->i_mutex);
2460         if (!err) {
2461                 shrink_dcache_sb(root->fs_info->sb);
2462                 btrfs_invalidate_inodes(dest);
2463                 d_delete(dentry);
2464                 ASSERT(dest->send_in_progress == 0);
2465
2466                 /* the last ref */
2467                 if (dest->cache_inode) {
2468                         iput(dest->cache_inode);
2469                         dest->cache_inode = NULL;
2470                 }
2471         }
2472 out_dput:
2473         dput(dentry);
2474 out_unlock_dir:
2475         mutex_unlock(&dir->i_mutex);
2476 out_drop_write:
2477         mnt_drop_write_file(file);
2478 out:
2479         kfree(vol_args);
2480         return err;
2481 }
2482
2483 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2484 {
2485         struct inode *inode = file_inode(file);
2486         struct btrfs_root *root = BTRFS_I(inode)->root;
2487         struct btrfs_ioctl_defrag_range_args *range;
2488         int ret;
2489
2490         ret = mnt_want_write_file(file);
2491         if (ret)
2492                 return ret;
2493
2494         if (btrfs_root_readonly(root)) {
2495                 ret = -EROFS;
2496                 goto out;
2497         }
2498
2499         switch (inode->i_mode & S_IFMT) {
2500         case S_IFDIR:
2501                 if (!capable(CAP_SYS_ADMIN)) {
2502                         ret = -EPERM;
2503                         goto out;
2504                 }
2505                 ret = btrfs_defrag_root(root);
2506                 if (ret)
2507                         goto out;
2508                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2509                 break;
2510         case S_IFREG:
2511                 if (!(file->f_mode & FMODE_WRITE)) {
2512                         ret = -EINVAL;
2513                         goto out;
2514                 }
2515
2516                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2517                 if (!range) {
2518                         ret = -ENOMEM;
2519                         goto out;
2520                 }
2521
2522                 if (argp) {
2523                         if (copy_from_user(range, argp,
2524                                            sizeof(*range))) {
2525                                 ret = -EFAULT;
2526                                 kfree(range);
2527                                 goto out;
2528                         }
2529                         /* compression requires us to start the IO */
2530                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2531                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2532                                 range->extent_thresh = (u32)-1;
2533                         }
2534                 } else {
2535                         /* the rest are all set to zero by kzalloc */
2536                         range->len = (u64)-1;
2537                 }
2538                 ret = btrfs_defrag_file(file_inode(file), file,
2539                                         range, 0, 0);
2540                 if (ret > 0)
2541                         ret = 0;
2542                 kfree(range);
2543                 break;
2544         default:
2545                 ret = -EINVAL;
2546         }
2547 out:
2548         mnt_drop_write_file(file);
2549         return ret;
2550 }
2551
2552 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2553 {
2554         struct btrfs_ioctl_vol_args *vol_args;
2555         int ret;
2556
2557         if (!capable(CAP_SYS_ADMIN))
2558                 return -EPERM;
2559
2560         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2561                         1)) {
2562                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2563         }
2564
2565         mutex_lock(&root->fs_info->volume_mutex);
2566         vol_args = memdup_user(arg, sizeof(*vol_args));
2567         if (IS_ERR(vol_args)) {
2568                 ret = PTR_ERR(vol_args);
2569                 goto out;
2570         }
2571
2572         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2573         ret = btrfs_init_new_device(root, vol_args->name);
2574
2575         kfree(vol_args);
2576 out:
2577         mutex_unlock(&root->fs_info->volume_mutex);
2578         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2579         return ret;
2580 }
2581
2582 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2583 {
2584         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2585         struct btrfs_ioctl_vol_args *vol_args;
2586         int ret;
2587
2588         if (!capable(CAP_SYS_ADMIN))
2589                 return -EPERM;
2590
2591         ret = mnt_want_write_file(file);
2592         if (ret)
2593                 return ret;
2594
2595         vol_args = memdup_user(arg, sizeof(*vol_args));
2596         if (IS_ERR(vol_args)) {
2597                 ret = PTR_ERR(vol_args);
2598                 goto out;
2599         }
2600
2601         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2602
2603         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2604                         1)) {
2605                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2606                 goto out;
2607         }
2608
2609         mutex_lock(&root->fs_info->volume_mutex);
2610         ret = btrfs_rm_device(root, vol_args->name);
2611         mutex_unlock(&root->fs_info->volume_mutex);
2612         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2613
2614 out:
2615         kfree(vol_args);
2616         mnt_drop_write_file(file);
2617         return ret;
2618 }
2619
2620 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2621 {
2622         struct btrfs_ioctl_fs_info_args *fi_args;
2623         struct btrfs_device *device;
2624         struct btrfs_device *next;
2625         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2626         int ret = 0;
2627
2628         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2629         if (!fi_args)
2630                 return -ENOMEM;
2631
2632         mutex_lock(&fs_devices->device_list_mutex);
2633         fi_args->num_devices = fs_devices->num_devices;
2634         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2635
2636         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2637                 if (device->devid > fi_args->max_id)
2638                         fi_args->max_id = device->devid;
2639         }
2640         mutex_unlock(&fs_devices->device_list_mutex);
2641
2642         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2643         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2644         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2645
2646         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2647                 ret = -EFAULT;
2648
2649         kfree(fi_args);
2650         return ret;
2651 }
2652
2653 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2654 {
2655         struct btrfs_ioctl_dev_info_args *di_args;
2656         struct btrfs_device *dev;
2657         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2658         int ret = 0;
2659         char *s_uuid = NULL;
2660
2661         di_args = memdup_user(arg, sizeof(*di_args));
2662         if (IS_ERR(di_args))
2663                 return PTR_ERR(di_args);
2664
2665         if (!btrfs_is_empty_uuid(di_args->uuid))
2666                 s_uuid = di_args->uuid;
2667
2668         mutex_lock(&fs_devices->device_list_mutex);
2669         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2670
2671         if (!dev) {
2672                 ret = -ENODEV;
2673                 goto out;
2674         }
2675
2676         di_args->devid = dev->devid;
2677         di_args->bytes_used = dev->bytes_used;
2678         di_args->total_bytes = dev->total_bytes;
2679         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2680         if (dev->name) {
2681                 struct rcu_string *name;
2682
2683                 rcu_read_lock();
2684                 name = rcu_dereference(dev->name);
2685                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2686                 rcu_read_unlock();
2687                 di_args->path[sizeof(di_args->path) - 1] = 0;
2688         } else {
2689                 di_args->path[0] = '\0';
2690         }
2691
2692 out:
2693         mutex_unlock(&fs_devices->device_list_mutex);
2694         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2695                 ret = -EFAULT;
2696
2697         kfree(di_args);
2698         return ret;
2699 }
2700
2701 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2702 {
2703         struct page *page;
2704         pgoff_t index;
2705         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2706
2707         index = off >> PAGE_CACHE_SHIFT;
2708
2709         page = grab_cache_page(inode->i_mapping, index);
2710         if (!page)
2711                 return NULL;
2712
2713         if (!PageUptodate(page)) {
2714                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2715                                                  0))
2716                         return NULL;
2717                 lock_page(page);
2718                 if (!PageUptodate(page)) {
2719                         unlock_page(page);
2720                         page_cache_release(page);
2721                         return NULL;
2722                 }
2723         }
2724         unlock_page(page);
2725
2726         return page;
2727 }
2728
2729 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2730 {
2731         /* do any pending delalloc/csum calc on src, one way or
2732            another, and lock file content */
2733         while (1) {
2734                 struct btrfs_ordered_extent *ordered;
2735                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2736                 ordered = btrfs_lookup_first_ordered_extent(inode,
2737                                                             off + len - 1);
2738                 if ((!ordered ||
2739                      ordered->file_offset + ordered->len <= off ||
2740                      ordered->file_offset >= off + len) &&
2741                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2742                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2743                         if (ordered)
2744                                 btrfs_put_ordered_extent(ordered);
2745                         break;
2746                 }
2747                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2748                 if (ordered)
2749                         btrfs_put_ordered_extent(ordered);
2750                 btrfs_wait_ordered_range(inode, off, len);
2751         }
2752 }
2753
2754 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2755                                 struct inode *inode2, u64 loff2, u64 len)
2756 {
2757         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2758         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2759
2760         mutex_unlock(&inode1->i_mutex);
2761         mutex_unlock(&inode2->i_mutex);
2762 }
2763
2764 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2765                               struct inode *inode2, u64 loff2, u64 len)
2766 {
2767         if (inode1 < inode2) {
2768                 swap(inode1, inode2);
2769                 swap(loff1, loff2);
2770         }
2771
2772         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2773         lock_extent_range(inode1, loff1, len);
2774         if (inode1 != inode2) {
2775                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2776                 lock_extent_range(inode2, loff2, len);
2777         }
2778 }
2779
2780 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2781                           u64 dst_loff, u64 len)
2782 {
2783         int ret = 0;
2784         struct page *src_page, *dst_page;
2785         unsigned int cmp_len = PAGE_CACHE_SIZE;
2786         void *addr, *dst_addr;
2787
2788         while (len) {
2789                 if (len < PAGE_CACHE_SIZE)
2790                         cmp_len = len;
2791
2792                 src_page = extent_same_get_page(src, loff);
2793                 if (!src_page)
2794                         return -EINVAL;
2795                 dst_page = extent_same_get_page(dst, dst_loff);
2796                 if (!dst_page) {
2797                         page_cache_release(src_page);
2798                         return -EINVAL;
2799                 }
2800                 addr = kmap_atomic(src_page);
2801                 dst_addr = kmap_atomic(dst_page);
2802
2803                 flush_dcache_page(src_page);
2804                 flush_dcache_page(dst_page);
2805
2806                 if (memcmp(addr, dst_addr, cmp_len))
2807                         ret = BTRFS_SAME_DATA_DIFFERS;
2808
2809                 kunmap_atomic(addr);
2810                 kunmap_atomic(dst_addr);
2811                 page_cache_release(src_page);
2812                 page_cache_release(dst_page);
2813
2814                 if (ret)
2815                         break;
2816
2817                 loff += cmp_len;
2818                 dst_loff += cmp_len;
2819                 len -= cmp_len;
2820         }
2821
2822         return ret;
2823 }
2824
2825 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2826 {
2827         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2828
2829         if (off + len > inode->i_size || off + len < off)
2830                 return -EINVAL;
2831         /* Check that we are block aligned - btrfs_clone() requires this */
2832         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2833                 return -EINVAL;
2834
2835         return 0;
2836 }
2837
2838 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2839                              struct inode *dst, u64 dst_loff)
2840 {
2841         int ret;
2842
2843         /*
2844          * btrfs_clone() can't handle extents in the same file
2845          * yet. Once that works, we can drop this check and replace it
2846          * with a check for the same inode, but overlapping extents.
