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         /*
 460          * Don't create subvolume whose level is not zero. Or qgroup will be
 461          * screwed up since it assume subvolme qgroup's level to be 0.
 462          */
 463         if (btrfs_qgroup_level(objectid))
 464                 return -ENOSPC;
 465
 466         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 467         /*
 468          * The same as the snapshot creation, please see the comment
 469          * of create_snapshot().
 470          */
 471         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 472                                                8, &qgroup_reserved, false);
 473         if (ret)
 474                 return ret;
 475
 476         trans = btrfs_start_transaction(root, 0);
 477         if (IS_ERR(trans)) {
 478                 ret = PTR_ERR(trans);
 479                 btrfs_subvolume_release_metadata(root, &block_rsv,
 480                                                  qgroup_reserved);
 481                 return ret;
 482         }
 483         trans->block_rsv = &block_rsv;
 484         trans->bytes_reserved = block_rsv.size;
 485
 486         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 487         if (ret)
 488                 goto fail;
 489
 490         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 491         if (IS_ERR(leaf)) {
 492                 ret = PTR_ERR(leaf);
 493                 goto fail;
 494         }
 495
 496         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 497         btrfs_set_header_bytenr(leaf, leaf->start);
 498         btrfs_set_header_generation(leaf, trans->transid);
 499         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 500         btrfs_set_header_owner(leaf, objectid);
 501
 502         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 503                             BTRFS_FSID_SIZE);
 504         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 505                             btrfs_header_chunk_tree_uuid(leaf),
 506                             BTRFS_UUID_SIZE);
 507         btrfs_mark_buffer_dirty(leaf);
 508
 509         memset(&root_item, 0, sizeof(root_item));
 510
 511         inode_item = &root_item.inode;
 512         btrfs_set_stack_inode_generation(inode_item, 1);
 513         btrfs_set_stack_inode_size(inode_item, 3);
 514         btrfs_set_stack_inode_nlink(inode_item, 1);
 515         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 516         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 517
 518         btrfs_set_root_flags(&root_item, 0);
 519         btrfs_set_root_limit(&root_item, 0);
 520         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 521
 522         btrfs_set_root_bytenr(&root_item, leaf->start);
 523         btrfs_set_root_generation(&root_item, trans->transid);
 524         btrfs_set_root_level(&root_item, 0);
 525         btrfs_set_root_refs(&root_item, 1);
 526         btrfs_set_root_used(&root_item, leaf->len);
 527         btrfs_set_root_last_snapshot(&root_item, 0);
 528
 529         btrfs_set_root_generation_v2(&root_item,
 530                         btrfs_root_generation(&root_item));
 531         uuid_le_gen(&new_uuid);
 532         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 533         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 534         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 535         root_item.ctime = root_item.otime;
 536         btrfs_set_root_ctransid(&root_item, trans->transid);
 537         btrfs_set_root_otransid(&root_item, trans->transid);
 538
 539         btrfs_tree_unlock(leaf);
 540         free_extent_buffer(leaf);
 541         leaf = NULL;
 542
 543         btrfs_set_root_dirid(&root_item, new_dirid);
 544
 545         key.objectid = objectid;
 546         key.offset = 0;
 547         key.type = BTRFS_ROOT_ITEM_KEY;
 548         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 549                                 &root_item);
 550         if (ret)
 551                 goto fail;
 552
 553         key.offset = (u64)-1;
 554         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 555         if (IS_ERR(new_root)) {
 556                 ret = PTR_ERR(new_root);
 557                 btrfs_abort_transaction(trans, root, ret);
 558                 goto fail;
 559         }
 560
 561         btrfs_record_root_in_trans(trans, new_root);
 562
 563         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 564         if (ret) {
 565                 /* We potentially lose an unused inode item here */
 566                 btrfs_abort_transaction(trans, root, ret);
 567                 goto fail;
 568         }
 569
 570         /*
 571          * insert the directory item
 572          */
 573         ret = btrfs_set_inode_index(dir, &index);
 574         if (ret) {
 575                 btrfs_abort_transaction(trans, root, ret);
 576                 goto fail;
 577         }
 578
 579         ret = btrfs_insert_dir_item(trans, root,
 580                                     name, namelen, dir, &key,
 581                                     BTRFS_FT_DIR, index);
 582         if (ret) {
 583                 btrfs_abort_transaction(trans, root, ret);
 584                 goto fail;
 585         }
 586
 587         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 588         ret = btrfs_update_inode(trans, root, dir);
 589         BUG_ON(ret);
 590
 591         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 592                                  objectid, root->root_key.objectid,
 593                                  btrfs_ino(dir), index, name, namelen);
 594         BUG_ON(ret);
 595
 596         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 597                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 598                                   objectid);
 599         if (ret)
 600                 btrfs_abort_transaction(trans, root, ret);
 601
 602 fail:
 603         trans->block_rsv = NULL;
 604         trans->bytes_reserved = 0;
 605         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 606
 607         if (async_transid) {
 608                 *async_transid = trans->transid;
 609                 err = btrfs_commit_transaction_async(trans, root, 1);
 610                 if (err)
 611                         err = btrfs_commit_transaction(trans, root);
 612         } else {
 613                 err = btrfs_commit_transaction(trans, root);
 614         }
 615         if (err && !ret)
 616                 ret = err;
 617
 618         if (!ret) {
 619                 inode = btrfs_lookup_dentry(dir, dentry);
 620                 if (IS_ERR(inode))
 621                         return PTR_ERR(inode);
 622                 d_instantiate(dentry, inode);
 623         }
 624         return ret;
 625 }
 626
 627 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 628 {
 629         s64 writers;
 630         DEFINE_WAIT(wait);
 631
 632         do {
 633                 prepare_to_wait(&root->subv_writers->wait, &wait,
 634                                 TASK_UNINTERRUPTIBLE);
 635
 636                 writers = percpu_counter_sum(&root->subv_writers->counter);
 637                 if (writers)
 638                         schedule();
 639
 640                 finish_wait(&root->subv_writers->wait, &wait);
 641         } while (writers);
 642 }
 643
 644 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 645                            struct dentry *dentry, char *name, int namelen,
 646                            u64 *async_transid, bool readonly,
 647                            struct btrfs_qgroup_inherit *inherit)
 648 {
 649         struct inode *inode;
 650         struct btrfs_pending_snapshot *pending_snapshot;
 651         struct btrfs_trans_handle *trans;
 652         int ret;
 653
 654         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 655                 return -EINVAL;
 656
 657         atomic_inc(&root->will_be_snapshoted);
 658         smp_mb__after_atomic();
 659         btrfs_wait_for_no_snapshoting_writes(root);
 660
 661         ret = btrfs_start_delalloc_inodes(root, 0);
 662         if (ret)
 663                 goto out;
 664
 665         btrfs_wait_ordered_extents(root, -1);
 666
 667         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 668         if (!pending_snapshot) {
 669                 ret = -ENOMEM;
 670                 goto out;
 671         }
 672
 673         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 674                              BTRFS_BLOCK_RSV_TEMP);
 675         /*
 676          * 1 - parent dir inode
 677          * 2 - dir entries
 678          * 1 - root item
 679          * 2 - root ref/backref
 680          * 1 - root of snapshot
 681          * 1 - UUID item
 682          */
 683         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 684                                         &pending_snapshot->block_rsv, 8,
 685                                         &pending_snapshot->qgroup_reserved,
 686                                         false);
 687         if (ret)
 688                 goto free;
 689
 690         pending_snapshot->dentry = dentry;
 691         pending_snapshot->root = root;
 692         pending_snapshot->readonly = readonly;
 693         pending_snapshot->dir = dir;
 694         pending_snapshot->inherit = inherit;
 695
 696         trans = btrfs_start_transaction(root, 0);
 697         if (IS_ERR(trans)) {
 698                 ret = PTR_ERR(trans);
 699                 goto fail;
 700         }
 701
 702         spin_lock(&root->fs_info->trans_lock);
 703         list_add(&pending_snapshot->list,
 704                  &trans->transaction->pending_snapshots);
 705         spin_unlock(&root->fs_info->trans_lock);
 706         if (async_transid) {
 707                 *async_transid = trans->transid;
 708                 ret = btrfs_commit_transaction_async(trans,
 709                                      root->fs_info->extent_root, 1);
 710                 if (ret)
 711                         ret = btrfs_commit_transaction(trans, root);
 712         } else {
 713                 ret = btrfs_commit_transaction(trans,
 714                                                root->fs_info->extent_root);
 715         }
 716         if (ret)
 717                 goto fail;
 718
 719         ret = pending_snapshot->error;
 720         if (ret)
 721                 goto fail;
 722
 723         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 724         if (ret)
 725                 goto fail;
 726
 727         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 728         if (IS_ERR(inode)) {
 729                 ret = PTR_ERR(inode);
 730                 goto fail;
 731         }
 732
 733         d_instantiate(dentry, inode);
 734         ret = 0;
 735 fail:
 736         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 737                                          &pending_snapshot->block_rsv,
 738                                          pending_snapshot->qgroup_reserved);
 739 free:
 740         kfree(pending_snapshot);
 741 out:
 742         if (atomic_dec_and_test(&root->will_be_snapshoted))
 743                 wake_up_atomic_t(&root->will_be_snapshoted);
 744         return ret;
 745 }
 746
 747 /*  copy of may_delete in fs/namei.c()
 748  *      Check whether we can remove a link victim from directory dir, check
 749  *  whether the type of victim is right.
 750  *  1. We can't do it if dir is read-only (done in permission())
 751  *  2. We should have write and exec permissions on dir
 752  *  3. We can't remove anything from append-only dir
 753  *  4. We can't do anything with immutable dir (done in permission())
 754  *  5. If the sticky bit on dir is set we should either
 755  *      a. be owner of dir, or
 756  *      b. be owner of victim, or
 757  *      c. have CAP_FOWNER capability
 758  *  6. If the victim is append-only or immutable we can't do antyhing with
 759  *     links pointing to it.
 760  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 761  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 762  *  9. We can't remove a root or mountpoint.
 763  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 764  *     nfs_async_unlink().
 765  */
 766
 767 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 768 {
 769         int error;
 770
 771         if (d_really_is_negative(victim))
 772                 return -ENOENT;
 773
 774         BUG_ON(d_inode(victim->d_parent) != dir);
 775         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 776
 777         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 778         if (error)
 779                 return error;
 780         if (IS_APPEND(dir))
 781                 return -EPERM;
 782         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 783             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 784                 return -EPERM;
 785         if (isdir) {
 786                 if (!d_is_dir(victim))
 787                         return -ENOTDIR;
 788                 if (IS_ROOT(victim))
 789                         return -EBUSY;
 790         } else if (d_is_dir(victim))
 791                 return -EISDIR;
 792         if (IS_DEADDIR(dir))
 793                 return -ENOENT;
 794         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 795                 return -EBUSY;
 796         return 0;
 797 }
 798
 799 /* copy of may_create in fs/namei.c() */
 800 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 801 {
 802         if (d_really_is_positive(child))
 803                 return -EEXIST;
 804         if (IS_DEADDIR(dir))
 805                 return -ENOENT;
 806         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 807 }
 808
 809 /*
 810  * Create a new subvolume below @parent.  This is largely modeled after
 811  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 812  * inside this filesystem so it's quite a bit simpler.
 813  */
 814 static noinline int btrfs_mksubvol(struct path *parent,
 815                                    char *name, int namelen,
 816                                    struct btrfs_root *snap_src,
 817                                    u64 *async_transid, bool readonly,
 818                                    struct btrfs_qgroup_inherit *inherit)
 819 {
 820         struct inode *dir  = d_inode(parent->dentry);
 821         struct dentry *dentry;
 822         int error;
 823
 824         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 825         if (error == -EINTR)
 826                 return error;
 827
 828         dentry = lookup_one_len(name, parent->dentry, namelen);
 829         error = PTR_ERR(dentry);
 830         if (IS_ERR(dentry))
 831                 goto out_unlock;
 832
 833         error = -EEXIST;
 834         if (d_really_is_positive(dentry))
 835                 goto out_dput;
 836
 837         error = btrfs_may_create(dir, dentry);
 838         if (error)
 839                 goto out_dput;
 840
 841         /*
 842          * even if this name doesn't exist, we may get hash collisions.
