fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62
  63 #ifdef CONFIG_64BIT
  64 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  65  * structures are incorrect, as the timespec structure from userspace
  66  * is 4 bytes too small. We define these alternatives here to teach
  67  * the kernel about the 32-bit struct packing.
  68  */
  69 struct btrfs_ioctl_timespec_32 {
  70         __u64 sec;
  71         __u32 nsec;
  72 } __attribute__ ((__packed__));
  73
  74 struct btrfs_ioctl_received_subvol_args_32 {
  75         char    uuid[BTRFS_UUID_SIZE];  /* in */
  76         __u64   stransid;               /* in */
  77         __u64   rtransid;               /* out */
  78         struct btrfs_ioctl_timespec_32 stime; /* in */
  79         struct btrfs_ioctl_timespec_32 rtime; /* out */
  80         __u64   flags;                  /* in */
  81         __u64   reserved[16];           /* in */
  82 } __attribute__ ((__packed__));
  83
  84 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  85                                 struct btrfs_ioctl_received_subvol_args_32)
  86 #endif
  87
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  91
  92 /* Mask out flags that are inappropriate for the given type of inode. */
  93 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  94 {
  95         if (S_ISDIR(mode))
  96                 return flags;
  97         else if (S_ISREG(mode))
  98                 return flags & ~FS_DIRSYNC_FL;
  99         else
 100                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 101 }
 102
 103 /*
 104  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 105  */
 106 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 126                 iflags |= FS_COMPR_FL;
 127         else if (flags & BTRFS_INODE_NOCOMPRESS)
 128                 iflags |= FS_NOCOMP_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_update_iflags(struct inode *inode)
 137 {
 138         struct btrfs_inode *ip = BTRFS_I(inode);
 139         unsigned int new_fl = 0;
 140
 141         if (ip->flags & BTRFS_INODE_SYNC)
 142                 new_fl |= S_SYNC;
 143         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 144                 new_fl |= S_IMMUTABLE;
 145         if (ip->flags & BTRFS_INODE_APPEND)
 146                 new_fl |= S_APPEND;
 147         if (ip->flags & BTRFS_INODE_NOATIME)
 148                 new_fl |= S_NOATIME;
 149         if (ip->flags & BTRFS_INODE_DIRSYNC)
 150                 new_fl |= S_DIRSYNC;
 151
 152         set_mask_bits(&inode->i_flags,
 153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 154                       new_fl);
 155 }
 156
 157 /*
 158  * Inherit flags from the parent inode.
 159  *
 160  * Currently only the compression flags and the cow flags are inherited.
 161  */
 162 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 163 {
 164         unsigned int flags;
 165
 166         if (!dir)
 167                 return;
 168
 169         flags = BTRFS_I(dir)->flags;
 170
 171         if (flags & BTRFS_INODE_NOCOMPRESS) {
 172                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 173                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 174         } else if (flags & BTRFS_INODE_COMPRESS) {
 175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 177         }
 178
 179         if (flags & BTRFS_INODE_NODATACOW) {
 180                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 181                 if (S_ISREG(inode->i_mode))
 182                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 183         }
 184
 185         btrfs_update_iflags(inode);
 186 }
 187
 188 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 189 {
 190         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 191         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 192
 193         if (copy_to_user(arg, &flags, sizeof(flags)))
 194                 return -EFAULT;
 195         return 0;
 196 }
 197
 198 static int check_flags(unsigned int flags)
 199 {
 200         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 201                       FS_NOATIME_FL | FS_NODUMP_FL | \
 202                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 203                       FS_NOCOMP_FL | FS_COMPR_FL |
 204                       FS_NOCOW_FL))
 205                 return -EOPNOTSUPP;
 206
 207         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 208                 return -EINVAL;
 209
 210         return 0;
 211 }
 212
 213 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 214 {
 215         struct inode *inode = file_inode(file);
 216         struct btrfs_inode *ip = BTRFS_I(inode);
 217         struct btrfs_root *root = ip->root;
 218         struct btrfs_trans_handle *trans;
 219         unsigned int flags, oldflags;
 220         int ret;
 221         u64 ip_oldflags;
 222         unsigned int i_oldflags;
 223         umode_t mode;
 224
 225         if (!inode_owner_or_capable(inode))
 226                 return -EPERM;
 227
 228         if (btrfs_root_readonly(root))
 229                 return -EROFS;
 230
 231         if (copy_from_user(&flags, arg, sizeof(flags)))
 232                 return -EFAULT;
 233
 234         ret = check_flags(flags);
 235         if (ret)
 236                 return ret;
 237
 238         ret = mnt_want_write_file(file);
 239         if (ret)
 240                 return ret;
 241
 242         mutex_lock(&inode->i_mutex);
 243
 244         ip_oldflags = ip->flags;
 245         i_oldflags = inode->i_flags;
 246         mode = inode->i_mode;
 247
 248         flags = btrfs_mask_flags(inode->i_mode, flags);
 249         oldflags = btrfs_flags_to_ioctl(ip->flags);
 250         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 251                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 252                         ret = -EPERM;
 253                         goto out_unlock;
 254                 }
 255         }
 256
 257         if (flags & FS_SYNC_FL)
 258                 ip->flags |= BTRFS_INODE_SYNC;
 259         else
 260                 ip->flags &= ~BTRFS_INODE_SYNC;
 261         if (flags & FS_IMMUTABLE_FL)
 262                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 263         else
 264                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 265         if (flags & FS_APPEND_FL)
 266                 ip->flags |= BTRFS_INODE_APPEND;
 267         else
 268                 ip->flags &= ~BTRFS_INODE_APPEND;
 269         if (flags & FS_NODUMP_FL)
 270                 ip->flags |= BTRFS_INODE_NODUMP;
 271         else
 272                 ip->flags &= ~BTRFS_INODE_NODUMP;
 273         if (flags & FS_NOATIME_FL)
 274                 ip->flags |= BTRFS_INODE_NOATIME;
 275         else
 276                 ip->flags &= ~BTRFS_INODE_NOATIME;
 277         if (flags & FS_DIRSYNC_FL)
 278                 ip->flags |= BTRFS_INODE_DIRSYNC;
 279         else
 280                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 281         if (flags & FS_NOCOW_FL) {
 282                 if (S_ISREG(mode)) {
 283                         /*
 284                          * It's safe to turn csums off here, no extents exist.
 285                          * Otherwise we want the flag to reflect the real COW
 286                          * status of the file and will not set it.
 287                          */
 288                         if (inode->i_size == 0)
 289                                 ip->flags |= BTRFS_INODE_NODATACOW
 290                                            | BTRFS_INODE_NODATASUM;
 291                 } else {
 292                         ip->flags |= BTRFS_INODE_NODATACOW;
 293                 }
 294         } else {
 295                 /*
 296                  * Revert back under same assuptions as above
 297                  */
 298                 if (S_ISREG(mode)) {
 299                         if (inode->i_size == 0)
 300                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 301                                              | BTRFS_INODE_NODATASUM);
 302                 } else {
 303                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 304                 }
 305         }
 306
 307         /*
 308          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 309          * flag may be changed automatically if compression code won't make
 310          * things smaller.
 311          */
 312         if (flags & FS_NOCOMP_FL) {
 313                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 314                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 315
 316                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 317                 if (ret && ret != -ENODATA)
 318                         goto out_drop;
 319         } else if (flags & FS_COMPR_FL) {
 320                 const char *comp;
 321
 322                 ip->flags |= BTRFS_INODE_COMPRESS;
 323                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 324
 325                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 326                         comp = "lzo";
 327                 else
 328                         comp = "zlib";
 329                 ret = btrfs_set_prop(inode, "btrfs.compression",
 330                                      comp, strlen(comp), 0);
 331                 if (ret)
 332                         goto out_drop;
 333
 334         } else {
 335                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 336                 if (ret && ret != -ENODATA)
 337                         goto out_drop;
 338                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 339         }
 340
 341         trans = btrfs_start_transaction(root, 1);
 342         if (IS_ERR(trans)) {
 343                 ret = PTR_ERR(trans);
 344                 goto out_drop;
 345         }
 346
 347         btrfs_update_iflags(inode);
 348         inode_inc_iversion(inode);
 349         inode->i_ctime = CURRENT_TIME;
 350         ret = btrfs_update_inode(trans, root, inode);
 351
 352         btrfs_end_transaction(trans, root);
 353  out_drop:
 354         if (ret) {
 355                 ip->flags = ip_oldflags;
 356                 inode->i_flags = i_oldflags;
 357         }
 358
 359  out_unlock:
 360         mutex_unlock(&inode->i_mutex);
 361         mnt_drop_write_file(file);
 362         return ret;
 363 }
 364
 365 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 366 {
 367         struct inode *inode = file_inode(file);
 368
 369         return put_user(inode->i_generation, arg);
 370 }
 371
 372 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 373 {
 374         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 375         struct btrfs_device *device;
 376         struct request_queue *q;
 377         struct fstrim_range range;
 378         u64 minlen = ULLONG_MAX;
 379         u64 num_devices = 0;
 380         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 381         int ret;
 382
 383         if (!capable(CAP_SYS_ADMIN))
 384                 return -EPERM;
 385
 386         rcu_read_lock();
 387         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 388                                 dev_list) {
 389                 if (!device->bdev)
 390                         continue;
 391                 q = bdev_get_queue(device->bdev);
 392                 if (blk_queue_discard(q)) {
 393                         num_devices++;
 394                         minlen = min((u64)q->limits.discard_granularity,
 395                                      minlen);
 396                 }
 397         }
 398         rcu_read_unlock();
 399
 400         if (!num_devices)
 401                 return -EOPNOTSUPP;
 402         if (copy_from_user(&range, arg, sizeof(range)))
 403                 return -EFAULT;
 404         if (range.start > total_bytes ||
 405             range.len < fs_info->sb->s_blocksize)
 406                 return -EINVAL;
 407
 408         range.len = min(range.len, total_bytes - range.start);
 409         range.minlen = max(range.minlen, minlen);
 410         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 411         if (ret < 0)
 412                 return ret;
 413
 414         if (copy_to_user(arg, &range, sizeof(range)))
 415                 return -EFAULT;
 416
 417         return 0;
 418 }
 419
 420 int btrfs_is_empty_uuid(u8 *uuid)
 421 {
 422         int i;
 423
 424         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 425                 if (uuid[i])
 426                         return 0;
 427         }
 428         return 1;
 429 }
 430
 431 static noinline int create_subvol(struct inode *dir,
 432                                   struct dentry *dentry,
 433                                   char *name, int namelen,
 434                                   u64 *async_transid,
 435                                   struct btrfs_qgroup_inherit *inherit)
 436 {
 437         struct btrfs_trans_handle *trans;
 438         struct btrfs_key key;
 439         struct btrfs_root_item root_item;
 440         struct btrfs_inode_item *inode_item;
 441         struct extent_buffer *leaf;
 442         struct btrfs_root *root = BTRFS_I(dir)->root;
 443         struct btrfs_root *new_root;
 444         struct btrfs_block_rsv block_rsv;
 445         struct timespec cur_time = CURRENT_TIME;
 446         struct inode *inode;
 447         int ret;
 448         int err;
 449         u64 objectid;
 450         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 451         u64 index = 0;
 452         u64 qgroup_reserved;
 453         uuid_le new_uuid;
 454
 455         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 456         if (ret)
 457                 return ret;
 458
 459         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 460         /*
 461          * The same as the snapshot creation, please see the comment
 462          * of create_snapshot().
 463          */
 464         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 465                                                8, &qgroup_reserved, false);
 466         if (ret)
 467                 return ret;
 468
 469         trans = btrfs_start_transaction(root, 0);
 470         if (IS_ERR(trans)) {
 471                 ret = PTR_ERR(trans);
 472                 btrfs_subvolume_release_metadata(root, &block_rsv,
 473                                                  qgroup_reserved);
 474                 return ret;
 475         }
 476         trans->block_rsv = &block_rsv;
 477         trans->bytes_reserved = block_rsv.size;
 478
 479         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 480         if (ret)
 481                 goto fail;
 482
 483         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 484         if (IS_ERR(leaf)) {
 485                 ret = PTR_ERR(leaf);
 486                 goto fail;
 487         }
 488
 489         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 490         btrfs_set_header_bytenr(leaf, leaf->start);
 491         btrfs_set_header_generation(leaf, trans->transid);
 492         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 493         btrfs_set_header_owner(leaf, objectid);
 494
 495         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 496                             BTRFS_FSID_SIZE);
 497         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 498                             btrfs_header_chunk_tree_uuid(leaf),
 499                             BTRFS_UUID_SIZE);
 500         btrfs_mark_buffer_dirty(leaf);
 501
 502         memset(&root_item, 0, sizeof(root_item));
 503
 504         inode_item = &root_item.inode;
 505         btrfs_set_stack_inode_generation(inode_item, 1);
 506         btrfs_set_stack_inode_size(inode_item, 3);
 507         btrfs_set_stack_inode_nlink(inode_item, 1);
 508         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 509         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 510
 511         btrfs_set_root_flags(&root_item, 0);
 512         btrfs_set_root_limit(&root_item, 0);
 513         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 514
 515         btrfs_set_root_bytenr(&root_item, leaf->start);
 516         btrfs_set_root_generation(&root_item, trans->transid);
 517         btrfs_set_root_level(&root_item, 0);
 518         btrfs_set_root_refs(&root_item, 1);
 519         btrfs_set_root_used(&root_item, leaf->len);
 520         btrfs_set_root_last_snapshot(&root_item, 0);
 521
 522         btrfs_set_root_generation_v2(&root_item,
 523                         btrfs_root_generation(&root_item));
 524         uuid_le_gen(&new_uuid);
 525         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 526         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 527         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 528         root_item.ctime = root_item.otime;
 529         btrfs_set_root_ctransid(&root_item, trans->transid);
 530         btrfs_set_root_otransid(&root_item, trans->transid);
 531
 532         btrfs_tree_unlock(leaf);
 533         free_extent_buffer(leaf);
 534         leaf = NULL;
 535
 536         btrfs_set_root_dirid(&root_item, new_dirid);
 537
 538         key.objectid = objectid;
 539         key.offset = 0;
 540         key.type = BTRFS_ROOT_ITEM_KEY;
 541         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 542                                 &root_item);
 543         if (ret)
 544                 goto fail;
 545
 546         key.offset = (u64)-1;
 547         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 548         if (IS_ERR(new_root)) {
 549                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 550                 ret = PTR_ERR(new_root);
 551                 goto fail;
 552         }
 553
 554         btrfs_record_root_in_trans(trans, new_root);
 555
 556         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 557         if (ret) {
 558                 /* We potentially lose an unused inode item here */
 559                 btrfs_abort_transaction(trans, root, ret);
 560                 goto fail;
 561         }
 562
 563         /*
 564          * insert the directory item
 565          */
 566         ret = btrfs_set_inode_index(dir, &index);
 567         if (ret) {
 568                 btrfs_abort_transaction(trans, root, ret);
 569                 goto fail;
 570         }
 571
 572         ret = btrfs_insert_dir_item(trans, root,
 573                                     name, namelen, dir, &key,
 574                                     BTRFS_FT_DIR, index);
 575         if (ret) {
 576                 btrfs_abort_transaction(trans, root, ret);
 577                 goto fail;
 578         }
 579
 580         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 581         ret = btrfs_update_inode(trans, root, dir);
 582         BUG_ON(ret);
 583
 584         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 585                                  objectid, root->root_key.objectid,
 586                                  btrfs_ino(dir), index, name, namelen);
 587         BUG_ON(ret);
 588
 589         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 590                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 591                                   objectid);
 592         if (ret)
 593                 btrfs_abort_transaction(trans, root, ret);
 594
 595 fail:
 596         trans->block_rsv = NULL;
 597         trans->bytes_reserved = 0;
 598         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 599
 600         if (async_transid) {
 601                 *async_transid = trans->transid;
 602                 err = btrfs_commit_transaction_async(trans, root, 1);
 603                 if (err)
 604                         err = btrfs_commit_transaction(trans, root);
 605         } else {
 606                 err = btrfs_commit_transaction(trans, root);
 607         }
 608         if (err && !ret)
 609                 ret = err;
 610
 611         if (!ret) {
 612                 inode = btrfs_lookup_dentry(dir, dentry);
 613                 if (IS_ERR(inode))
 614                         return PTR_ERR(inode);
 615                 d_instantiate(dentry, inode);
 616         }
 617         return ret;
 618 }
 619
 620 static void btrfs_wait_nocow_write(struct btrfs_root *root)
 621 {
 622         s64 writers;
 623         DEFINE_WAIT(wait);
 624
 625         do {
 626                 prepare_to_wait(&root->subv_writers->wait, &wait,
 627                                 TASK_UNINTERRUPTIBLE);
 628
 629                 writers = percpu_counter_sum(&root->subv_writers->counter);
 630                 if (writers)
 631                         schedule();
 632
 633                 finish_wait(&root->subv_writers->wait, &wait);
 634         } while (writers);
 635 }
 636
 637 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 638                            struct dentry *dentry, char *name, int namelen,
 639                            u64 *async_transid, bool readonly,
 640                            struct btrfs_qgroup_inherit *inherit)
 641 {
 642         struct inode *inode;
 643         struct btrfs_pending_snapshot *pending_snapshot;
 644         struct btrfs_trans_handle *trans;
 645         int ret;
 646
 647         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 648                 return -EINVAL;
 649
 650         atomic_inc(&root->will_be_snapshoted);
 651         smp_mb__after_atomic();
 652         btrfs_wait_nocow_write(root);
 653
 654         ret = btrfs_start_delalloc_inodes(root, 0);
 655         if (ret)
 656                 goto out;
 657
 658         btrfs_wait_ordered_extents(root, -1);
 659
 660         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 661         if (!pending_snapshot) {
 662                 ret = -ENOMEM;
 663                 goto out;
 664         }
 665
 666         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 667                              BTRFS_BLOCK_RSV_TEMP);
 668         /*
 669          * 1 - parent dir inode
 670          * 2 - dir entries
 671          * 1 - root item
 672          * 2 - root ref/backref
 673          * 1 - root of snapshot
 674          * 1 - UUID item
 675          */
 676         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 677                                         &pending_snapshot->block_rsv, 8,
 678                                         &pending_snapshot->qgroup_reserved,
 679                                         false);
 680         if (ret)
 681                 goto free;
 682
 683         pending_snapshot->dentry = dentry;
 684         pending_snapshot->root = root;
 685         pending_snapshot->readonly = readonly;
 686         pending_snapshot->dir = dir;
 687         pending_snapshot->inherit = inherit;
 688
 689         trans = btrfs_start_transaction(root, 0);
 690         if (IS_ERR(trans)) {
 691                 ret = PTR_ERR(trans);
 692                 goto fail;
 693         }
 694
 695         spin_lock(&root->fs_info->trans_lock);
 696         list_add(&pending_snapshot->list,
 697                  &trans->transaction->pending_snapshots);
 698         spin_unlock(&root->fs_info->trans_lock);
 699         if (async_transid) {
 700                 *async_transid = trans->transid;
 701                 ret = btrfs_commit_transaction_async(trans,
 702                                      root->fs_info->extent_root, 1);
 703                 if (ret)
 704                         ret = btrfs_commit_transaction(trans, root);
 705         } else {
 706                 ret = btrfs_commit_transaction(trans,
 707                                                root->fs_info->extent_root);
 708         }
 709         if (ret)
 710                 goto fail;
 711
 712         ret = pending_snapshot->error;
 713         if (ret)
 714                 goto fail;
 715
 716         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 717         if (IS_ERR(inode)) {
 718                 ret = PTR_ERR(inode);
 719                 goto fail;
 720         }
 721
 722         d_instantiate(dentry, inode);
 723         ret = 0;
 724 fail:
 725         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 726                                          &pending_snapshot->block_rsv,
 727                                          pending_snapshot->qgroup_reserved);
 728 free:
 729         kfree(pending_snapshot);
 730 out:
 731         atomic_dec(&root->will_be_snapshoted);
 732         return ret;
 733 }
 734
 735 /*  copy of check_sticky in fs/namei.c()
 736 * It's inline, so penalty for filesystems that don't use sticky bit is
 737 * minimal.