2847          */
2848         if (src == dst)
2849                 return -EINVAL;
2850
2851         btrfs_double_lock(src, loff, dst, dst_loff, len);
2852
2853         ret = extent_same_check_offsets(src, loff, len);
2854         if (ret)
2855                 goto out_unlock;
2856
2857         ret = extent_same_check_offsets(dst, dst_loff, len);
2858         if (ret)
2859                 goto out_unlock;
2860
2861         /* don't make the dst file partly checksummed */
2862         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2863             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2864                 ret = -EINVAL;
2865                 goto out_unlock;
2866         }
2867
2868         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2869         if (ret == 0)
2870                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2871
2872 out_unlock:
2873         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2874
2875         return ret;
2876 }
2877
2878 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2879
2880 static long btrfs_ioctl_file_extent_same(struct file *file,
2881                         struct btrfs_ioctl_same_args __user *argp)
2882 {
2883         struct btrfs_ioctl_same_args *same;
2884         struct btrfs_ioctl_same_extent_info *info;
2885         struct inode *src = file_inode(file);
2886         u64 off;
2887         u64 len;
2888         int i;
2889         int ret;
2890         unsigned long size;
2891         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2892         bool is_admin = capable(CAP_SYS_ADMIN);
2893         u16 count;
2894
2895         if (!(file->f_mode & FMODE_READ))
2896                 return -EINVAL;
2897
2898         ret = mnt_want_write_file(file);
2899         if (ret)
2900                 return ret;
2901
2902         if (get_user(count, &argp->dest_count)) {
2903                 ret = -EFAULT;
2904                 goto out;
2905         }
2906
2907         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2908
2909         same = memdup_user(argp, size);
2910
2911         if (IS_ERR(same)) {
2912                 ret = PTR_ERR(same);
2913                 goto out;
2914         }
2915
2916         off = same->logical_offset;
2917         len = same->length;
2918
2919         /*
2920          * Limit the total length we will dedupe for each operation.
2921          * This is intended to bound the total time spent in this
2922          * ioctl to something sane.
2923          */
2924         if (len > BTRFS_MAX_DEDUPE_LEN)
2925                 len = BTRFS_MAX_DEDUPE_LEN;
2926
2927         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2928                 /*
2929                  * Btrfs does not support blocksize < page_size. As a
2930                  * result, btrfs_cmp_data() won't correctly handle
2931                  * this situation without an update.
2932                  */
2933                 ret = -EINVAL;
2934                 goto out;
2935         }
2936
2937         ret = -EISDIR;
2938         if (S_ISDIR(src->i_mode))
2939                 goto out;
2940
2941         ret = -EACCES;
2942         if (!S_ISREG(src->i_mode))
2943                 goto out;
2944
2945         /* pre-format output fields to sane values */
2946         for (i = 0; i < count; i++) {
2947                 same->info[i].bytes_deduped = 0ULL;
2948                 same->info[i].status = 0;
2949         }
2950
2951         for (i = 0, info = same->info; i < count; i++, info++) {
2952                 struct inode *dst;
2953                 struct fd dst_file = fdget(info->fd);
2954                 if (!dst_file.file) {
2955                         info->status = -EBADF;
2956                         continue;
2957                 }
2958                 dst = file_inode(dst_file.file);
2959
2960                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
2961                         info->status = -EINVAL;
2962                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
2963                         info->status = -EXDEV;
2964                 } else if (S_ISDIR(dst->i_mode)) {
2965                         info->status = -EISDIR;
2966                 } else if (!S_ISREG(dst->i_mode)) {
2967                         info->status = -EACCES;
2968                 } else {
2969                         info->status = btrfs_extent_same(src, off, len, dst,
2970                                                         info->logical_offset);
2971                         if (info->status == 0)
2972                                 info->bytes_deduped += len;
2973                 }
2974                 fdput(dst_file);
2975         }
2976
2977         ret = copy_to_user(argp, same, size);
2978         if (ret)
2979                 ret = -EFAULT;
2980
2981 out:
2982         mnt_drop_write_file(file);
2983         return ret;
2984 }
2985
2986 /* Helper to check and see if this root currently has a ref on the given disk
2987  * bytenr.  If it does then we need to update the quota for this root.  This
2988  * doesn't do anything if quotas aren't enabled.
2989  */
2990 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2991                      u64 disko)
2992 {
2993         struct seq_list tree_mod_seq_elem = {};
2994         struct ulist *roots;
2995         struct ulist_iterator uiter;
2996         struct ulist_node *root_node = NULL;
2997         int ret;
2998
2999         if (!root->fs_info->quota_enabled)
3000                 return 1;
3001
3002         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3003         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3004                                    tree_mod_seq_elem.seq, &roots);
3005         if (ret < 0)
3006                 goto out;
3007         ret = 0;
3008         ULIST_ITER_INIT(&uiter);
3009         while ((root_node = ulist_next(roots, &uiter))) {
3010                 if (root_node->val == root->objectid) {
3011                         ret = 1;
3012                         break;
3013                 }
3014         }
3015         ulist_free(roots);
3016 out:
3017         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3018         return ret;
3019 }
3020
3021 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3022                                      struct inode *inode,
3023                                      u64 endoff,
3024                                      const u64 destoff,
3025                                      const u64 olen)
3026 {
3027         struct btrfs_root *root = BTRFS_I(inode)->root;
3028         int ret;
3029
3030         inode_inc_iversion(inode);
3031         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3032         /*
3033          * We round up to the block size at eof when determining which
3034          * extents to clone above, but shouldn't round up the file size.
3035          */
3036         if (endoff > destoff + olen)
3037                 endoff = destoff + olen;
3038         if (endoff > inode->i_size)
3039                 btrfs_i_size_write(inode, endoff);
3040
3041         ret = btrfs_update_inode(trans, root, inode);
3042         if (ret) {
3043                 btrfs_abort_transaction(trans, root, ret);
3044                 btrfs_end_transaction(trans, root);
3045                 goto out;
3046         }
3047         ret = btrfs_end_transaction(trans, root);
3048 out:
3049         return ret;
3050 }
3051
3052 static void clone_update_extent_map(struct inode *inode,
3053                                     const struct btrfs_trans_handle *trans,
3054                                     const struct btrfs_path *path,
3055                                     struct btrfs_file_extent_item *fi,
3056                                     const u64 hole_offset,
3057                                     const u64 hole_len)
3058 {
3059         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3060         struct extent_map *em;
3061         int ret;
3062
3063         em = alloc_extent_map();
3064         if (!em) {
3065                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3066                         &BTRFS_I(inode)->runtime_flags);
3067                 return;
3068         }
3069
3070         if (fi) {
3071                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3072                 em->generation = -1;
3073                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3074                     BTRFS_FILE_EXTENT_INLINE)
3075                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3076                                 &BTRFS_I(inode)->runtime_flags);
3077         } else {
3078                 em->start = hole_offset;
3079                 em->len = hole_len;
3080                 em->ram_bytes = em->len;
3081                 em->orig_start = hole_offset;
3082                 em->block_start = EXTENT_MAP_HOLE;
3083                 em->block_len = 0;
3084                 em->orig_block_len = 0;
3085                 em->compress_type = BTRFS_COMPRESS_NONE;
3086                 em->generation = trans->transid;
3087         }
3088
3089         while (1) {
3090                 write_lock(&em_tree->lock);
3091                 ret = add_extent_mapping(em_tree, em, 1);
3092                 write_unlock(&em_tree->lock);
3093                 if (ret != -EEXIST) {
3094                         free_extent_map(em);
3095                         break;
3096                 }
3097                 btrfs_drop_extent_cache(inode, em->start,
3098                                         em->start + em->len - 1, 0);
3099         }
3100
3101         if (unlikely(ret))
3102                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3103                         &BTRFS_I(inode)->runtime_flags);
3104 }
3105
3106 /**
3107  * btrfs_clone() - clone a range from inode file to another
3108  *
3109  * @src: Inode to clone from
3110  * @inode: Inode to clone to
3111  * @off: Offset within source to start clone from
3112  * @olen: Original length, passed by user, of range to clone
3113  * @olen_aligned: Block-aligned value of olen, extent_same uses
3114  *               identical values here
3115  * @destoff: Offset within @inode to start clone
3116  */
3117 static int btrfs_clone(struct inode *src, struct inode *inode,
3118                        const u64 off, const u64 olen, const u64 olen_aligned,
3119                        const u64 destoff)
3120 {
3121         struct btrfs_root *root = BTRFS_I(inode)->root;
3122         struct btrfs_path *path = NULL;
3123         struct extent_buffer *leaf;
3124         struct btrfs_trans_handle *trans;
3125         char *buf = NULL;
3126         struct btrfs_key key;
3127         u32 nritems;
3128         int slot;
3129         int ret;
3130         int no_quota;
3131         const u64 len = olen_aligned;
3132         u64 last_disko = 0;
3133         u64 last_dest_end = destoff;
3134
3135         ret = -ENOMEM;
3136         buf = vmalloc(btrfs_level_size(root, 0));
3137         if (!buf)
3138                 return ret;
3139
3140         path = btrfs_alloc_path();
3141         if (!path) {
3142                 vfree(buf);
3143                 return ret;
3144         }
3145
3146         path->reada = 2;
3147         /* clone data */
3148         key.objectid = btrfs_ino(src);
3149         key.type = BTRFS_EXTENT_DATA_KEY;
3150         key.offset = off;
3151
3152         while (1) {
3153                 /*
3154                  * note the key will change type as we walk through the
3155                  * tree.
3156                  */
3157                 path->leave_spinning = 1;
3158                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3159                                 0, 0);
3160                 if (ret < 0)
3161                         goto out;
3162                 /*
3163                  * First search, if no extent item that starts at offset off was
3164                  * found but the previous item is an extent item, it's possible
3165                  * it might overlap our target range, therefore process it.
3166                  */
3167                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3168                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3169                                               path->slots[0] - 1);
3170                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3171                                 path->slots[0]--;
3172                 }
3173
3174                 nritems = btrfs_header_nritems(path->nodes[0]);
3175 process_slot:
3176                 no_quota = 1;
3177                 if (path->slots[0] >= nritems) {
3178                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3179                         if (ret < 0)
3180                                 goto out;
3181                         if (ret > 0)
3182                                 break;
3183                         nritems = btrfs_header_nritems(path->nodes[0]);
3184                 }
3185                 leaf = path->nodes[0];
3186                 slot = path->slots[0];
3187
3188                 btrfs_item_key_to_cpu(leaf, &key, slot);
3189                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
3190                     key.objectid != btrfs_ino(src))
3191                         break;
3192
3193                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
3194                         struct btrfs_file_extent_item *extent;
3195                         int type;
3196                         u32 size;
3197                         struct btrfs_key new_key;
3198                         u64 disko = 0, diskl = 0;
3199                         u64 datao = 0, datal = 0;
3200                         u8 comp;
3201                         u64 drop_start;
3202
3203                         extent = btrfs_item_ptr(leaf, slot,
3204                                                 struct btrfs_file_extent_item);
3205                         comp = btrfs_file_extent_compression(leaf, extent);
3206                         type = btrfs_file_extent_type(leaf, extent);
3207                         if (type == BTRFS_FILE_EXTENT_REG ||
3208                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3209                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3210                                                                       extent);
3211                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3212                                                                  extent);
3213                                 datao = btrfs_file_extent_offset(leaf, extent);
3214                                 datal = btrfs_file_extent_num_bytes(leaf,
3215                                                                     extent);
3216                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3217                                 /* take upper bound, may be compressed */
3218                                 datal = btrfs_file_extent_ram_bytes(leaf,
3219                                                                     extent);
3220                         }
3221
3222                         /*
3223                          * The first search might have left us at an extent
3224                          * item that ends before our target range's start, can
3225                          * happen if we have holes and NO_HOLES feature enabled.