 843          * check for them now when we can safely fail
 844          */
 845         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 846                                                dir->i_ino, name,
 847                                                namelen);
 848         if (error)
 849                 goto out_dput;
 850
 851         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 852
 853         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 854                 goto out_up_read;
 855
 856         if (snap_src) {
 857                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 858                                         async_transid, readonly, inherit);
 859         } else {
 860                 error = create_subvol(dir, dentry, name, namelen,
 861                                       async_transid, inherit);
 862         }
 863         if (!error)
 864                 fsnotify_mkdir(dir, dentry);
 865 out_up_read:
 866         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 867 out_dput:
 868         dput(dentry);
 869 out_unlock:
 870         mutex_unlock(&dir->i_mutex);
 871         return error;
 872 }
 873
 874 /*
 875  * When we're defragging a range, we don't want to kick it off again
 876  * if it is really just waiting for delalloc to send it down.
 877  * If we find a nice big extent or delalloc range for the bytes in the
 878  * file you want to defrag, we return 0 to let you know to skip this
 879  * part of the file
 880  */
 881 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 882 {
 883         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 884         struct extent_map *em = NULL;
 885         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 886         u64 end;
 887
 888         read_lock(&em_tree->lock);
 889         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 890         read_unlock(&em_tree->lock);
 891
 892         if (em) {
 893                 end = extent_map_end(em);
 894                 free_extent_map(em);
 895                 if (end - offset > thresh)
 896                         return 0;
 897         }
 898         /* if we already have a nice delalloc here, just stop */
 899         thresh /= 2;
 900         end = count_range_bits(io_tree, &offset, offset + thresh,
 901                                thresh, EXTENT_DELALLOC, 1);
 902         if (end >= thresh)
 903                 return 0;
 904         return 1;
 905 }
 906
 907 /*
 908  * helper function to walk through a file and find extents
 909  * newer than a specific transid, and smaller than thresh.
 910  *
 911  * This is used by the defragging code to find new and small
 912  * extents
 913  */
 914 static int find_new_extents(struct btrfs_root *root,
 915                             struct inode *inode, u64 newer_than,
 916                             u64 *off, u32 thresh)
 917 {
 918         struct btrfs_path *path;
 919         struct btrfs_key min_key;
 920         struct extent_buffer *leaf;
 921         struct btrfs_file_extent_item *extent;
 922         int type;
 923         int ret;
 924         u64 ino = btrfs_ino(inode);
 925
 926         path = btrfs_alloc_path();
 927         if (!path)
 928                 return -ENOMEM;
 929
 930         min_key.objectid = ino;
 931         min_key.type = BTRFS_EXTENT_DATA_KEY;
 932         min_key.offset = *off;
 933
 934         while (1) {
 935                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 936                 if (ret != 0)
 937                         goto none;
 938 process_slot:
 939                 if (min_key.objectid != ino)
 940                         goto none;
 941                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 942                         goto none;
 943
 944                 leaf = path->nodes[0];
 945                 extent = btrfs_item_ptr(leaf, path->slots[0],
 946                                         struct btrfs_file_extent_item);
 947
 948                 type = btrfs_file_extent_type(leaf, extent);
 949                 if (type == BTRFS_FILE_EXTENT_REG &&
 950                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 951                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 952                         *off = min_key.offset;
 953                         btrfs_free_path(path);
 954                         return 0;
 955                 }
 956
 957                 path->slots[0]++;
 958                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 959                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 960                         goto process_slot;
 961                 }
 962
 963                 if (min_key.offset == (u64)-1)
 964                         goto none;
 965
 966                 min_key.offset++;
 967                 btrfs_release_path(path);
 968         }
 969 none:
 970         btrfs_free_path(path);
 971         return -ENOENT;
 972 }
 973
 974 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 975 {
 976         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 977         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 978         struct extent_map *em;
 979         u64 len = PAGE_CACHE_SIZE;
 980
 981         /*
 982          * hopefully we have this extent in the tree already, try without
 983          * the full extent lock
 984          */
 985         read_lock(&em_tree->lock);
 986         em = lookup_extent_mapping(em_tree, start, len);
 987         read_unlock(&em_tree->lock);
 988
 989         if (!em) {
 990                 struct extent_state *cached = NULL;
 991                 u64 end = start + len - 1;
 992
 993                 /* get the big lock and read metadata off disk */
 994                 lock_extent_bits(io_tree, start, end, 0, &cached);
 995                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 996                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 997
 998                 if (IS_ERR(em))
 999                         return NULL;
1000         }
1001
1002         return em;
1003 }
1004
1005 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1006 {
1007         struct extent_map *next;
1008         bool ret = true;
1009
1010         /* this is the last extent */
1011         if (em->start + em->len >= i_size_read(inode))
1012                 return false;
1013
1014         next = defrag_lookup_extent(inode, em->start + em->len);
1015         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1016                 ret = false;
1017         else if ((em->block_start + em->block_len == next->block_start) &&
1018                  (em->block_len > 128 * 1024 && next->block_len > 128 * 1024))
1019                 ret = false;
1020
1021         free_extent_map(next);
1022         return ret;
1023 }
1024
1025 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1026                                u64 *last_len, u64 *skip, u64 *defrag_end,
1027                                int compress)
1028 {
1029         struct extent_map *em;
1030         int ret = 1;
1031         bool next_mergeable = true;
1032
1033         /*
1034          * make sure that once we start defragging an extent, we keep on
1035          * defragging it
1036          */
1037         if (start < *defrag_end)
1038                 return 1;
1039
1040         *skip = 0;
1041
1042         em = defrag_lookup_extent(inode, start);
1043         if (!em)
1044                 return 0;
1045
1046         /* this will cover holes, and inline extents */
1047         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1048                 ret = 0;
1049                 goto out;
1050         }
1051
1052         next_mergeable = defrag_check_next_extent(inode, em);
1053         /*
1054          * we hit a real extent, if it is big or the next extent is not a
1055          * real extent, don't bother defragging it
1056          */
1057         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1058             (em->len >= thresh || !next_mergeable))
1059                 ret = 0;
1060 out:
1061         /*
1062          * last_len ends up being a counter of how many bytes we've defragged.
1063          * every time we choose not to defrag an extent, we reset *last_len
1064          * so that the next tiny extent will force a defrag.
1065          *
1066          * The end result of this is that tiny extents before a single big
1067          * extent will force at least part of that big extent to be defragged.
1068          */
1069         if (ret) {
1070                 *defrag_end = extent_map_end(em);
1071         } else {
1072                 *last_len = 0;
1073                 *skip = extent_map_end(em);
1074                 *defrag_end = 0;
1075         }
1076
1077         free_extent_map(em);
1078         return ret;
1079 }
1080
1081 /*
1082  * it doesn't do much good to defrag one or two pages
1083  * at a time.  This pulls in a nice chunk of pages
1084  * to COW and defrag.
1085  *
1086  * It also makes sure the delalloc code has enough
1087  * dirty data to avoid making new small extents as part
1088  * of the defrag
1089  *
1090  * It's a good idea to start RA on this range
1091  * before calling this.
1092  */
1093 static int cluster_pages_for_defrag(struct inode *inode,
1094                                     struct page **pages,
1095                                     unsigned long start_index,
1096                                     unsigned long num_pages)
1097 {
1098         unsigned long file_end;
1099         u64 isize = i_size_read(inode);
1100         u64 page_start;
1101         u64 page_end;
1102         u64 page_cnt;
1103         int ret;
1104         int i;
1105         int i_done;
1106         struct btrfs_ordered_extent *ordered;
1107         struct extent_state *cached_state = NULL;
1108         struct extent_io_tree *tree;
1109         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1110
1111         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1112         if (!isize || start_index > file_end)
1113                 return 0;
1114
1115         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1116
1117         ret = btrfs_delalloc_reserve_space(inode,
1118                                            page_cnt << PAGE_CACHE_SHIFT);
1119         if (ret)
1120                 return ret;
1121         i_done = 0;
1122         tree = &BTRFS_I(inode)->io_tree;
1123
1124         /* step one, lock all the pages */
1125         for (i = 0; i < page_cnt; i++) {
1126                 struct page *page;
1127 again:
1128                 page = find_or_create_page(inode->i_mapping,
1129                                            start_index + i, mask);
1130                 if (!page)
1131                         break;
1132
1133                 page_start = page_offset(page);
1134                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1135                 while (1) {
1136                         lock_extent_bits(tree, page_start, page_end,
1137                                          0, &cached_state);
1138                         ordered = btrfs_lookup_ordered_extent(inode,
1139                                                               page_start);
1140                         unlock_extent_cached(tree, page_start, page_end,
1141                                              &cached_state, GFP_NOFS);
1142                         if (!ordered)
1143                                 break;
1144
1145                         unlock_page(page);
1146                         btrfs_start_ordered_extent(inode, ordered, 1);
1147                         btrfs_put_ordered_extent(ordered);
1148                         lock_page(page);
1149                         /*
1150                          * we unlocked the page above, so we need check if
1151                          * it was released or not.