 738 */
 739 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 740 {
 741         kuid_t fsuid = current_fsuid();
 742
 743         if (!(dir->i_mode & S_ISVTX))
 744                 return 0;
 745         if (uid_eq(inode->i_uid, fsuid))
 746                 return 0;
 747         if (uid_eq(dir->i_uid, fsuid))
 748                 return 0;
 749         return !capable(CAP_FOWNER);
 750 }
 751
 752 /*  copy of may_delete in fs/namei.c()
 753  *      Check whether we can remove a link victim from directory dir, check
 754  *  whether the type of victim is right.
 755  *  1. We can't do it if dir is read-only (done in permission())
 756  *  2. We should have write and exec permissions on dir
 757  *  3. We can't remove anything from append-only dir
 758  *  4. We can't do anything with immutable dir (done in permission())
 759  *  5. If the sticky bit on dir is set we should either
 760  *      a. be owner of dir, or
 761  *      b. be owner of victim, or
 762  *      c. have CAP_FOWNER capability
 763  *  6. If the victim is append-only or immutable we can't do antyhing with
 764  *     links pointing to it.
 765  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 766  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 767  *  9. We can't remove a root or mountpoint.
 768  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 769  *     nfs_async_unlink().
 770  */
 771
 772 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 773 {
 774         int error;
 775
 776         if (!victim->d_inode)
 777                 return -ENOENT;
 778
 779         BUG_ON(victim->d_parent->d_inode != dir);
 780         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 781
 782         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 783         if (error)
 784                 return error;
 785         if (IS_APPEND(dir))
 786                 return -EPERM;
 787         if (btrfs_check_sticky(dir, victim->d_inode)||
 788                 IS_APPEND(victim->d_inode)||
 789             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 790                 return -EPERM;
 791         if (isdir) {
 792                 if (!S_ISDIR(victim->d_inode->i_mode))
 793                         return -ENOTDIR;
 794                 if (IS_ROOT(victim))
 795                         return -EBUSY;
 796         } else if (S_ISDIR(victim->d_inode->i_mode))
 797                 return -EISDIR;
 798         if (IS_DEADDIR(dir))
 799                 return -ENOENT;
 800         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 801                 return -EBUSY;
 802         return 0;
 803 }
 804
 805 /* copy of may_create in fs/namei.c() */
 806 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 807 {
 808         if (child->d_inode)
 809                 return -EEXIST;
 810         if (IS_DEADDIR(dir))
 811                 return -ENOENT;
 812         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 813 }
 814
 815 /*
 816  * Create a new subvolume below @parent.  This is largely modeled after
 817  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 818  * inside this filesystem so it's quite a bit simpler.
 819  */
 820 static noinline int btrfs_mksubvol(struct path *parent,
 821                                    char *name, int namelen,
 822                                    struct btrfs_root *snap_src,
 823                                    u64 *async_transid, bool readonly,
 824                                    struct btrfs_qgroup_inherit *inherit)
 825 {
 826         struct inode *dir  = parent->dentry->d_inode;
 827         struct dentry *dentry;
 828         int error;
 829
 830         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 831         if (error == -EINTR)
 832                 return error;
 833
 834         dentry = lookup_one_len(name, parent->dentry, namelen);
 835         error = PTR_ERR(dentry);
 836         if (IS_ERR(dentry))
 837                 goto out_unlock;
 838
 839         error = -EEXIST;
 840         if (dentry->d_inode)
 841                 goto out_dput;
 842
 843         error = btrfs_may_create(dir, dentry);
 844         if (error)
 845                 goto out_dput;
 846
 847         /*
 848          * even if this name doesn't exist, we may get hash collisions.
 849          * check for them now when we can safely fail
 850          */
 851         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 852                                                dir->i_ino, name,
 853                                                namelen);
 854         if (error)
 855                 goto out_dput;
 856
 857         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 858
 859         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 860                 goto out_up_read;
 861
 862         if (snap_src) {
 863                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 864                                         async_transid, readonly, inherit);
 865         } else {
 866                 error = create_subvol(dir, dentry, name, namelen,
 867                                       async_transid, inherit);
 868         }
 869         if (!error)
 870                 fsnotify_mkdir(dir, dentry);
 871 out_up_read:
 872         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 873 out_dput:
 874         dput(dentry);
 875 out_unlock:
 876         mutex_unlock(&dir->i_mutex);
 877         return error;
 878 }
 879
 880 /*
 881  * When we're defragging a range, we don't want to kick it off again
 882  * if it is really just waiting for delalloc to send it down.
 883  * If we find a nice big extent or delalloc range for the bytes in the
 884  * file you want to defrag, we return 0 to let you know to skip this
 885  * part of the file
 886  */
 887 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 888 {
 889         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 890         struct extent_map *em = NULL;
 891         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 892         u64 end;
 893
 894         read_lock(&em_tree->lock);
 895         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 896         read_unlock(&em_tree->lock);
 897
 898         if (em) {
 899                 end = extent_map_end(em);
 900                 free_extent_map(em);
 901                 if (end - offset > thresh)
 902                         return 0;
 903         }
 904         /* if we already have a nice delalloc here, just stop */
 905         thresh /= 2;
 906         end = count_range_bits(io_tree, &offset, offset + thresh,
 907                                thresh, EXTENT_DELALLOC, 1);
 908         if (end >= thresh)
 909                 return 0;
 910         return 1;
 911 }
 912
 913 /*
 914  * helper function to walk through a file and find extents
 915  * newer than a specific transid, and smaller than thresh.
 916  *
 917  * This is used by the defragging code to find new and small
 918  * extents
 919  */
 920 static int find_new_extents(struct btrfs_root *root,
 921                             struct inode *inode, u64 newer_than,
 922                             u64 *off, u32 thresh)
 923 {
 924         struct btrfs_path *path;
 925         struct btrfs_key min_key;
 926         struct extent_buffer *leaf;
 927         struct btrfs_file_extent_item *extent;
 928         int type;
 929         int ret;
 930         u64 ino = btrfs_ino(inode);
 931
 932         path = btrfs_alloc_path();
 933         if (!path)
 934                 return -ENOMEM;
 935
 936         min_key.objectid = ino;
 937         min_key.type = BTRFS_EXTENT_DATA_KEY;
 938         min_key.offset = *off;
 939
 940         while (1) {
 941                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 942                 if (ret != 0)
 943                         goto none;
 944 process_slot:
 945                 if (min_key.objectid != ino)
 946                         goto none;
 947                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 948                         goto none;
 949
 950                 leaf = path->nodes[0];
 951                 extent = btrfs_item_ptr(leaf, path->slots[0],
 952                                         struct btrfs_file_extent_item);
 953
 954                 type = btrfs_file_extent_type(leaf, extent);
 955                 if (type == BTRFS_FILE_EXTENT_REG &&
 956                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 957                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 958                         *off = min_key.offset;
 959                         btrfs_free_path(path);
 960                         return 0;
 961                 }
 962
 963                 path->slots[0]++;
 964                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 965                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 966                         goto process_slot;
 967                 }
 968
 969                 if (min_key.offset == (u64)-1)
 970                         goto none;
 971
 972                 min_key.offset++;
 973                 btrfs_release_path(path);
 974         }
 975 none:
 976         btrfs_free_path(path);
 977         return -ENOENT;
 978 }
 979
 980 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 981 {
 982         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 983         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 984         struct extent_map *em;
 985         u64 len = PAGE_CACHE_SIZE;
 986
 987         /*
 988          * hopefully we have this extent in the tree already, try without
 989          * the full extent lock
 990          */
 991         read_lock(&em_tree->lock);
 992         em = lookup_extent_mapping(em_tree, start, len);
 993         read_unlock(&em_tree->lock);
 994
 995         if (!em) {
 996                 struct extent_state *cached = NULL;
 997                 u64 end = start + len - 1;
 998
 999                 /* get the big lock and read metadata off disk */
1000                 lock_extent_bits(io_tree, start, end, 0, &cached);
1001                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1002                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1003
1004                 if (IS_ERR(em))
1005                         return NULL;
1006         }
1007
1008         return em;
1009 }
1010
1011 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1012 {
1013         struct extent_map *next;
1014         bool ret = true;
1015
1016         /* this is the last extent */
1017         if (em->start + em->len >= i_size_read(inode))
1018                 return false;
1019
1020         next = defrag_lookup_extent(inode, em->start + em->len);
1021         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1022                 ret = false;
1023         else if ((em->block_start + em->block_len == next->block_start) &&
1024                  (em->block_len > 128 * 1024 && next->block_len > 128 * 1024))
1025                 ret = false;
1026
1027         free_extent_map(next);
1028         return ret;
1029 }
1030
1031 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1032                                u64 *last_len, u64 *skip, u64 *defrag_end,
1033                                int compress)
1034 {
1035         struct extent_map *em;
1036         int ret = 1;
1037         bool next_mergeable = true;
1038
1039         /*
1040          * make sure that once we start defragging an extent, we keep on
1041          * defragging it
1042          */
1043         if (start < *defrag_end)
1044                 return 1;
1045
1046         *skip = 0;
1047
1048         em = defrag_lookup_extent(inode, start);
1049         if (!em)
1050                 return 0;
1051
1052         /* this will cover holes, and inline extents */
1053         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1054                 ret = 0;
1055                 goto out;
1056         }
1057
1058         next_mergeable = defrag_check_next_extent(inode, em);
1059         /*
1060          * we hit a real extent, if it is big or the next extent is not a
1061          * real extent, don't bother defragging it
1062          */
1063         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1064             (em->len >= thresh || !next_mergeable))
1065                 ret = 0;
1066 out:
1067         /*
1068          * last_len ends up being a counter of how many bytes we've defragged.
1069          * every time we choose not to defrag an extent, we reset *last_len
1070          * so that the next tiny extent will force a defrag.
1071          *
1072          * The end result of this is that tiny extents before a single big
1073          * extent will force at least part of that big extent to be defragged.
1074          */
1075         if (ret) {
1076                 *defrag_end = extent_map_end(em);
1077         } else {
1078                 *last_len = 0;
1079                 *skip = extent_map_end(em);
1080                 *defrag_end = 0;
1081         }
1082
1083         free_extent_map(em);
1084         return ret;
1085 }
1086
1087 /*
1088  * it doesn't do much good to defrag one or two pages
1089  * at a time.  This pulls in a nice chunk of pages
1090  * to COW and defrag.
1091  *
1092  * It also makes sure the delalloc code has enough
1093  * dirty data to avoid making new small extents as part
1094  * of the defrag
1095  *
1096  * It's a good idea to start RA on this range
1097  * before calling this.
1098  */
1099 static int cluster_pages_for_defrag(struct inode *inode,
1100                                     struct page **pages,
1101                                     unsigned long start_index,
1102                                     unsigned long num_pages)
1103 {
1104         unsigned long file_end;
1105         u64 isize = i_size_read(inode);
1106         u64 page_start;
1107         u64 page_end;
1108         u64 page_cnt;
1109         int ret;
1110         int i;
1111         int i_done;
1112         struct btrfs_ordered_extent *ordered;
1113         struct extent_state *cached_state = NULL;
1114         struct extent_io_tree *tree;
1115         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1116
1117         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1118         if (!isize || start_index > file_end)
1119                 return 0;
1120
1121         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1122
1123         ret = btrfs_delalloc_reserve_space(inode,
1124                                            page_cnt << PAGE_CACHE_SHIFT);
1125         if (ret)
1126                 return ret;
1127         i_done = 0;
1128         tree = &BTRFS_I(inode)->io_tree;
1129
1130         /* step one, lock all the pages */
1131         for (i = 0; i < page_cnt; i++) {
1132                 struct page *page;
1133 again:
1134                 page = find_or_create_page(inode->i_mapping,
1135                                            start_index + i, mask);
1136                 if (!page)
1137                         break;
1138
1139                 page_start = page_offset(page);
1140                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1141                 while (1) {
1142                         lock_extent_bits(tree, page_start, page_end,
1143                                          0, &cached_state);
1144                         ordered = btrfs_lookup_ordered_extent(inode,
1145                                                               page_start);
1146                         unlock_extent_cached(tree, page_start, page_end,
1147                                              &cached_state, GFP_NOFS);
1148                         if (!ordered)
1149                                 break;
1150
1151                         unlock_page(page);
1152                         btrfs_start_ordered_extent(inode, ordered, 1);
1153                         btrfs_put_ordered_extent(ordered);
1154                         lock_page(page);
1155                         /*
1156                          * we unlocked the page above, so we need check if
1157                          * it was released or not.
1158                          */
1159                         if (page->mapping != inode->i_mapping) {
1160                                 unlock_page(page);
1161                                 page_cache_release(page);
1162                                 goto again;
1163                         }
1164                 }
1165
1166                 if (!PageUptodate(page)) {
1167                         btrfs_readpage(NULL, page);
1168                         lock_page(page);
1169                         if (!PageUptodate(page)) {
1170                                 unlock_page(page);
1171                                 page_cache_release(page);
1172                                 ret = -EIO;
1173                                 break;
1174                         }
1175                 }
1176
1177                 if (page->mapping != inode->i_mapping) {
1178                         unlock_page(page);
1179                         page_cache_release(page);
1180                         goto again;
1181                 }
1182
1183                 pages[i] = page;
1184                 i_done++;
1185         }
1186         if (!i_done || ret)
1187                 goto out;
1188
1189         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1190                 goto out;
1191
1192         /*
1193          * so now we have a nice long stream of locked
1194          * and up to date pages, lets wait on them
1195          */
1196         for (i = 0; i < i_done; i++)
1197                 wait_on_page_writeback(pages[i]);
1198
1199         page_start = page_offset(pages[0]);
1200         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1201
1202         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1203                          page_start, page_end - 1, 0, &cached_state);
1204         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1205                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1206                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1207                           &cached_state, GFP_NOFS);
1208
1209         if (i_done != page_cnt) {
1210                 spin_lock(&BTRFS_I(inode)->lock);
1211                 BTRFS_I(inode)->outstanding_extents++;
1212                 spin_unlock(&BTRFS_I(inode)->lock);
1213                 btrfs_delalloc_release_space(inode,
1214                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1215         }
1216
1217
1218         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1219                           &cached_state, GFP_NOFS);
1220
1221         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1222                              page_start, page_end - 1, &cached_state,
1223                              GFP_NOFS);
1224
1225         for (i = 0; i < i_done; i++) {
1226                 clear_page_dirty_for_io(pages[i]);
1227                 ClearPageChecked(pages[i]);
1228                 set_page_extent_mapped(pages[i]);
1229                 set_page_dirty(pages[i]);
1230                 unlock_page(pages[i]);
1231                 page_cache_release(pages[i]);
1232         }
1233         return i_done;
1234 out:
1235         for (i = 0; i < i_done; i++) {
1236                 unlock_page(pages[i]);
1237                 page_cache_release(pages[i]);
1238         }
1239         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1240         return ret;
1241
1242 }
1243
1244 int btrfs_defrag_file(struct inode *inode, struct file *file,
1245                       struct btrfs_ioctl_defrag_range_args *range,
1246                       u64 newer_than, unsigned long max_to_defrag)
1247 {
1248         struct btrfs_root *root = BTRFS_I(inode)->root;
1249         struct file_ra_state *ra = NULL;
1250         unsigned long last_index;
1251         u64 isize = i_size_read(inode);
1252         u64 last_len = 0;
1253         u64 skip = 0;
1254         u64 defrag_end = 0;
1255         u64 newer_off = range->start;
1256         unsigned long i;
1257         unsigned long ra_index = 0;
1258         int ret;
1259         int defrag_count = 0;
1260         int compress_type = BTRFS_COMPRESS_ZLIB;
1261         u32 extent_thresh = range->extent_thresh;
1262         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1263         unsigned long cluster = max_cluster;
1264         u64 new_align = ~((u64)128 * 1024 - 1);
1265         struct page **pages = NULL;
1266
1267         if (isize == 0)
1268                 return 0;
1269
1270         if (range->start >= isize)
1271                 return -EINVAL;
1272
1273         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1274                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1275                         return -EINVAL;
1276                 if (range->compress_type)
1277                         compress_type = range->compress_type;
1278         }
1279
1280         if (extent_thresh == 0)
1281                 extent_thresh = 256 * 1024;
1282
1283         /*
1284          * if we were not given a file, allocate a readahead
1285          * context
1286          */
1287         if (!file) {
1288                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1289                 if (!ra)
1290                         return -ENOMEM;
1291                 file_ra_state_init(ra, inode->i_mapping);
1292         } else {
1293                 ra = &file->f_ra;
1294         }
1295
1296         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1297                         GFP_NOFS);
1298         if (!pages) {
1299                 ret = -ENOMEM;
1300                 goto out_ra;
1301         }
1302
1303         /* find the last page to defrag */
1304         if (range->start + range->len > range->start) {
1305                 last_index = min_t(u64, isize - 1,
1306                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1307         } else {
1308                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1309         }
1310
1311         if (newer_than) {
1312                 ret = find_new_extents(root, inode, newer_than,
1313                                        &newer_off, 64 * 1024);
1314                 if (!ret) {
1315                         range->start = newer_off;
1316                         /*
1317                          * we always align our defrag to help keep
1318                          * the extents in the file evenly spaced
1319                          */
1320                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1321                 } else
1322                         goto out_ra;
1323         } else {
1324                 i = range->start >> PAGE_CACHE_SHIFT;
1325         }
1326         if (!max_to_defrag)
1327                 max_to_defrag = last_index + 1;
1328
1329         /*
1330          * make writeback starts from i, so the defrag range can be
1331          * written sequentially.