3226                          */
3227                         if (key.offset + datal <= off) {
3228                                 path->slots[0]++;
3229                                 goto process_slot;
3230                         } else if (key.offset >= off + len) {
3231                                 break;
3232                         }
3233
3234                         size = btrfs_item_size_nr(leaf, slot);
3235                         read_extent_buffer(leaf, buf,
3236                                            btrfs_item_ptr_offset(leaf, slot),
3237                                            size);
3238
3239                         btrfs_release_path(path);
3240                         path->leave_spinning = 0;
3241
3242                         memcpy(&new_key, &key, sizeof(new_key));
3243                         new_key.objectid = btrfs_ino(inode);
3244                         if (off <= key.offset)
3245                                 new_key.offset = key.offset + destoff - off;
3246                         else
3247                                 new_key.offset = destoff;
3248
3249                         /*
3250                          * Deal with a hole that doesn't have an extent item
3251                          * that represents it (NO_HOLES feature enabled).
3252                          * This hole is either in the middle of the cloning
3253                          * range or at the beginning (fully overlaps it or
3254                          * partially overlaps it).
3255                          */
3256                         if (new_key.offset != last_dest_end)
3257                                 drop_start = last_dest_end;
3258                         else
3259                                 drop_start = new_key.offset;
3260
3261                         /*
3262                          * 1 - adjusting old extent (we may have to split it)
3263                          * 1 - add new extent
3264                          * 1 - inode update
3265                          */
3266                         trans = btrfs_start_transaction(root, 3);
3267                         if (IS_ERR(trans)) {
3268                                 ret = PTR_ERR(trans);
3269                                 goto out;
3270                         }
3271
3272                         if (type == BTRFS_FILE_EXTENT_REG ||
3273                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3274                                 /*
3275                                  *    a  | --- range to clone ---|  b
3276                                  * | ------------- extent ------------- |
3277                                  */
3278
3279                                 /* subtract range b */
3280                                 if (key.offset + datal > off + len)
3281                                         datal = off + len - key.offset;
3282
3283                                 /* subtract range a */
3284                                 if (off > key.offset) {
3285                                         datao += off - key.offset;
3286                                         datal -= off - key.offset;
3287                                 }
3288
3289                                 ret = btrfs_drop_extents(trans, root, inode,
3290                                                          drop_start,
3291                                                          new_key.offset + datal,
3292                                                          1);
3293                                 if (ret) {
3294                                         if (ret != -EOPNOTSUPP)
3295                                                 btrfs_abort_transaction(trans,
3296                                                                 root, ret);
3297                                         btrfs_end_transaction(trans, root);
3298                                         goto out;
3299                                 }
3300
3301                                 ret = btrfs_insert_empty_item(trans, root, path,
3302                                                               &new_key, size);
3303                                 if (ret) {
3304                                         btrfs_abort_transaction(trans, root,
3305                                                                 ret);
3306                                         btrfs_end_transaction(trans, root);
3307                                         goto out;
3308                                 }
3309
3310                                 leaf = path->nodes[0];
3311                                 slot = path->slots[0];
3312                                 write_extent_buffer(leaf, buf,
3313                                             btrfs_item_ptr_offset(leaf, slot),
3314                                             size);
3315
3316                                 extent = btrfs_item_ptr(leaf, slot,
3317                                                 struct btrfs_file_extent_item);
3318
3319                                 /* disko == 0 means it's a hole */
3320                                 if (!disko)
3321                                         datao = 0;
3322
3323                                 btrfs_set_file_extent_offset(leaf, extent,
3324                                                              datao);
3325                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3326                                                                 datal);
3327
3328                                 /*
3329                                  * We need to look up the roots that point at
3330                                  * this bytenr and see if the new root does.  If
3331                                  * it does not we need to make sure we update
3332                                  * quotas appropriately.
3333                                  */
3334                                 if (disko && root != BTRFS_I(src)->root &&
3335                                     disko != last_disko) {
3336                                         no_quota = check_ref(trans, root,
3337                                                              disko);
3338                                         if (no_quota < 0) {
3339                                                 btrfs_abort_transaction(trans,
3340                                                                         root,
3341                                                                         ret);
3342                                                 btrfs_end_transaction(trans,
3343                                                                       root);
3344                                                 ret = no_quota;
3345                                                 goto out;
3346                                         }
3347                                 }
3348
3349                                 if (disko) {
3350                                         inode_add_bytes(inode, datal);
3351                                         ret = btrfs_inc_extent_ref(trans, root,
3352                                                         disko, diskl, 0,
3353                                                         root->root_key.objectid,
3354                                                         btrfs_ino(inode),
3355                                                         new_key.offset - datao,
3356                                                         no_quota);
3357                                         if (ret) {
3358                                                 btrfs_abort_transaction(trans,
3359                                                                         root,
3360                                                                         ret);
3361                                                 btrfs_end_transaction(trans,
3362                                                                       root);
3363                                                 goto out;
3364
3365                                         }
3366                                 }
3367                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3368                                 u64 skip = 0;
3369                                 u64 trim = 0;
3370                                 u64 aligned_end = 0;
3371
3372                                 if (off > key.offset) {
3373                                         skip = off - key.offset;
3374                                         new_key.offset += skip;
3375                                 }
3376
3377                                 if (key.offset + datal > off + len)
3378                                         trim = key.offset + datal - (off + len);
3379
3380                                 if (comp && (skip || trim)) {
3381                                         ret = -EINVAL;
3382                                         btrfs_end_transaction(trans, root);
3383                                         goto out;
3384                                 }
3385                                 size -= skip + trim;
3386                                 datal -= skip + trim;
3387
3388                                 aligned_end = ALIGN(new_key.offset + datal,
3389                                                     root->sectorsize);
3390                                 ret = btrfs_drop_extents(trans, root, inode,
3391                                                          drop_start,
3392                                                          aligned_end,
3393                                                          1);
3394                                 if (ret) {
3395                                         if (ret != -EOPNOTSUPP)
3396                                                 btrfs_abort_transaction(trans,
3397                                                         root, ret);
3398                                         btrfs_end_transaction(trans, root);
3399                                         goto out;
3400                                 }
3401
3402                                 ret = btrfs_insert_empty_item(trans, root, path,
3403                                                               &new_key, size);
3404                                 if (ret) {
3405                                         btrfs_abort_transaction(trans, root,
3406                                                                 ret);
3407                                         btrfs_end_transaction(trans, root);
3408                                         goto out;
3409                                 }
3410
3411                                 if (skip) {
3412                                         u32 start =
3413                                           btrfs_file_extent_calc_inline_size(0);
3414                                         memmove(buf+start, buf+start+skip,
3415                                                 datal);
3416                                 }
3417
3418                                 leaf = path->nodes[0];
3419                                 slot = path->slots[0];
3420                                 write_extent_buffer(leaf, buf,
3421                                             btrfs_item_ptr_offset(leaf, slot),
3422                                             size);
3423                                 inode_add_bytes(inode, datal);
3424                                 extent = btrfs_item_ptr(leaf, slot,
3425                                                 struct btrfs_file_extent_item);
3426                         }
3427
3428                         /* If we have an implicit hole (NO_HOLES feature). */
3429                         if (drop_start < new_key.offset)
3430                                 clone_update_extent_map(inode, trans,
3431                                                 path, NULL, drop_start,
3432                                                 new_key.offset - drop_start);
3433
3434                         clone_update_extent_map(inode, trans, path,
3435                                                 extent, 0, 0);
3436
3437                         btrfs_mark_buffer_dirty(leaf);
3438                         btrfs_release_path(path);
3439
3440                         last_dest_end = new_key.offset + datal;
3441                         ret = clone_finish_inode_update(trans, inode,
3442                                                         last_dest_end,
3443                                                         destoff, olen);
3444                         if (ret)
3445                                 goto out;
3446                         if (new_key.offset + datal >= destoff + len)
3447                                 break;
3448                 }
3449                 btrfs_release_path(path);
3450                 key.offset++;
3451         }
3452         ret = 0;
3453
3454         if (last_dest_end < destoff + len) {
3455                 /*
3456                  * We have an implicit hole (NO_HOLES feature is enabled) that
3457                  * fully or partially overlaps our cloning range at its end.
3458                  */
3459                 btrfs_release_path(path);
3460
3461                 /*
3462                  * 1 - remove extent(s)
3463                  * 1 - inode update
3464                  */
3465                 trans = btrfs_start_transaction(root, 2);
3466                 if (IS_ERR(trans)) {
3467                         ret = PTR_ERR(trans);
3468                         goto out;
3469                 }
3470                 ret = btrfs_drop_extents(trans, root, inode,
3471                                          last_dest_end, destoff + len, 1);
3472                 if (ret) {
3473                         if (ret != -EOPNOTSUPP)
3474                                 btrfs_abort_transaction(trans, root, ret);
3475                         btrfs_end_transaction(trans, root);
3476                         goto out;
3477                 }
3478                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3479                                                 destoff, olen);
3480                 if (ret)
3481                         goto out;
3482                 clone_update_extent_map(inode, trans, path, NULL, last_dest_end,
3483                                         destoff + len - last_dest_end);
3484         }
3485
3486 out:
3487         btrfs_free_path(path);
3488         vfree(buf);
3489         return ret;
3490 }
3491
3492 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3493                                        u64 off, u64 olen, u64 destoff)
3494 {
3495         struct inode *inode = file_inode(file);
3496         struct btrfs_root *root = BTRFS_I(inode)->root;
3497         struct fd src_file;
3498         struct inode *src;
3499         int ret;
3500         u64 len = olen;
3501         u64 bs = root->fs_info->sb->s_blocksize;
3502         int same_inode = 0;
3503
3504         /*
3505          * TODO:
3506          * - split compressed inline extents.  annoying: we need to
3507          *   decompress into destination's address_space (the file offset
3508          *   may change, so source mapping won't do), then recompress (or
3509          *   otherwise reinsert) a subrange.
3510          *
3511          * - split destination inode's inline extents.  The inline extents can
3512          *   be either compressed or non-compressed.