1152                          */
1153                         if (page->mapping != inode->i_mapping) {
1154                                 unlock_page(page);
1155                                 page_cache_release(page);
1156                                 goto again;
1157                         }
1158                 }
1159
1160                 if (!PageUptodate(page)) {
1161                         btrfs_readpage(NULL, page);
1162                         lock_page(page);
1163                         if (!PageUptodate(page)) {
1164                                 unlock_page(page);
1165                                 page_cache_release(page);
1166                                 ret = -EIO;
1167                                 break;
1168                         }
1169                 }
1170
1171                 if (page->mapping != inode->i_mapping) {
1172                         unlock_page(page);
1173                         page_cache_release(page);
1174                         goto again;
1175                 }
1176
1177                 pages[i] = page;
1178                 i_done++;
1179         }
1180         if (!i_done || ret)
1181                 goto out;
1182
1183         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1184                 goto out;
1185
1186         /*
1187          * so now we have a nice long stream of locked
1188          * and up to date pages, lets wait on them
1189          */
1190         for (i = 0; i < i_done; i++)
1191                 wait_on_page_writeback(pages[i]);
1192
1193         page_start = page_offset(pages[0]);
1194         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1195
1196         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1197                          page_start, page_end - 1, 0, &cached_state);
1198         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1199                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1200                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1201                           &cached_state, GFP_NOFS);
1202
1203         if (i_done != page_cnt) {
1204                 spin_lock(&BTRFS_I(inode)->lock);
1205                 BTRFS_I(inode)->outstanding_extents++;
1206                 spin_unlock(&BTRFS_I(inode)->lock);
1207                 btrfs_delalloc_release_space(inode,
1208                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1209         }
1210
1211
1212         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1213                           &cached_state, GFP_NOFS);
1214
1215         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1216                              page_start, page_end - 1, &cached_state,
1217                              GFP_NOFS);
1218
1219         for (i = 0; i < i_done; i++) {
1220                 clear_page_dirty_for_io(pages[i]);
1221                 ClearPageChecked(pages[i]);
1222                 set_page_extent_mapped(pages[i]);
1223                 set_page_dirty(pages[i]);
1224                 unlock_page(pages[i]);
1225                 page_cache_release(pages[i]);
1226         }
1227         return i_done;
1228 out:
1229         for (i = 0; i < i_done; i++) {
1230                 unlock_page(pages[i]);
1231                 page_cache_release(pages[i]);
1232         }
1233         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1234         return ret;
1235
1236 }
1237
1238 int btrfs_defrag_file(struct inode *inode, struct file *file,
1239                       struct btrfs_ioctl_defrag_range_args *range,
1240                       u64 newer_than, unsigned long max_to_defrag)
1241 {
1242         struct btrfs_root *root = BTRFS_I(inode)->root;
1243         struct file_ra_state *ra = NULL;
1244         unsigned long last_index;
1245         u64 isize = i_size_read(inode);
1246         u64 last_len = 0;
1247         u64 skip = 0;
1248         u64 defrag_end = 0;
1249         u64 newer_off = range->start;
1250         unsigned long i;
1251         unsigned long ra_index = 0;
1252         int ret;
1253         int defrag_count = 0;
1254         int compress_type = BTRFS_COMPRESS_ZLIB;
1255         u32 extent_thresh = range->extent_thresh;
1256         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1257         unsigned long cluster = max_cluster;
1258         u64 new_align = ~((u64)128 * 1024 - 1);
1259         struct page **pages = NULL;
1260
1261         if (isize == 0)
1262                 return 0;
1263
1264         if (range->start >= isize)
1265                 return -EINVAL;
1266
1267         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1268                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1269                         return -EINVAL;
1270                 if (range->compress_type)
1271                         compress_type = range->compress_type;
1272         }
1273
1274         if (extent_thresh == 0)
1275                 extent_thresh = 256 * 1024;
1276
1277         /*
1278          * if we were not given a file, allocate a readahead
1279          * context
1280          */
1281         if (!file) {
1282                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1283                 if (!ra)
1284                         return -ENOMEM;
1285                 file_ra_state_init(ra, inode->i_mapping);
1286         } else {
1287                 ra = &file->f_ra;
1288         }
1289
1290         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1291                         GFP_NOFS);
1292         if (!pages) {
1293                 ret = -ENOMEM;
1294                 goto out_ra;
1295         }
1296
1297         /* find the last page to defrag */
1298         if (range->start + range->len > range->start) {
1299                 last_index = min_t(u64, isize - 1,
1300                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1301         } else {
1302                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1303         }
1304
1305         if (newer_than) {
1306                 ret = find_new_extents(root, inode, newer_than,
1307                                        &newer_off, 64 * 1024);
1308                 if (!ret) {
1309                         range->start = newer_off;
1310                         /*
1311                          * we always align our defrag to help keep
1312                          * the extents in the file evenly spaced
1313                          */
1314                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1315                 } else
1316                         goto out_ra;
1317         } else {
1318                 i = range->start >> PAGE_CACHE_SHIFT;
1319         }
1320         if (!max_to_defrag)
1321                 max_to_defrag = last_index - i + 1;
1322
1323         /*
1324          * make writeback starts from i, so the defrag range can be
1325          * written sequentially.
1326          */
1327         if (i < inode->i_mapping->writeback_index)
1328                 inode->i_mapping->writeback_index = i;
1329
1330         while (i <= last_index && defrag_count < max_to_defrag &&
1331                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1332                 /*
1333                  * make sure we stop running if someone unmounts
1334                  * the FS
1335                  */
1336                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1337                         break;
1338
1339                 if (btrfs_defrag_cancelled(root->fs_info)) {
1340                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1341                         ret = -EAGAIN;
1342                         break;
1343                 }
1344
1345                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1346                                          extent_thresh, &last_len, &skip,
1347                                          &defrag_end, range->flags &
1348                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1349                         unsigned long next;
1350                         /*
1351                          * the should_defrag function tells us how much to skip
1352                          * bump our counter by the suggested amount
1353                          */
1354                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1355                         i = max(i + 1, next);
1356                         continue;
1357                 }
1358
1359                 if (!newer_than) {
1360                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1361                                    PAGE_CACHE_SHIFT) - i;
1362                         cluster = min(cluster, max_cluster);
1363                 } else {
1364                         cluster = max_cluster;
1365                 }
1366
1367                 if (i + cluster > ra_index) {
1368                         ra_index = max(i, ra_index);
1369                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1370                                        cluster);
1371                         ra_index += cluster;
1372                 }
1373
1374                 mutex_lock(&inode->i_mutex);
1375                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1376                         BTRFS_I(inode)->force_compress = compress_type;
1377                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1378                 if (ret < 0) {
1379                         mutex_unlock(&inode->i_mutex);
1380                         goto out_ra;
1381                 }
1382
1383                 defrag_count += ret;
1384                 balance_dirty_pages_ratelimited(inode->i_mapping);
1385                 mutex_unlock(&inode->i_mutex);
1386
1387                 if (newer_than) {
1388                         if (newer_off == (u64)-1)
1389                                 break;
1390
1391                         if (ret > 0)
1392                                 i += ret;
1393
1394                         newer_off = max(newer_off + 1,
1395                                         (u64)i << PAGE_CACHE_SHIFT);
1396
1397                         ret = find_new_extents(root, inode,
1398                                                newer_than, &newer_off,
1399                                                64 * 1024);
1400                         if (!ret) {
1401                                 range->start = newer_off;
1402                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1403                         } else {
1404                                 break;
1405                         }
1406                 } else {
1407                         if (ret > 0) {
1408                                 i += ret;
1409                                 last_len += ret << PAGE_CACHE_SHIFT;
1410                         } else {
1411                                 i++;
1412                                 last_len = 0;
1413                         }
1414                 }
1415         }
1416
1417         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1418                 filemap_flush(inode->i_mapping);
1419                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1420                              &BTRFS_I(inode)->runtime_flags))
1421                         filemap_flush(inode->i_mapping);
1422         }
1423
1424         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1425                 /* the filemap_flush will queue IO into the worker threads, but
1426                  * we have to make sure the IO is actually started and that
1427                  * ordered extents get created before we return
1428                  */
1429                 atomic_inc(&root->fs_info->async_submit_draining);
1430                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1431                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1432                         wait_event(root->fs_info->async_submit_wait,
1433                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1434                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1435                 }
1436                 atomic_dec(&root->fs_info->async_submit_draining);
1437         }
1438
1439         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1440                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1441         }
1442
1443         ret = defrag_count;
1444
1445 out_ra:
1446         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1447                 mutex_lock(&inode->i_mutex);
1448                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1449                 mutex_unlock(&inode->i_mutex);
1450         }
1451         if (!file)
1452                 kfree(ra);
1453         kfree(pages);
1454         return ret;
1455 }
1456
1457 static noinline int btrfs_ioctl_resize(struct file *file,
1458                                         void __user *arg)
1459 {
1460         u64 new_size;
1461         u64 old_size;
1462         u64 devid = 1;
1463         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1464         struct btrfs_ioctl_vol_args *vol_args;
1465         struct btrfs_trans_handle *trans;
1466         struct btrfs_device *device = NULL;
1467         char *sizestr;
1468         char *retptr;
1469         char *devstr = NULL;
1470         int ret = 0;
1471         int mod = 0;
1472
1473         if (!capable(CAP_SYS_ADMIN))
1474                 return -EPERM;
1475
1476         ret = mnt_want_write_file(file);
1477         if (ret)
1478                 return ret;
1479
1480         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1481                         1)) {
1482                 mnt_drop_write_file(file);
1483                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1484         }
1485
1486         mutex_lock(&root->fs_info->volume_mutex);
1487         vol_args = memdup_user(arg, sizeof(*vol_args));
1488         if (IS_ERR(vol_args)) {
1489                 ret = PTR_ERR(vol_args);
1490                 goto out;
1491         }
1492
1493         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1494
1495         sizestr = vol_args->name;
1496         devstr = strchr(sizestr, ':');
1497         if (devstr) {
1498                 sizestr = devstr + 1;
1499                 *devstr = '\0';
1500                 devstr = vol_args->name;
1501                 ret = kstrtoull(devstr, 10, &devid);
1502                 if (ret)
1503                         goto out_free;
1504                 if (!devid) {
1505                         ret = -EINVAL;
1506                         goto out_free;
1507                 }
1508                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1509         }
1510
1511         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1512         if (!device) {
1513                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1514                        devid);
1515                 ret = -ENODEV;
1516                 goto out_free;
1517         }
1518
1519         if (!device->writeable) {
1520                 btrfs_info(root->fs_info,
1521                            "resizer unable to apply on readonly device %llu",
1522                        devid);
1523                 ret = -EPERM;
1524                 goto out_free;
1525         }
1526
1527         if (!strcmp(sizestr, "max"))
1528                 new_size = device->bdev->bd_inode->i_size;
1529         else {
1530                 if (sizestr[0] == '-') {
1531                         mod = -1;
1532                         sizestr++;
1533                 } else if (sizestr[0] == '+') {
1534                         mod = 1;
1535                         sizestr++;
1536                 }
1537                 new_size = memparse(sizestr, &retptr);
1538                 if (*retptr != '\0' || new_size == 0) {
1539                         ret = -EINVAL;
1540                         goto out_free;
1541                 }
1542         }
1543
1544         if (device->is_tgtdev_for_dev_replace) {
1545                 ret = -EPERM;
1546                 goto out_free;
1547         }
1548
1549         old_size = btrfs_device_get_total_bytes(device);
1550
1551         if (mod < 0) {
1552                 if (new_size > old_size) {
1553                         ret = -EINVAL;
1554                         goto out_free;
1555                 }
1556                 new_size = old_size - new_size;
1557         } else if (mod > 0) {
1558                 if (new_size > ULLONG_MAX - old_size) {
1559                         ret = -ERANGE;
1560                         goto out_free;
1561                 }
1562                 new_size = old_size + new_size;
1563         }
1564
1565         if (new_size < 256 * 1024 * 1024) {
1566                 ret = -EINVAL;
1567                 goto out_free;
1568         }
1569         if (new_size > device->bdev->bd_inode->i_size) {
1570                 ret = -EFBIG;
1571                 goto out_free;
1572         }
1573
1574         new_size = div_u64(new_size, root->sectorsize);
1575         new_size *= root->sectorsize;
1576
1577         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1578                       rcu_str_deref(device->name), new_size);
1579
1580         if (new_size > old_size) {
1581                 trans = btrfs_start_transaction(root, 0);
1582                 if (IS_ERR(trans)) {
1583                         ret = PTR_ERR(trans);
1584                         goto out_free;
1585                 }
1586                 ret = btrfs_grow_device(trans, device, new_size);
1587                 btrfs_commit_transaction(trans, root);
1588         } else if (new_size < old_size) {
1589                 ret = btrfs_shrink_device(device, new_size);
1590         } /* equal, nothing need to do */
1591
1592 out_free:
1593         kfree(vol_args);
1594 out:
1595         mutex_unlock(&root->fs_info->volume_mutex);
1596         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1597         mnt_drop_write_file(file);
1598         return ret;
1599 }
1600
1601 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1602                                 char *name, unsigned long fd, int subvol,
1603                                 u64 *transid, bool readonly,
1604                                 struct btrfs_qgroup_inherit *inherit)
1605 {
1606         int namelen;
1607         int ret = 0;
1608
1609         ret = mnt_want_write_file(file);
1610         if (ret)
1611                 goto out;
1612
1613         namelen = strlen(name);
1614         if (strchr(name, '/')) {
1615                 ret = -EINVAL;
1616                 goto out_drop_write;
1617         }
1618
1619         if (name[0] == '.' &&
1620            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1621                 ret = -EEXIST;
1622                 goto out_drop_write;
1623         }
1624
1625         if (subvol) {
1626                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1627                                      NULL, transid, readonly, inherit);