1332          */
1333         if (i < inode->i_mapping->writeback_index)
1334                 inode->i_mapping->writeback_index = i;
1335
1336         while (i <= last_index && defrag_count < max_to_defrag &&
1337                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1338                 /*
1339                  * make sure we stop running if someone unmounts
1340                  * the FS
1341                  */
1342                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1343                         break;
1344
1345                 if (btrfs_defrag_cancelled(root->fs_info)) {
1346                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1347                         ret = -EAGAIN;
1348                         break;
1349                 }
1350
1351                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1352                                          extent_thresh, &last_len, &skip,
1353                                          &defrag_end, range->flags &
1354                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1355                         unsigned long next;
1356                         /*
1357                          * the should_defrag function tells us how much to skip
1358                          * bump our counter by the suggested amount
1359                          */
1360                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1361                         i = max(i + 1, next);
1362                         continue;
1363                 }
1364
1365                 if (!newer_than) {
1366                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1367                                    PAGE_CACHE_SHIFT) - i;
1368                         cluster = min(cluster, max_cluster);
1369                 } else {
1370                         cluster = max_cluster;
1371                 }
1372
1373                 if (i + cluster > ra_index) {
1374                         ra_index = max(i, ra_index);
1375                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1376                                        cluster);
1377                         ra_index += max_cluster;
1378                 }
1379
1380                 mutex_lock(&inode->i_mutex);
1381                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1382                         BTRFS_I(inode)->force_compress = compress_type;
1383                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1384                 if (ret < 0) {
1385                         mutex_unlock(&inode->i_mutex);
1386                         goto out_ra;
1387                 }
1388
1389                 defrag_count += ret;
1390                 balance_dirty_pages_ratelimited(inode->i_mapping);
1391                 mutex_unlock(&inode->i_mutex);
1392
1393                 if (newer_than) {
1394                         if (newer_off == (u64)-1)
1395                                 break;
1396
1397                         if (ret > 0)
1398                                 i += ret;
1399
1400                         newer_off = max(newer_off + 1,
1401                                         (u64)i << PAGE_CACHE_SHIFT);
1402
1403                         ret = find_new_extents(root, inode,
1404                                                newer_than, &newer_off,
1405                                                64 * 1024);
1406                         if (!ret) {
1407                                 range->start = newer_off;
1408                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1409                         } else {
1410                                 break;
1411                         }
1412                 } else {
1413                         if (ret > 0) {
1414                                 i += ret;
1415                                 last_len += ret << PAGE_CACHE_SHIFT;
1416                         } else {
1417                                 i++;
1418                                 last_len = 0;
1419                         }
1420                 }
1421         }
1422
1423         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1424                 filemap_flush(inode->i_mapping);
1425                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1426                              &BTRFS_I(inode)->runtime_flags))
1427                         filemap_flush(inode->i_mapping);
1428         }
1429
1430         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1431                 /* the filemap_flush will queue IO into the worker threads, but
1432                  * we have to make sure the IO is actually started and that
1433                  * ordered extents get created before we return
1434                  */
1435                 atomic_inc(&root->fs_info->async_submit_draining);
1436                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1437                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1438                         wait_event(root->fs_info->async_submit_wait,
1439                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1440                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1441                 }
1442                 atomic_dec(&root->fs_info->async_submit_draining);
1443         }
1444
1445         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1446                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1447         }
1448
1449         ret = defrag_count;
1450
1451 out_ra:
1452         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1453                 mutex_lock(&inode->i_mutex);
1454                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1455                 mutex_unlock(&inode->i_mutex);
1456         }
1457         if (!file)
1458                 kfree(ra);
1459         kfree(pages);
1460         return ret;
1461 }
1462
1463 static noinline int btrfs_ioctl_resize(struct file *file,
1464                                         void __user *arg)
1465 {
1466         u64 new_size;
1467         u64 old_size;
1468         u64 devid = 1;
1469         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1470         struct btrfs_ioctl_vol_args *vol_args;
1471         struct btrfs_trans_handle *trans;
1472         struct btrfs_device *device = NULL;
1473         char *sizestr;
1474         char *retptr;
1475         char *devstr = NULL;
1476         int ret = 0;
1477         int mod = 0;
1478
1479         if (!capable(CAP_SYS_ADMIN))
1480                 return -EPERM;
1481
1482         ret = mnt_want_write_file(file);
1483         if (ret)
1484                 return ret;
1485
1486         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1487                         1)) {
1488                 mnt_drop_write_file(file);
1489                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1490         }
1491
1492         mutex_lock(&root->fs_info->volume_mutex);
1493         vol_args = memdup_user(arg, sizeof(*vol_args));
1494         if (IS_ERR(vol_args)) {
1495                 ret = PTR_ERR(vol_args);
1496                 goto out;
1497         }
1498
1499         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1500
1501         sizestr = vol_args->name;
1502         devstr = strchr(sizestr, ':');
1503         if (devstr) {
1504                 sizestr = devstr + 1;
1505                 *devstr = '\0';
1506                 devstr = vol_args->name;
1507                 ret = kstrtoull(devstr, 10, &devid);
1508                 if (ret)
1509                         goto out_free;
1510                 if (!devid) {
1511                         ret = -EINVAL;
1512                         goto out_free;
1513                 }
1514                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1515         }
1516
1517         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1518         if (!device) {
1519                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1520                        devid);
1521                 ret = -ENODEV;
1522                 goto out_free;
1523         }
1524
1525         if (!device->writeable) {
1526                 btrfs_info(root->fs_info,
1527                            "resizer unable to apply on readonly device %llu",
1528                        devid);
1529                 ret = -EPERM;
1530                 goto out_free;
1531         }
1532
1533         if (!strcmp(sizestr, "max"))
1534                 new_size = device->bdev->bd_inode->i_size;
1535         else {
1536                 if (sizestr[0] == '-') {
1537                         mod = -1;
1538                         sizestr++;
1539                 } else if (sizestr[0] == '+') {
1540                         mod = 1;
1541                         sizestr++;
1542                 }
1543                 new_size = memparse(sizestr, &retptr);
1544                 if (*retptr != '\0' || new_size == 0) {
1545                         ret = -EINVAL;
1546                         goto out_free;
1547                 }
1548         }
1549
1550         if (device->is_tgtdev_for_dev_replace) {
1551                 ret = -EPERM;
1552                 goto out_free;
1553         }
1554
1555         old_size = btrfs_device_get_total_bytes(device);
1556
1557         if (mod < 0) {
1558                 if (new_size > old_size) {
1559                         ret = -EINVAL;
1560                         goto out_free;
1561                 }
1562                 new_size = old_size - new_size;
1563         } else if (mod > 0) {
1564                 if (new_size > ULLONG_MAX - old_size) {
1565                         ret = -ERANGE;
1566                         goto out_free;
1567                 }
1568                 new_size = old_size + new_size;
1569         }
1570
1571         if (new_size < 256 * 1024 * 1024) {
1572                 ret = -EINVAL;
1573                 goto out_free;
1574         }
1575         if (new_size > device->bdev->bd_inode->i_size) {
1576                 ret = -EFBIG;
1577                 goto out_free;
1578         }
1579
1580         do_div(new_size, root->sectorsize);
1581         new_size *= root->sectorsize;
1582
1583         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1584                       rcu_str_deref(device->name), new_size);
1585
1586         if (new_size > old_size) {
1587                 trans = btrfs_start_transaction(root, 0);
1588                 if (IS_ERR(trans)) {
1589                         ret = PTR_ERR(trans);
1590                         goto out_free;
1591                 }
1592                 ret = btrfs_grow_device(trans, device, new_size);
1593                 btrfs_commit_transaction(trans, root);
1594         } else if (new_size < old_size) {
1595                 ret = btrfs_shrink_device(device, new_size);
1596         } /* equal, nothing need to do */
1597
1598 out_free:
1599         kfree(vol_args);
1600 out:
1601         mutex_unlock(&root->fs_info->volume_mutex);
1602         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1603         mnt_drop_write_file(file);
1604         return ret;
1605 }
1606
1607 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1608                                 char *name, unsigned long fd, int subvol,
1609                                 u64 *transid, bool readonly,
1610                                 struct btrfs_qgroup_inherit *inherit)
1611 {
1612         int namelen;
1613         int ret = 0;
1614
1615         ret = mnt_want_write_file(file);
1616         if (ret)
1617                 goto out;
1618
1619         namelen = strlen(name);
1620         if (strchr(name, '/')) {
1621                 ret = -EINVAL;
1622                 goto out_drop_write;
1623         }
1624
1625         if (name[0] == '.' &&
1626            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1627                 ret = -EEXIST;
1628                 goto out_drop_write;
1629         }
1630
1631         if (subvol) {
1632                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1633                                      NULL, transid, readonly, inherit);
1634         } else {
1635                 struct fd src = fdget(fd);
1636                 struct inode *src_inode;
1637                 if (!src.file) {
1638                         ret = -EINVAL;
1639                         goto out_drop_write;
1640                 }
1641
1642                 src_inode = file_inode(src.file);
1643                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1644                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1645                                    "Snapshot src from another FS");
1646                         ret = -EXDEV;
1647                 } else if (!inode_owner_or_capable(src_inode)) {
1648                         /*
1649                          * Subvolume creation is not restricted, but snapshots
1650                          * are limited to own subvolumes only
1651                          */
1652                         ret = -EPERM;
1653                 } else {
1654                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1655                                              BTRFS_I(src_inode)->root,
1656                                              transid, readonly, inherit);
1657                 }
1658                 fdput(src);
1659         }
1660 out_drop_write:
1661         mnt_drop_write_file(file);
1662 out:
1663         return ret;
1664 }
1665
1666 static noinline int btrfs_ioctl_snap_create(struct file *file,
1667                                             void __user *arg, int subvol)
1668 {
1669         struct btrfs_ioctl_vol_args *vol_args;
1670         int ret;
1671
1672         vol_args = memdup_user(arg, sizeof(*vol_args));
1673         if (IS_ERR(vol_args))
1674                 return PTR_ERR(vol_args);
1675         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1676
1677         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1678                                               vol_args->fd, subvol,
1679                                               NULL, false, NULL);
1680
1681         kfree(vol_args);
1682         return ret;
1683 }
1684
1685 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1686                                                void __user *arg, int subvol)
1687 {
1688         struct btrfs_ioctl_vol_args_v2 *vol_args;
1689         int ret;
1690         u64 transid = 0;
1691         u64 *ptr = NULL;
1692         bool readonly = false;
1693         struct btrfs_qgroup_inherit *inherit = NULL;
1694
1695         vol_args = memdup_user(arg, sizeof(*vol_args));
1696         if (IS_ERR(vol_args))
1697                 return PTR_ERR(vol_args);
1698         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1699
1700         if (vol_args->flags &
1701             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1702               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1703                 ret = -EOPNOTSUPP;
1704                 goto free_args;
1705         }
1706
1707         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1708                 ptr = &transid;
1709         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1710                 readonly = true;
1711         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1712                 if (vol_args->size > PAGE_CACHE_SIZE) {
1713                         ret = -EINVAL;
1714                         goto free_args;
1715                 }
1716                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1717                 if (IS_ERR(inherit)) {
1718                         ret = PTR_ERR(inherit);
1719                         goto free_args;
1720                 }
1721         }
1722
1723         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1724                                               vol_args->fd, subvol, ptr,
1725                                               readonly, inherit);
1726         if (ret)
1727                 goto free_inherit;
1728
1729         if (ptr && copy_to_user(arg +
1730                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1731                                         transid),
1732                                 ptr, sizeof(*ptr)))
1733                 ret = -EFAULT;
1734
1735 free_inherit:
1736         kfree(inherit);
1737 free_args:
1738         kfree(vol_args);
1739         return ret;
1740 }
1741
1742 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1743                                                 void __user *arg)
1744 {
1745         struct inode *inode = file_inode(file);
1746         struct btrfs_root *root = BTRFS_I(inode)->root;
1747         int ret = 0;
1748         u64 flags = 0;
1749
1750         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1751                 return -EINVAL;
1752
1753         down_read(&root->fs_info->subvol_sem);
1754         if (btrfs_root_readonly(root))
1755                 flags |= BTRFS_SUBVOL_RDONLY;
1756         up_read(&root->fs_info->subvol_sem);
1757
1758         if (copy_to_user(arg, &flags, sizeof(flags)))
1759                 ret = -EFAULT;
1760
1761         return ret;
1762 }
1763
1764 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1765                                               void __user *arg)
1766 {
1767         struct inode *inode = file_inode(file);
1768         struct btrfs_root *root = BTRFS_I(inode)->root;
1769         struct btrfs_trans_handle *trans;
1770         u64 root_flags;
1771         u64 flags;
1772         int ret = 0;
1773
1774         if (!inode_owner_or_capable(inode))
1775                 return -EPERM;
1776
1777         ret = mnt_want_write_file(file);
1778         if (ret)
1779                 goto out;
1780
1781         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1782                 ret = -EINVAL;
1783                 goto out_drop_write;
1784         }
1785
1786         if (copy_from_user(&flags, arg, sizeof(flags))) {
1787                 ret = -EFAULT;
1788                 goto out_drop_write;
1789         }
1790
1791         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1792                 ret = -EINVAL;
1793                 goto out_drop_write;
1794         }
1795
1796         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1797                 ret = -EOPNOTSUPP;
1798                 goto out_drop_write;
1799         }
1800
1801         down_write(&root->fs_info->subvol_sem);
1802
1803         /* nothing to do */
1804         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1805                 goto out_drop_sem;
1806
1807         root_flags = btrfs_root_flags(&root->root_item);
1808         if (flags & BTRFS_SUBVOL_RDONLY) {
1809                 btrfs_set_root_flags(&root->root_item,
1810                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1811         } else {
1812                 /*
1813                  * Block RO -> RW transition if this subvolume is involved in
1814                  * send
1815                  */
1816                 spin_lock(&root->root_item_lock);
1817                 if (root->send_in_progress == 0) {
1818                         btrfs_set_root_flags(&root->root_item,
1819                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1820                         spin_unlock(&root->root_item_lock);
1821                 } else {
1822                         spin_unlock(&root->root_item_lock);
1823                         btrfs_warn(root->fs_info,
1824                         "Attempt to set subvolume %llu read-write during send",
1825                                         root->root_key.objectid);
1826                         ret = -EPERM;
1827                         goto out_drop_sem;
1828                 }
1829         }
1830
1831         trans = btrfs_start_transaction(root, 1);
1832         if (IS_ERR(trans)) {
1833                 ret = PTR_ERR(trans);
1834                 goto out_reset;
1835         }
1836
1837         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1838                                 &root->root_key, &root->root_item);
1839
1840         btrfs_commit_transaction(trans, root);
1841 out_reset:
1842         if (ret)
1843                 btrfs_set_root_flags(&root->root_item, root_flags);
1844 out_drop_sem:
1845         up_write(&root->fs_info->subvol_sem);
1846 out_drop_write:
1847         mnt_drop_write_file(file);
1848 out:
1849         return ret;
1850 }
1851
1852 /*
1853  * helper to check if the subvolume references other subvolumes
1854  */
1855 static noinline int may_destroy_subvol(struct btrfs_root *root)
1856 {
1857         struct btrfs_path *path;
1858         struct btrfs_dir_item *di;
1859         struct btrfs_key key;
1860         u64 dir_id;
1861         int ret;
1862
1863         path = btrfs_alloc_path();
1864         if (!path)
1865                 return -ENOMEM;
1866
1867         /* Make sure this root isn't set as the default subvol */
1868         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1869         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1870                                    dir_id, "default", 7, 0);
1871         if (di && !IS_ERR(di)) {
1872                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1873                 if (key.objectid == root->root_key.objectid) {
1874                         ret = -EPERM;
1875                         btrfs_err(root->fs_info, "deleting default subvolume "
1876                                   "%llu is not allowed", key.objectid);
1877                         goto out;
1878                 }
1879                 btrfs_release_path(path);
1880         }
1881
1882         key.objectid = root->root_key.objectid;
1883         key.type = BTRFS_ROOT_REF_KEY;
1884         key.offset = (u64)-1;
1885
1886         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1887                                 &key, path, 0, 0);
1888         if (ret < 0)
1889                 goto out;
1890         BUG_ON(ret == 0);
1891
1892         ret = 0;
1893         if (path->slots[0] > 0) {
1894                 path->slots[0]--;
1895                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1896                 if (key.objectid == root->root_key.objectid &&
1897                     key.type == BTRFS_ROOT_REF_KEY)
1898                         ret = -ENOTEMPTY;
1899         }
1900 out:
1901         btrfs_free_path(path);
1902         return ret;
1903 }
1904
1905 static noinline int key_in_sk(struct btrfs_key *key,
1906                               struct btrfs_ioctl_search_key *sk)
1907 {
1908         struct btrfs_key test;
1909         int ret;
1910
1911         test.objectid = sk->min_objectid;
1912         test.type = sk->min_type;
1913         test.offset = sk->min_offset;
1914
1915         ret = btrfs_comp_cpu_keys(key, &test);
1916         if (ret < 0)
1917                 return 0;
1918
1919         test.objectid = sk->max_objectid;
1920         test.type = sk->max_type;
1921         test.offset = sk->max_offset;
1922
1923         ret = btrfs_comp_cpu_keys(key, &test);
1924         if (ret > 0)
1925                 return 0;
1926         return 1;
1927 }
1928
1929 static noinline int copy_to_sk(struct btrfs_root *root,
1930                                struct btrfs_path *path,
1931                                struct btrfs_key *key,
1932                                struct btrfs_ioctl_search_key *sk,
1933                                size_t *buf_size,
1934                                char __user *ubuf,
1935                                unsigned long *sk_offset,
1936                                int *num_found)
1937 {
1938         u64 found_transid;
1939         struct extent_buffer *leaf;
1940         struct btrfs_ioctl_search_header sh;
1941         unsigned long item_off;
1942         unsigned long item_len;
1943         int nritems;
1944         int i;
1945         int slot;
1946         int ret = 0;
1947
1948         leaf = path->nodes[0];
1949         slot = path->slots[0];
1950         nritems = btrfs_header_nritems(leaf);
1951
1952         if (btrfs_header_generation(leaf) > sk->max_transid) {
1953                 i = nritems;
1954                 goto advance_key;
1955         }
1956         found_transid = btrfs_header_generation(leaf);
1957
1958         for (i = slot; i < nritems; i++) {
1959                 item_off = btrfs_item_ptr_offset(leaf, i);
1960                 item_len = btrfs_item_size_nr(leaf, i);
1961
1962                 btrfs_item_key_to_cpu(leaf, key, i);
1963                 if (!key_in_sk(key, sk))
1964                         continue;
1965
1966                 if (sizeof(sh) + item_len > *buf_size) {
1967                         if (*num_found) {
1968                                 ret = 1;
1969                                 goto out;
1970                         }
1971
1972                         /*
1973                          * return one empty item back for v1, which does not
1974                          * handle -EOVERFLOW
1975                          */
1976
1977                         *buf_size = sizeof(sh) + item_len;
1978                         item_len = 0;
1979                         ret = -EOVERFLOW;
1980                 }
1981
1982                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1983                         ret = 1;
1984                         goto out;
1985                 }
1986
1987                 sh.objectid = key->objectid;
1988                 sh.offset = key->offset;
1989                 sh.type = key->type;
1990                 sh.len = item_len;
1991                 sh.transid = found_transid;
1992
1993                 /* copy search result header */
1994                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1995                         ret = -EFAULT;
1996                         goto out;
1997                 }
1998
1999                 *sk_offset += sizeof(sh);
2000
2001                 if (item_len) {
2002                         char __user *up = ubuf + *sk_offset;
2003                         /* copy the item */
2004                         if (read_extent_buffer_to_user(leaf, up,
2005                                                        item_off, item_len)) {
2006                                 ret = -EFAULT;
2007                                 goto out;
2008                         }
2009
2010                         *sk_offset += item_len;
2011                 }
2012                 (*num_found)++;
2013
2014                 if (ret) /* -EOVERFLOW from above */
2015                         goto out;
2016
2017                 if (*num_found >= sk->nr_items) {
2018                         ret = 1;
2019                         goto out;
2020                 }
2021         }
2022 advance_key:
2023         ret = 0;
2024         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2025                 key->offset++;
2026         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2027                 key->offset = 0;
2028                 key->type++;
2029         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2030                 key->offset = 0;
2031                 key->type = 0;
2032                 key->objectid++;
2033         } else
2034                 ret = 1;
2035 out:
2036         /*
2037          *  0: all items from this leaf copied, continue with next
2038          *  1: * more items can be copied, but unused buffer is too small
2039          *     * all items were found
2040          *     Either way, it will stops the loop which iterates to the next
2041          *     leaf
2042          *  -EOVERFLOW: item was to large for buffer
2043          *  -EFAULT: could not copy extent buffer back to userspace
2044          */
2045         return ret;
2046 }
2047
2048 static noinline int search_ioctl(struct inode *inode,
2049                                  struct btrfs_ioctl_search_key *sk,
2050                                  size_t *buf_size,
2051                                  char __user *ubuf)
2052 {
2053         struct btrfs_root *root;
2054         struct btrfs_key key;
2055         struct btrfs_path *path;
2056         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2057         int ret;
2058         int num_found = 0;
2059         unsigned long sk_offset = 0;
2060
2061         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2062                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2063                 return -EOVERFLOW;
2064         }
2065
2066         path = btrfs_alloc_path();
2067         if (!path)
2068                 return -ENOMEM;
2069
2070         if (sk->tree_id == 0) {
2071                 /* search the root of the inode that was passed */
2072                 root = BTRFS_I(inode)->root;
2073         } else {
2074                 key.objectid = sk->tree_id;
2075                 key.type = BTRFS_ROOT_ITEM_KEY;
2076                 key.offset = (u64)-1;
2077                 root = btrfs_read_fs_root_no_name(info, &key);
2078                 if (IS_ERR(root)) {
2079                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2080                                sk->tree_id);
2081                         btrfs_free_path(path);
2082                         return -ENOENT;
2083                 }
2084         }
2085
2086         key.objectid = sk->min_objectid;
2087         key.type = sk->min_type;
2088         key.offset = sk->min_offset;
2089
2090         while (1) {
2091                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2092                 if (ret != 0) {
2093                         if (ret > 0)
2094                                 ret = 0;
2095                         goto err;
2096                 }
2097                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2098                                  &sk_offset, &num_found);
2099                 btrfs_release_path(path);
2100                 if (ret)
2101                         break;
2102
2103         }
2104         if (ret > 0)
2105                 ret = 0;
2106 err:
2107         sk->nr_items = num_found;
2108         btrfs_free_path(path);
2109         return ret;
2110 }
2111
2112 static noinline int btrfs_ioctl_tree_search(struct file *file,
2113                                            void __user *argp)
2114 {
2115         struct btrfs_ioctl_search_args __user *uargs;
2116         struct btrfs_ioctl_search_key sk;
2117         struct inode *inode;
2118         int ret;
2119         size_t buf_size;
2120
2121         if (!capable(CAP_SYS_ADMIN))
2122                 return -EPERM;
2123
2124         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2125
2126         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2127                 return -EFAULT;
2128
2129         buf_size = sizeof(uargs->buf);
2130
2131         inode = file_inode(file);
2132         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2133
2134         /*
2135          * In the origin implementation an overflow is handled by returning a
2136          * search header with a len of zero, so reset ret.