3513          */
3514
3515         /* the destination must be opened for writing */
3516         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3517                 return -EINVAL;
3518
3519         if (btrfs_root_readonly(root))
3520                 return -EROFS;
3521
3522         ret = mnt_want_write_file(file);
3523         if (ret)
3524                 return ret;
3525
3526         src_file = fdget(srcfd);
3527         if (!src_file.file) {
3528                 ret = -EBADF;
3529                 goto out_drop_write;
3530         }
3531
3532         ret = -EXDEV;
3533         if (src_file.file->f_path.mnt != file->f_path.mnt)
3534                 goto out_fput;
3535
3536         src = file_inode(src_file.file);
3537
3538         ret = -EINVAL;
3539         if (src == inode)
3540                 same_inode = 1;
3541
3542         /* the src must be open for reading */
3543         if (!(src_file.file->f_mode & FMODE_READ))
3544                 goto out_fput;
3545
3546         /* don't make the dst file partly checksummed */
3547         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3548             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3549                 goto out_fput;
3550
3551         ret = -EISDIR;
3552         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3553                 goto out_fput;
3554
3555         ret = -EXDEV;
3556         if (src->i_sb != inode->i_sb)
3557                 goto out_fput;
3558
3559         if (!same_inode) {
3560                 if (inode < src) {
3561                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3562                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3563                 } else {
3564                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3565                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3566                 }
3567         } else {
3568                 mutex_lock(&src->i_mutex);
3569         }
3570
3571         /* determine range to clone */
3572         ret = -EINVAL;
3573         if (off + len > src->i_size || off + len < off)
3574                 goto out_unlock;
3575         if (len == 0)
3576                 olen = len = src->i_size - off;
3577         /* if we extend to eof, continue to block boundary */
3578         if (off + len == src->i_size)
3579                 len = ALIGN(src->i_size, bs) - off;
3580
3581         /* verify the end result is block aligned */
3582         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3583             !IS_ALIGNED(destoff, bs))
3584                 goto out_unlock;
3585
3586         /* verify if ranges are overlapped within the same file */
3587         if (same_inode) {
3588                 if (destoff + len > off && destoff < off + len)
3589                         goto out_unlock;
3590         }
3591
3592         if (destoff > inode->i_size) {
3593                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3594                 if (ret)
3595                         goto out_unlock;
3596         }
3597
3598         /*
3599          * Lock the target range too. Right after we replace the file extent
3600          * items in the fs tree (which now point to the cloned data), we might
3601          * have a worker replace them with extent items relative to a write
3602          * operation that was issued before this clone operation (i.e. confront
3603          * with inode.c:btrfs_finish_ordered_io).
3604          */
3605         if (same_inode) {
3606                 u64 lock_start = min_t(u64, off, destoff);
3607                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3608
3609                 lock_extent_range(src, lock_start, lock_len);
3610         } else {
3611                 lock_extent_range(src, off, len);
3612                 lock_extent_range(inode, destoff, len);
3613         }
3614
3615         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3616
3617         if (same_inode) {
3618                 u64 lock_start = min_t(u64, off, destoff);
3619                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3620
3621                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3622         } else {
3623                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3624                 unlock_extent(&BTRFS_I(inode)->io_tree, destoff,
3625                               destoff + len - 1);
3626         }
3627         /*
3628          * Truncate page cache pages so that future reads will see the cloned
3629          * data immediately and not the previous data.
3630          */
3631         truncate_inode_pages_range(&inode->i_data, destoff,
3632                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3633 out_unlock:
3634         if (!same_inode) {
3635                 if (inode < src) {
3636                         mutex_unlock(&src->i_mutex);
3637                         mutex_unlock(&inode->i_mutex);
3638                 } else {
3639                         mutex_unlock(&inode->i_mutex);
3640                         mutex_unlock(&src->i_mutex);
3641                 }
3642         } else {
3643                 mutex_unlock(&src->i_mutex);
3644         }
3645 out_fput:
3646         fdput(src_file);
3647 out_drop_write:
3648         mnt_drop_write_file(file);
3649         return ret;
3650 }
3651
3652 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3653 {
3654         struct btrfs_ioctl_clone_range_args args;
3655
3656         if (copy_from_user(&args, argp, sizeof(args)))
3657                 return -EFAULT;
3658         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3659                                  args.src_length, args.dest_offset);
3660 }
3661
3662 /*
3663  * there are many ways the trans_start and trans_end ioctls can lead
3664  * to deadlocks.  They should only be used by applications that
3665  * basically own the machine, and have a very in depth understanding
3666  * of all the possible deadlocks and enospc problems.
3667  */
3668 static long btrfs_ioctl_trans_start(struct file *file)
3669 {
3670         struct inode *inode = file_inode(file);
3671         struct btrfs_root *root = BTRFS_I(inode)->root;
3672         struct btrfs_trans_handle *trans;
3673         int ret;
3674
3675         ret = -EPERM;
3676         if (!capable(CAP_SYS_ADMIN))
3677                 goto out;
3678
3679         ret = -EINPROGRESS;
3680         if (file->private_data)
3681                 goto out;
3682
3683         ret = -EROFS;
3684         if (btrfs_root_readonly(root))
3685                 goto out;
3686
3687         ret = mnt_want_write_file(file);
3688         if (ret)
3689                 goto out;
3690
3691         atomic_inc(&root->fs_info->open_ioctl_trans);
3692
3693         ret = -ENOMEM;
3694         trans = btrfs_start_ioctl_transaction(root);
3695         if (IS_ERR(trans))
3696                 goto out_drop;
3697
3698         file->private_data = trans;
3699         return 0;
3700
3701 out_drop:
3702         atomic_dec(&root->fs_info->open_ioctl_trans);
3703         mnt_drop_write_file(file);
3704 out:
3705         return ret;
3706 }
3707
3708 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3709 {
3710         struct inode *inode = file_inode(file);
3711         struct btrfs_root *root = BTRFS_I(inode)->root;
3712         struct btrfs_root *new_root;
3713         struct btrfs_dir_item *di;
3714         struct btrfs_trans_handle *trans;
3715         struct btrfs_path *path;
3716         struct btrfs_key location;
3717         struct btrfs_disk_key disk_key;
3718         u64 objectid = 0;
3719         u64 dir_id;
3720         int ret;
3721
3722         if (!capable(CAP_SYS_ADMIN))
3723                 return -EPERM;
3724
3725         ret = mnt_want_write_file(file);
3726         if (ret)
3727                 return ret;
3728
3729         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3730                 ret = -EFAULT;
3731                 goto out;
3732         }
3733
3734         if (!objectid)
3735                 objectid = BTRFS_FS_TREE_OBJECTID;
3736
3737         location.objectid = objectid;
3738         location.type = BTRFS_ROOT_ITEM_KEY;
3739         location.offset = (u64)-1;
3740
3741         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3742         if (IS_ERR(new_root)) {
3743                 ret = PTR_ERR(new_root);
3744                 goto out;
3745         }
3746
3747         path = btrfs_alloc_path();
3748         if (!path) {
3749                 ret = -ENOMEM;
3750                 goto out;
3751         }
3752         path->leave_spinning = 1;
3753
3754         trans = btrfs_start_transaction(root, 1);
3755         if (IS_ERR(trans)) {
3756                 btrfs_free_path(path);
3757                 ret = PTR_ERR(trans);
3758                 goto out;
3759         }
3760
3761         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3762         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3763                                    dir_id, "default", 7, 1);
3764         if (IS_ERR_OR_NULL(di)) {
3765                 btrfs_free_path(path);
3766                 btrfs_end_transaction(trans, root);
3767                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3768                            "item, this isn't going to work");
3769                 ret = -ENOENT;
3770                 goto out;
3771         }
3772
3773         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3774         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3775         btrfs_mark_buffer_dirty(path->nodes[0]);
3776         btrfs_free_path(path);
3777
3778         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3779         btrfs_end_transaction(trans, root);
3780 out:
3781         mnt_drop_write_file(file);
3782         return ret;
3783 }
3784
3785 void btrfs_get_block_group_info(struct list_head *groups_list,
3786                                 struct btrfs_ioctl_space_info *space)
3787 {
3788         struct btrfs_block_group_cache *block_group;
3789
3790         space->total_bytes = 0;
3791         space->used_bytes = 0;
3792         space->flags = 0;
3793         list_for_each_entry(block_group, groups_list, list) {
3794                 space->flags = block_group->flags;
3795                 space->total_bytes += block_group->key.offset;
3796                 space->used_bytes +=
3797                         btrfs_block_group_used(&block_group->item);
3798         }
3799 }
3800
3801 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3802 {
3803         struct btrfs_ioctl_space_args space_args;
3804         struct btrfs_ioctl_space_info space;
3805         struct btrfs_ioctl_space_info *dest;
3806         struct btrfs_ioctl_space_info *dest_orig;
3807         struct btrfs_ioctl_space_info __user *user_dest;
3808         struct btrfs_space_info *info;
3809         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3810                        BTRFS_BLOCK_GROUP_SYSTEM,
3811                        BTRFS_BLOCK_GROUP_METADATA,
3812                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3813         int num_types = 4;
3814         int alloc_size;
3815         int ret = 0;
3816         u64 slot_count = 0;
3817         int i, c;
3818
3819         if (copy_from_user(&space_args,
3820                            (struct btrfs_ioctl_space_args __user *)arg,
3821                            sizeof(space_args)))
3822                 return -EFAULT;
3823
3824         for (i = 0; i < num_types; i++) {
3825                 struct btrfs_space_info *tmp;
3826
3827                 info = NULL;
3828                 rcu_read_lock();
3829                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3830                                         list) {
3831                         if (tmp->flags == types[i]) {
3832                                 info = tmp;
3833                                 break;
3834                         }
3835                 }
3836                 rcu_read_unlock();
3837
3838                 if (!info)
3839                         continue;
3840
3841                 down_read(&info->groups_sem);
3842                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3843                         if (!list_empty(&info->block_groups[c]))
3844                                 slot_count++;
3845                 }
3846                 up_read(&info->groups_sem);
3847         }
3848
3849         /*
3850          * Global block reserve, exported as a space_info
3851          */
3852         slot_count++;
3853
3854         /* space_slots == 0 means they are asking for a count */
3855         if (space_args.space_slots == 0) {
3856                 space_args.total_spaces = slot_count;
3857                 goto out;
3858         }
3859
3860         slot_count = min_t(u64, space_args.space_slots, slot_count);
3861
3862         alloc_size = sizeof(*dest) * slot_count;
3863
3864         /* we generally have at most 6 or so space infos, one for each raid
3865          * level.  So, a whole page should be more than enough for everyone
3866          */
3867         if (alloc_size > PAGE_CACHE_SIZE)
3868                 return -ENOMEM;
3869
3870         space_args.total_spaces = 0;
3871         dest = kmalloc(alloc_size, GFP_NOFS);
3872         if (!dest)
3873                 return -ENOMEM;
3874         dest_orig = dest;
3875
3876         /* now we have a buffer to copy into */
3877         for (i = 0; i < num_types; i++) {
3878                 struct btrfs_space_info *tmp;
3879
3880                 if (!slot_count)
3881                         break;
3882
3883                 info = NULL;
3884                 rcu_read_lock();
3885                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3886                                         list) {
3887                         if (tmp->flags == types[i]) {
3888                                 info = tmp;
3889                                 break;
3890                         }
3891                 }
3892                 rcu_read_unlock();
3893
3894                 if (!info)
3895                         continue;
3896                 down_read(&info->groups_sem);
3897                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3898                         if (!list_empty(&info->block_groups[c])) {
3899                                 btrfs_get_block_group_info(
3900                                         &info->block_groups[c], &space);
3901                                 memcpy(dest, &space, sizeof(space));
3902                                 dest++;
3903                                 space_args.total_spaces++;
3904                                 slot_count--;
3905                         }
3906                         if (!slot_count)
3907                                 break;
3908                 }
3909                 up_read(&info->groups_sem);
3910         }
3911
3912         /*
3913          * Add global block reserve
3914          */
3915         if (slot_count) {
3916                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3917
3918                 spin_lock(&block_rsv->lock);
3919                 space.total_bytes = block_rsv->size;
3920                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3921                 spin_unlock(&block_rsv->lock);
3922                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3923                 memcpy(dest, &space, sizeof(space));
3924                 space_args.total_spaces++;
3925         }
3926
3927         user_dest = (struct btrfs_ioctl_space_info __user *)
3928                 (arg + sizeof(struct btrfs_ioctl_space_args));
3929
3930         if (copy_to_user(user_dest, dest_orig, alloc_size))
3931                 ret = -EFAULT;
3932
3933         kfree(dest_orig);
3934 out:
3935         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3936                 ret = -EFAULT;
3937
3938         return ret;
3939 }
3940
3941 /*
3942  * there are many ways the trans_start and trans_end ioctls can lead
3943  * to deadlocks.  They should only be used by applications that
3944  * basically own the machine, and have a very in depth understanding
3945  * of all the possible deadlocks and enospc problems.