1628         } else {
1629                 struct fd src = fdget(fd);
1630                 struct inode *src_inode;
1631                 if (!src.file) {
1632                         ret = -EINVAL;
1633                         goto out_drop_write;
1634                 }
1635
1636                 src_inode = file_inode(src.file);
1637                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1638                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1639                                    "Snapshot src from another FS");
1640                         ret = -EXDEV;
1641                 } else if (!inode_owner_or_capable(src_inode)) {
1642                         /*
1643                          * Subvolume creation is not restricted, but snapshots
1644                          * are limited to own subvolumes only
1645                          */
1646                         ret = -EPERM;
1647                 } else {
1648                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1649                                              BTRFS_I(src_inode)->root,
1650                                              transid, readonly, inherit);
1651                 }
1652                 fdput(src);
1653         }
1654 out_drop_write:
1655         mnt_drop_write_file(file);
1656 out:
1657         return ret;
1658 }
1659
1660 static noinline int btrfs_ioctl_snap_create(struct file *file,
1661                                             void __user *arg, int subvol)
1662 {
1663         struct btrfs_ioctl_vol_args *vol_args;
1664         int ret;
1665
1666         vol_args = memdup_user(arg, sizeof(*vol_args));
1667         if (IS_ERR(vol_args))
1668                 return PTR_ERR(vol_args);
1669         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1670
1671         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1672                                               vol_args->fd, subvol,
1673                                               NULL, false, NULL);
1674
1675         kfree(vol_args);
1676         return ret;
1677 }
1678
1679 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1680                                                void __user *arg, int subvol)
1681 {
1682         struct btrfs_ioctl_vol_args_v2 *vol_args;
1683         int ret;
1684         u64 transid = 0;
1685         u64 *ptr = NULL;
1686         bool readonly = false;
1687         struct btrfs_qgroup_inherit *inherit = NULL;
1688
1689         vol_args = memdup_user(arg, sizeof(*vol_args));
1690         if (IS_ERR(vol_args))
1691                 return PTR_ERR(vol_args);
1692         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1693
1694         if (vol_args->flags &
1695             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1696               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1697                 ret = -EOPNOTSUPP;
1698                 goto free_args;
1699         }
1700
1701         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1702                 ptr = &transid;
1703         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1704                 readonly = true;
1705         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1706                 if (vol_args->size > PAGE_CACHE_SIZE) {
1707                         ret = -EINVAL;
1708                         goto free_args;
1709                 }
1710                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1711                 if (IS_ERR(inherit)) {
1712                         ret = PTR_ERR(inherit);
1713                         goto free_args;
1714                 }
1715         }
1716
1717         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1718                                               vol_args->fd, subvol, ptr,
1719                                               readonly, inherit);
1720         if (ret)
1721                 goto free_inherit;
1722
1723         if (ptr && copy_to_user(arg +
1724                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1725                                         transid),
1726                                 ptr, sizeof(*ptr)))
1727                 ret = -EFAULT;
1728
1729 free_inherit:
1730         kfree(inherit);
1731 free_args:
1732         kfree(vol_args);
1733         return ret;
1734 }
1735
1736 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1737                                                 void __user *arg)
1738 {
1739         struct inode *inode = file_inode(file);
1740         struct btrfs_root *root = BTRFS_I(inode)->root;
1741         int ret = 0;
1742         u64 flags = 0;
1743
1744         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1745                 return -EINVAL;
1746
1747         down_read(&root->fs_info->subvol_sem);
1748         if (btrfs_root_readonly(root))
1749                 flags |= BTRFS_SUBVOL_RDONLY;
1750         up_read(&root->fs_info->subvol_sem);
1751
1752         if (copy_to_user(arg, &flags, sizeof(flags)))
1753                 ret = -EFAULT;
1754
1755         return ret;
1756 }
1757
1758 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1759                                               void __user *arg)
1760 {
1761         struct inode *inode = file_inode(file);
1762         struct btrfs_root *root = BTRFS_I(inode)->root;
1763         struct btrfs_trans_handle *trans;
1764         u64 root_flags;
1765         u64 flags;
1766         int ret = 0;
1767
1768         if (!inode_owner_or_capable(inode))
1769                 return -EPERM;
1770
1771         ret = mnt_want_write_file(file);
1772         if (ret)
1773                 goto out;
1774
1775         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1776                 ret = -EINVAL;
1777                 goto out_drop_write;
1778         }
1779
1780         if (copy_from_user(&flags, arg, sizeof(flags))) {
1781                 ret = -EFAULT;
1782                 goto out_drop_write;
1783         }
1784
1785         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1786                 ret = -EINVAL;
1787                 goto out_drop_write;
1788         }
1789
1790         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1791                 ret = -EOPNOTSUPP;
1792                 goto out_drop_write;
1793         }
1794
1795         down_write(&root->fs_info->subvol_sem);
1796
1797         /* nothing to do */
1798         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1799                 goto out_drop_sem;
1800
1801         root_flags = btrfs_root_flags(&root->root_item);
1802         if (flags & BTRFS_SUBVOL_RDONLY) {
1803                 btrfs_set_root_flags(&root->root_item,
1804                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1805         } else {
1806                 /*
1807                  * Block RO -> RW transition if this subvolume is involved in
1808                  * send
1809                  */
1810                 spin_lock(&root->root_item_lock);
1811                 if (root->send_in_progress == 0) {
1812                         btrfs_set_root_flags(&root->root_item,
1813                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1814                         spin_unlock(&root->root_item_lock);
1815                 } else {
1816                         spin_unlock(&root->root_item_lock);
1817                         btrfs_warn(root->fs_info,
1818                         "Attempt to set subvolume %llu read-write during send",
1819                                         root->root_key.objectid);
1820                         ret = -EPERM;
1821                         goto out_drop_sem;
1822                 }
1823         }
1824
1825         trans = btrfs_start_transaction(root, 1);
1826         if (IS_ERR(trans)) {
1827                 ret = PTR_ERR(trans);
1828                 goto out_reset;
1829         }
1830
1831         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1832                                 &root->root_key, &root->root_item);
1833
1834         btrfs_commit_transaction(trans, root);
1835 out_reset:
1836         if (ret)
1837                 btrfs_set_root_flags(&root->root_item, root_flags);
1838 out_drop_sem:
1839         up_write(&root->fs_info->subvol_sem);
1840 out_drop_write:
1841         mnt_drop_write_file(file);
1842 out:
1843         return ret;
1844 }
1845
1846 /*
1847  * helper to check if the subvolume references other subvolumes
1848  */
1849 static noinline int may_destroy_subvol(struct btrfs_root *root)
1850 {
1851         struct btrfs_path *path;
1852         struct btrfs_dir_item *di;
1853         struct btrfs_key key;
1854         u64 dir_id;
1855         int ret;
1856
1857         path = btrfs_alloc_path();
1858         if (!path)
1859                 return -ENOMEM;
1860
1861         /* Make sure this root isn't set as the default subvol */
1862         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1863         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1864                                    dir_id, "default", 7, 0);
1865         if (di && !IS_ERR(di)) {
1866                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1867                 if (key.objectid == root->root_key.objectid) {
1868                         ret = -EPERM;
1869                         btrfs_err(root->fs_info, "deleting default subvolume "
1870                                   "%llu is not allowed", key.objectid);
1871                         goto out;
1872                 }
1873                 btrfs_release_path(path);
1874         }
1875
1876         key.objectid = root->root_key.objectid;
1877         key.type = BTRFS_ROOT_REF_KEY;
1878         key.offset = (u64)-1;
1879
1880         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1881                                 &key, path, 0, 0);
1882         if (ret < 0)
1883                 goto out;
1884         BUG_ON(ret == 0);
1885
1886         ret = 0;
1887         if (path->slots[0] > 0) {
1888                 path->slots[0]--;
1889                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1890                 if (key.objectid == root->root_key.objectid &&
1891                     key.type == BTRFS_ROOT_REF_KEY)
1892                         ret = -ENOTEMPTY;
1893         }
1894 out:
1895         btrfs_free_path(path);
1896         return ret;
1897 }
1898
1899 static noinline int key_in_sk(struct btrfs_key *key,
1900                               struct btrfs_ioctl_search_key *sk)
1901 {
1902         struct btrfs_key test;
1903         int ret;
1904
1905         test.objectid = sk->min_objectid;
1906         test.type = sk->min_type;
1907         test.offset = sk->min_offset;
1908
1909         ret = btrfs_comp_cpu_keys(key, &test);
1910         if (ret < 0)
1911                 return 0;
1912
1913         test.objectid = sk->max_objectid;
1914         test.type = sk->max_type;
1915         test.offset = sk->max_offset;
1916
1917         ret = btrfs_comp_cpu_keys(key, &test);
1918         if (ret > 0)
1919                 return 0;
1920         return 1;
1921 }
1922
1923 static noinline int copy_to_sk(struct btrfs_root *root,
1924                                struct btrfs_path *path,
1925                                struct btrfs_key *key,
1926                                struct btrfs_ioctl_search_key *sk,
1927                                size_t *buf_size,
1928                                char __user *ubuf,
1929                                unsigned long *sk_offset,
1930                                int *num_found)
1931 {
1932         u64 found_transid;
1933         struct extent_buffer *leaf;
1934         struct btrfs_ioctl_search_header sh;
1935         unsigned long item_off;
1936         unsigned long item_len;
1937         int nritems;
1938         int i;
1939         int slot;
1940         int ret = 0;
1941
1942         leaf = path->nodes[0];
1943         slot = path->slots[0];
1944         nritems = btrfs_header_nritems(leaf);
1945
1946         if (btrfs_header_generation(leaf) > sk->max_transid) {
1947                 i = nritems;
1948                 goto advance_key;
1949         }
1950         found_transid = btrfs_header_generation(leaf);
1951
1952         for (i = slot; i < nritems; i++) {
1953                 item_off = btrfs_item_ptr_offset(leaf, i);
1954                 item_len = btrfs_item_size_nr(leaf, i);
1955
1956                 btrfs_item_key_to_cpu(leaf, key, i);
1957                 if (!key_in_sk(key, sk))
1958                         continue;
1959
1960                 if (sizeof(sh) + item_len > *buf_size) {
1961                         if (*num_found) {
1962                                 ret = 1;
1963                                 goto out;
1964                         }
1965
1966                         /*
1967                          * return one empty item back for v1, which does not
1968                          * handle -EOVERFLOW
1969                          */
1970
1971                         *buf_size = sizeof(sh) + item_len;
1972                         item_len = 0;
1973                         ret = -EOVERFLOW;
1974                 }
1975
1976                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1977                         ret = 1;
1978                         goto out;
1979                 }
1980
1981                 sh.objectid = key->objectid;
1982                 sh.offset = key->offset;
1983                 sh.type = key->type;
1984                 sh.len = item_len;
1985                 sh.transid = found_transid;
1986
1987                 /* copy search result header */
1988                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1989                         ret = -EFAULT;
1990                         goto out;
1991                 }
1992
1993                 *sk_offset += sizeof(sh);
1994
1995                 if (item_len) {
1996                         char __user *up = ubuf + *sk_offset;
1997                         /* copy the item */
1998                         if (read_extent_buffer_to_user(leaf, up,
1999                                                        item_off, item_len)) {
2000                                 ret = -EFAULT;
2001                                 goto out;
2002                         }
2003
2004                         *sk_offset += item_len;
2005                 }
2006                 (*num_found)++;
2007
2008                 if (ret) /* -EOVERFLOW from above */
2009                         goto out;
2010
2011                 if (*num_found >= sk->nr_items) {
2012                         ret = 1;
2013                         goto out;
2014                 }
2015         }
2016 advance_key:
2017         ret = 0;
2018         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2019                 key->offset++;
2020         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2021                 key->offset = 0;
2022                 key->type++;
2023         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2024                 key->offset = 0;
2025                 key->type = 0;
2026                 key->objectid++;
2027         } else
2028                 ret = 1;
2029 out:
2030         /*
2031          *  0: all items from this leaf copied, continue with next
2032          *  1: * more items can be copied, but unused buffer is too small
2033          *     * all items were found
2034          *     Either way, it will stops the loop which iterates to the next
2035          *     leaf
2036          *  -EOVERFLOW: item was to large for buffer
2037          *  -EFAULT: could not copy extent buffer back to userspace
2038          */
2039         return ret;
2040 }
2041
2042 static noinline int search_ioctl(struct inode *inode,
2043                                  struct btrfs_ioctl_search_key *sk,
2044                                  size_t *buf_size,
2045                                  char __user *ubuf)
2046 {
2047         struct btrfs_root *root;
2048         struct btrfs_key key;
2049         struct btrfs_path *path;
2050         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2051         int ret;
2052         int num_found = 0;
2053         unsigned long sk_offset = 0;
2054
2055         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2056                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2057                 return -EOVERFLOW;
2058         }
2059
2060         path = btrfs_alloc_path();
2061         if (!path)
2062                 return -ENOMEM;
2063
2064         if (sk->tree_id == 0) {
2065                 /* search the root of the inode that was passed */
2066                 root = BTRFS_I(inode)->root;
2067         } else {
2068                 key.objectid = sk->tree_id;
2069                 key.type = BTRFS_ROOT_ITEM_KEY;
2070                 key.offset = (u64)-1;
2071                 root = btrfs_read_fs_root_no_name(info, &key);
2072                 if (IS_ERR(root)) {
2073                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2074                                sk->tree_id);
2075                         btrfs_free_path(path);
2076                         return -ENOENT;
2077                 }
2078         }
2079
2080         key.objectid = sk->min_objectid;
2081         key.type = sk->min_type;
2082         key.offset = sk->min_offset;
2083
2084         while (1) {
2085                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2086                 if (ret != 0) {
2087                         if (ret > 0)
2088                                 ret = 0;
2089                         goto err;
2090                 }
2091                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2092                                  &sk_offset, &num_found);
2093                 btrfs_release_path(path);
2094                 if (ret)
2095                         break;
2096
2097         }
2098         if (ret > 0)
2099                 ret = 0;
2100 err:
2101         sk->nr_items = num_found;
2102         btrfs_free_path(path);
2103         return ret;
2104 }
2105
2106 static noinline int btrfs_ioctl_tree_search(struct file *file,
2107                                            void __user *argp)
2108 {
2109         struct btrfs_ioctl_search_args __user *uargs;
2110         struct btrfs_ioctl_search_key sk;
2111         struct inode *inode;
2112         int ret;
2113         size_t buf_size;
2114
2115         if (!capable(CAP_SYS_ADMIN))
2116                 return -EPERM;
2117
2118         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2119
2120         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2121                 return -EFAULT;
2122
2123         buf_size = sizeof(uargs->buf);
2124
2125         inode = file_inode(file);
2126         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2127
2128         /*
2129          * In the origin implementation an overflow is handled by returning a
2130          * search header with a len of zero, so reset ret.