2137          */
2138         if (ret == -EOVERFLOW)
2139                 ret = 0;
2140
2141         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2142                 ret = -EFAULT;
2143         return ret;
2144 }
2145
2146 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2147                                                void __user *argp)
2148 {
2149         struct btrfs_ioctl_search_args_v2 __user *uarg;
2150         struct btrfs_ioctl_search_args_v2 args;
2151         struct inode *inode;
2152         int ret;
2153         size_t buf_size;
2154         const size_t buf_limit = 16 * 1024 * 1024;
2155
2156         if (!capable(CAP_SYS_ADMIN))
2157                 return -EPERM;
2158
2159         /* copy search header and buffer size */
2160         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2161         if (copy_from_user(&args, uarg, sizeof(args)))
2162                 return -EFAULT;
2163
2164         buf_size = args.buf_size;
2165
2166         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2167                 return -EOVERFLOW;
2168
2169         /* limit result size to 16MB */
2170         if (buf_size > buf_limit)
2171                 buf_size = buf_limit;
2172
2173         inode = file_inode(file);
2174         ret = search_ioctl(inode, &args.key, &buf_size,
2175                            (char *)(&uarg->buf[0]));
2176         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2177                 ret = -EFAULT;
2178         else if (ret == -EOVERFLOW &&
2179                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2180                 ret = -EFAULT;
2181
2182         return ret;
2183 }
2184
2185 /*
2186  * Search INODE_REFs to identify path name of 'dirid' directory
2187  * in a 'tree_id' tree. and sets path name to 'name'.
2188  */
2189 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2190                                 u64 tree_id, u64 dirid, char *name)
2191 {
2192         struct btrfs_root *root;
2193         struct btrfs_key key;
2194         char *ptr;
2195         int ret = -1;
2196         int slot;
2197         int len;
2198         int total_len = 0;
2199         struct btrfs_inode_ref *iref;
2200         struct extent_buffer *l;
2201         struct btrfs_path *path;
2202
2203         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2204                 name[0]='\0';
2205                 return 0;
2206         }
2207
2208         path = btrfs_alloc_path();
2209         if (!path)
2210                 return -ENOMEM;
2211
2212         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2213
2214         key.objectid = tree_id;
2215         key.type = BTRFS_ROOT_ITEM_KEY;
2216         key.offset = (u64)-1;
2217         root = btrfs_read_fs_root_no_name(info, &key);
2218         if (IS_ERR(root)) {
2219                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2220                 ret = -ENOENT;
2221                 goto out;
2222         }
2223
2224         key.objectid = dirid;
2225         key.type = BTRFS_INODE_REF_KEY;
2226         key.offset = (u64)-1;
2227
2228         while (1) {
2229                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2230                 if (ret < 0)
2231                         goto out;
2232                 else if (ret > 0) {
2233                         ret = btrfs_previous_item(root, path, dirid,
2234                                                   BTRFS_INODE_REF_KEY);
2235                         if (ret < 0)
2236                                 goto out;
2237                         else if (ret > 0) {
2238                                 ret = -ENOENT;
2239                                 goto out;
2240                         }
2241                 }
2242
2243                 l = path->nodes[0];
2244                 slot = path->slots[0];
2245                 btrfs_item_key_to_cpu(l, &key, slot);
2246
2247                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2248                 len = btrfs_inode_ref_name_len(l, iref);
2249                 ptr -= len + 1;
2250                 total_len += len + 1;
2251                 if (ptr < name) {
2252                         ret = -ENAMETOOLONG;
2253                         goto out;
2254                 }
2255
2256                 *(ptr + len) = '/';
2257                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2258
2259                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2260                         break;
2261
2262                 btrfs_release_path(path);
2263                 key.objectid = key.offset;
2264                 key.offset = (u64)-1;
2265                 dirid = key.objectid;
2266         }
2267         memmove(name, ptr, total_len);
2268         name[total_len] = '\0';
2269         ret = 0;
2270 out:
2271         btrfs_free_path(path);
2272         return ret;
2273 }
2274
2275 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2276                                            void __user *argp)
2277 {
2278          struct btrfs_ioctl_ino_lookup_args *args;
2279          struct inode *inode;
2280          int ret;
2281
2282         if (!capable(CAP_SYS_ADMIN))
2283                 return -EPERM;
2284
2285         args = memdup_user(argp, sizeof(*args));
2286         if (IS_ERR(args))
2287                 return PTR_ERR(args);
2288
2289         inode = file_inode(file);
2290
2291         if (args->treeid == 0)
2292                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2293
2294         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2295                                         args->treeid, args->objectid,
2296                                         args->name);
2297
2298         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2299                 ret = -EFAULT;
2300
2301         kfree(args);
2302         return ret;
2303 }
2304
2305 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2306                                              void __user *arg)
2307 {
2308         struct dentry *parent = file->f_path.dentry;
2309         struct dentry *dentry;
2310         struct inode *dir = parent->d_inode;
2311         struct inode *inode;
2312         struct btrfs_root *root = BTRFS_I(dir)->root;
2313         struct btrfs_root *dest = NULL;
2314         struct btrfs_ioctl_vol_args *vol_args;
2315         struct btrfs_trans_handle *trans;
2316         struct btrfs_block_rsv block_rsv;
2317         u64 root_flags;
2318         u64 qgroup_reserved;
2319         int namelen;
2320         int ret;
2321         int err = 0;
2322
2323         vol_args = memdup_user(arg, sizeof(*vol_args));
2324         if (IS_ERR(vol_args))
2325                 return PTR_ERR(vol_args);
2326
2327         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2328         namelen = strlen(vol_args->name);
2329         if (strchr(vol_args->name, '/') ||
2330             strncmp(vol_args->name, "..", namelen) == 0) {
2331                 err = -EINVAL;
2332                 goto out;
2333         }
2334
2335         err = mnt_want_write_file(file);
2336         if (err)
2337                 goto out;
2338
2339
2340         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2341         if (err == -EINTR)
2342                 goto out_drop_write;
2343         dentry = lookup_one_len(vol_args->name, parent, namelen);
2344         if (IS_ERR(dentry)) {
2345                 err = PTR_ERR(dentry);
2346                 goto out_unlock_dir;
2347         }
2348
2349         if (!dentry->d_inode) {
2350                 err = -ENOENT;
2351                 goto out_dput;
2352         }
2353
2354         inode = dentry->d_inode;
2355         dest = BTRFS_I(inode)->root;
2356         if (!capable(CAP_SYS_ADMIN)) {
2357                 /*
2358                  * Regular user.  Only allow this with a special mount
2359                  * option, when the user has write+exec access to the
2360                  * subvol root, and when rmdir(2) would have been
2361                  * allowed.
2362                  *
2363                  * Note that this is _not_ check that the subvol is
2364                  * empty or doesn't contain data that we wouldn't
2365                  * otherwise be able to delete.
2366                  *
2367                  * Users who want to delete empty subvols should try
2368                  * rmdir(2).
2369                  */
2370                 err = -EPERM;
2371                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2372                         goto out_dput;
2373
2374                 /*
2375                  * Do not allow deletion if the parent dir is the same
2376                  * as the dir to be deleted.  That means the ioctl
2377                  * must be called on the dentry referencing the root
2378                  * of the subvol, not a random directory contained
2379                  * within it.
2380                  */
2381                 err = -EINVAL;
2382                 if (root == dest)
2383                         goto out_dput;
2384
2385                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2386                 if (err)
2387                         goto out_dput;
2388         }
2389
2390         /* check if subvolume may be deleted by a user */
2391         err = btrfs_may_delete(dir, dentry, 1);
2392         if (err)
2393                 goto out_dput;
2394
2395         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2396                 err = -EINVAL;
2397                 goto out_dput;
2398         }
2399
2400         mutex_lock(&inode->i_mutex);
2401
2402         /*
2403          * Don't allow to delete a subvolume with send in progress. This is
2404          * inside the i_mutex so the error handling that has to drop the bit
2405          * again is not run concurrently.
2406          */
2407         spin_lock(&dest->root_item_lock);
2408         root_flags = btrfs_root_flags(&dest->root_item);
2409         if (dest->send_in_progress == 0) {
2410                 btrfs_set_root_flags(&dest->root_item,
2411                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2412                 spin_unlock(&dest->root_item_lock);
2413         } else {
2414                 spin_unlock(&dest->root_item_lock);
2415                 btrfs_warn(root->fs_info,
2416                         "Attempt to delete subvolume %llu during send",
2417                         dest->root_key.objectid);
2418                 err = -EPERM;
2419                 goto out_dput;
2420         }
2421
2422         d_invalidate(dentry);
2423
2424         down_write(&root->fs_info->subvol_sem);
2425
2426         err = may_destroy_subvol(dest);
2427         if (err)
2428                 goto out_up_write;
2429
2430         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2431         /*
2432          * One for dir inode, two for dir entries, two for root
2433          * ref/backref.
2434          */
2435         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2436                                                5, &qgroup_reserved, true);
2437         if (err)
2438                 goto out_up_write;
2439
2440         trans = btrfs_start_transaction(root, 0);
2441         if (IS_ERR(trans)) {
2442                 err = PTR_ERR(trans);
2443                 goto out_release;
2444         }
2445         trans->block_rsv = &block_rsv;
2446         trans->bytes_reserved = block_rsv.size;
2447
2448         ret = btrfs_unlink_subvol(trans, root, dir,
2449                                 dest->root_key.objectid,
2450                                 dentry->d_name.name,
2451                                 dentry->d_name.len);
2452         if (ret) {
2453                 err = ret;
2454                 btrfs_abort_transaction(trans, root, ret);
2455                 goto out_end_trans;
2456         }
2457
2458         btrfs_record_root_in_trans(trans, dest);
2459
2460         memset(&dest->root_item.drop_progress, 0,
2461                 sizeof(dest->root_item.drop_progress));
2462         dest->root_item.drop_level = 0;
2463         btrfs_set_root_refs(&dest->root_item, 0);
2464
2465         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2466                 ret = btrfs_insert_orphan_item(trans,
2467                                         root->fs_info->tree_root,
2468                                         dest->root_key.objectid);
2469                 if (ret) {
2470                         btrfs_abort_transaction(trans, root, ret);
2471                         err = ret;
2472                         goto out_end_trans;
2473                 }
2474         }
2475
2476         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2477                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2478                                   dest->root_key.objectid);
2479         if (ret && ret != -ENOENT) {
2480                 btrfs_abort_transaction(trans, root, ret);
2481                 err = ret;
2482                 goto out_end_trans;
2483         }
2484         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2485                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2486                                           dest->root_item.received_uuid,
2487                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2488                                           dest->root_key.objectid);
2489                 if (ret && ret != -ENOENT) {
2490                         btrfs_abort_transaction(trans, root, ret);
2491                         err = ret;
2492                         goto out_end_trans;
2493                 }
2494         }
2495
2496 out_end_trans:
2497         trans->block_rsv = NULL;
2498         trans->bytes_reserved = 0;
2499         ret = btrfs_end_transaction(trans, root);
2500         if (ret && !err)
2501                 err = ret;
2502         inode->i_flags |= S_DEAD;
2503 out_release:
2504         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2505 out_up_write:
2506         up_write(&root->fs_info->subvol_sem);
2507         if (err) {
2508                 spin_lock(&dest->root_item_lock);
2509                 root_flags = btrfs_root_flags(&dest->root_item);
2510                 btrfs_set_root_flags(&dest->root_item,
2511                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2512                 spin_unlock(&dest->root_item_lock);
2513         }
2514         mutex_unlock(&inode->i_mutex);
2515         if (!err) {
2516                 shrink_dcache_sb(root->fs_info->sb);
2517                 btrfs_invalidate_inodes(dest);
2518                 d_delete(dentry);
2519                 ASSERT(dest->send_in_progress == 0);
2520
2521                 /* the last ref */
2522                 if (dest->ino_cache_inode) {
2523                         iput(dest->ino_cache_inode);
2524                         dest->ino_cache_inode = NULL;
2525                 }
2526         }
2527 out_dput:
2528         dput(dentry);
2529 out_unlock_dir:
2530         mutex_unlock(&dir->i_mutex);
2531 out_drop_write:
2532         mnt_drop_write_file(file);
2533 out:
2534         kfree(vol_args);
2535         return err;
2536 }
2537
2538 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2539 {
2540         struct inode *inode = file_inode(file);
2541         struct btrfs_root *root = BTRFS_I(inode)->root;
2542         struct btrfs_ioctl_defrag_range_args *range;
2543         int ret;
2544
2545         ret = mnt_want_write_file(file);
2546         if (ret)
2547                 return ret;
2548
2549         if (btrfs_root_readonly(root)) {
2550                 ret = -EROFS;
2551                 goto out;
2552         }
2553
2554         switch (inode->i_mode & S_IFMT) {
2555         case S_IFDIR:
2556                 if (!capable(CAP_SYS_ADMIN)) {
2557                         ret = -EPERM;
2558                         goto out;
2559                 }
2560                 ret = btrfs_defrag_root(root);
2561                 if (ret)
2562                         goto out;
2563                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2564                 break;
2565         case S_IFREG:
2566                 if (!(file->f_mode & FMODE_WRITE)) {
2567                         ret = -EINVAL;
2568                         goto out;
2569                 }
2570
2571                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2572                 if (!range) {
2573                         ret = -ENOMEM;
2574                         goto out;
2575                 }
2576
2577                 if (argp) {
2578                         if (copy_from_user(range, argp,
2579                                            sizeof(*range))) {
2580                                 ret = -EFAULT;
2581                                 kfree(range);
2582                                 goto out;
2583                         }
2584                         /* compression requires us to start the IO */
2585                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2586                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2587                                 range->extent_thresh = (u32)-1;
2588                         }
2589                 } else {
2590                         /* the rest are all set to zero by kzalloc */
2591                         range->len = (u64)-1;
2592                 }
2593                 ret = btrfs_defrag_file(file_inode(file), file,
2594                                         range, 0, 0);
2595                 if (ret > 0)
2596                         ret = 0;
2597                 kfree(range);
2598                 break;
2599         default:
2600                 ret = -EINVAL;
2601         }
2602 out:
2603         mnt_drop_write_file(file);
2604         return ret;
2605 }
2606
2607 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2608 {
2609         struct btrfs_ioctl_vol_args *vol_args;
2610         int ret;
2611
2612         if (!capable(CAP_SYS_ADMIN))
2613                 return -EPERM;
2614
2615         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2616                         1)) {
2617                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2618         }
2619
2620         mutex_lock(&root->fs_info->volume_mutex);
2621         vol_args = memdup_user(arg, sizeof(*vol_args));
2622         if (IS_ERR(vol_args)) {
2623                 ret = PTR_ERR(vol_args);
2624                 goto out;
2625         }
2626
2627         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2628         ret = btrfs_init_new_device(root, vol_args->name);
2629
2630         if (!ret)
2631                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2632
2633         kfree(vol_args);
2634 out:
2635         mutex_unlock(&root->fs_info->volume_mutex);
2636         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2637         return ret;
2638 }
2639
2640 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2641 {
2642         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2643         struct btrfs_ioctl_vol_args *vol_args;
2644         int ret;
2645
2646         if (!capable(CAP_SYS_ADMIN))
2647                 return -EPERM;
2648
2649         ret = mnt_want_write_file(file);
2650         if (ret)
2651                 return ret;
2652
2653         vol_args = memdup_user(arg, sizeof(*vol_args));
2654         if (IS_ERR(vol_args)) {
2655                 ret = PTR_ERR(vol_args);
2656                 goto err_drop;
2657         }
2658
2659         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2660
2661         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2662                         1)) {
2663                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2664                 goto out;
2665         }
2666
2667         mutex_lock(&root->fs_info->volume_mutex);
2668         ret = btrfs_rm_device(root, vol_args->name);
2669         mutex_unlock(&root->fs_info->volume_mutex);
2670         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2671
2672         if (!ret)
2673                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2674
2675 out:
2676         kfree(vol_args);
2677 err_drop:
2678         mnt_drop_write_file(file);
2679         return ret;
2680 }
2681
2682 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2683 {
2684         struct btrfs_ioctl_fs_info_args *fi_args;
2685         struct btrfs_device *device;
2686         struct btrfs_device *next;
2687         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2688         int ret = 0;
2689
2690         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2691         if (!fi_args)
2692                 return -ENOMEM;
2693
2694         mutex_lock(&fs_devices->device_list_mutex);
2695         fi_args->num_devices = fs_devices->num_devices;
2696         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2697
2698         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2699                 if (device->devid > fi_args->max_id)
2700                         fi_args->max_id = device->devid;
2701         }
2702         mutex_unlock(&fs_devices->device_list_mutex);
2703
2704         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2705         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2706         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2707
2708         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2709                 ret = -EFAULT;
2710
2711         kfree(fi_args);
2712         return ret;
2713 }
2714
2715 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2716 {
2717         struct btrfs_ioctl_dev_info_args *di_args;
2718         struct btrfs_device *dev;
2719         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2720         int ret = 0;
2721         char *s_uuid = NULL;
2722
2723         di_args = memdup_user(arg, sizeof(*di_args));
2724         if (IS_ERR(di_args))
2725                 return PTR_ERR(di_args);
2726
2727         if (!btrfs_is_empty_uuid(di_args->uuid))
2728                 s_uuid = di_args->uuid;
2729
2730         mutex_lock(&fs_devices->device_list_mutex);
2731         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2732
2733         if (!dev) {
2734                 ret = -ENODEV;
2735                 goto out;
2736         }
2737
2738         di_args->devid = dev->devid;
2739         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2740         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2741         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2742         if (dev->name) {
2743                 struct rcu_string *name;
2744
2745                 rcu_read_lock();
2746                 name = rcu_dereference(dev->name);
2747                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2748                 rcu_read_unlock();
2749                 di_args->path[sizeof(di_args->path) - 1] = 0;
2750         } else {
2751                 di_args->path[0] = '\0';
2752         }
2753
2754 out:
2755         mutex_unlock(&fs_devices->device_list_mutex);
2756         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2757                 ret = -EFAULT;
2758
2759         kfree(di_args);
2760         return ret;
2761 }
2762
2763 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2764 {
2765         struct page *page;
2766         pgoff_t index;
2767         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2768
2769         index = off >> PAGE_CACHE_SHIFT;
2770
2771         page = grab_cache_page(inode->i_mapping, index);
2772         if (!page)
2773                 return NULL;
2774
2775         if (!PageUptodate(page)) {
2776                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2777                                                  0))
2778                         return NULL;
2779                 lock_page(page);
2780                 if (!PageUptodate(page)) {
2781                         unlock_page(page);
2782                         page_cache_release(page);
2783                         return NULL;
2784                 }
2785         }
2786         unlock_page(page);
2787
2788         return page;
2789 }
2790
2791 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2792 {
2793         /* do any pending delalloc/csum calc on src, one way or
2794            another, and lock file content */
2795         while (1) {
2796                 struct btrfs_ordered_extent *ordered;
2797                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2798                 ordered = btrfs_lookup_first_ordered_extent(inode,
2799                                                             off + len - 1);
2800                 if ((!ordered ||
2801                      ordered->file_offset + ordered->len <= off ||
2802                      ordered->file_offset >= off + len) &&
2803                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2804                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2805                         if (ordered)
2806                                 btrfs_put_ordered_extent(ordered);
2807                         break;
2808                 }
2809                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2810                 if (ordered)
2811                         btrfs_put_ordered_extent(ordered);
2812                 btrfs_wait_ordered_range(inode, off, len);
2813         }
2814 }
2815
2816 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2817                                 struct inode *inode2, u64 loff2, u64 len)
2818 {
2819         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2820         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2821
2822         mutex_unlock(&inode1->i_mutex);
2823         mutex_unlock(&inode2->i_mutex);
2824 }
2825
2826 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2827                               struct inode *inode2, u64 loff2, u64 len)
2828 {
2829         if (inode1 < inode2) {
2830                 swap(inode1, inode2);
2831                 swap(loff1, loff2);
2832         }
2833
2834         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2835         lock_extent_range(inode1, loff1, len);
2836         if (inode1 != inode2) {
2837                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2838                 lock_extent_range(inode2, loff2, len);
2839         }
2840 }
2841
2842 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2843                           u64 dst_loff, u64 len)
2844 {
2845         int ret = 0;
2846         struct page *src_page, *dst_page;
2847         unsigned int cmp_len = PAGE_CACHE_SIZE;
2848         void *addr, *dst_addr;
2849
2850         while (len) {
2851                 if (len < PAGE_CACHE_SIZE)
2852                         cmp_len = len;
2853
2854                 src_page = extent_same_get_page(src, loff);
2855                 if (!src_page)
2856                         return -EINVAL;
2857                 dst_page = extent_same_get_page(dst, dst_loff);
2858                 if (!dst_page) {
2859                         page_cache_release(src_page);
2860                         return -EINVAL;
2861                 }
2862                 addr = kmap_atomic(src_page);
2863                 dst_addr = kmap_atomic(dst_page);
2864
2865                 flush_dcache_page(src_page);
2866                 flush_dcache_page(dst_page);
2867
2868                 if (memcmp(addr, dst_addr, cmp_len))
2869                         ret = BTRFS_SAME_DATA_DIFFERS;
2870
2871                 kunmap_atomic(addr);
2872                 kunmap_atomic(dst_addr);
2873                 page_cache_release(src_page);
2874                 page_cache_release(dst_page);
2875
2876                 if (ret)
2877                         break;
2878
2879                 loff += cmp_len;
2880                 dst_loff += cmp_len;
2881                 len -= cmp_len;
2882         }
2883
2884         return ret;
2885 }
2886
2887 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2888 {
2889         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2890
2891         if (off + len > inode->i_size || off + len < off)
2892                 return -EINVAL;
2893         /* Check that we are block aligned - btrfs_clone() requires this */
2894         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2895                 return -EINVAL;
2896
2897         return 0;
2898 }
2899
2900 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2901                              struct inode *dst, u64 dst_loff)
2902 {
2903         int ret;
2904
2905         /*
2906          * btrfs_clone() can't handle extents in the same file
2907          * yet. Once that works, we can drop this check and replace it
2908          * with a check for the same inode, but overlapping extents.