3946  */
3947 long btrfs_ioctl_trans_end(struct file *file)
3948 {
3949         struct inode *inode = file_inode(file);
3950         struct btrfs_root *root = BTRFS_I(inode)->root;
3951         struct btrfs_trans_handle *trans;
3952
3953         trans = file->private_data;
3954         if (!trans)
3955                 return -EINVAL;
3956         file->private_data = NULL;
3957
3958         btrfs_end_transaction(trans, root);
3959
3960         atomic_dec(&root->fs_info->open_ioctl_trans);
3961
3962         mnt_drop_write_file(file);
3963         return 0;
3964 }
3965
3966 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3967                                             void __user *argp)
3968 {
3969         struct btrfs_trans_handle *trans;
3970         u64 transid;
3971         int ret;
3972
3973         trans = btrfs_attach_transaction_barrier(root);
3974         if (IS_ERR(trans)) {
3975                 if (PTR_ERR(trans) != -ENOENT)
3976                         return PTR_ERR(trans);
3977
3978                 /* No running transaction, don't bother */
3979                 transid = root->fs_info->last_trans_committed;
3980                 goto out;
3981         }
3982         transid = trans->transid;
3983         ret = btrfs_commit_transaction_async(trans, root, 0);
3984         if (ret) {
3985                 btrfs_end_transaction(trans, root);
3986                 return ret;
3987         }
3988 out:
3989         if (argp)
3990                 if (copy_to_user(argp, &transid, sizeof(transid)))
3991                         return -EFAULT;
3992         return 0;
3993 }
3994
3995 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3996                                            void __user *argp)
3997 {
3998         u64 transid;
3999
4000         if (argp) {
4001                 if (copy_from_user(&transid, argp, sizeof(transid)))
4002                         return -EFAULT;
4003         } else {
4004                 transid = 0;  /* current trans */
4005         }
4006         return btrfs_wait_for_commit(root, transid);
4007 }
4008
4009 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4010 {
4011         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4012         struct btrfs_ioctl_scrub_args *sa;
4013         int ret;
4014
4015         if (!capable(CAP_SYS_ADMIN))
4016                 return -EPERM;
4017
4018         sa = memdup_user(arg, sizeof(*sa));
4019         if (IS_ERR(sa))
4020                 return PTR_ERR(sa);
4021
4022         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4023                 ret = mnt_want_write_file(file);
4024                 if (ret)
4025                         goto out;
4026         }
4027
4028         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4029                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4030                               0);
4031
4032         if (copy_to_user(arg, sa, sizeof(*sa)))
4033                 ret = -EFAULT;
4034
4035         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4036                 mnt_drop_write_file(file);
4037 out:
4038         kfree(sa);
4039         return ret;
4040 }
4041
4042 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4043 {
4044         if (!capable(CAP_SYS_ADMIN))
4045                 return -EPERM;
4046
4047         return btrfs_scrub_cancel(root->fs_info);
4048 }
4049
4050 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4051                                        void __user *arg)
4052 {
4053         struct btrfs_ioctl_scrub_args *sa;
4054         int ret;
4055
4056         if (!capable(CAP_SYS_ADMIN))
4057                 return -EPERM;
4058
4059         sa = memdup_user(arg, sizeof(*sa));
4060         if (IS_ERR(sa))
4061                 return PTR_ERR(sa);
4062
4063         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4064
4065         if (copy_to_user(arg, sa, sizeof(*sa)))
4066                 ret = -EFAULT;
4067
4068         kfree(sa);
4069         return ret;
4070 }
4071
4072 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4073                                       void __user *arg)
4074 {
4075         struct btrfs_ioctl_get_dev_stats *sa;
4076         int ret;
4077
4078         sa = memdup_user(arg, sizeof(*sa));
4079         if (IS_ERR(sa))
4080                 return PTR_ERR(sa);
4081
4082         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4083                 kfree(sa);
4084                 return -EPERM;
4085         }
4086
4087         ret = btrfs_get_dev_stats(root, sa);
4088
4089         if (copy_to_user(arg, sa, sizeof(*sa)))
4090                 ret = -EFAULT;
4091
4092         kfree(sa);
4093         return ret;
4094 }
4095
4096 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4097 {
4098         struct btrfs_ioctl_dev_replace_args *p;
4099         int ret;
4100
4101         if (!capable(CAP_SYS_ADMIN))
4102                 return -EPERM;
4103
4104         p = memdup_user(arg, sizeof(*p));
4105         if (IS_ERR(p))
4106                 return PTR_ERR(p);
4107
4108         switch (p->cmd) {
4109         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4110                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4111                         ret = -EROFS;
4112                         goto out;
4113                 }
4114                 if (atomic_xchg(
4115                         &root->fs_info->mutually_exclusive_operation_running,
4116                         1)) {
4117                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4118                 } else {
4119                         ret = btrfs_dev_replace_start(root, p);
4120                         atomic_set(
4121                          &root->fs_info->mutually_exclusive_operation_running,
4122                          0);
4123                 }
4124                 break;
4125         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4126                 btrfs_dev_replace_status(root->fs_info, p);
4127                 ret = 0;
4128                 break;
4129         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4130                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4131                 break;
4132         default:
4133                 ret = -EINVAL;
4134                 break;
4135         }
4136
4137         if (copy_to_user(arg, p, sizeof(*p)))
4138                 ret = -EFAULT;
4139 out:
4140         kfree(p);
4141         return ret;
4142 }
4143
4144 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4145 {
4146         int ret = 0;
4147         int i;
4148         u64 rel_ptr;
4149         int size;
4150         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4151         struct inode_fs_paths *ipath = NULL;
4152         struct btrfs_path *path;
4153
4154         if (!capable(CAP_DAC_READ_SEARCH))
4155                 return -EPERM;
4156
4157         path = btrfs_alloc_path();
4158         if (!path) {
4159                 ret = -ENOMEM;
4160                 goto out;
4161         }
4162
4163         ipa = memdup_user(arg, sizeof(*ipa));
4164         if (IS_ERR(ipa)) {
4165                 ret = PTR_ERR(ipa);
4166                 ipa = NULL;
4167                 goto out;
4168         }
4169
4170         size = min_t(u32, ipa->size, 4096);
4171         ipath = init_ipath(size, root, path);
4172         if (IS_ERR(ipath)) {
4173                 ret = PTR_ERR(ipath);
4174                 ipath = NULL;
4175                 goto out;
4176         }
4177
4178         ret = paths_from_inode(ipa->inum, ipath);
4179         if (ret < 0)
4180                 goto out;
4181
4182         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4183                 rel_ptr = ipath->fspath->val[i] -
4184                           (u64)(unsigned long)ipath->fspath->val;
4185                 ipath->fspath->val[i] = rel_ptr;
4186         }
4187
4188         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4189                            (void *)(unsigned long)ipath->fspath, size);
4190         if (ret) {
4191                 ret = -EFAULT;
4192                 goto out;
4193         }
4194
4195 out:
4196         btrfs_free_path(path);
4197         free_ipath(ipath);
4198         kfree(ipa);
4199
4200         return ret;
4201 }
4202
4203 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4204 {
4205         struct btrfs_data_container *inodes = ctx;
4206         const size_t c = 3 * sizeof(u64);
4207
4208         if (inodes->bytes_left >= c) {
4209                 inodes->bytes_left -= c;
4210                 inodes->val[inodes->elem_cnt] = inum;
4211                 inodes->val[inodes->elem_cnt + 1] = offset;
4212                 inodes->val[inodes->elem_cnt + 2] = root;
4213                 inodes->elem_cnt += 3;
4214         } else {
4215                 inodes->bytes_missing += c - inodes->bytes_left;
4216                 inodes->bytes_left = 0;
4217                 inodes->elem_missed += 3;
4218         }
4219
4220         return 0;
4221 }
4222
4223 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4224                                         void __user *arg)
4225 {
4226         int ret = 0;
4227         int size;
4228         struct btrfs_ioctl_logical_ino_args *loi;
4229         struct btrfs_data_container *inodes = NULL;
4230         struct btrfs_path *path = NULL;
4231
4232         if (!capable(CAP_SYS_ADMIN))
4233                 return -EPERM;
4234
4235         loi = memdup_user(arg, sizeof(*loi));
4236         if (IS_ERR(loi)) {
4237                 ret = PTR_ERR(loi);
4238                 loi = NULL;
4239                 goto out;
4240         }
4241
4242         path = btrfs_alloc_path();
4243         if (!path) {
4244                 ret = -ENOMEM;
4245                 goto out;
4246         }
4247
4248         size = min_t(u32, loi->size, 64 * 1024);
4249         inodes = init_data_container(size);
4250         if (IS_ERR(inodes)) {
4251                 ret = PTR_ERR(inodes);
4252                 inodes = NULL;
4253                 goto out;
4254         }
4255
4256         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4257                                           build_ino_list, inodes);
4258         if (ret == -EINVAL)
4259                 ret = -ENOENT;
4260         if (ret < 0)
4261                 goto out;
4262
4263         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4264                            (void *)(unsigned long)inodes, size);
4265         if (ret)
4266                 ret = -EFAULT;
4267
4268 out:
4269         btrfs_free_path(path);
4270         vfree(inodes);
4271         kfree(loi);
4272
4273         return ret;
4274 }
4275
4276 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4277                                struct btrfs_ioctl_balance_args *bargs)
4278 {
4279         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4280
4281         bargs->flags = bctl->flags;
4282
4283         if (atomic_read(&fs_info->balance_running))
4284                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4285         if (atomic_read(&fs_info->balance_pause_req))
4286                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4287         if (atomic_read(&fs_info->balance_cancel_req))
4288                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4289
4290         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4291         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4292         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4293
4294         if (lock) {
4295                 spin_lock(&fs_info->balance_lock);
4296                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4297                 spin_unlock(&fs_info->balance_lock);
4298         } else {
4299                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4300         }
4301 }
4302
4303 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4304 {
4305         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4306         struct btrfs_fs_info *fs_info = root->fs_info;
4307         struct btrfs_ioctl_balance_args *bargs;
4308         struct btrfs_balance_control *bctl;