2131          */
2132         if (ret == -EOVERFLOW)
2133                 ret = 0;
2134
2135         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2136                 ret = -EFAULT;
2137         return ret;
2138 }
2139
2140 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2141                                                void __user *argp)
2142 {
2143         struct btrfs_ioctl_search_args_v2 __user *uarg;
2144         struct btrfs_ioctl_search_args_v2 args;
2145         struct inode *inode;
2146         int ret;
2147         size_t buf_size;
2148         const size_t buf_limit = 16 * 1024 * 1024;
2149
2150         if (!capable(CAP_SYS_ADMIN))
2151                 return -EPERM;
2152
2153         /* copy search header and buffer size */
2154         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2155         if (copy_from_user(&args, uarg, sizeof(args)))
2156                 return -EFAULT;
2157
2158         buf_size = args.buf_size;
2159
2160         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2161                 return -EOVERFLOW;
2162
2163         /* limit result size to 16MB */
2164         if (buf_size > buf_limit)
2165                 buf_size = buf_limit;
2166
2167         inode = file_inode(file);
2168         ret = search_ioctl(inode, &args.key, &buf_size,
2169                            (char *)(&uarg->buf[0]));
2170         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2171                 ret = -EFAULT;
2172         else if (ret == -EOVERFLOW &&
2173                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2174                 ret = -EFAULT;
2175
2176         return ret;
2177 }
2178
2179 /*
2180  * Search INODE_REFs to identify path name of 'dirid' directory
2181  * in a 'tree_id' tree. and sets path name to 'name'.
2182  */
2183 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2184                                 u64 tree_id, u64 dirid, char *name)
2185 {
2186         struct btrfs_root *root;
2187         struct btrfs_key key;
2188         char *ptr;
2189         int ret = -1;
2190         int slot;
2191         int len;
2192         int total_len = 0;
2193         struct btrfs_inode_ref *iref;
2194         struct extent_buffer *l;
2195         struct btrfs_path *path;
2196
2197         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2198                 name[0]='\0';
2199                 return 0;
2200         }
2201
2202         path = btrfs_alloc_path();
2203         if (!path)
2204                 return -ENOMEM;
2205
2206         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2207
2208         key.objectid = tree_id;
2209         key.type = BTRFS_ROOT_ITEM_KEY;
2210         key.offset = (u64)-1;
2211         root = btrfs_read_fs_root_no_name(info, &key);
2212         if (IS_ERR(root)) {
2213                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2214                 ret = -ENOENT;
2215                 goto out;
2216         }
2217
2218         key.objectid = dirid;
2219         key.type = BTRFS_INODE_REF_KEY;
2220         key.offset = (u64)-1;
2221
2222         while (1) {
2223                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2224                 if (ret < 0)
2225                         goto out;
2226                 else if (ret > 0) {
2227                         ret = btrfs_previous_item(root, path, dirid,
2228                                                   BTRFS_INODE_REF_KEY);
2229                         if (ret < 0)
2230                                 goto out;
2231                         else if (ret > 0) {
2232                                 ret = -ENOENT;
2233                                 goto out;
2234                         }
2235                 }
2236
2237                 l = path->nodes[0];
2238                 slot = path->slots[0];
2239                 btrfs_item_key_to_cpu(l, &key, slot);
2240
2241                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2242                 len = btrfs_inode_ref_name_len(l, iref);
2243                 ptr -= len + 1;
2244                 total_len += len + 1;
2245                 if (ptr < name) {
2246                         ret = -ENAMETOOLONG;
2247                         goto out;
2248                 }
2249
2250                 *(ptr + len) = '/';
2251                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2252
2253                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2254                         break;
2255
2256                 btrfs_release_path(path);
2257                 key.objectid = key.offset;
2258                 key.offset = (u64)-1;
2259                 dirid = key.objectid;
2260         }
2261         memmove(name, ptr, total_len);
2262         name[total_len] = '\0';
2263         ret = 0;
2264 out:
2265         btrfs_free_path(path);
2266         return ret;
2267 }
2268
2269 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2270                                            void __user *argp)
2271 {
2272          struct btrfs_ioctl_ino_lookup_args *args;
2273          struct inode *inode;
2274         int ret = 0;
2275
2276         args = memdup_user(argp, sizeof(*args));
2277         if (IS_ERR(args))
2278                 return PTR_ERR(args);
2279
2280         inode = file_inode(file);
2281
2282         /*
2283          * Unprivileged query to obtain the containing subvolume root id. The
2284          * path is reset so it's consistent with btrfs_search_path_in_tree.
2285          */
2286         if (args->treeid == 0)
2287                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2288
2289         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2290                 args->name[0] = 0;
2291                 goto out;
2292         }
2293
2294         if (!capable(CAP_SYS_ADMIN)) {
2295                 ret = -EPERM;
2296                 goto out;
2297         }
2298
2299         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2300                                         args->treeid, args->objectid,
2301                                         args->name);
2302
2303 out:
2304         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2305                 ret = -EFAULT;
2306
2307         kfree(args);
2308         return ret;
2309 }
2310
2311 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2312                                              void __user *arg)
2313 {
2314         struct dentry *parent = file->f_path.dentry;
2315         struct dentry *dentry;
2316         struct inode *dir = d_inode(parent);
2317         struct inode *inode;
2318         struct btrfs_root *root = BTRFS_I(dir)->root;
2319         struct btrfs_root *dest = NULL;
2320         struct btrfs_ioctl_vol_args *vol_args;
2321         struct btrfs_trans_handle *trans;
2322         struct btrfs_block_rsv block_rsv;
2323         u64 root_flags;
2324         u64 qgroup_reserved;
2325         int namelen;
2326         int ret;
2327         int err = 0;
2328
2329         vol_args = memdup_user(arg, sizeof(*vol_args));
2330         if (IS_ERR(vol_args))
2331                 return PTR_ERR(vol_args);
2332
2333         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2334         namelen = strlen(vol_args->name);
2335         if (strchr(vol_args->name, '/') ||
2336             strncmp(vol_args->name, "..", namelen) == 0) {
2337                 err = -EINVAL;
2338                 goto out;
2339         }
2340
2341         err = mnt_want_write_file(file);
2342         if (err)
2343                 goto out;
2344
2345
2346         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2347         if (err == -EINTR)
2348                 goto out_drop_write;
2349         dentry = lookup_one_len(vol_args->name, parent, namelen);
2350         if (IS_ERR(dentry)) {
2351                 err = PTR_ERR(dentry);
2352                 goto out_unlock_dir;
2353         }
2354
2355         if (d_really_is_negative(dentry)) {
2356                 err = -ENOENT;
2357                 goto out_dput;
2358         }
2359
2360         inode = d_inode(dentry);
2361         dest = BTRFS_I(inode)->root;
2362         if (!capable(CAP_SYS_ADMIN)) {
2363                 /*
2364                  * Regular user.  Only allow this with a special mount
2365                  * option, when the user has write+exec access to the
2366                  * subvol root, and when rmdir(2) would have been
2367                  * allowed.
2368                  *
2369                  * Note that this is _not_ check that the subvol is
2370                  * empty or doesn't contain data that we wouldn't
2371                  * otherwise be able to delete.
2372                  *
2373                  * Users who want to delete empty subvols should try
2374                  * rmdir(2).
2375                  */
2376                 err = -EPERM;
2377                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2378                         goto out_dput;
2379
2380                 /*
2381                  * Do not allow deletion if the parent dir is the same
2382                  * as the dir to be deleted.  That means the ioctl
2383                  * must be called on the dentry referencing the root
2384                  * of the subvol, not a random directory contained
2385                  * within it.
2386                  */
2387                 err = -EINVAL;
2388                 if (root == dest)
2389                         goto out_dput;
2390
2391                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2392                 if (err)
2393                         goto out_dput;
2394         }
2395
2396         /* check if subvolume may be deleted by a user */
2397         err = btrfs_may_delete(dir, dentry, 1);
2398         if (err)
2399                 goto out_dput;
2400
2401         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2402                 err = -EINVAL;
2403                 goto out_dput;
2404         }
2405
2406         mutex_lock(&inode->i_mutex);
2407
2408         /*
2409          * Don't allow to delete a subvolume with send in progress. This is
2410          * inside the i_mutex so the error handling that has to drop the bit
2411          * again is not run concurrently.
2412          */
2413         spin_lock(&dest->root_item_lock);
2414         root_flags = btrfs_root_flags(&dest->root_item);
2415         if (dest->send_in_progress == 0) {
2416                 btrfs_set_root_flags(&dest->root_item,
2417                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2418                 spin_unlock(&dest->root_item_lock);
2419         } else {
2420                 spin_unlock(&dest->root_item_lock);
2421                 btrfs_warn(root->fs_info,
2422                         "Attempt to delete subvolume %llu during send",
2423                         dest->root_key.objectid);
2424                 err = -EPERM;
2425                 goto out_unlock_inode;
2426         }
2427
2428         down_write(&root->fs_info->subvol_sem);
2429
2430         err = may_destroy_subvol(dest);
2431         if (err)
2432                 goto out_up_write;
2433
2434         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2435         /*
2436          * One for dir inode, two for dir entries, two for root
2437          * ref/backref.