2909          */
2910         if (src == dst)
2911                 return -EINVAL;
2912
2913         btrfs_double_lock(src, loff, dst, dst_loff, len);
2914
2915         ret = extent_same_check_offsets(src, loff, len);
2916         if (ret)
2917                 goto out_unlock;
2918
2919         ret = extent_same_check_offsets(dst, dst_loff, len);
2920         if (ret)
2921                 goto out_unlock;
2922
2923         /* don't make the dst file partly checksummed */
2924         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2925             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2926                 ret = -EINVAL;
2927                 goto out_unlock;
2928         }
2929
2930         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2931         if (ret == 0)
2932                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2933
2934 out_unlock:
2935         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2936
2937         return ret;
2938 }
2939
2940 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2941
2942 static long btrfs_ioctl_file_extent_same(struct file *file,
2943                         struct btrfs_ioctl_same_args __user *argp)
2944 {
2945         struct btrfs_ioctl_same_args *same;
2946         struct btrfs_ioctl_same_extent_info *info;
2947         struct inode *src = file_inode(file);
2948         u64 off;
2949         u64 len;
2950         int i;
2951         int ret;
2952         unsigned long size;
2953         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2954         bool is_admin = capable(CAP_SYS_ADMIN);
2955         u16 count;
2956
2957         if (!(file->f_mode & FMODE_READ))
2958                 return -EINVAL;
2959
2960         ret = mnt_want_write_file(file);
2961         if (ret)
2962                 return ret;
2963
2964         if (get_user(count, &argp->dest_count)) {
2965                 ret = -EFAULT;
2966                 goto out;
2967         }
2968
2969         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2970
2971         same = memdup_user(argp, size);
2972
2973         if (IS_ERR(same)) {
2974                 ret = PTR_ERR(same);
2975                 goto out;
2976         }
2977
2978         off = same->logical_offset;
2979         len = same->length;
2980
2981         /*
2982          * Limit the total length we will dedupe for each operation.
2983          * This is intended to bound the total time spent in this
2984          * ioctl to something sane.
2985          */
2986         if (len > BTRFS_MAX_DEDUPE_LEN)
2987                 len = BTRFS_MAX_DEDUPE_LEN;
2988
2989         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2990                 /*
2991                  * Btrfs does not support blocksize < page_size. As a
2992                  * result, btrfs_cmp_data() won't correctly handle
2993                  * this situation without an update.
2994                  */
2995                 ret = -EINVAL;
2996                 goto out;
2997         }
2998
2999         ret = -EISDIR;
3000         if (S_ISDIR(src->i_mode))
3001                 goto out;
3002
3003         ret = -EACCES;
3004         if (!S_ISREG(src->i_mode))
3005                 goto out;
3006
3007         /* pre-format output fields to sane values */
3008         for (i = 0; i < count; i++) {
3009                 same->info[i].bytes_deduped = 0ULL;
3010                 same->info[i].status = 0;
3011         }
3012
3013         for (i = 0, info = same->info; i < count; i++, info++) {
3014                 struct inode *dst;
3015                 struct fd dst_file = fdget(info->fd);
3016                 if (!dst_file.file) {
3017                         info->status = -EBADF;
3018                         continue;
3019                 }
3020                 dst = file_inode(dst_file.file);
3021
3022                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3023                         info->status = -EINVAL;
3024                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3025                         info->status = -EXDEV;
3026                 } else if (S_ISDIR(dst->i_mode)) {
3027                         info->status = -EISDIR;
3028                 } else if (!S_ISREG(dst->i_mode)) {
3029                         info->status = -EACCES;
3030                 } else {
3031                         info->status = btrfs_extent_same(src, off, len, dst,
3032                                                         info->logical_offset);
3033                         if (info->status == 0)
3034                                 info->bytes_deduped += len;
3035                 }
3036                 fdput(dst_file);
3037         }
3038
3039         ret = copy_to_user(argp, same, size);
3040         if (ret)
3041                 ret = -EFAULT;
3042
3043 out:
3044         mnt_drop_write_file(file);
3045         return ret;
3046 }
3047
3048 /* Helper to check and see if this root currently has a ref on the given disk
3049  * bytenr.  If it does then we need to update the quota for this root.  This
3050  * doesn't do anything if quotas aren't enabled.
3051  */
3052 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3053                      u64 disko)
3054 {
3055         struct seq_list tree_mod_seq_elem = {};
3056         struct ulist *roots;
3057         struct ulist_iterator uiter;
3058         struct ulist_node *root_node = NULL;
3059         int ret;
3060
3061         if (!root->fs_info->quota_enabled)
3062                 return 1;
3063
3064         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3065         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3066                                    tree_mod_seq_elem.seq, &roots);
3067         if (ret < 0)
3068                 goto out;
3069         ret = 0;
3070         ULIST_ITER_INIT(&uiter);
3071         while ((root_node = ulist_next(roots, &uiter))) {
3072                 if (root_node->val == root->objectid) {
3073                         ret = 1;
3074                         break;
3075                 }
3076         }
3077         ulist_free(roots);
3078 out:
3079         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3080         return ret;
3081 }
3082
3083 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3084                                      struct inode *inode,
3085                                      u64 endoff,
3086                                      const u64 destoff,
3087                                      const u64 olen)
3088 {
3089         struct btrfs_root *root = BTRFS_I(inode)->root;
3090         int ret;
3091
3092         inode_inc_iversion(inode);
3093         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3094         /*
3095          * We round up to the block size at eof when determining which
3096          * extents to clone above, but shouldn't round up the file size.
3097          */
3098         if (endoff > destoff + olen)
3099                 endoff = destoff + olen;
3100         if (endoff > inode->i_size)
3101                 btrfs_i_size_write(inode, endoff);
3102
3103         ret = btrfs_update_inode(trans, root, inode);
3104         if (ret) {
3105                 btrfs_abort_transaction(trans, root, ret);
3106                 btrfs_end_transaction(trans, root);
3107                 goto out;
3108         }
3109         ret = btrfs_end_transaction(trans, root);
3110 out:
3111         return ret;
3112 }
3113
3114 static void clone_update_extent_map(struct inode *inode,
3115                                     const struct btrfs_trans_handle *trans,
3116                                     const struct btrfs_path *path,
3117                                     const u64 hole_offset,
3118                                     const u64 hole_len)
3119 {
3120         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3121         struct extent_map *em;
3122         int ret;
3123
3124         em = alloc_extent_map();
3125         if (!em) {
3126                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3127                         &BTRFS_I(inode)->runtime_flags);
3128                 return;
3129         }
3130
3131         if (path) {
3132                 struct btrfs_file_extent_item *fi;
3133
3134                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3135                                     struct btrfs_file_extent_item);
3136                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3137                 em->generation = -1;
3138                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3139                     BTRFS_FILE_EXTENT_INLINE)
3140                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3141                                 &BTRFS_I(inode)->runtime_flags);
3142         } else {
3143                 em->start = hole_offset;
3144                 em->len = hole_len;
3145                 em->ram_bytes = em->len;
3146                 em->orig_start = hole_offset;
3147                 em->block_start = EXTENT_MAP_HOLE;
3148                 em->block_len = 0;
3149                 em->orig_block_len = 0;
3150                 em->compress_type = BTRFS_COMPRESS_NONE;
3151                 em->generation = trans->transid;
3152         }
3153
3154         while (1) {
3155                 write_lock(&em_tree->lock);
3156                 ret = add_extent_mapping(em_tree, em, 1);
3157                 write_unlock(&em_tree->lock);
3158                 if (ret != -EEXIST) {
3159                         free_extent_map(em);
3160                         break;
3161                 }
3162                 btrfs_drop_extent_cache(inode, em->start,
3163                                         em->start + em->len - 1, 0);
3164         }
3165
3166         if (ret)
3167                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3168                         &BTRFS_I(inode)->runtime_flags);
3169 }
3170
3171 /**
3172  * btrfs_clone() - clone a range from inode file to another
3173  *
3174  * @src: Inode to clone from
3175  * @inode: Inode to clone to
3176  * @off: Offset within source to start clone from
3177  * @olen: Original length, passed by user, of range to clone
3178  * @olen_aligned: Block-aligned value of olen, extent_same uses
3179  *               identical values here
3180  * @destoff: Offset within @inode to start clone
3181  */
3182 static int btrfs_clone(struct inode *src, struct inode *inode,
3183                        const u64 off, const u64 olen, const u64 olen_aligned,
3184                        const u64 destoff)
3185 {
3186         struct btrfs_root *root = BTRFS_I(inode)->root;
3187         struct btrfs_path *path = NULL;
3188         struct extent_buffer *leaf;
3189         struct btrfs_trans_handle *trans;
3190         char *buf = NULL;
3191         struct btrfs_key key;
3192         u32 nritems;
3193         int slot;
3194         int ret;
3195         int no_quota;
3196         const u64 len = olen_aligned;
3197         u64 last_disko = 0;
3198         u64 last_dest_end = destoff;
3199
3200         ret = -ENOMEM;
3201         buf = vmalloc(root->nodesize);
3202         if (!buf)
3203                 return ret;
3204
3205         path = btrfs_alloc_path();
3206         if (!path) {
3207                 vfree(buf);
3208                 return ret;
3209         }
3210
3211         path->reada = 2;
3212         /* clone data */
3213         key.objectid = btrfs_ino(src);
3214         key.type = BTRFS_EXTENT_DATA_KEY;
3215         key.offset = off;
3216
3217         while (1) {
3218                 /*
3219                  * note the key will change type as we walk through the
3220                  * tree.
3221                  */
3222                 path->leave_spinning = 1;
3223                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3224                                 0, 0);
3225                 if (ret < 0)
3226                         goto out;
3227                 /*
3228                  * First search, if no extent item that starts at offset off was
3229                  * found but the previous item is an extent item, it's possible
3230                  * it might overlap our target range, therefore process it.
3231                  */
3232                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3233                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3234                                               path->slots[0] - 1);
3235                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3236                                 path->slots[0]--;
3237                 }
3238
3239                 nritems = btrfs_header_nritems(path->nodes[0]);
3240 process_slot:
3241                 no_quota = 1;
3242                 if (path->slots[0] >= nritems) {
3243                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3244                         if (ret < 0)
3245                                 goto out;
3246                         if (ret > 0)
3247                                 break;
3248                         nritems = btrfs_header_nritems(path->nodes[0]);
3249                 }
3250                 leaf = path->nodes[0];
3251                 slot = path->slots[0];
3252
3253                 btrfs_item_key_to_cpu(leaf, &key, slot);
3254                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3255                     key.objectid != btrfs_ino(src))
3256                         break;
3257
3258                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3259                         struct btrfs_file_extent_item *extent;
3260                         int type;
3261                         u32 size;
3262                         struct btrfs_key new_key;
3263                         u64 disko = 0, diskl = 0;
3264                         u64 datao = 0, datal = 0;
3265                         u8 comp;
3266                         u64 drop_start;
3267
3268                         extent = btrfs_item_ptr(leaf, slot,
3269                                                 struct btrfs_file_extent_item);
3270                         comp = btrfs_file_extent_compression(leaf, extent);
3271                         type = btrfs_file_extent_type(leaf, extent);
3272                         if (type == BTRFS_FILE_EXTENT_REG ||
3273                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3274                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3275                                                                       extent);
3276                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3277                                                                  extent);
3278                                 datao = btrfs_file_extent_offset(leaf, extent);
3279                                 datal = btrfs_file_extent_num_bytes(leaf,
3280                                                                     extent);
3281                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3282                                 /* take upper bound, may be compressed */
3283                                 datal = btrfs_file_extent_ram_bytes(leaf,
3284                                                                     extent);
3285                         }
3286
3287                         /*
3288                          * The first search might have left us at an extent
3289                          * item that ends before our target range's start, can
3290                          * happen if we have holes and NO_HOLES feature enabled.
3291                          */
3292                         if (key.offset + datal <= off) {
3293                                 path->slots[0]++;
3294                                 goto process_slot;
3295                         } else if (key.offset >= off + len) {
3296                                 break;
3297                         }
3298
3299                         size = btrfs_item_size_nr(leaf, slot);
3300                         read_extent_buffer(leaf, buf,
3301                                            btrfs_item_ptr_offset(leaf, slot),
3302                                            size);
3303
3304                         btrfs_release_path(path);
3305                         path->leave_spinning = 0;
3306
3307                         memcpy(&new_key, &key, sizeof(new_key));
3308                         new_key.objectid = btrfs_ino(inode);
3309                         if (off <= key.offset)
3310                                 new_key.offset = key.offset + destoff - off;
3311                         else
3312                                 new_key.offset = destoff;
3313
3314                         /*
3315                          * Deal with a hole that doesn't have an extent item
3316                          * that represents it (NO_HOLES feature enabled).
3317                          * This hole is either in the middle of the cloning
3318                          * range or at the beginning (fully overlaps it or
3319                          * partially overlaps it).
3320                          */
3321                         if (new_key.offset != last_dest_end)
3322                                 drop_start = last_dest_end;
3323                         else
3324                                 drop_start = new_key.offset;
3325
3326                         /*
3327                          * 1 - adjusting old extent (we may have to split it)
3328                          * 1 - add new extent
3329                          * 1 - inode update
3330                          */
3331                         trans = btrfs_start_transaction(root, 3);
3332                         if (IS_ERR(trans)) {
3333                                 ret = PTR_ERR(trans);
3334                                 goto out;
3335                         }
3336
3337                         if (type == BTRFS_FILE_EXTENT_REG ||
3338                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3339                                 /*
3340                                  *    a  | --- range to clone ---|  b
3341                                  * | ------------- extent ------------- |
3342                                  */
3343
3344                                 /* subtract range b */
3345                                 if (key.offset + datal > off + len)
3346                                         datal = off + len - key.offset;
3347
3348                                 /* subtract range a */
3349                                 if (off > key.offset) {
3350                                         datao += off - key.offset;
3351                                         datal -= off - key.offset;
3352                                 }
3353
3354                                 ret = btrfs_drop_extents(trans, root, inode,
3355                                                          drop_start,
3356                                                          new_key.offset + datal,
3357                                                          1);
3358                                 if (ret) {
3359                                         if (ret != -EOPNOTSUPP)
3360                                                 btrfs_abort_transaction(trans,
3361                                                                 root, ret);
3362                                         btrfs_end_transaction(trans, root);
3363                                         goto out;
3364                                 }
3365
3366                                 ret = btrfs_insert_empty_item(trans, root, path,
3367                                                               &new_key, size);
3368                                 if (ret) {
3369                                         btrfs_abort_transaction(trans, root,
3370                                                                 ret);
3371                                         btrfs_end_transaction(trans, root);
3372                                         goto out;
3373                                 }
3374
3375                                 leaf = path->nodes[0];
3376                                 slot = path->slots[0];
3377                                 write_extent_buffer(leaf, buf,
3378                                             btrfs_item_ptr_offset(leaf, slot),
3379                                             size);
3380
3381                                 extent = btrfs_item_ptr(leaf, slot,
3382                                                 struct btrfs_file_extent_item);
3383
3384                                 /* disko == 0 means it's a hole */
3385                                 if (!disko)
3386                                         datao = 0;
3387
3388                                 btrfs_set_file_extent_offset(leaf, extent,
3389                                                              datao);
3390                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3391                                                                 datal);
3392
3393                                 /*
3394                                  * We need to look up the roots that point at
3395                                  * this bytenr and see if the new root does.  If
3396                                  * it does not we need to make sure we update
3397                                  * quotas appropriately.