4309         bool need_unlock; /* for mut. excl. ops lock */
4310         int ret;
4311
4312         if (!capable(CAP_SYS_ADMIN))
4313                 return -EPERM;
4314
4315         ret = mnt_want_write_file(file);
4316         if (ret)
4317                 return ret;
4318
4319 again:
4320         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4321                 mutex_lock(&fs_info->volume_mutex);
4322                 mutex_lock(&fs_info->balance_mutex);
4323                 need_unlock = true;
4324                 goto locked;
4325         }
4326
4327         /*
4328          * mut. excl. ops lock is locked.  Three possibilites:
4329          *   (1) some other op is running
4330          *   (2) balance is running
4331          *   (3) balance is paused -- special case (think resume)
4332          */
4333         mutex_lock(&fs_info->balance_mutex);
4334         if (fs_info->balance_ctl) {
4335                 /* this is either (2) or (3) */
4336                 if (!atomic_read(&fs_info->balance_running)) {
4337                         mutex_unlock(&fs_info->balance_mutex);
4338                         if (!mutex_trylock(&fs_info->volume_mutex))
4339                                 goto again;
4340                         mutex_lock(&fs_info->balance_mutex);
4341
4342                         if (fs_info->balance_ctl &&
4343                             !atomic_read(&fs_info->balance_running)) {
4344                                 /* this is (3) */
4345                                 need_unlock = false;
4346                                 goto locked;
4347                         }
4348
4349                         mutex_unlock(&fs_info->balance_mutex);
4350                         mutex_unlock(&fs_info->volume_mutex);
4351                         goto again;
4352                 } else {
4353                         /* this is (2) */
4354                         mutex_unlock(&fs_info->balance_mutex);
4355                         ret = -EINPROGRESS;
4356                         goto out;
4357                 }
4358         } else {
4359                 /* this is (1) */
4360                 mutex_unlock(&fs_info->balance_mutex);
4361                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4362                 goto out;
4363         }
4364
4365 locked:
4366         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4367
4368         if (arg) {
4369                 bargs = memdup_user(arg, sizeof(*bargs));
4370                 if (IS_ERR(bargs)) {
4371                         ret = PTR_ERR(bargs);
4372                         goto out_unlock;
4373                 }
4374
4375                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4376                         if (!fs_info->balance_ctl) {
4377                                 ret = -ENOTCONN;
4378                                 goto out_bargs;
4379                         }
4380
4381                         bctl = fs_info->balance_ctl;
4382                         spin_lock(&fs_info->balance_lock);
4383                         bctl->flags |= BTRFS_BALANCE_RESUME;
4384                         spin_unlock(&fs_info->balance_lock);
4385
4386                         goto do_balance;
4387                 }
4388         } else {
4389                 bargs = NULL;
4390         }
4391
4392         if (fs_info->balance_ctl) {
4393                 ret = -EINPROGRESS;
4394                 goto out_bargs;
4395         }
4396
4397         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4398         if (!bctl) {
4399                 ret = -ENOMEM;
4400                 goto out_bargs;
4401         }
4402
4403         bctl->fs_info = fs_info;
4404         if (arg) {
4405                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4406                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4407                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4408
4409                 bctl->flags = bargs->flags;
4410         } else {
4411                 /* balance everything - no filters */
4412                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4413         }
4414
4415 do_balance:
4416         /*
4417          * Ownership of bctl and mutually_exclusive_operation_running
4418          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4419          * or, if restriper was paused all the way until unmount, in
4420          * free_fs_info.  mutually_exclusive_operation_running is
4421          * cleared in __cancel_balance.
4422          */
4423         need_unlock = false;
4424
4425         ret = btrfs_balance(bctl, bargs);
4426
4427         if (arg) {
4428                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4429                         ret = -EFAULT;
4430         }
4431
4432 out_bargs:
4433         kfree(bargs);
4434 out_unlock:
4435         mutex_unlock(&fs_info->balance_mutex);
4436         mutex_unlock(&fs_info->volume_mutex);
4437         if (need_unlock)
4438                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4439 out:
4440         mnt_drop_write_file(file);
4441         return ret;
4442 }
4443
4444 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4445 {
4446         if (!capable(CAP_SYS_ADMIN))
4447                 return -EPERM;
4448
4449         switch (cmd) {
4450         case BTRFS_BALANCE_CTL_PAUSE:
4451                 return btrfs_pause_balance(root->fs_info);
4452         case BTRFS_BALANCE_CTL_CANCEL:
4453                 return btrfs_cancel_balance(root->fs_info);
4454         }
4455
4456         return -EINVAL;
4457 }
4458
4459 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4460                                          void __user *arg)
4461 {
4462         struct btrfs_fs_info *fs_info = root->fs_info;
4463         struct btrfs_ioctl_balance_args *bargs;
4464         int ret = 0;
4465
4466         if (!capable(CAP_SYS_ADMIN))
4467                 return -EPERM;
4468
4469         mutex_lock(&fs_info->balance_mutex);
4470         if (!fs_info->balance_ctl) {
4471                 ret = -ENOTCONN;
4472                 goto out;
4473         }
4474
4475         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4476         if (!bargs) {
4477                 ret = -ENOMEM;
4478                 goto out;
4479         }
4480
4481         update_ioctl_balance_args(fs_info, 1, bargs);
4482
4483         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4484                 ret = -EFAULT;
4485
4486         kfree(bargs);
4487 out:
4488         mutex_unlock(&fs_info->balance_mutex);
4489         return ret;
4490 }
4491
4492 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4493 {
4494         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4495         struct btrfs_ioctl_quota_ctl_args *sa;
4496         struct btrfs_trans_handle *trans = NULL;
4497         int ret;
4498         int err;
4499
4500         if (!capable(CAP_SYS_ADMIN))
4501                 return -EPERM;
4502
4503         ret = mnt_want_write_file(file);
4504         if (ret)
4505                 return ret;
4506
4507         sa = memdup_user(arg, sizeof(*sa));
4508         if (IS_ERR(sa)) {
4509                 ret = PTR_ERR(sa);
4510                 goto drop_write;
4511         }
4512
4513         down_write(&root->fs_info->subvol_sem);
4514         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4515         if (IS_ERR(trans)) {
4516                 ret = PTR_ERR(trans);
4517                 goto out;
4518         }
4519
4520         switch (sa->cmd) {
4521         case BTRFS_QUOTA_CTL_ENABLE:
4522                 ret = btrfs_quota_enable(trans, root->fs_info);
4523                 break;
4524         case BTRFS_QUOTA_CTL_DISABLE:
4525                 ret = btrfs_quota_disable(trans, root->fs_info);
4526                 break;
4527         default:
4528                 ret = -EINVAL;
4529                 break;
4530         }
4531
4532         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4533         if (err && !ret)
4534                 ret = err;
4535 out:
4536         kfree(sa);
4537         up_write(&root->fs_info->subvol_sem);
4538 drop_write:
4539         mnt_drop_write_file(file);
4540         return ret;
4541 }
4542
4543 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4544 {
4545         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4546         struct btrfs_ioctl_qgroup_assign_args *sa;
4547         struct btrfs_trans_handle *trans;
4548         int ret;
4549         int err;
4550
4551         if (!capable(CAP_SYS_ADMIN))
4552                 return -EPERM;
4553
4554         ret = mnt_want_write_file(file);
4555         if (ret)
4556                 return ret;
4557
4558         sa = memdup_user(arg, sizeof(*sa));
4559         if (IS_ERR(sa)) {
4560                 ret = PTR_ERR(sa);
4561                 goto drop_write;
4562         }
4563
4564         trans = btrfs_join_transaction(root);
4565         if (IS_ERR(trans)) {
4566                 ret = PTR_ERR(trans);
4567                 goto out;
4568         }
4569
4570         /* FIXME: check if the IDs really exist */
4571         if (sa->assign) {
4572                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4573                                                 sa->src, sa->dst);
4574         } else {
4575                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4576                                                 sa->src, sa->dst);
4577         }
4578
4579         err = btrfs_end_transaction(trans, root);
4580         if (err && !ret)
4581                 ret = err;
4582
4583 out:
4584         kfree(sa);
4585 drop_write:
4586         mnt_drop_write_file(file);
4587         return ret;
4588 }
4589
4590 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4591 {
4592         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4593         struct btrfs_ioctl_qgroup_create_args *sa;
4594         struct btrfs_trans_handle *trans;
4595         int ret;
4596         int err;
4597
4598         if (!capable(CAP_SYS_ADMIN))
4599                 return -EPERM;
4600
4601         ret = mnt_want_write_file(file);
4602         if (ret)
4603                 return ret;
4604
4605         sa = memdup_user(arg, sizeof(*sa));
4606         if (IS_ERR(sa)) {
4607                 ret = PTR_ERR(sa);
4608                 goto drop_write;
4609         }
4610
4611         if (!sa->qgroupid) {
4612                 ret = -EINVAL;
4613                 goto out;
4614         }
4615
4616         trans = btrfs_join_transaction(root);
4617         if (IS_ERR(trans)) {
4618                 ret = PTR_ERR(trans);
4619                 goto out;
4620         }
4621
4622         /* FIXME: check if the IDs really exist */
4623         if (sa->create) {
4624                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4625                                           NULL);
4626         } else {
4627                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4628         }
4629
4630         err = btrfs_end_transaction(trans, root);
4631         if (err && !ret)
4632                 ret = err;
4633
4634 out:
4635         kfree(sa);
4636 drop_write:
4637         mnt_drop_write_file(file);
4638         return ret;
4639 }
4640
4641 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4642 {
4643         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4644         struct btrfs_ioctl_qgroup_limit_args *sa;
4645         struct btrfs_trans_handle *trans;
4646         int ret;
4647         int err;
4648         u64 qgroupid;
4649
4650         if (!capable(CAP_SYS_ADMIN))
4651                 return -EPERM;
4652
4653         ret = mnt_want_write_file(file);
4654         if (ret)
4655                 return ret;
4656
4657         sa = memdup_user(arg, sizeof(*sa));
4658         if (IS_ERR(sa)) {
4659                 ret = PTR_ERR(sa);
4660                 goto drop_write;
4661         }
4662
4663         trans = btrfs_join_transaction(root);
4664         if (IS_ERR(trans)) {
4665                 ret = PTR_ERR(trans);
4666                 goto out;
4667         }
4668
4669         qgroupid = sa->qgroupid;
4670         if (!qgroupid) {
4671                 /* take the current subvol as qgroup */