2438          */
2439         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2440                                                5, &qgroup_reserved, true);
2441         if (err)
2442                 goto out_up_write;
2443
2444         trans = btrfs_start_transaction(root, 0);
2445         if (IS_ERR(trans)) {
2446                 err = PTR_ERR(trans);
2447                 goto out_release;
2448         }
2449         trans->block_rsv = &block_rsv;
2450         trans->bytes_reserved = block_rsv.size;
2451
2452         ret = btrfs_unlink_subvol(trans, root, dir,
2453                                 dest->root_key.objectid,
2454                                 dentry->d_name.name,
2455                                 dentry->d_name.len);
2456         if (ret) {
2457                 err = ret;
2458                 btrfs_abort_transaction(trans, root, ret);
2459                 goto out_end_trans;
2460         }
2461
2462         btrfs_record_root_in_trans(trans, dest);
2463
2464         memset(&dest->root_item.drop_progress, 0,
2465                 sizeof(dest->root_item.drop_progress));
2466         dest->root_item.drop_level = 0;
2467         btrfs_set_root_refs(&dest->root_item, 0);
2468
2469         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2470                 ret = btrfs_insert_orphan_item(trans,
2471                                         root->fs_info->tree_root,
2472                                         dest->root_key.objectid);
2473                 if (ret) {
2474                         btrfs_abort_transaction(trans, root, ret);
2475                         err = ret;
2476                         goto out_end_trans;
2477                 }
2478         }
2479
2480         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2481                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2482                                   dest->root_key.objectid);
2483         if (ret && ret != -ENOENT) {
2484                 btrfs_abort_transaction(trans, root, ret);
2485                 err = ret;
2486                 goto out_end_trans;
2487         }
2488         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2489                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2490                                           dest->root_item.received_uuid,
2491                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2492                                           dest->root_key.objectid);
2493                 if (ret && ret != -ENOENT) {
2494                         btrfs_abort_transaction(trans, root, ret);
2495                         err = ret;
2496                         goto out_end_trans;
2497                 }
2498         }
2499
2500 out_end_trans:
2501         trans->block_rsv = NULL;
2502         trans->bytes_reserved = 0;
2503         ret = btrfs_end_transaction(trans, root);
2504         if (ret && !err)
2505                 err = ret;
2506         inode->i_flags |= S_DEAD;
2507 out_release:
2508         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2509 out_up_write:
2510         up_write(&root->fs_info->subvol_sem);
2511         if (err) {
2512                 spin_lock(&dest->root_item_lock);
2513                 root_flags = btrfs_root_flags(&dest->root_item);
2514                 btrfs_set_root_flags(&dest->root_item,
2515                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2516                 spin_unlock(&dest->root_item_lock);
2517         }
2518 out_unlock_inode:
2519         mutex_unlock(&inode->i_mutex);
2520         if (!err) {
2521                 d_invalidate(dentry);
2522                 btrfs_invalidate_inodes(dest);
2523                 d_delete(dentry);
2524                 ASSERT(dest->send_in_progress == 0);
2525
2526                 /* the last ref */
2527                 if (dest->ino_cache_inode) {
2528                         iput(dest->ino_cache_inode);
2529                         dest->ino_cache_inode = NULL;
2530                 }
2531         }
2532 out_dput:
2533         dput(dentry);
2534 out_unlock_dir:
2535         mutex_unlock(&dir->i_mutex);
2536 out_drop_write:
2537         mnt_drop_write_file(file);
2538 out:
2539         kfree(vol_args);
2540         return err;
2541 }
2542
2543 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2544 {
2545         struct inode *inode = file_inode(file);
2546         struct btrfs_root *root = BTRFS_I(inode)->root;
2547         struct btrfs_ioctl_defrag_range_args *range;
2548         int ret;
2549
2550         ret = mnt_want_write_file(file);
2551         if (ret)
2552                 return ret;
2553
2554         if (btrfs_root_readonly(root)) {
2555                 ret = -EROFS;
2556                 goto out;
2557         }
2558
2559         switch (inode->i_mode & S_IFMT) {
2560         case S_IFDIR:
2561                 if (!capable(CAP_SYS_ADMIN)) {
2562                         ret = -EPERM;
2563                         goto out;
2564                 }
2565                 ret = btrfs_defrag_root(root);
2566                 if (ret)
2567                         goto out;
2568                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2569                 break;
2570         case S_IFREG:
2571                 if (!(file->f_mode & FMODE_WRITE)) {
2572                         ret = -EINVAL;
2573                         goto out;
2574                 }
2575
2576                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2577                 if (!range) {
2578                         ret = -ENOMEM;
2579                         goto out;
2580                 }
2581
2582                 if (argp) {
2583                         if (copy_from_user(range, argp,
2584                                            sizeof(*range))) {
2585                                 ret = -EFAULT;
2586                                 kfree(range);
2587                                 goto out;
2588                         }
2589                         /* compression requires us to start the IO */
2590                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2591                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2592                                 range->extent_thresh = (u32)-1;
2593                         }
2594                 } else {
2595                         /* the rest are all set to zero by kzalloc */
2596                         range->len = (u64)-1;
2597                 }
2598                 ret = btrfs_defrag_file(file_inode(file), file,
2599                                         range, 0, 0);
2600                 if (ret > 0)
2601                         ret = 0;
2602                 kfree(range);
2603                 break;
2604         default:
2605                 ret = -EINVAL;
2606         }
2607 out:
2608         mnt_drop_write_file(file);
2609         return ret;
2610 }
2611
2612 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2613 {
2614         struct btrfs_ioctl_vol_args *vol_args;
2615         int ret;
2616
2617         if (!capable(CAP_SYS_ADMIN))
2618                 return -EPERM;
2619
2620         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2621                         1)) {
2622                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2623         }
2624
2625         mutex_lock(&root->fs_info->volume_mutex);
2626         vol_args = memdup_user(arg, sizeof(*vol_args));
2627         if (IS_ERR(vol_args)) {
2628                 ret = PTR_ERR(vol_args);
2629                 goto out;
2630         }
2631
2632         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2633         ret = btrfs_init_new_device(root, vol_args->name);
2634
2635         if (!ret)
2636                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2637
2638         kfree(vol_args);
2639 out:
2640         mutex_unlock(&root->fs_info->volume_mutex);
2641         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2642         return ret;
2643 }
2644
2645 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2646 {
2647         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2648         struct btrfs_ioctl_vol_args *vol_args;
2649         int ret;
2650
2651         if (!capable(CAP_SYS_ADMIN))
2652                 return -EPERM;
2653
2654         ret = mnt_want_write_file(file);
2655         if (ret)
2656                 return ret;
2657
2658         vol_args = memdup_user(arg, sizeof(*vol_args));
2659         if (IS_ERR(vol_args)) {
2660                 ret = PTR_ERR(vol_args);
2661                 goto err_drop;
2662         }
2663
2664         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2665
2666         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2667                         1)) {
2668                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2669                 goto out;
2670         }
2671
2672         mutex_lock(&root->fs_info->volume_mutex);
2673         ret = btrfs_rm_device(root, vol_args->name);
2674         mutex_unlock(&root->fs_info->volume_mutex);
2675         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2676
2677         if (!ret)
2678                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2679
2680 out:
2681         kfree(vol_args);
2682 err_drop:
2683         mnt_drop_write_file(file);
2684         return ret;
2685 }
2686
2687 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2688 {
2689         struct btrfs_ioctl_fs_info_args *fi_args;
2690         struct btrfs_device *device;
2691         struct btrfs_device *next;
2692         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2693         int ret = 0;
2694
2695         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2696         if (!fi_args)
2697                 return -ENOMEM;
2698
2699         mutex_lock(&fs_devices->device_list_mutex);
2700         fi_args->num_devices = fs_devices->num_devices;
2701         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2702
2703         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2704                 if (device->devid > fi_args->max_id)
2705                         fi_args->max_id = device->devid;
2706         }
2707         mutex_unlock(&fs_devices->device_list_mutex);
2708
2709         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2710         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2711         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2712
2713         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2714                 ret = -EFAULT;
2715
2716         kfree(fi_args);
2717         return ret;
2718 }
2719
2720 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2721 {
2722         struct btrfs_ioctl_dev_info_args *di_args;
2723         struct btrfs_device *dev;
2724         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2725         int ret = 0;
2726         char *s_uuid = NULL;
2727
2728         di_args = memdup_user(arg, sizeof(*di_args));
2729         if (IS_ERR(di_args))
2730                 return PTR_ERR(di_args);
2731
2732         if (!btrfs_is_empty_uuid(di_args->uuid))
2733                 s_uuid = di_args->uuid;
2734
2735         mutex_lock(&fs_devices->device_list_mutex);
2736         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2737
2738         if (!dev) {
2739                 ret = -ENODEV;
2740                 goto out;
2741         }
2742
2743         di_args->devid = dev->devid;
2744         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2745         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2746         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2747         if (dev->name) {
2748                 struct rcu_string *name;
2749
2750                 rcu_read_lock();
2751                 name = rcu_dereference(dev->name);
2752                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2753                 rcu_read_unlock();
2754                 di_args->path[sizeof(di_args->path) - 1] = 0;
2755         } else {
2756                 di_args->path[0] = '\0';
2757         }
2758
2759 out:
2760         mutex_unlock(&fs_devices->device_list_mutex);
2761         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2762                 ret = -EFAULT;
2763
2764         kfree(di_args);
2765         return ret;
2766 }
2767
2768 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2769 {
2770         struct page *page;
2771         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2772
2773         page = grab_cache_page(inode->i_mapping, index);
2774         if (!page)
2775                 return NULL;
2776
2777         if (!PageUptodate(page)) {
2778                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2779                                                  0))
2780                         return NULL;
2781                 lock_page(page);
2782                 if (!PageUptodate(page)) {
2783                         unlock_page(page);
2784                         page_cache_release(page);
2785                         return NULL;
2786                 }
2787         }
2788         unlock_page(page);
2789
2790         return page;
2791 }
2792
2793 static int gather_extent_pages(struct inode *inode, struct page **pages,
2794                                int num_pages, u64 off)
2795 {
2796         int i;
2797         pgoff_t index = off >> PAGE_CACHE_SHIFT;
2798
2799         for (i = 0; i < num_pages; i++) {
2800                 pages[i] = extent_same_get_page(inode, index + i);
2801                 if (!pages[i])
2802                         return -ENOMEM;
2803         }
2804         return 0;
2805 }
2806
2807 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2808 {
2809         /* do any pending delalloc/csum calc on src, one way or
2810            another, and lock file content */
2811         while (1) {
2812                 struct btrfs_ordered_extent *ordered;
2813                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2814                 ordered = btrfs_lookup_first_ordered_extent(inode,
2815                                                             off + len - 1);
2816                 if ((!ordered ||
2817                      ordered->file_offset + ordered->len <= off ||
2818                      ordered->file_offset >= off + len) &&
2819                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2820                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2821                         if (ordered)
2822                                 btrfs_put_ordered_extent(ordered);
2823                         break;
2824                 }
2825                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2826                 if (ordered)
2827                         btrfs_put_ordered_extent(ordered);
2828                 btrfs_wait_ordered_range(inode, off, len);
2829         }
2830 }
2831
2832 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2833 {
2834         mutex_unlock(&inode1->i_mutex);
2835         mutex_unlock(&inode2->i_mutex);
2836 }
2837
2838 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2839 {
2840         if (inode1 < inode2)
2841                 swap(inode1, inode2);
2842
2843         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2844         if (inode1 != inode2)
2845                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2846 }
2847
2848 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2849                                       struct inode *inode2, u64 loff2, u64 len)
2850 {
2851         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2852         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2853 }
2854
2855 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2856                                      struct inode *inode2, u64 loff2, u64 len)
2857 {
2858         if (inode1 < inode2) {
2859                 swap(inode1, inode2);
2860                 swap(loff1, loff2);
2861         }
2862         lock_extent_range(inode1, loff1, len);
2863         if (inode1 != inode2)
2864                 lock_extent_range(inode2, loff2, len);
2865 }
2866
2867 struct cmp_pages {
2868         int             num_pages;
2869         struct page     **src_pages;
2870         struct page     **dst_pages;
2871 };
2872
2873 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2874 {
2875         int i;
2876         struct page *pg;
2877
2878         for (i = 0; i < cmp->num_pages; i++) {
2879                 pg = cmp->src_pages[i];
2880                 if (pg)
2881                         page_cache_release(pg);
2882                 pg = cmp->dst_pages[i];
2883                 if (pg)
2884                         page_cache_release(pg);
2885         }
2886         kfree(cmp->src_pages);
2887         kfree(cmp->dst_pages);
2888 }
2889
2890 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2891                                   struct inode *dst, u64 dst_loff,
2892                                   u64 len, struct cmp_pages *cmp)
2893 {
2894         int ret;
2895         int num_pages = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT;
2896         struct page **src_pgarr, **dst_pgarr;
2897
2898         /*
2899          * We must gather up all the pages before we initiate our
2900          * extent locking. We use an array for the page pointers. Size
2901          * of the array is bounded by len, which is in turn bounded by
2902          * BTRFS_MAX_DEDUPE_LEN.