3398                                  */
3399                                 if (disko && root != BTRFS_I(src)->root &&
3400                                     disko != last_disko) {
3401                                         no_quota = check_ref(trans, root,
3402                                                              disko);
3403                                         if (no_quota < 0) {
3404                                                 btrfs_abort_transaction(trans,
3405                                                                         root,
3406                                                                         ret);
3407                                                 btrfs_end_transaction(trans,
3408                                                                       root);
3409                                                 ret = no_quota;
3410                                                 goto out;
3411                                         }
3412                                 }
3413
3414                                 if (disko) {
3415                                         inode_add_bytes(inode, datal);
3416                                         ret = btrfs_inc_extent_ref(trans, root,
3417                                                         disko, diskl, 0,
3418                                                         root->root_key.objectid,
3419                                                         btrfs_ino(inode),
3420                                                         new_key.offset - datao,
3421                                                         no_quota);
3422                                         if (ret) {
3423                                                 btrfs_abort_transaction(trans,
3424                                                                         root,
3425                                                                         ret);
3426                                                 btrfs_end_transaction(trans,
3427                                                                       root);
3428                                                 goto out;
3429
3430                                         }
3431                                 }
3432                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3433                                 u64 skip = 0;
3434                                 u64 trim = 0;
3435                                 u64 aligned_end = 0;
3436
3437                                 if (off > key.offset) {
3438                                         skip = off - key.offset;
3439                                         new_key.offset += skip;
3440                                 }
3441
3442                                 if (key.offset + datal > off + len)
3443                                         trim = key.offset + datal - (off + len);
3444
3445                                 if (comp && (skip || trim)) {
3446                                         ret = -EINVAL;
3447                                         btrfs_end_transaction(trans, root);
3448                                         goto out;
3449                                 }
3450                                 size -= skip + trim;
3451                                 datal -= skip + trim;
3452
3453                                 aligned_end = ALIGN(new_key.offset + datal,
3454                                                     root->sectorsize);
3455                                 ret = btrfs_drop_extents(trans, root, inode,
3456                                                          drop_start,
3457                                                          aligned_end,
3458                                                          1);
3459                                 if (ret) {
3460                                         if (ret != -EOPNOTSUPP)
3461                                                 btrfs_abort_transaction(trans,
3462                                                         root, ret);
3463                                         btrfs_end_transaction(trans, root);
3464                                         goto out;
3465                                 }
3466
3467                                 ret = btrfs_insert_empty_item(trans, root, path,
3468                                                               &new_key, size);
3469                                 if (ret) {
3470                                         btrfs_abort_transaction(trans, root,
3471                                                                 ret);
3472                                         btrfs_end_transaction(trans, root);
3473                                         goto out;
3474                                 }
3475
3476                                 if (skip) {
3477                                         u32 start =
3478                                           btrfs_file_extent_calc_inline_size(0);
3479                                         memmove(buf+start, buf+start+skip,
3480                                                 datal);
3481                                 }
3482
3483                                 leaf = path->nodes[0];
3484                                 slot = path->slots[0];
3485                                 write_extent_buffer(leaf, buf,
3486                                             btrfs_item_ptr_offset(leaf, slot),
3487                                             size);
3488                                 inode_add_bytes(inode, datal);
3489                         }
3490
3491                         /* If we have an implicit hole (NO_HOLES feature). */
3492                         if (drop_start < new_key.offset)
3493                                 clone_update_extent_map(inode, trans,
3494                                                 NULL, drop_start,
3495                                                 new_key.offset - drop_start);
3496
3497                         clone_update_extent_map(inode, trans, path, 0, 0);
3498
3499                         btrfs_mark_buffer_dirty(leaf);
3500                         btrfs_release_path(path);
3501
3502                         last_dest_end = ALIGN(new_key.offset + datal,
3503                                               root->sectorsize);
3504                         ret = clone_finish_inode_update(trans, inode,
3505                                                         last_dest_end,
3506                                                         destoff, olen);
3507                         if (ret)
3508                                 goto out;
3509                         if (new_key.offset + datal >= destoff + len)
3510                                 break;
3511                 }
3512                 btrfs_release_path(path);
3513                 key.offset++;
3514         }
3515         ret = 0;
3516
3517         if (last_dest_end < destoff + len) {
3518                 /*
3519                  * We have an implicit hole (NO_HOLES feature is enabled) that
3520                  * fully or partially overlaps our cloning range at its end.
3521                  */
3522                 btrfs_release_path(path);
3523
3524                 /*
3525                  * 1 - remove extent(s)
3526                  * 1 - inode update
3527                  */
3528                 trans = btrfs_start_transaction(root, 2);
3529                 if (IS_ERR(trans)) {
3530                         ret = PTR_ERR(trans);
3531                         goto out;
3532                 }
3533                 ret = btrfs_drop_extents(trans, root, inode,
3534                                          last_dest_end, destoff + len, 1);
3535                 if (ret) {
3536                         if (ret != -EOPNOTSUPP)
3537                                 btrfs_abort_transaction(trans, root, ret);
3538                         btrfs_end_transaction(trans, root);
3539                         goto out;
3540                 }
3541                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3542                                         destoff + len - last_dest_end);
3543                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3544                                                 destoff, olen);
3545         }
3546
3547 out:
3548         btrfs_free_path(path);
3549         vfree(buf);
3550         return ret;
3551 }
3552
3553 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3554                                        u64 off, u64 olen, u64 destoff)
3555 {
3556         struct inode *inode = file_inode(file);
3557         struct btrfs_root *root = BTRFS_I(inode)->root;
3558         struct fd src_file;
3559         struct inode *src;
3560         int ret;
3561         u64 len = olen;
3562         u64 bs = root->fs_info->sb->s_blocksize;
3563         int same_inode = 0;
3564
3565         /*
3566          * TODO:
3567          * - split compressed inline extents.  annoying: we need to
3568          *   decompress into destination's address_space (the file offset
3569          *   may change, so source mapping won't do), then recompress (or
3570          *   otherwise reinsert) a subrange.
3571          *
3572          * - split destination inode's inline extents.  The inline extents can
3573          *   be either compressed or non-compressed.
3574          */
3575
3576         /* the destination must be opened for writing */
3577         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3578                 return -EINVAL;
3579
3580         if (btrfs_root_readonly(root))
3581                 return -EROFS;
3582
3583         ret = mnt_want_write_file(file);
3584         if (ret)
3585                 return ret;
3586
3587         src_file = fdget(srcfd);
3588         if (!src_file.file) {
3589                 ret = -EBADF;
3590                 goto out_drop_write;
3591         }
3592
3593         ret = -EXDEV;
3594         if (src_file.file->f_path.mnt != file->f_path.mnt)
3595                 goto out_fput;
3596
3597         src = file_inode(src_file.file);
3598
3599         ret = -EINVAL;
3600         if (src == inode)
3601                 same_inode = 1;
3602
3603         /* the src must be open for reading */
3604         if (!(src_file.file->f_mode & FMODE_READ))
3605                 goto out_fput;
3606
3607         /* don't make the dst file partly checksummed */
3608         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3609             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3610                 goto out_fput;
3611
3612         ret = -EISDIR;
3613         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3614                 goto out_fput;
3615
3616         ret = -EXDEV;
3617         if (src->i_sb != inode->i_sb)
3618                 goto out_fput;
3619
3620         if (!same_inode) {
3621                 if (inode < src) {
3622                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3623                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3624                 } else {
3625                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3626                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3627                 }
3628         } else {
3629                 mutex_lock(&src->i_mutex);
3630         }
3631
3632         /* determine range to clone */
3633         ret = -EINVAL;
3634         if (off + len > src->i_size || off + len < off)
3635                 goto out_unlock;
3636         if (len == 0)
3637                 olen = len = src->i_size - off;
3638         /* if we extend to eof, continue to block boundary */
3639         if (off + len == src->i_size)
3640                 len = ALIGN(src->i_size, bs) - off;
3641
3642         /* verify the end result is block aligned */
3643         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3644             !IS_ALIGNED(destoff, bs))
3645                 goto out_unlock;
3646
3647         /* verify if ranges are overlapped within the same file */
3648         if (same_inode) {
3649                 if (destoff + len > off && destoff < off + len)
3650                         goto out_unlock;
3651         }
3652
3653         if (destoff > inode->i_size) {
3654                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3655                 if (ret)
3656                         goto out_unlock;
3657         }
3658
3659         /*
3660          * Lock the target range too. Right after we replace the file extent
3661          * items in the fs tree (which now point to the cloned data), we might
3662          * have a worker replace them with extent items relative to a write
3663          * operation that was issued before this clone operation (i.e. confront
3664          * with inode.c:btrfs_finish_ordered_io).
3665          */
3666         if (same_inode) {
3667                 u64 lock_start = min_t(u64, off, destoff);
3668                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3669
3670                 lock_extent_range(src, lock_start, lock_len);
3671         } else {
3672                 lock_extent_range(src, off, len);
3673                 lock_extent_range(inode, destoff, len);
3674         }
3675
3676         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3677
3678         if (same_inode) {
3679                 u64 lock_start = min_t(u64, off, destoff);
3680                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3681
3682                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3683         } else {
3684                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3685                 unlock_extent(&BTRFS_I(inode)->io_tree, destoff,
3686                               destoff + len - 1);
3687         }
3688         /*
3689          * Truncate page cache pages so that future reads will see the cloned
3690          * data immediately and not the previous data.
3691          */
3692         truncate_inode_pages_range(&inode->i_data, destoff,
3693                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3694 out_unlock:
3695         if (!same_inode) {
3696                 if (inode < src) {
3697                         mutex_unlock(&src->i_mutex);
3698                         mutex_unlock(&inode->i_mutex);
3699                 } else {
3700                         mutex_unlock(&inode->i_mutex);
3701                         mutex_unlock(&src->i_mutex);
3702                 }
3703         } else {
3704                 mutex_unlock(&src->i_mutex);
3705         }
3706 out_fput:
3707         fdput(src_file);
3708 out_drop_write:
3709         mnt_drop_write_file(file);
3710         return ret;
3711 }
3712
3713 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3714 {
3715         struct btrfs_ioctl_clone_range_args args;
3716
3717         if (copy_from_user(&args, argp, sizeof(args)))
3718                 return -EFAULT;
3719         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3720                                  args.src_length, args.dest_offset);
3721 }
3722
3723 /*
3724  * there are many ways the trans_start and trans_end ioctls can lead
3725  * to deadlocks.  They should only be used by applications that
3726  * basically own the machine, and have a very in depth understanding
3727  * of all the possible deadlocks and enospc problems.
3728  */
3729 static long btrfs_ioctl_trans_start(struct file *file)
3730 {
3731         struct inode *inode = file_inode(file);
3732         struct btrfs_root *root = BTRFS_I(inode)->root;
3733         struct btrfs_trans_handle *trans;
3734         int ret;
3735
3736         ret = -EPERM;
3737         if (!capable(CAP_SYS_ADMIN))
3738                 goto out;
3739
3740         ret = -EINPROGRESS;
3741         if (file->private_data)
3742                 goto out;
3743
3744         ret = -EROFS;
3745         if (btrfs_root_readonly(root))
3746                 goto out;
3747
3748         ret = mnt_want_write_file(file);
3749         if (ret)
3750                 goto out;
3751
3752         atomic_inc(&root->fs_info->open_ioctl_trans);
3753
3754         ret = -ENOMEM;
3755         trans = btrfs_start_ioctl_transaction(root);
3756         if (IS_ERR(trans))
3757                 goto out_drop;
3758
3759         file->private_data = trans;
3760         return 0;
3761
3762 out_drop:
3763         atomic_dec(&root->fs_info->open_ioctl_trans);
3764         mnt_drop_write_file(file);
3765 out:
3766         return ret;
3767 }
3768
3769 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3770 {
3771         struct inode *inode = file_inode(file);
3772         struct btrfs_root *root = BTRFS_I(inode)->root;
3773         struct btrfs_root *new_root;
3774         struct btrfs_dir_item *di;
3775         struct btrfs_trans_handle *trans;
3776         struct btrfs_path *path;
3777         struct btrfs_key location;
3778         struct btrfs_disk_key disk_key;
3779         u64 objectid = 0;
3780         u64 dir_id;
3781         int ret;
3782
3783         if (!capable(CAP_SYS_ADMIN))
3784                 return -EPERM;
3785
3786         ret = mnt_want_write_file(file);
3787         if (ret)
3788                 return ret;
3789
3790         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3791                 ret = -EFAULT;
3792                 goto out;
3793         }
3794
3795         if (!objectid)
3796                 objectid = BTRFS_FS_TREE_OBJECTID;
3797
3798         location.objectid = objectid;
3799         location.type = BTRFS_ROOT_ITEM_KEY;
3800         location.offset = (u64)-1;
3801
3802         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3803         if (IS_ERR(new_root)) {
3804                 ret = PTR_ERR(new_root);
3805                 goto out;
3806         }
3807
3808         path = btrfs_alloc_path();
3809         if (!path) {
3810                 ret = -ENOMEM;
3811                 goto out;
3812         }
3813         path->leave_spinning = 1;
3814
3815         trans = btrfs_start_transaction(root, 1);
3816         if (IS_ERR(trans)) {
3817                 btrfs_free_path(path);
3818                 ret = PTR_ERR(trans);
3819                 goto out;
3820         }
3821
3822         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3823         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3824                                    dir_id, "default", 7, 1);
3825         if (IS_ERR_OR_NULL(di)) {
3826                 btrfs_free_path(path);
3827                 btrfs_end_transaction(trans, root);
3828                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3829                            "item, this isn't going to work");
3830                 ret = -ENOENT;
3831                 goto out;
3832         }
3833
3834         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3835         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3836         btrfs_mark_buffer_dirty(path->nodes[0]);
3837         btrfs_free_path(path);
3838
3839         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3840         btrfs_end_transaction(trans, root);
3841 out:
3842         mnt_drop_write_file(file);
3843         return ret;
3844 }
3845
3846 void btrfs_get_block_group_info(struct list_head *groups_list,
3847                                 struct btrfs_ioctl_space_info *space)
3848 {
3849         struct btrfs_block_group_cache *block_group;
3850
3851         space->total_bytes = 0;
3852         space->used_bytes = 0;
3853         space->flags = 0;
3854         list_for_each_entry(block_group, groups_list, list) {
3855                 space->flags = block_group->flags;
3856                 space->total_bytes += block_group->key.offset;
3857                 space->used_bytes +=
3858                         btrfs_block_group_used(&block_group->item);
3859         }
3860 }
3861
3862 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3863 {
3864         struct btrfs_ioctl_space_args space_args;
3865         struct btrfs_ioctl_space_info space;
3866         struct btrfs_ioctl_space_info *dest;
3867         struct btrfs_ioctl_space_info *dest_orig;
3868         struct btrfs_ioctl_space_info __user *user_dest;
3869         struct btrfs_space_info *info;
3870         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3871                        BTRFS_BLOCK_GROUP_SYSTEM,
3872                        BTRFS_BLOCK_GROUP_METADATA,
3873                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3874         int num_types = 4;
3875         int alloc_size;
3876         int ret = 0;
3877         u64 slot_count = 0;
3878         int i, c;
3879
3880         if (copy_from_user(&space_args,
3881                            (struct btrfs_ioctl_space_args __user *)arg,
3882                            sizeof(space_args)))
3883                 return -EFAULT;
3884
3885         for (i = 0; i < num_types; i++) {
3886                 struct btrfs_space_info *tmp;
3887
3888                 info = NULL;
3889                 rcu_read_lock();
3890                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3891                                         list) {
3892                         if (tmp->flags == types[i]) {
3893                                 info = tmp;
3894                                 break;
3895                         }
3896                 }
3897                 rcu_read_unlock();
3898
3899                 if (!info)
3900                         continue;
3901
3902                 down_read(&info->groups_sem);
3903                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3904                         if (!list_empty(&info->block_groups[c]))
3905                                 slot_count++;
3906                 }
3907                 up_read(&info->groups_sem);
3908         }
3909
3910         /*
3911          * Global block reserve, exported as a space_info
3912          */
3913         slot_count++;
3914
3915         /* space_slots == 0 means they are asking for a count */
3916         if (space_args.space_slots == 0) {
3917                 space_args.total_spaces = slot_count;
3918                 goto out;
3919         }
3920
3921         slot_count = min_t(u64, space_args.space_slots, slot_count);
3922
3923         alloc_size = sizeof(*dest) * slot_count;
3924
3925         /* we generally have at most 6 or so space infos, one for each raid
3926          * level.  So, a whole page should be more than enough for everyone
3927          */
3928         if (alloc_size > PAGE_CACHE_SIZE)
3929                 return -ENOMEM;
3930
3931         space_args.total_spaces = 0;
3932         dest = kmalloc(alloc_size, GFP_NOFS);
3933         if (!dest)
3934                 return -ENOMEM;
3935         dest_orig = dest;
3936
3937         /* now we have a buffer to copy into */
3938         for (i = 0; i < num_types; i++) {
3939                 struct btrfs_space_info *tmp;
3940
3941                 if (!slot_count)
3942                         break;
3943
3944                 info = NULL;
3945                 rcu_read_lock();
3946                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3947                                         list) {
3948                         if (tmp->flags == types[i]) {
3949                                 info = tmp;
3950                                 break;
3951                         }
3952                 }
3953                 rcu_read_unlock();
3954
3955                 if (!info)
3956                         continue;
3957                 down_read(&info->groups_sem);
3958                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3959                         if (!list_empty(&info->block_groups[c])) {
3960                                 btrfs_get_block_group_info(
3961                                         &info->block_groups[c], &space);
3962                                 memcpy(dest, &space, sizeof(space));
3963                                 dest++;
3964                                 space_args.total_spaces++;
3965                                 slot_count--;
3966                         }
3967                         if (!slot_count)
3968                                 break;
3969                 }
3970                 up_read(&info->groups_sem);
3971         }
3972
3973         /*
3974          * Add global block reserve
3975          */
3976         if (slot_count) {
3977                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3978
3979                 spin_lock(&block_rsv->lock);
3980                 space.total_bytes = block_rsv->size;
3981                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3982                 spin_unlock(&block_rsv->lock);
3983                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3984                 memcpy(dest, &space, sizeof(space));
3985                 space_args.total_spaces++;
3986         }
3987
3988         user_dest = (struct btrfs_ioctl_space_info __user *)
3989                 (arg + sizeof(struct btrfs_ioctl_space_args));
3990
3991         if (copy_to_user(user_dest, dest_orig, alloc_size))
3992                 ret = -EFAULT;
3993
3994         kfree(dest_orig);
3995 out:
3996         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3997                 ret = -EFAULT;
3998
3999         return ret;
4000 }
4001
4002 /*
4003  * there are many ways the trans_start and trans_end ioctls can lead
4004  * to deadlocks.  They should only be used by applications that
4005  * basically own the machine, and have a very in depth understanding
4006  * of all the possible deadlocks and enospc problems.