4672                 qgroupid = root->root_key.objectid;
4673         }
4674
4675         /* FIXME: check if the IDs really exist */
4676         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4677
4678         err = btrfs_end_transaction(trans, root);
4679         if (err && !ret)
4680                 ret = err;
4681
4682 out:
4683         kfree(sa);
4684 drop_write:
4685         mnt_drop_write_file(file);
4686         return ret;
4687 }
4688
4689 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4690 {
4691         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4692         struct btrfs_ioctl_quota_rescan_args *qsa;
4693         int ret;
4694
4695         if (!capable(CAP_SYS_ADMIN))
4696                 return -EPERM;
4697
4698         ret = mnt_want_write_file(file);
4699         if (ret)
4700                 return ret;
4701
4702         qsa = memdup_user(arg, sizeof(*qsa));
4703         if (IS_ERR(qsa)) {
4704                 ret = PTR_ERR(qsa);
4705                 goto drop_write;
4706         }
4707
4708         if (qsa->flags) {
4709                 ret = -EINVAL;
4710                 goto out;
4711         }
4712
4713         ret = btrfs_qgroup_rescan(root->fs_info);
4714
4715 out:
4716         kfree(qsa);
4717 drop_write:
4718         mnt_drop_write_file(file);
4719         return ret;
4720 }
4721
4722 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4723 {
4724         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4725         struct btrfs_ioctl_quota_rescan_args *qsa;
4726         int ret = 0;
4727
4728         if (!capable(CAP_SYS_ADMIN))
4729                 return -EPERM;
4730
4731         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4732         if (!qsa)
4733                 return -ENOMEM;
4734
4735         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4736                 qsa->flags = 1;
4737                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4738         }
4739
4740         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4741                 ret = -EFAULT;
4742
4743         kfree(qsa);
4744         return ret;
4745 }
4746
4747 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4748 {
4749         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4750
4751         if (!capable(CAP_SYS_ADMIN))
4752                 return -EPERM;
4753
4754         return btrfs_qgroup_wait_for_completion(root->fs_info);
4755 }
4756
4757 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4758                                             struct btrfs_ioctl_received_subvol_args *sa)
4759 {
4760         struct inode *inode = file_inode(file);
4761         struct btrfs_root *root = BTRFS_I(inode)->root;
4762         struct btrfs_root_item *root_item = &root->root_item;
4763         struct btrfs_trans_handle *trans;
4764         struct timespec ct = CURRENT_TIME;
4765         int ret = 0;
4766         int received_uuid_changed;
4767
4768         if (!inode_owner_or_capable(inode))
4769                 return -EPERM;
4770
4771         ret = mnt_want_write_file(file);
4772         if (ret < 0)
4773                 return ret;
4774
4775         down_write(&root->fs_info->subvol_sem);
4776
4777         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4778                 ret = -EINVAL;
4779                 goto out;
4780         }
4781
4782         if (btrfs_root_readonly(root)) {
4783                 ret = -EROFS;
4784                 goto out;
4785         }
4786
4787         /*
4788          * 1 - root item
4789          * 2 - uuid items (received uuid + subvol uuid)
4790          */
4791         trans = btrfs_start_transaction(root, 3);
4792         if (IS_ERR(trans)) {
4793                 ret = PTR_ERR(trans);
4794                 trans = NULL;
4795                 goto out;
4796         }
4797
4798         sa->rtransid = trans->transid;
4799         sa->rtime.sec = ct.tv_sec;
4800         sa->rtime.nsec = ct.tv_nsec;
4801
4802         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4803                                        BTRFS_UUID_SIZE);
4804         if (received_uuid_changed &&
4805             !btrfs_is_empty_uuid(root_item->received_uuid))
4806                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4807                                     root_item->received_uuid,
4808                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4809                                     root->root_key.objectid);
4810         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4811         btrfs_set_root_stransid(root_item, sa->stransid);
4812         btrfs_set_root_rtransid(root_item, sa->rtransid);
4813         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4814         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4815         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4816         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4817
4818         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4819                                 &root->root_key, &root->root_item);
4820         if (ret < 0) {
4821                 btrfs_end_transaction(trans, root);
4822                 goto out;
4823         }
4824         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4825                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4826                                           sa->uuid,
4827                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4828                                           root->root_key.objectid);
4829                 if (ret < 0 && ret != -EEXIST) {
4830                         btrfs_abort_transaction(trans, root, ret);
4831                         goto out;
4832                 }
4833         }
4834         ret = btrfs_commit_transaction(trans, root);
4835         if (ret < 0) {
4836                 btrfs_abort_transaction(trans, root, ret);
4837                 goto out;
4838         }
4839
4840 out:
4841         up_write(&root->fs_info->subvol_sem);
4842         mnt_drop_write_file(file);
4843         return ret;
4844 }
4845
4846 #ifdef CONFIG_64BIT
4847 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4848                                                 void __user *arg)
4849 {
4850         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4851         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4852         int ret = 0;
4853
4854         args32 = memdup_user(arg, sizeof(*args32));
4855         if (IS_ERR(args32)) {
4856                 ret = PTR_ERR(args32);
4857                 args32 = NULL;
4858                 goto out;
4859         }
4860
4861         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4862         if (!args64) {
4863                 ret = -ENOMEM;
4864                 goto out;
4865         }
4866
4867         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4868         args64->stransid = args32->stransid;
4869         args64->rtransid = args32->rtransid;
4870         args64->stime.sec = args32->stime.sec;
4871         args64->stime.nsec = args32->stime.nsec;
4872         args64->rtime.sec = args32->rtime.sec;
4873         args64->rtime.nsec = args32->rtime.nsec;
4874         args64->flags = args32->flags;
4875
4876         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4877         if (ret)
4878                 goto out;
4879
4880         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4881         args32->stransid = args64->stransid;
4882         args32->rtransid = args64->rtransid;
4883         args32->stime.sec = args64->stime.sec;
4884         args32->stime.nsec = args64->stime.nsec;
4885         args32->rtime.sec = args64->rtime.sec;
4886         args32->rtime.nsec = args64->rtime.nsec;
4887         args32->flags = args64->flags;
4888
4889         ret = copy_to_user(arg, args32, sizeof(*args32));
4890         if (ret)
4891                 ret = -EFAULT;
4892
4893 out:
4894         kfree(args32);
4895         kfree(args64);
4896         return ret;
4897 }
4898 #endif
4899
4900 static long btrfs_ioctl_set_received_subvol(struct file *file,
4901                                             void __user *arg)
4902 {
4903         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4904         int ret = 0;
4905
4906         sa = memdup_user(arg, sizeof(*sa));
4907         if (IS_ERR(sa)) {
4908                 ret = PTR_ERR(sa);
4909                 sa = NULL;
4910                 goto out;
4911         }
4912
4913         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4914
4915         if (ret)
4916                 goto out;
4917
4918         ret = copy_to_user(arg, sa, sizeof(*sa));
4919         if (ret)
4920                 ret = -EFAULT;
4921
4922 out:
4923         kfree(sa);
4924         return ret;
4925 }
4926
4927 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4928 {
4929         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4930         size_t len;
4931         int ret;
4932         char label[BTRFS_LABEL_SIZE];
4933
4934         spin_lock(&root->fs_info->super_lock);
4935         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4936         spin_unlock(&root->fs_info->super_lock);
4937
4938         len = strnlen(label, BTRFS_LABEL_SIZE);
4939
4940         if (len == BTRFS_LABEL_SIZE) {
4941                 btrfs_warn(root->fs_info,
4942                         "label is too long, return the first %zu bytes", --len);
4943         }
4944
4945         ret = copy_to_user(arg, label, len);
4946
4947         return ret ? -EFAULT : 0;
4948 }
4949
4950 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4951 {
4952         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4953         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4954         struct btrfs_trans_handle *trans;
4955         char label[BTRFS_LABEL_SIZE];
4956         int ret;
4957
4958         if (!capable(CAP_SYS_ADMIN))
4959                 return -EPERM;
4960
4961         if (copy_from_user(label, arg, sizeof(label)))
4962                 return -EFAULT;
4963
4964         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4965                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
4966                        BTRFS_LABEL_SIZE - 1);
4967                 return -EINVAL;
4968         }
4969
4970         ret = mnt_want_write_file(file);
4971         if (ret)
4972                 return ret;
4973
4974         trans = btrfs_start_transaction(root, 0);
4975         if (IS_ERR(trans)) {
4976                 ret = PTR_ERR(trans);
4977                 goto out_unlock;
4978         }
4979
4980         spin_lock(&root->fs_info->super_lock);
4981         strcpy(super_block->label, label);
4982         spin_unlock(&root->fs_info->super_lock);
4983         ret = btrfs_commit_transaction(trans, root);
4984
4985 out_unlock:
4986         mnt_drop_write_file(file);
4987         return ret;
4988 }
4989
4990 #define INIT_FEATURE_FLAGS(suffix) \
4991         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4992           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4993           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4994
4995 static int btrfs_ioctl_get_supported_features(struct file *file,
4996                                               void __user *arg)
4997 {
4998         static struct btrfs_ioctl_feature_flags features[3] = {
4999                 INIT_FEATURE_FLAGS(SUPP),
5000                 INIT_FEATURE_FLAGS(SAFE_SET),
5001                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5002         };
5003
5004         if (copy_to_user(arg, &features, sizeof(features)))
5005                 return -EFAULT;
5006
5007         return 0;
5008 }
5009
5010 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5011 {
5012         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5013         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5014         struct btrfs_ioctl_feature_flags features;
5015
5016         features.compat_flags = btrfs_super_compat_flags(super_block);
5017         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5018         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5019