2903          */
2904         src_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2905         dst_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2906         if (!src_pgarr || !dst_pgarr) {
2907                 kfree(src_pgarr);
2908                 kfree(dst_pgarr);
2909                 return -ENOMEM;
2910         }
2911         cmp->num_pages = num_pages;
2912         cmp->src_pages = src_pgarr;
2913         cmp->dst_pages = dst_pgarr;
2914
2915         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
2916         if (ret)
2917                 goto out;
2918
2919         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
2920
2921 out:
2922         if (ret)
2923                 btrfs_cmp_data_free(cmp);
2924         return 0;
2925 }
2926
2927 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2928                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
2929 {
2930         int ret = 0;
2931         int i;
2932         struct page *src_page, *dst_page;
2933         unsigned int cmp_len = PAGE_CACHE_SIZE;
2934         void *addr, *dst_addr;
2935
2936         i = 0;
2937         while (len) {
2938                 if (len < PAGE_CACHE_SIZE)
2939                         cmp_len = len;
2940
2941                 BUG_ON(i >= cmp->num_pages);
2942
2943                 src_page = cmp->src_pages[i];
2944                 dst_page = cmp->dst_pages[i];
2945
2946                 addr = kmap_atomic(src_page);
2947                 dst_addr = kmap_atomic(dst_page);
2948
2949                 flush_dcache_page(src_page);
2950                 flush_dcache_page(dst_page);
2951
2952                 if (memcmp(addr, dst_addr, cmp_len))
2953                         ret = BTRFS_SAME_DATA_DIFFERS;
2954
2955                 kunmap_atomic(addr);
2956                 kunmap_atomic(dst_addr);
2957
2958                 if (ret)
2959                         break;
2960
2961                 len -= cmp_len;
2962                 i++;
2963         }
2964
2965         return ret;
2966 }
2967
2968 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
2969                                      u64 olen)
2970 {
2971         u64 len = *plen;
2972         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2973
2974         if (off + olen > inode->i_size || off + olen < off)
2975                 return -EINVAL;
2976
2977         /* if we extend to eof, continue to block boundary */
2978         if (off + len == inode->i_size)
2979                 *plen = len = ALIGN(inode->i_size, bs) - off;
2980
2981         /* Check that we are block aligned - btrfs_clone() requires this */
2982         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2983                 return -EINVAL;
2984
2985         return 0;
2986 }
2987
2988 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
2989                              struct inode *dst, u64 dst_loff)
2990 {
2991         int ret;
2992         u64 len = olen;
2993         struct cmp_pages cmp;
2994         int same_inode = 0;
2995         u64 same_lock_start = 0;
2996         u64 same_lock_len = 0;
2997
2998         if (src == dst)
2999                 same_inode = 1;
3000
3001         if (len == 0)
3002                 return 0;
3003
3004         if (same_inode) {
3005                 mutex_lock(&src->i_mutex);
3006
3007                 ret = extent_same_check_offsets(src, loff, &len, olen);
3008                 if (ret)
3009                         goto out_unlock;
3010
3011                 /*
3012                  * Single inode case wants the same checks, except we
3013                  * don't want our length pushed out past i_size as
3014                  * comparing that data range makes no sense.
3015                  *
3016                  * extent_same_check_offsets() will do this for an
3017                  * unaligned length at i_size, so catch it here and
3018                  * reject the request.
3019                  *
3020                  * This effectively means we require aligned extents
3021                  * for the single-inode case, whereas the other cases
3022                  * allow an unaligned length so long as it ends at
3023                  * i_size.
3024                  */
3025                 if (len != olen) {
3026                         ret = -EINVAL;
3027                         goto out_unlock;
3028                 }
3029
3030                 /* Check for overlapping ranges */
3031                 if (dst_loff + len > loff && dst_loff < loff + len) {
3032                         ret = -EINVAL;
3033                         goto out_unlock;
3034                 }
3035
3036                 same_lock_start = min_t(u64, loff, dst_loff);
3037                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3038         } else {
3039                 btrfs_double_inode_lock(src, dst);
3040
3041                 ret = extent_same_check_offsets(src, loff, &len, olen);
3042                 if (ret)
3043                         goto out_unlock;
3044
3045                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3046                 if (ret)
3047                         goto out_unlock;
3048         }
3049
3050         /* don't make the dst file partly checksummed */
3051         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3052             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3053                 ret = -EINVAL;
3054                 goto out_unlock;
3055         }
3056
3057         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3058         if (ret)
3059                 goto out_unlock;
3060
3061         if (same_inode)
3062                 lock_extent_range(src, same_lock_start, same_lock_len);
3063         else
3064                 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3065
3066         /* pass original length for comparison so we stay within i_size */
3067         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3068         if (ret == 0)
3069                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff);
3070
3071         if (same_inode)
3072                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3073                               same_lock_start + same_lock_len - 1);
3074         else
3075                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3076
3077         btrfs_cmp_data_free(&cmp);
3078 out_unlock:
3079         if (same_inode)
3080                 mutex_unlock(&src->i_mutex);
3081         else
3082                 btrfs_double_inode_unlock(src, dst);
3083
3084         return ret;
3085 }
3086
3087 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
3088
3089 static long btrfs_ioctl_file_extent_same(struct file *file,
3090                         struct btrfs_ioctl_same_args __user *argp)
3091 {
3092         struct btrfs_ioctl_same_args *same;
3093         struct btrfs_ioctl_same_extent_info *info;
3094         struct inode *src = file_inode(file);
3095         u64 off;
3096         u64 len;
3097         int i;
3098         int ret;
3099         unsigned long size;
3100         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3101         bool is_admin = capable(CAP_SYS_ADMIN);
3102         u16 count;
3103
3104         if (!(file->f_mode & FMODE_READ))
3105                 return -EINVAL;
3106
3107         ret = mnt_want_write_file(file);
3108         if (ret)
3109                 return ret;
3110
3111         if (get_user(count, &argp->dest_count)) {
3112                 ret = -EFAULT;
3113                 goto out;
3114         }
3115
3116         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
3117
3118         same = memdup_user(argp, size);
3119
3120         if (IS_ERR(same)) {
3121                 ret = PTR_ERR(same);
3122                 goto out;
3123         }
3124
3125         off = same->logical_offset;
3126         len = same->length;
3127
3128         /*
3129          * Limit the total length we will dedupe for each operation.
3130          * This is intended to bound the total time spent in this
3131          * ioctl to something sane.
3132          */
3133         if (len > BTRFS_MAX_DEDUPE_LEN)
3134                 len = BTRFS_MAX_DEDUPE_LEN;
3135
3136         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3137                 /*
3138                  * Btrfs does not support blocksize < page_size. As a
3139                  * result, btrfs_cmp_data() won't correctly handle
3140                  * this situation without an update.
3141                  */
3142                 ret = -EINVAL;
3143                 goto out;
3144         }
3145
3146         ret = -EISDIR;
3147         if (S_ISDIR(src->i_mode))
3148                 goto out;
3149
3150         ret = -EACCES;
3151         if (!S_ISREG(src->i_mode))
3152                 goto out;
3153
3154         /* pre-format output fields to sane values */
3155         for (i = 0; i < count; i++) {
3156                 same->info[i].bytes_deduped = 0ULL;
3157                 same->info[i].status = 0;
3158         }
3159
3160         for (i = 0, info = same->info; i < count; i++, info++) {
3161                 struct inode *dst;
3162                 struct fd dst_file = fdget(info->fd);
3163                 if (!dst_file.file) {
3164                         info->status = -EBADF;
3165                         continue;
3166                 }
3167                 dst = file_inode(dst_file.file);
3168
3169                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3170                         info->status = -EINVAL;
3171                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3172                         info->status = -EXDEV;
3173                 } else if (S_ISDIR(dst->i_mode)) {
3174                         info->status = -EISDIR;
3175                 } else if (!S_ISREG(dst->i_mode)) {
3176                         info->status = -EACCES;
3177                 } else {
3178                         info->status = btrfs_extent_same(src, off, len, dst,
3179                                                         info->logical_offset);
3180                         if (info->status == 0)
3181                                 info->bytes_deduped += len;
3182                 }
3183                 fdput(dst_file);
3184         }
3185
3186         ret = copy_to_user(argp, same, size);
3187         if (ret)
3188                 ret = -EFAULT;
3189
3190 out:
3191         mnt_drop_write_file(file);
3192         return ret;
3193 }
3194
3195 /* Helper to check and see if this root currently has a ref on the given disk
3196  * bytenr.  If it does then we need to update the quota for this root.  This
3197  * doesn't do anything if quotas aren't enabled.
3198  */
3199 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3200                      u64 disko)
3201 {
3202         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
3203         struct ulist *roots;
3204         struct ulist_iterator uiter;
3205         struct ulist_node *root_node = NULL;
3206         int ret;
3207
3208         if (!root->fs_info->quota_enabled)
3209                 return 1;
3210
3211         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3212         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3213                                    tree_mod_seq_elem.seq, &roots);
3214         if (ret < 0)
3215                 goto out;
3216         ret = 0;
3217         ULIST_ITER_INIT(&uiter);
3218         while ((root_node = ulist_next(roots, &uiter))) {
3219                 if (root_node->val == root->objectid) {
3220                         ret = 1;
3221                         break;
3222                 }
3223         }
3224         ulist_free(roots);
3225 out:
3226         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3227         return ret;
3228 }
3229
3230 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3231                                      struct inode *inode,
3232                                      u64 endoff,
3233                                      const u64 destoff,
3234                                      const u64 olen)
3235 {
3236         struct btrfs_root *root = BTRFS_I(inode)->root;
3237         int ret;
3238
3239         inode_inc_iversion(inode);
3240         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3241         /*
3242          * We round up to the block size at eof when determining which
3243          * extents to clone above, but shouldn't round up the file size.