4007  */
4008 long btrfs_ioctl_trans_end(struct file *file)
4009 {
4010         struct inode *inode = file_inode(file);
4011         struct btrfs_root *root = BTRFS_I(inode)->root;
4012         struct btrfs_trans_handle *trans;
4013
4014         trans = file->private_data;
4015         if (!trans)
4016                 return -EINVAL;
4017         file->private_data = NULL;
4018
4019         btrfs_end_transaction(trans, root);
4020
4021         atomic_dec(&root->fs_info->open_ioctl_trans);
4022
4023         mnt_drop_write_file(file);
4024         return 0;
4025 }
4026
4027 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4028                                             void __user *argp)
4029 {
4030         struct btrfs_trans_handle *trans;
4031         u64 transid;
4032         int ret;
4033
4034         trans = btrfs_attach_transaction_barrier(root);
4035         if (IS_ERR(trans)) {
4036                 if (PTR_ERR(trans) != -ENOENT)
4037                         return PTR_ERR(trans);
4038
4039                 /* No running transaction, don't bother */
4040                 transid = root->fs_info->last_trans_committed;
4041                 goto out;
4042         }
4043         transid = trans->transid;
4044         ret = btrfs_commit_transaction_async(trans, root, 0);
4045         if (ret) {
4046                 btrfs_end_transaction(trans, root);
4047                 return ret;
4048         }
4049 out:
4050         if (argp)
4051                 if (copy_to_user(argp, &transid, sizeof(transid)))
4052                         return -EFAULT;
4053         return 0;
4054 }
4055
4056 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4057                                            void __user *argp)
4058 {
4059         u64 transid;
4060
4061         if (argp) {
4062                 if (copy_from_user(&transid, argp, sizeof(transid)))
4063                         return -EFAULT;
4064         } else {
4065                 transid = 0;  /* current trans */
4066         }
4067         return btrfs_wait_for_commit(root, transid);
4068 }
4069
4070 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4071 {
4072         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4073         struct btrfs_ioctl_scrub_args *sa;
4074         int ret;
4075
4076         if (!capable(CAP_SYS_ADMIN))
4077                 return -EPERM;
4078
4079         sa = memdup_user(arg, sizeof(*sa));
4080         if (IS_ERR(sa))
4081                 return PTR_ERR(sa);
4082
4083         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4084                 ret = mnt_want_write_file(file);
4085                 if (ret)
4086                         goto out;
4087         }
4088
4089         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4090                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4091                               0);
4092
4093         if (copy_to_user(arg, sa, sizeof(*sa)))
4094                 ret = -EFAULT;
4095
4096         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4097                 mnt_drop_write_file(file);
4098 out:
4099         kfree(sa);
4100         return ret;
4101 }
4102
4103 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4104 {
4105         if (!capable(CAP_SYS_ADMIN))
4106                 return -EPERM;
4107
4108         return btrfs_scrub_cancel(root->fs_info);
4109 }
4110
4111 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4112                                        void __user *arg)
4113 {
4114         struct btrfs_ioctl_scrub_args *sa;
4115         int ret;
4116
4117         if (!capable(CAP_SYS_ADMIN))
4118                 return -EPERM;
4119
4120         sa = memdup_user(arg, sizeof(*sa));
4121         if (IS_ERR(sa))
4122                 return PTR_ERR(sa);
4123
4124         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4125
4126         if (copy_to_user(arg, sa, sizeof(*sa)))
4127                 ret = -EFAULT;
4128
4129         kfree(sa);
4130         return ret;
4131 }
4132
4133 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4134                                       void __user *arg)
4135 {
4136         struct btrfs_ioctl_get_dev_stats *sa;
4137         int ret;
4138
4139         sa = memdup_user(arg, sizeof(*sa));
4140         if (IS_ERR(sa))
4141                 return PTR_ERR(sa);
4142
4143         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4144                 kfree(sa);
4145                 return -EPERM;
4146         }
4147
4148         ret = btrfs_get_dev_stats(root, sa);
4149
4150         if (copy_to_user(arg, sa, sizeof(*sa)))
4151                 ret = -EFAULT;
4152
4153         kfree(sa);
4154         return ret;
4155 }
4156
4157 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4158 {
4159         struct btrfs_ioctl_dev_replace_args *p;
4160         int ret;
4161
4162         if (!capable(CAP_SYS_ADMIN))
4163                 return -EPERM;
4164
4165         p = memdup_user(arg, sizeof(*p));
4166         if (IS_ERR(p))
4167                 return PTR_ERR(p);
4168
4169         switch (p->cmd) {
4170         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4171                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4172                         ret = -EROFS;
4173                         goto out;
4174                 }
4175                 if (atomic_xchg(
4176                         &root->fs_info->mutually_exclusive_operation_running,
4177                         1)) {
4178                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4179                 } else {
4180                         ret = btrfs_dev_replace_start(root, p);
4181                         atomic_set(
4182                          &root->fs_info->mutually_exclusive_operation_running,
4183                          0);
4184                 }
4185                 break;
4186         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4187                 btrfs_dev_replace_status(root->fs_info, p);
4188                 ret = 0;
4189                 break;
4190         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4191                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4192                 break;
4193         default:
4194                 ret = -EINVAL;
4195                 break;
4196         }
4197
4198         if (copy_to_user(arg, p, sizeof(*p)))
4199                 ret = -EFAULT;
4200 out:
4201         kfree(p);
4202         return ret;
4203 }
4204
4205 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4206 {
4207         int ret = 0;
4208         int i;
4209         u64 rel_ptr;
4210         int size;
4211         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4212         struct inode_fs_paths *ipath = NULL;
4213         struct btrfs_path *path;
4214
4215         if (!capable(CAP_DAC_READ_SEARCH))
4216                 return -EPERM;
4217
4218         path = btrfs_alloc_path();
4219         if (!path) {
4220                 ret = -ENOMEM;
4221                 goto out;
4222         }
4223
4224         ipa = memdup_user(arg, sizeof(*ipa));
4225         if (IS_ERR(ipa)) {
4226                 ret = PTR_ERR(ipa);
4227                 ipa = NULL;
4228                 goto out;
4229         }
4230
4231         size = min_t(u32, ipa->size, 4096);
4232         ipath = init_ipath(size, root, path);
4233         if (IS_ERR(ipath)) {
4234                 ret = PTR_ERR(ipath);
4235                 ipath = NULL;
4236                 goto out;
4237         }
4238
4239         ret = paths_from_inode(ipa->inum, ipath);
4240         if (ret < 0)
4241                 goto out;
4242
4243         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4244                 rel_ptr = ipath->fspath->val[i] -
4245                           (u64)(unsigned long)ipath->fspath->val;
4246                 ipath->fspath->val[i] = rel_ptr;
4247         }
4248
4249         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4250                            (void *)(unsigned long)ipath->fspath, size);
4251         if (ret) {
4252                 ret = -EFAULT;
4253                 goto out;
4254         }
4255
4256 out:
4257         btrfs_free_path(path);
4258         free_ipath(ipath);
4259         kfree(ipa);
4260
4261         return ret;
4262 }
4263
4264 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4265 {
4266         struct btrfs_data_container *inodes = ctx;
4267         const size_t c = 3 * sizeof(u64);
4268
4269         if (inodes->bytes_left >= c) {
4270                 inodes->bytes_left -= c;
4271                 inodes->val[inodes->elem_cnt] = inum;
4272                 inodes->val[inodes->elem_cnt + 1] = offset;
4273                 inodes->val[inodes->elem_cnt + 2] = root;
4274                 inodes->elem_cnt += 3;
4275         } else {
4276                 inodes->bytes_missing += c - inodes->bytes_left;
4277                 inodes->bytes_left = 0;
4278                 inodes->elem_missed += 3;
4279         }
4280
4281         return 0;
4282 }
4283
4284 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4285                                         void __user *arg)
4286 {
4287         int ret = 0;
4288         int size;
4289         struct btrfs_ioctl_logical_ino_args *loi;
4290         struct btrfs_data_container *inodes = NULL;
4291         struct btrfs_path *path = NULL;
4292
4293         if (!capable(CAP_SYS_ADMIN))
4294                 return -EPERM;
4295
4296         loi = memdup_user(arg, sizeof(*loi));
4297         if (IS_ERR(loi)) {
4298                 ret = PTR_ERR(loi);
4299                 loi = NULL;
4300                 goto out;
4301         }
4302
4303         path = btrfs_alloc_path();
4304         if (!path) {
4305                 ret = -ENOMEM;
4306                 goto out;
4307         }
4308
4309         size = min_t(u32, loi->size, 64 * 1024);
4310         inodes = init_data_container(size);
4311         if (IS_ERR(inodes)) {
4312                 ret = PTR_ERR(inodes);
4313                 inodes = NULL;
4314                 goto out;
4315         }
4316
4317         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4318                                           build_ino_list, inodes);
4319         if (ret == -EINVAL)
4320                 ret = -ENOENT;
4321         if (ret < 0)
4322                 goto out;
4323
4324         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4325                            (void *)(unsigned long)inodes, size);
4326         if (ret)
4327                 ret = -EFAULT;
4328
4329 out:
4330         btrfs_free_path(path);
4331         vfree(inodes);
4332         kfree(loi);
4333
4334         return ret;
4335 }
4336
4337 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4338                                struct btrfs_ioctl_balance_args *bargs)
4339 {
4340         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4341
4342         bargs->flags = bctl->flags;
4343
4344         if (atomic_read(&fs_info->balance_running))
4345                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4346         if (atomic_read(&fs_info->balance_pause_req))
4347                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4348         if (atomic_read(&fs_info->balance_cancel_req))
4349                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4350
4351         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4352         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4353         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4354
4355         if (lock) {
4356                 spin_lock(&fs_info->balance_lock);
4357                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4358                 spin_unlock(&fs_info->balance_lock);
4359         } else {
4360                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4361         }
4362 }
4363
4364 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4365 {
4366         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4367         struct btrfs_fs_info *fs_info = root->fs_info;
4368         struct btrfs_ioctl_balance_args *bargs;
4369         struct btrfs_balance_control *bctl;
4370         bool need_unlock; /* for mut. excl. ops lock */
4371         int ret;
4372
4373         if (!capable(CAP_SYS_ADMIN))
4374                 return -EPERM;
4375
4376         ret = mnt_want_write_file(file);
4377         if (ret)
4378                 return ret;
4379
4380 again:
4381         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4382                 mutex_lock(&fs_info->volume_mutex);
4383                 mutex_lock(&fs_info->balance_mutex);
4384                 need_unlock = true;
4385                 goto locked;
4386         }
4387
4388         /*
4389          * mut. excl. ops lock is locked.  Three possibilites:
4390          *   (1) some other op is running
4391          *   (2) balance is running
4392          *   (3) balance is paused -- special case (think resume)
4393          */
4394         mutex_lock(&fs_info->balance_mutex);
4395         if (fs_info->balance_ctl) {
4396                 /* this is either (2) or (3) */
4397                 if (!atomic_read(&fs_info->balance_running)) {
4398                         mutex_unlock(&fs_info->balance_mutex);
4399                         if (!mutex_trylock(&fs_info->volume_mutex))
4400                                 goto again;
4401                         mutex_lock(&fs_info->balance_mutex);
4402
4403                         if (fs_info->balance_ctl &&
4404                             !atomic_read(&fs_info->balance_running)) {
4405                                 /* this is (3) */
4406                                 need_unlock = false;
4407                                 goto locked;
4408                         }
4409
4410                         mutex_unlock(&fs_info->balance_mutex);
4411                         mutex_unlock(&fs_info->volume_mutex);
4412                         goto again;
4413                 } else {
4414                         /* this is (2) */
4415                         mutex_unlock(&fs_info->balance_mutex);
4416                         ret = -EINPROGRESS;
4417                         goto out;
4418                 }
4419         } else {
4420                 /* this is (1) */
4421                 mutex_unlock(&fs_info->balance_mutex);
4422                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4423                 goto out;
4424         }
4425
4426 locked:
4427         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4428
4429         if (arg) {
4430                 bargs = memdup_user(arg, sizeof(*bargs));
4431                 if (IS_ERR(bargs)) {
4432                         ret = PTR_ERR(bargs);
4433                         goto out_unlock;
4434                 }
4435
4436                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4437                         if (!fs_info->balance_ctl) {
4438                                 ret = -ENOTCONN;
4439                                 goto out_bargs;
4440                         }
4441
4442                         bctl = fs_info->balance_ctl;
4443                         spin_lock(&fs_info->balance_lock);
4444                         bctl->flags |= BTRFS_BALANCE_RESUME;
4445                         spin_unlock(&fs_info->balance_lock);
4446
4447                         goto do_balance;
4448                 }
4449         } else {
4450                 bargs = NULL;
4451         }
4452
4453         if (fs_info->balance_ctl) {
4454                 ret = -EINPROGRESS;
4455                 goto out_bargs;
4456         }
4457
4458         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4459         if (!bctl) {
4460                 ret = -ENOMEM;
4461                 goto out_bargs;
4462         }
4463
4464         bctl->fs_info = fs_info;
4465         if (arg) {
4466                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4467                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4468                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4469
4470                 bctl->flags = bargs->flags;
4471         } else {
4472                 /* balance everything - no filters */
4473                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4474         }
4475
4476 do_balance:
4477         /*
4478          * Ownership of bctl and mutually_exclusive_operation_running
4479          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4480          * or, if restriper was paused all the way until unmount, in
4481          * free_fs_info.  mutually_exclusive_operation_running is
4482          * cleared in __cancel_balance.
4483          */
4484         need_unlock = false;
4485
4486         ret = btrfs_balance(bctl, bargs);
4487
4488         if (arg) {
4489                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4490                         ret = -EFAULT;
4491         }
4492
4493 out_bargs:
4494         kfree(bargs);
4495 out_unlock:
4496         mutex_unlock(&fs_info->balance_mutex);
4497         mutex_unlock(&fs_info->volume_mutex);
4498         if (need_unlock)
4499                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4500 out:
4501         mnt_drop_write_file(file);
4502         return ret;
4503 }
4504
4505 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4506 {
4507         if (!capable(CAP_SYS_ADMIN))
4508                 return -EPERM;
4509
4510         switch (cmd) {
4511         case BTRFS_BALANCE_CTL_PAUSE:
4512                 return btrfs_pause_balance(root->fs_info);
4513         case BTRFS_BALANCE_CTL_CANCEL:
4514                 return btrfs_cancel_balance(root->fs_info);
4515         }
4516
4517         return -EINVAL;
4518 }
4519
4520 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4521                                          void __user *arg)
4522 {
4523         struct btrfs_fs_info *fs_info = root->fs_info;
4524         struct btrfs_ioctl_balance_args *bargs;
4525         int ret = 0;
4526
4527         if (!capable(CAP_SYS_ADMIN))
4528                 return -EPERM;
4529
4530         mutex_lock(&fs_info->balance_mutex);
4531         if (!fs_info->balance_ctl) {
4532                 ret = -ENOTCONN;
4533                 goto out;
4534         }
4535
4536         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4537         if (!bargs) {
4538                 ret = -ENOMEM;
4539                 goto out;
4540         }
4541
4542         update_ioctl_balance_args(fs_info, 1, bargs);
4543
4544         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4545                 ret = -EFAULT;
4546
4547         kfree(bargs);
4548 out:
4549         mutex_unlock(&fs_info->balance_mutex);
4550         return ret;
4551 }
4552
4553 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4554 {
4555         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4556         struct btrfs_ioctl_quota_ctl_args *sa;
4557         struct btrfs_trans_handle *trans = NULL;
4558         int ret;
4559         int err;
4560
4561         if (!capable(CAP_SYS_ADMIN))
4562                 return -EPERM;
4563
4564         ret = mnt_want_write_file(file);
4565         if (ret)
4566                 return ret;
4567
4568         sa = memdup_user(arg, sizeof(*sa));
4569         if (IS_ERR(sa)) {
4570                 ret = PTR_ERR(sa);
4571                 goto drop_write;
4572         }
4573
4574         down_write(&root->fs_info->subvol_sem);
4575         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4576         if (IS_ERR(trans)) {
4577                 ret = PTR_ERR(trans);
4578                 goto out;
4579         }
4580
4581         switch (sa->cmd) {
4582         case BTRFS_QUOTA_CTL_ENABLE:
4583                 ret = btrfs_quota_enable(trans, root->fs_info);
4584                 break;
4585         case BTRFS_QUOTA_CTL_DISABLE:
4586                 ret = btrfs_quota_disable(trans, root->fs_info);
4587                 break;
4588         default:
4589                 ret = -EINVAL;
4590                 break;
4591         }
4592
4593         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4594         if (err && !ret)
4595                 ret = err;
4596 out:
4597         kfree(sa);
4598         up_write(&root->fs_info->subvol_sem);
4599 drop_write:
4600         mnt_drop_write_file(file);
4601         return ret;
4602 }
4603
4604 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4605 {
4606         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4607         struct btrfs_ioctl_qgroup_assign_args *sa;
4608         struct btrfs_trans_handle *trans;
4609         int ret;
4610         int err;
4611
4612         if (!capable(CAP_SYS_ADMIN))
4613                 return -EPERM;
4614
4615         ret = mnt_want_write_file(file);
4616         if (ret)
4617                 return ret;
4618
4619         sa = memdup_user(arg, sizeof(*sa));
4620         if (IS_ERR(sa)) {
4621                 ret = PTR_ERR(sa);
4622                 goto drop_write;
4623         }
4624
4625         trans = btrfs_join_transaction(root);
4626         if (IS_ERR(trans)) {
4627                 ret = PTR_ERR(trans);
4628                 goto out;
4629         }
4630
4631         /* FIXME: check if the IDs really exist */
4632         if (sa->assign) {
4633                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4634                                                 sa->src, sa->dst);
4635         } else {
4636                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4637                                                 sa->src, sa->dst);
4638         }
4639
4640         err = btrfs_end_transaction(trans, root);
4641         if (err && !ret)
4642                 ret = err;
4643
4644 out:
4645         kfree(sa);
4646 drop_write:
4647         mnt_drop_write_file(file);
4648         return ret;
4649 }
4650
4651 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4652 {
4653         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4654         struct btrfs_ioctl_qgroup_create_args *sa;
4655         struct btrfs_trans_handle *trans;
4656         int ret;
4657         int err;
4658
4659         if (!capable(CAP_SYS_ADMIN))
4660                 return -EPERM;
4661
4662         ret = mnt_want_write_file(file);
4663         if (ret)
4664                 return ret;
4665
4666         sa = memdup_user(arg, sizeof(*sa));
4667         if (IS_ERR(sa)) {
4668                 ret = PTR_ERR(sa);
4669                 goto drop_write;
4670         }
4671
4672         if (!sa->qgroupid) {
4673                 ret = -EINVAL;
4674                 goto out;
4675         }
4676
4677         trans = btrfs_join_transaction(root);
4678         if (IS_ERR(trans)) {
4679                 ret = PTR_ERR(trans);
4680                 goto out;
4681         }
4682
4683         /* FIXME: check if the IDs really exist */
4684         if (sa->create) {
4685                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4686                                           NULL);
4687         } else {
4688                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4689         }
4690
4691         err = btrfs_end_transaction(trans, root);
4692         if (err && !ret)
4693                 ret = err;
4694
4695 out:
4696         kfree(sa);
4697 drop_write:
4698         mnt_drop_write_file(file);
4699         return ret;
4700 }
4701
4702 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4703 {
4704         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4705         struct btrfs_ioctl_qgroup_limit_args *sa;
4706         struct btrfs_trans_handle *trans;
4707         int ret;
4708         int err;
4709         u64 qgroupid;
4710
4711         if (!capable(CAP_SYS_ADMIN))
4712                 return -EPERM;
4713
4714         ret = mnt_want_write_file(file);
4715         if (ret)
4716                 return ret;
4717
4718         sa = memdup_user(arg, sizeof(*sa));
4719         if (IS_ERR(sa)) {
4720                 ret = PTR_ERR(sa);
4721                 goto drop_write;
4722         }
4723
4724         trans = btrfs_join_transaction(root);
4725         if (IS_ERR(trans)) {
4726                 ret = PTR_ERR(trans);
4727                 goto out;
4728         }
4729
4730         qgroupid = sa->qgroupid;
4731         if (!qgroupid) {
4732                 /* take the current subvol as qgroup */
4733                 qgroupid = root->root_key.objectid;
4734         }
4735
4736         /* FIXME: check if the IDs really exist */
4737         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4738