5020         if (copy_to_user(arg, &features, sizeof(features)))
5021                 return -EFAULT;
5022
5023         return 0;
5024 }
5025
5026 static int check_feature_bits(struct btrfs_root *root,
5027                               enum btrfs_feature_set set,
5028                               u64 change_mask, u64 flags, u64 supported_flags,
5029                               u64 safe_set, u64 safe_clear)
5030 {
5031         const char *type = btrfs_feature_set_names[set];
5032         char *names;
5033         u64 disallowed, unsupported;
5034         u64 set_mask = flags & change_mask;
5035         u64 clear_mask = ~flags & change_mask;
5036
5037         unsupported = set_mask & ~supported_flags;
5038         if (unsupported) {
5039                 names = btrfs_printable_features(set, unsupported);
5040                 if (names) {
5041                         btrfs_warn(root->fs_info,
5042                            "this kernel does not support the %s feature bit%s",
5043                            names, strchr(names, ',') ? "s" : "");
5044                         kfree(names);
5045                 } else
5046                         btrfs_warn(root->fs_info,
5047                            "this kernel does not support %s bits 0x%llx",
5048                            type, unsupported);
5049                 return -EOPNOTSUPP;
5050         }
5051
5052         disallowed = set_mask & ~safe_set;
5053         if (disallowed) {
5054                 names = btrfs_printable_features(set, disallowed);
5055                 if (names) {
5056                         btrfs_warn(root->fs_info,
5057                            "can't set the %s feature bit%s while mounted",
5058                            names, strchr(names, ',') ? "s" : "");
5059                         kfree(names);
5060                 } else
5061                         btrfs_warn(root->fs_info,
5062                            "can't set %s bits 0x%llx while mounted",
5063                            type, disallowed);
5064                 return -EPERM;
5065         }
5066
5067         disallowed = clear_mask & ~safe_clear;
5068         if (disallowed) {
5069                 names = btrfs_printable_features(set, disallowed);
5070                 if (names) {
5071                         btrfs_warn(root->fs_info,
5072                            "can't clear the %s feature bit%s while mounted",
5073                            names, strchr(names, ',') ? "s" : "");
5074                         kfree(names);
5075                 } else
5076                         btrfs_warn(root->fs_info,
5077                            "can't clear %s bits 0x%llx while mounted",
5078                            type, disallowed);
5079                 return -EPERM;
5080         }
5081
5082         return 0;
5083 }
5084
5085 #define check_feature(root, change_mask, flags, mask_base)      \
5086 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5087                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5088                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5089                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5090
5091 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5092 {
5093         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5094         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5095         struct btrfs_ioctl_feature_flags flags[2];
5096         struct btrfs_trans_handle *trans;
5097         u64 newflags;
5098         int ret;
5099
5100         if (!capable(CAP_SYS_ADMIN))
5101                 return -EPERM;
5102
5103         if (copy_from_user(flags, arg, sizeof(flags)))
5104                 return -EFAULT;
5105
5106         /* Nothing to do */
5107         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5108             !flags[0].incompat_flags)
5109                 return 0;
5110
5111         ret = check_feature(root, flags[0].compat_flags,
5112                             flags[1].compat_flags, COMPAT);
5113         if (ret)
5114                 return ret;
5115
5116         ret = check_feature(root, flags[0].compat_ro_flags,
5117                             flags[1].compat_ro_flags, COMPAT_RO);
5118         if (ret)
5119                 return ret;
5120
5121         ret = check_feature(root, flags[0].incompat_flags,
5122                             flags[1].incompat_flags, INCOMPAT);
5123         if (ret)
5124                 return ret;
5125
5126         trans = btrfs_start_transaction(root, 0);
5127         if (IS_ERR(trans))
5128                 return PTR_ERR(trans);
5129
5130         spin_lock(&root->fs_info->super_lock);
5131         newflags = btrfs_super_compat_flags(super_block);
5132         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5133         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5134         btrfs_set_super_compat_flags(super_block, newflags);
5135
5136         newflags = btrfs_super_compat_ro_flags(super_block);
5137         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5138         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5139         btrfs_set_super_compat_ro_flags(super_block, newflags);
5140
5141         newflags = btrfs_super_incompat_flags(super_block);
5142         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5143         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5144         btrfs_set_super_incompat_flags(super_block, newflags);
5145         spin_unlock(&root->fs_info->super_lock);
5146
5147         return btrfs_commit_transaction(trans, root);
5148 }
5149
5150 long btrfs_ioctl(struct file *file, unsigned int
5151                 cmd, unsigned long arg)
5152 {
5153         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5154         void __user *argp = (void __user *)arg;
5155
5156         switch (cmd) {
5157         case FS_IOC_GETFLAGS:
5158                 return btrfs_ioctl_getflags(file, argp);
5159         case FS_IOC_SETFLAGS:
5160                 return btrfs_ioctl_setflags(file, argp);
5161         case FS_IOC_GETVERSION:
5162                 return btrfs_ioctl_getversion(file, argp);
5163         case FITRIM:
5164                 return btrfs_ioctl_fitrim(file, argp);
5165         case BTRFS_IOC_SNAP_CREATE:
5166                 return btrfs_ioctl_snap_create(file, argp, 0);
5167         case BTRFS_IOC_SNAP_CREATE_V2:
5168                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5169         case BTRFS_IOC_SUBVOL_CREATE:
5170                 return btrfs_ioctl_snap_create(file, argp, 1);
5171         case BTRFS_IOC_SUBVOL_CREATE_V2:
5172                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5173         case BTRFS_IOC_SNAP_DESTROY:
5174                 return btrfs_ioctl_snap_destroy(file, argp);
5175         case BTRFS_IOC_SUBVOL_GETFLAGS:
5176                 return btrfs_ioctl_subvol_getflags(file, argp);
5177         case BTRFS_IOC_SUBVOL_SETFLAGS:
5178                 return btrfs_ioctl_subvol_setflags(file, argp);
5179         case BTRFS_IOC_DEFAULT_SUBVOL:
5180                 return btrfs_ioctl_default_subvol(file, argp);
5181         case BTRFS_IOC_DEFRAG:
5182                 return btrfs_ioctl_defrag(file, NULL);
5183         case BTRFS_IOC_DEFRAG_RANGE:
5184                 return btrfs_ioctl_defrag(file, argp);
5185         case BTRFS_IOC_RESIZE:
5186                 return btrfs_ioctl_resize(file, argp);
5187         case BTRFS_IOC_ADD_DEV:
5188                 return btrfs_ioctl_add_dev(root, argp);
5189         case BTRFS_IOC_RM_DEV:
5190                 return btrfs_ioctl_rm_dev(file, argp);
5191         case BTRFS_IOC_FS_INFO:
5192                 return btrfs_ioctl_fs_info(root, argp);
5193         case BTRFS_IOC_DEV_INFO:
5194                 return btrfs_ioctl_dev_info(root, argp);
5195         case BTRFS_IOC_BALANCE:
5196                 return btrfs_ioctl_balance(file, NULL);
5197         case BTRFS_IOC_CLONE:
5198                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5199         case BTRFS_IOC_CLONE_RANGE:
5200                 return btrfs_ioctl_clone_range(file, argp);
5201         case BTRFS_IOC_TRANS_START:
5202                 return btrfs_ioctl_trans_start(file);
5203         case BTRFS_IOC_TRANS_END:
5204                 return btrfs_ioctl_trans_end(file);
5205         case BTRFS_IOC_TREE_SEARCH:
5206                 return btrfs_ioctl_tree_search(file, argp);
5207         case BTRFS_IOC_INO_LOOKUP:
5208                 return btrfs_ioctl_ino_lookup(file, argp);
5209         case BTRFS_IOC_INO_PATHS:
5210                 return btrfs_ioctl_ino_to_path(root, argp);
5211         case BTRFS_IOC_LOGICAL_INO:
5212                 return btrfs_ioctl_logical_to_ino(root, argp);
5213         case BTRFS_IOC_SPACE_INFO:
5214                 return btrfs_ioctl_space_info(root, argp);
5215         case BTRFS_IOC_SYNC: {
5216                 int ret;
5217
5218                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5219                 if (ret)
5220                         return ret;
5221                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
5222                 return ret;
5223         }
5224         case BTRFS_IOC_START_SYNC:
5225                 return btrfs_ioctl_start_sync(root, argp);
5226         case BTRFS_IOC_WAIT_SYNC:
5227                 return btrfs_ioctl_wait_sync(root, argp);
5228         case BTRFS_IOC_SCRUB:
5229                 return btrfs_ioctl_scrub(file, argp);
5230         case BTRFS_IOC_SCRUB_CANCEL:
5231                 return btrfs_ioctl_scrub_cancel(root, argp);
5232         case BTRFS_IOC_SCRUB_PROGRESS:
5233                 return btrfs_ioctl_scrub_progress(root, argp);
5234         case BTRFS_IOC_BALANCE_V2:
5235                 return btrfs_ioctl_balance(file, argp);
5236         case BTRFS_IOC_BALANCE_CTL:
5237                 return btrfs_ioctl_balance_ctl(root, arg);
5238         case BTRFS_IOC_BALANCE_PROGRESS:
5239                 return btrfs_ioctl_balance_progress(root, argp);
5240         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5241                 return btrfs_ioctl_set_received_subvol(file, argp);
5242 #ifdef CONFIG_64BIT
5243         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5244                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5245 #endif
5246         case BTRFS_IOC_SEND:
5247                 return btrfs_ioctl_send(file, argp);
5248         case BTRFS_IOC_GET_DEV_STATS:
5249                 return btrfs_ioctl_get_dev_stats(root, argp);
5250         case BTRFS_IOC_QUOTA_CTL:
5251                 return btrfs_ioctl_quota_ctl(file, argp);
5252         case BTRFS_IOC_QGROUP_ASSIGN:
5253                 return btrfs_ioctl_qgroup_assign(file, argp);
5254         case BTRFS_IOC_QGROUP_CREATE:
5255                 return btrfs_ioctl_qgroup_create(file, argp);
5256         case BTRFS_IOC_QGROUP_LIMIT:
5257                 return btrfs_ioctl_qgroup_limit(file, argp);
5258         case BTRFS_IOC_QUOTA_RESCAN:
5259                 return btrfs_ioctl_quota_rescan(file, argp);
5260         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5261                 return btrfs_ioctl_quota_rescan_status(file, argp);
5262         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5263                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5264         case BTRFS_IOC_DEV_REPLACE:
5265                 return btrfs_ioctl_dev_replace(root, argp);
5266         case BTRFS_IOC_GET_FSLABEL:
5267                 return btrfs_ioctl_get_fslabel(file, argp);
5268         case BTRFS_IOC_SET_FSLABEL:
5269                 return btrfs_ioctl_set_fslabel(file, argp);
5270         case BTRFS_IOC_FILE_EXTENT_SAME:
5271                 return btrfs_ioctl_file_extent_same(file, argp);
5272         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5273                 return btrfs_ioctl_get_supported_features(file, argp);
5274         case BTRFS_IOC_GET_FEATURES:
5275                 return btrfs_ioctl_get_features(file, argp);
5276         case BTRFS_IOC_SET_FEATURES:
5277                 return btrfs_ioctl_set_features(file, argp);
5278         }
5279
5280         return -ENOTTY;
5281 }