4739         err = btrfs_end_transaction(trans, root);
4740         if (err && !ret)
4741                 ret = err;
4742
4743 out:
4744         kfree(sa);
4745 drop_write:
4746         mnt_drop_write_file(file);
4747         return ret;
4748 }
4749
4750 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4751 {
4752         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4753         struct btrfs_ioctl_quota_rescan_args *qsa;
4754         int ret;
4755
4756         if (!capable(CAP_SYS_ADMIN))
4757                 return -EPERM;
4758
4759         ret = mnt_want_write_file(file);
4760         if (ret)
4761                 return ret;
4762
4763         qsa = memdup_user(arg, sizeof(*qsa));
4764         if (IS_ERR(qsa)) {
4765                 ret = PTR_ERR(qsa);
4766                 goto drop_write;
4767         }
4768
4769         if (qsa->flags) {
4770                 ret = -EINVAL;
4771                 goto out;
4772         }
4773
4774         ret = btrfs_qgroup_rescan(root->fs_info);
4775
4776 out:
4777         kfree(qsa);
4778 drop_write:
4779         mnt_drop_write_file(file);
4780         return ret;
4781 }
4782
4783 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4784 {
4785         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4786         struct btrfs_ioctl_quota_rescan_args *qsa;
4787         int ret = 0;
4788
4789         if (!capable(CAP_SYS_ADMIN))
4790                 return -EPERM;
4791
4792         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4793         if (!qsa)
4794                 return -ENOMEM;
4795
4796         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4797                 qsa->flags = 1;
4798                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4799         }
4800
4801         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4802                 ret = -EFAULT;
4803
4804         kfree(qsa);
4805         return ret;
4806 }
4807
4808 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4809 {
4810         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4811
4812         if (!capable(CAP_SYS_ADMIN))
4813                 return -EPERM;
4814
4815         return btrfs_qgroup_wait_for_completion(root->fs_info);
4816 }
4817
4818 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4819                                             struct btrfs_ioctl_received_subvol_args *sa)
4820 {
4821         struct inode *inode = file_inode(file);
4822         struct btrfs_root *root = BTRFS_I(inode)->root;
4823         struct btrfs_root_item *root_item = &root->root_item;
4824         struct btrfs_trans_handle *trans;
4825         struct timespec ct = CURRENT_TIME;
4826         int ret = 0;
4827         int received_uuid_changed;
4828
4829         if (!inode_owner_or_capable(inode))
4830                 return -EPERM;
4831
4832         ret = mnt_want_write_file(file);
4833         if (ret < 0)
4834                 return ret;
4835
4836         down_write(&root->fs_info->subvol_sem);
4837
4838         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4839                 ret = -EINVAL;
4840                 goto out;
4841         }
4842
4843         if (btrfs_root_readonly(root)) {
4844                 ret = -EROFS;
4845                 goto out;
4846         }
4847
4848         /*
4849          * 1 - root item
4850          * 2 - uuid items (received uuid + subvol uuid)
4851          */
4852         trans = btrfs_start_transaction(root, 3);
4853         if (IS_ERR(trans)) {
4854                 ret = PTR_ERR(trans);
4855                 trans = NULL;
4856                 goto out;
4857         }
4858
4859         sa->rtransid = trans->transid;
4860         sa->rtime.sec = ct.tv_sec;
4861         sa->rtime.nsec = ct.tv_nsec;
4862
4863         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4864                                        BTRFS_UUID_SIZE);
4865         if (received_uuid_changed &&
4866             !btrfs_is_empty_uuid(root_item->received_uuid))
4867                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4868                                     root_item->received_uuid,
4869                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4870                                     root->root_key.objectid);
4871         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4872         btrfs_set_root_stransid(root_item, sa->stransid);
4873         btrfs_set_root_rtransid(root_item, sa->rtransid);
4874         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4875         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4876         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4877         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4878
4879         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4880                                 &root->root_key, &root->root_item);
4881         if (ret < 0) {
4882                 btrfs_end_transaction(trans, root);
4883                 goto out;
4884         }
4885         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4886                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4887                                           sa->uuid,
4888                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4889                                           root->root_key.objectid);
4890                 if (ret < 0 && ret != -EEXIST) {
4891                         btrfs_abort_transaction(trans, root, ret);
4892                         goto out;
4893                 }
4894         }
4895         ret = btrfs_commit_transaction(trans, root);
4896         if (ret < 0) {
4897                 btrfs_abort_transaction(trans, root, ret);
4898                 goto out;
4899         }
4900
4901 out:
4902         up_write(&root->fs_info->subvol_sem);
4903         mnt_drop_write_file(file);
4904         return ret;
4905 }
4906
4907 #ifdef CONFIG_64BIT
4908 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4909                                                 void __user *arg)
4910 {
4911         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4912         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4913         int ret = 0;
4914
4915         args32 = memdup_user(arg, sizeof(*args32));
4916         if (IS_ERR(args32)) {
4917                 ret = PTR_ERR(args32);
4918                 args32 = NULL;
4919                 goto out;
4920         }
4921
4922         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4923         if (!args64) {
4924                 ret = -ENOMEM;
4925                 goto out;
4926         }
4927
4928         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4929         args64->stransid = args32->stransid;
4930         args64->rtransid = args32->rtransid;
4931         args64->stime.sec = args32->stime.sec;
4932         args64->stime.nsec = args32->stime.nsec;
4933         args64->rtime.sec = args32->rtime.sec;
4934         args64->rtime.nsec = args32->rtime.nsec;
4935         args64->flags = args32->flags;
4936
4937         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4938         if (ret)
4939                 goto out;
4940
4941         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4942         args32->stransid = args64->stransid;
4943         args32->rtransid = args64->rtransid;
4944         args32->stime.sec = args64->stime.sec;
4945         args32->stime.nsec = args64->stime.nsec;
4946         args32->rtime.sec = args64->rtime.sec;
4947         args32->rtime.nsec = args64->rtime.nsec;
4948         args32->flags = args64->flags;
4949
4950         ret = copy_to_user(arg, args32, sizeof(*args32));
4951         if (ret)
4952                 ret = -EFAULT;
4953
4954 out:
4955         kfree(args32);
4956         kfree(args64);
4957         return ret;
4958 }
4959 #endif
4960
4961 static long btrfs_ioctl_set_received_subvol(struct file *file,
4962                                             void __user *arg)
4963 {
4964         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4965         int ret = 0;
4966
4967         sa = memdup_user(arg, sizeof(*sa));
4968         if (IS_ERR(sa)) {
4969                 ret = PTR_ERR(sa);
4970                 sa = NULL;
4971                 goto out;
4972         }
4973
4974         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4975
4976         if (ret)
4977                 goto out;
4978
4979         ret = copy_to_user(arg, sa, sizeof(*sa));
4980         if (ret)
4981                 ret = -EFAULT;
4982
4983 out:
4984         kfree(sa);
4985         return ret;
4986 }
4987
4988 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4989 {
4990         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4991         size_t len;
4992         int ret;
4993         char label[BTRFS_LABEL_SIZE];
4994
4995         spin_lock(&root->fs_info->super_lock);
4996         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4997         spin_unlock(&root->fs_info->super_lock);
4998
4999         len = strnlen(label, BTRFS_LABEL_SIZE);
5000
5001         if (len == BTRFS_LABEL_SIZE) {
5002                 btrfs_warn(root->fs_info,
5003                         "label is too long, return the first %zu bytes", --len);
5004         }
5005
5006         ret = copy_to_user(arg, label, len);
5007
5008         return ret ? -EFAULT : 0;
5009 }
5010
5011 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5012 {
5013         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5014         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5015         struct btrfs_trans_handle *trans;
5016         char label[BTRFS_LABEL_SIZE];
5017         int ret;
5018
5019         if (!capable(CAP_SYS_ADMIN))
5020                 return -EPERM;
5021
5022         if (copy_from_user(label, arg, sizeof(label)))
5023                 return -EFAULT;
5024
5025         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5026                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5027                        BTRFS_LABEL_SIZE - 1);
5028                 return -EINVAL;
5029         }
5030
5031         ret = mnt_want_write_file(file);
5032         if (ret)
5033                 return ret;
5034
5035         trans = btrfs_start_transaction(root, 0);
5036         if (IS_ERR(trans)) {
5037                 ret = PTR_ERR(trans);
5038                 goto out_unlock;
5039         }
5040
5041         spin_lock(&root->fs_info->super_lock);
5042         strcpy(super_block->label, label);
5043         spin_unlock(&root->fs_info->super_lock);
5044         ret = btrfs_commit_transaction(trans, root);
5045
5046 out_unlock:
5047         mnt_drop_write_file(file);
5048         return ret;
5049 }
5050
5051 #define INIT_FEATURE_FLAGS(suffix) \
5052         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5053           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5054           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5055
5056 static int btrfs_ioctl_get_supported_features(struct file *file,
5057                                               void __user *arg)
5058 {
5059         static struct btrfs_ioctl_feature_flags features[3] = {
5060                 INIT_FEATURE_FLAGS(SUPP),
5061                 INIT_FEATURE_FLAGS(SAFE_SET),
5062                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5063         };
5064
5065         if (copy_to_user(arg, &features, sizeof(features)))
5066                 return -EFAULT;
5067
5068         return 0;
5069 }
5070
5071 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5072 {
5073         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5074         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5075         struct btrfs_ioctl_feature_flags features;
5076
5077         features.compat_flags = btrfs_super_compat_flags(super_block);
5078         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5079         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5080
5081         if (copy_to_user(arg, &features, sizeof(features)))
5082                 return -EFAULT;
5083
5084         return 0;
5085 }
5086
5087 static int check_feature_bits(struct btrfs_root *root,
5088                               enum btrfs_feature_set set,
5089                               u64 change_mask, u64 flags, u64 supported_flags,
5090                               u64 safe_set, u64 safe_clear)
5091 {
5092         const char *type = btrfs_feature_set_names[set];
5093         char *names;
5094         u64 disallowed, unsupported;
5095         u64 set_mask = flags & change_mask;
5096         u64 clear_mask = ~flags & change_mask;
5097
5098         unsupported = set_mask & ~supported_flags;
5099         if (unsupported) {
5100                 names = btrfs_printable_features(set, unsupported);
5101                 if (names) {
5102                         btrfs_warn(root->fs_info,
5103                            "this kernel does not support the %s feature bit%s",
5104                            names, strchr(names, ',') ? "s" : "");
5105                         kfree(names);
5106                 } else
5107                         btrfs_warn(root->fs_info,
5108                            "this kernel does not support %s bits 0x%llx",
5109                            type, unsupported);
5110                 return -EOPNOTSUPP;
5111         }
5112
5113         disallowed = set_mask & ~safe_set;
5114         if (disallowed) {
5115                 names = btrfs_printable_features(set, disallowed);
5116                 if (names) {
5117                         btrfs_warn(root->fs_info,
5118                            "can't set the %s feature bit%s while mounted",
5119                            names, strchr(names, ',') ? "s" : "");
5120                         kfree(names);
5121                 } else
5122                         btrfs_warn(root->fs_info,
5123                            "can't set %s bits 0x%llx while mounted",
5124                            type, disallowed);
5125                 return -EPERM;
5126         }
5127
5128         disallowed = clear_mask & ~safe_clear;
5129         if (disallowed) {
5130                 names = btrfs_printable_features(set, disallowed);
5131                 if (names) {
5132                         btrfs_warn(root->fs_info,
5133                            "can't clear the %s feature bit%s while mounted",
5134                            names, strchr(names, ',') ? "s" : "");
5135                         kfree(names);
5136                 } else
5137                         btrfs_warn(root->fs_info,
5138                            "can't clear %s bits 0x%llx while mounted",
5139                            type, disallowed);
5140                 return -EPERM;
5141         }
5142
5143         return 0;
5144 }
5145
5146 #define check_feature(root, change_mask, flags, mask_base)      \
5147 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5148                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5149                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5150                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5151
5152 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5153 {
5154         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5155         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5156         struct btrfs_ioctl_feature_flags flags[2];
5157         struct btrfs_trans_handle *trans;
5158         u64 newflags;
5159         int ret;
5160
5161         if (!capable(CAP_SYS_ADMIN))
5162                 return -EPERM;
5163
5164         if (copy_from_user(flags, arg, sizeof(flags)))
5165                 return -EFAULT;
5166
5167         /* Nothing to do */
5168         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5169             !flags[0].incompat_flags)
5170                 return 0;
5171
5172         ret = check_feature(root, flags[0].compat_flags,
5173                             flags[1].compat_flags, COMPAT);
5174         if (ret)
5175                 return ret;
5176
5177         ret = check_feature(root, flags[0].compat_ro_flags,
5178                             flags[1].compat_ro_flags, COMPAT_RO);
5179         if (ret)
5180                 return ret;
5181
5182         ret = check_feature(root, flags[0].incompat_flags,
5183                             flags[1].incompat_flags, INCOMPAT);
5184         if (ret)
5185                 return ret;
5186
5187         trans = btrfs_start_transaction(root, 0);
5188         if (IS_ERR(trans))
5189                 return PTR_ERR(trans);
5190
5191         spin_lock(&root->fs_info->super_lock);
5192         newflags = btrfs_super_compat_flags(super_block);
5193         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5194         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5195         btrfs_set_super_compat_flags(super_block, newflags);
5196
5197         newflags = btrfs_super_compat_ro_flags(super_block);
5198         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5199         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5200         btrfs_set_super_compat_ro_flags(super_block, newflags);
5201
5202         newflags = btrfs_super_incompat_flags(super_block);
5203         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5204         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5205         btrfs_set_super_incompat_flags(super_block, newflags);
5206         spin_unlock(&root->fs_info->super_lock);
5207
5208         return btrfs_commit_transaction(trans, root);
5209 }
5210
5211 long btrfs_ioctl(struct file *file, unsigned int
5212                 cmd, unsigned long arg)
5213 {
5214         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5215         void __user *argp = (void __user *)arg;
5216
5217         switch (cmd) {
5218         case FS_IOC_GETFLAGS:
5219                 return btrfs_ioctl_getflags(file, argp);
5220         case FS_IOC_SETFLAGS:
5221                 return btrfs_ioctl_setflags(file, argp);
5222         case FS_IOC_GETVERSION:
5223                 return btrfs_ioctl_getversion(file, argp);
5224         case FITRIM:
5225                 return btrfs_ioctl_fitrim(file, argp);
5226         case BTRFS_IOC_SNAP_CREATE:
5227                 return btrfs_ioctl_snap_create(file, argp, 0);
5228         case BTRFS_IOC_SNAP_CREATE_V2:
5229                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5230         case BTRFS_IOC_SUBVOL_CREATE:
5231                 return btrfs_ioctl_snap_create(file, argp, 1);
5232         case BTRFS_IOC_SUBVOL_CREATE_V2:
5233                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5234         case BTRFS_IOC_SNAP_DESTROY:
5235                 return btrfs_ioctl_snap_destroy(file, argp);
5236         case BTRFS_IOC_SUBVOL_GETFLAGS:
5237                 return btrfs_ioctl_subvol_getflags(file, argp);
5238         case BTRFS_IOC_SUBVOL_SETFLAGS:
5239                 return btrfs_ioctl_subvol_setflags(file, argp);
5240         case BTRFS_IOC_DEFAULT_SUBVOL:
5241                 return btrfs_ioctl_default_subvol(file, argp);
5242         case BTRFS_IOC_DEFRAG:
5243                 return btrfs_ioctl_defrag(file, NULL);
5244         case BTRFS_IOC_DEFRAG_RANGE:
5245                 return btrfs_ioctl_defrag(file, argp);
5246         case BTRFS_IOC_RESIZE:
5247                 return btrfs_ioctl_resize(file, argp);
5248         case BTRFS_IOC_ADD_DEV:
5249                 return btrfs_ioctl_add_dev(root, argp);
5250         case BTRFS_IOC_RM_DEV:
5251                 return btrfs_ioctl_rm_dev(file, argp);
5252         case BTRFS_IOC_FS_INFO:
5253                 return btrfs_ioctl_fs_info(root, argp);
5254         case BTRFS_IOC_DEV_INFO:
5255                 return btrfs_ioctl_dev_info(root, argp);
5256         case BTRFS_IOC_BALANCE:
5257                 return btrfs_ioctl_balance(file, NULL);
5258         case BTRFS_IOC_CLONE:
5259                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5260         case BTRFS_IOC_CLONE_RANGE:
5261                 return btrfs_ioctl_clone_range(file, argp);
5262         case BTRFS_IOC_TRANS_START:
5263                 return btrfs_ioctl_trans_start(file);
5264         case BTRFS_IOC_TRANS_END:
5265                 return btrfs_ioctl_trans_end(file);
5266         case BTRFS_IOC_TREE_SEARCH:
5267                 return btrfs_ioctl_tree_search(file, argp);
5268         case BTRFS_IOC_TREE_SEARCH_V2:
5269                 return btrfs_ioctl_tree_search_v2(file, argp);
5270         case BTRFS_IOC_INO_LOOKUP:
5271                 return btrfs_ioctl_ino_lookup(file, argp);
5272         case BTRFS_IOC_INO_PATHS:
5273                 return btrfs_ioctl_ino_to_path(root, argp);
5274         case BTRFS_IOC_LOGICAL_INO:
5275                 return btrfs_ioctl_logical_to_ino(root, argp);
5276         case BTRFS_IOC_SPACE_INFO:
5277                 return btrfs_ioctl_space_info(root, argp);
5278         case BTRFS_IOC_SYNC: {
5279                 int ret;
5280
5281                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5282                 if (ret)
5283                         return ret;
5284                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
5285                 /*
5286                  * The transaction thread may want to do more work,
5287                  * namely it pokes the cleaner ktread that will start
5288                  * processing uncleaned subvols.
5289                  */
5290                 wake_up_process(root->fs_info->transaction_kthread);
5291                 return ret;
5292         }
5293         case BTRFS_IOC_START_SYNC:
5294                 return btrfs_ioctl_start_sync(root, argp);
5295         case BTRFS_IOC_WAIT_SYNC:
5296                 return btrfs_ioctl_wait_sync(root, argp);
5297         case BTRFS_IOC_SCRUB:
5298                 return btrfs_ioctl_scrub(file, argp);
5299         case BTRFS_IOC_SCRUB_CANCEL:
5300                 return btrfs_ioctl_scrub_cancel(root, argp);
5301         case BTRFS_IOC_SCRUB_PROGRESS:
5302                 return btrfs_ioctl_scrub_progress(root, argp);
5303         case BTRFS_IOC_BALANCE_V2:
5304                 return btrfs_ioctl_balance(file, argp);
5305         case BTRFS_IOC_BALANCE_CTL:
5306                 return btrfs_ioctl_balance_ctl(root, arg);
5307         case BTRFS_IOC_BALANCE_PROGRESS:
5308                 return btrfs_ioctl_balance_progress(root, argp);
5309         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5310                 return btrfs_ioctl_set_received_subvol(file, argp);
5311 #ifdef CONFIG_64BIT
5312         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5313                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5314 #endif
5315         case BTRFS_IOC_SEND:
5316                 return btrfs_ioctl_send(file, argp);
5317         case BTRFS_IOC_GET_DEV_STATS:
5318                 return btrfs_ioctl_get_dev_stats(root, argp);
5319         case BTRFS_IOC_QUOTA_CTL:
5320                 return btrfs_ioctl_quota_ctl(file, argp);
5321         case BTRFS_IOC_QGROUP_ASSIGN:
5322                 return btrfs_ioctl_qgroup_assign(file, argp);
5323         case BTRFS_IOC_QGROUP_CREATE:
5324                 return btrfs_ioctl_qgroup_create(file, argp);
5325         case BTRFS_IOC_QGROUP_LIMIT:
5326                 return btrfs_ioctl_qgroup_limit(file, argp);
5327         case BTRFS_IOC_QUOTA_RESCAN:
5328                 return btrfs_ioctl_quota_rescan(file, argp);
5329         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5330                 return btrfs_ioctl_quota_rescan_status(file, argp);
5331         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5332                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5333         case BTRFS_IOC_DEV_REPLACE:
5334                 return btrfs_ioctl_dev_replace(root, argp);
5335         case BTRFS_IOC_GET_FSLABEL:
5336                 return btrfs_ioctl_get_fslabel(file, argp);
5337         case BTRFS_IOC_SET_FSLABEL:
5338                 return btrfs_ioctl_set_fslabel(file, argp);
5339         case BTRFS_IOC_FILE_EXTENT_SAME:
5340                 return btrfs_ioctl_file_extent_same(file, argp);
5341         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5342                 return btrfs_ioctl_get_supported_features(file, argp);
5343         case BTRFS_IOC_GET_FEATURES:
5344                 return btrfs_ioctl_get_features(file, argp);
5345         case BTRFS_IOC_SET_FEATURES:
5346                 return btrfs_ioctl_set_features(file, argp);
5347         }
5348
5349         return -ENOTTY;
5350 }