3244          */
3245         if (endoff > destoff + olen)
3246                 endoff = destoff + olen;
3247         if (endoff > inode->i_size)
3248                 btrfs_i_size_write(inode, endoff);
3249
3250         ret = btrfs_update_inode(trans, root, inode);
3251         if (ret) {
3252                 btrfs_abort_transaction(trans, root, ret);
3253                 btrfs_end_transaction(trans, root);
3254                 goto out;
3255         }
3256         ret = btrfs_end_transaction(trans, root);
3257 out:
3258         return ret;
3259 }
3260
3261 static void clone_update_extent_map(struct inode *inode,
3262                                     const struct btrfs_trans_handle *trans,
3263                                     const struct btrfs_path *path,
3264                                     const u64 hole_offset,
3265                                     const u64 hole_len)
3266 {
3267         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3268         struct extent_map *em;
3269         int ret;
3270
3271         em = alloc_extent_map();
3272         if (!em) {
3273                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3274                         &BTRFS_I(inode)->runtime_flags);
3275                 return;
3276         }
3277
3278         if (path) {
3279                 struct btrfs_file_extent_item *fi;
3280
3281                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3282                                     struct btrfs_file_extent_item);
3283                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3284                 em->generation = -1;
3285                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3286                     BTRFS_FILE_EXTENT_INLINE)
3287                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3288                                 &BTRFS_I(inode)->runtime_flags);
3289         } else {
3290                 em->start = hole_offset;
3291                 em->len = hole_len;
3292                 em->ram_bytes = em->len;
3293                 em->orig_start = hole_offset;
3294                 em->block_start = EXTENT_MAP_HOLE;
3295                 em->block_len = 0;
3296                 em->orig_block_len = 0;
3297                 em->compress_type = BTRFS_COMPRESS_NONE;
3298                 em->generation = trans->transid;
3299         }
3300
3301         while (1) {
3302                 write_lock(&em_tree->lock);
3303                 ret = add_extent_mapping(em_tree, em, 1);
3304                 write_unlock(&em_tree->lock);
3305                 if (ret != -EEXIST) {
3306                         free_extent_map(em);
3307                         break;
3308                 }
3309                 btrfs_drop_extent_cache(inode, em->start,
3310                                         em->start + em->len - 1, 0);
3311         }
3312
3313         if (ret)
3314                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3315                         &BTRFS_I(inode)->runtime_flags);
3316 }
3317
3318 /**
3319  * btrfs_clone() - clone a range from inode file to another
3320  *
3321  * @src: Inode to clone from
3322  * @inode: Inode to clone to
3323  * @off: Offset within source to start clone from
3324  * @olen: Original length, passed by user, of range to clone
3325  * @olen_aligned: Block-aligned value of olen, extent_same uses
3326  *               identical values here
3327  * @destoff: Offset within @inode to start clone
3328  */
3329 static int btrfs_clone(struct inode *src, struct inode *inode,
3330                        const u64 off, const u64 olen, const u64 olen_aligned,
3331                        const u64 destoff)
3332 {
3333         struct btrfs_root *root = BTRFS_I(inode)->root;
3334         struct btrfs_path *path = NULL;
3335         struct extent_buffer *leaf;
3336         struct btrfs_trans_handle *trans;
3337         char *buf = NULL;
3338         struct btrfs_key key;
3339         u32 nritems;
3340         int slot;
3341         int ret;
3342         int no_quota;
3343         const u64 len = olen_aligned;
3344         u64 last_disko = 0;
3345         u64 last_dest_end = destoff;
3346
3347         ret = -ENOMEM;
3348         buf = vmalloc(root->nodesize);
3349         if (!buf)
3350                 return ret;
3351
3352         path = btrfs_alloc_path();
3353         if (!path) {
3354                 vfree(buf);
3355                 return ret;
3356         }
3357
3358         path->reada = 2;
3359         /* clone data */
3360         key.objectid = btrfs_ino(src);
3361         key.type = BTRFS_EXTENT_DATA_KEY;
3362         key.offset = off;
3363
3364         while (1) {
3365                 u64 next_key_min_offset = key.offset + 1;
3366
3367                 /*
3368                  * note the key will change type as we walk through the
3369                  * tree.
3370                  */
3371                 path->leave_spinning = 1;
3372                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3373                                 0, 0);
3374                 if (ret < 0)
3375                         goto out;
3376                 /*
3377                  * First search, if no extent item that starts at offset off was
3378                  * found but the previous item is an extent item, it's possible
3379                  * it might overlap our target range, therefore process it.
3380                  */
3381                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3382                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3383                                               path->slots[0] - 1);
3384                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3385                                 path->slots[0]--;
3386                 }
3387
3388                 nritems = btrfs_header_nritems(path->nodes[0]);
3389 process_slot:
3390                 no_quota = 1;
3391                 if (path->slots[0] >= nritems) {
3392                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3393                         if (ret < 0)
3394                                 goto out;
3395                         if (ret > 0)
3396                                 break;
3397                         nritems = btrfs_header_nritems(path->nodes[0]);
3398                 }
3399                 leaf = path->nodes[0];
3400                 slot = path->slots[0];
3401
3402                 btrfs_item_key_to_cpu(leaf, &key, slot);
3403                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3404                     key.objectid != btrfs_ino(src))
3405                         break;
3406
3407                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3408                         struct btrfs_file_extent_item *extent;
3409                         int type;
3410                         u32 size;
3411                         struct btrfs_key new_key;
3412                         u64 disko = 0, diskl = 0;
3413                         u64 datao = 0, datal = 0;
3414                         u8 comp;
3415                         u64 drop_start;
3416
3417                         extent = btrfs_item_ptr(leaf, slot,
3418                                                 struct btrfs_file_extent_item);
3419                         comp = btrfs_file_extent_compression(leaf, extent);
3420                         type = btrfs_file_extent_type(leaf, extent);
3421                         if (type == BTRFS_FILE_EXTENT_REG ||
3422                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3423                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3424                                                                       extent);
3425                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3426                                                                  extent);
3427                                 datao = btrfs_file_extent_offset(leaf, extent);
3428                                 datal = btrfs_file_extent_num_bytes(leaf,
3429                                                                     extent);
3430                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3431                                 /* take upper bound, may be compressed */
3432                                 datal = btrfs_file_extent_ram_bytes(leaf,
3433                                                                     extent);
3434                         }
3435
3436                         /*
3437                          * The first search might have left us at an extent
3438                          * item that ends before our target range's start, can
3439                          * happen if we have holes and NO_HOLES feature enabled.
3440                          */
3441                         if (key.offset + datal <= off) {
3442                                 path->slots[0]++;
3443                                 goto process_slot;
3444                         } else if (key.offset >= off + len) {
3445                                 break;
3446                         }
3447                         next_key_min_offset = key.offset + datal;
3448                         size = btrfs_item_size_nr(leaf, slot);
3449                         read_extent_buffer(leaf, buf,
3450                                            btrfs_item_ptr_offset(leaf, slot),
3451                                            size);
3452
3453                         btrfs_release_path(path);
3454                         path->leave_spinning = 0;
3455
3456                         memcpy(&new_key, &key, sizeof(new_key));
3457                         new_key.objectid = btrfs_ino(inode);
3458                         if (off <= key.offset)
3459                                 new_key.offset = key.offset + destoff - off;
3460                         else
3461                                 new_key.offset = destoff;
3462
3463                         /*
3464                          * Deal with a hole that doesn't have an extent item
3465                          * that represents it (NO_HOLES feature enabled).
3466                          * This hole is either in the middle of the cloning
3467                          * range or at the beginning (fully overlaps it or
3468                          * partially overlaps it).
3469                          */
3470                         if (new_key.offset != last_dest_end)
3471                                 drop_start = last_dest_end;
3472                         else
3473                                 drop_start = new_key.offset;
3474
3475                         /*
3476                          * 1 - adjusting old extent (we may have to split it)
3477                          * 1 - add new extent
3478                          * 1 - inode update
3479                          */
3480                         trans = btrfs_start_transaction(root, 3);
3481                         if (IS_ERR(trans)) {
3482                                 ret = PTR_ERR(trans);
3483                                 goto out;
3484                         }
3485
3486                         if (type == BTRFS_FILE_EXTENT_REG ||
3487                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3488                                 /*
3489                                  *    a  | --- range to clone ---|  b
3490                                  * | ------------- extent ------------- |
3491                                  */
3492
3493                                 /* subtract range b */
3494                                 if (key.offset + datal > off + len)
3495                                         datal = off + len - key.offset;
3496
3497                                 /* subtract range a */
3498                                 if (off > key.offset) {
3499                                         datao += off - key.offset;
3500                                         datal -= off - key.offset;
3501                                 }
3502
3503                                 ret = btrfs_drop_extents(trans, root, inode,
3504                                                          drop_start,
3505                                                          new_key.offset + datal,
3506                                                          1);
3507                                 if (ret) {
3508                                         if (ret != -EOPNOTSUPP)
3509                                                 btrfs_abort_transaction(trans,
3510                                                                 root, ret);
3511                                         btrfs_end_transaction(trans, root);
3512                                         goto out;
3513                                 }
3514
3515                                 ret = btrfs_insert_empty_item(trans, root, path,
3516                                                               &new_key, size);
3517                                 if (ret) {
3518                                         btrfs_abort_transaction(trans, root,
3519                                                                 ret);
3520                                         btrfs_end_transaction(trans, root);
3521                                         goto out;
3522                                 }
3523
3524                                 leaf = path->nodes[0];
3525                                 slot = path->slots[0];
3526                                 write_extent_buffer(leaf, buf,
3527                                             btrfs_item_ptr_offset(leaf, slot),
3528                                             size);
3529
3530                                 extent = btrfs_item_ptr(leaf, slot,
3531                                                 struct btrfs_file_extent_item);
3532
3533                                 /* disko == 0 means it's a hole */
3534                                 if (!disko)
3535                                         datao = 0;
3536
3537                                 btrfs_set_file_extent_offset(leaf, extent,
3538                                                              datao);
3539                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3540                                                                 datal);
3541
3542                                 /*
3543                                  * We need to look up the roots that point at
3544                                  * this bytenr and see if the new root does.  If
3545                                  * it does not we need to make sure we update
3546                                  * quotas appropriately.
3547                                  */
3548                                 if (disko && root != BTRFS_I(src)->root &&
3549                                     disko != last_disko) {
3550                                         no_quota = check_ref(trans, root,
3551                                                              disko);
3552                                         if (no_quota < 0) {
3553                                                 btrfs_abort_transaction(trans,
3554                                                                         root,
3555                                                                         ret);
3556                                                 btrfs_end_transaction(trans,
3557                                                                       root);
3558                                                 ret = no_quota;
3559                                                 goto out;
3560                                         }
3561                                 }
3562
3563                                 if (disko) {
3564                                         inode_add_bytes(inode, datal);
3565                                         ret = btrfs_inc_extent_ref(trans, root,
3566                                                         disko, diskl, 0,
3567                                                         root->root_key.objectid,
3568                                                         btrfs_ino(inode),
3569                                                         new_key.offset - datao,
3570                                                         no_quota);
3571                                         if (ret) {
3572                                                 btrfs_abort_transaction(trans,
3573                                                                         root,
3574                                                                         ret);
3575                                                 btrfs_end_transaction(trans,
3576                                                                       root);
3577                                                 goto out;
3578
3579                                         }
3580                                 }
3581                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3582                                 u64 skip = 0;
3583                                 u64 trim = 0;
3584                                 u64 aligned_end = 0;
3585
3586                                 if (off > key.offset) {
3587                                         skip = off - key.offset;
3588                                         new_key.offset += skip;
3589                                 }
3590
3591                                 if (key.offset + datal > off + len)
3592                                         trim = key.offset + datal - (off + len);
3593
3594                                 if (comp && (skip || trim)) {
3595                                         ret = -EINVAL;
3596                                         btrfs_end_transaction(trans, root);
3597                                         goto out;
3598                                 }
3599                                 size -= skip + trim;
3600                                 datal -= skip + trim;
3601
3602                                 aligned_end = ALIGN(new_key.offset + datal,
3603                                                     root->sectorsize);
3604                                 ret = btrfs_drop_extents(trans, root, inode,
3605                                                          drop_start,
3606                                                          aligned_end,
3607                                                          1);
3608                                 if (ret) {
3609                                         if (ret != -EOPNOTSUPP)
3610                                                 btrfs_abort_transaction(trans,
3611                                                         root, ret);
3612                                         btrfs_end_transaction(trans, root);
3613                                         goto out;
3614                                 }
3615
3616                                 ret = btrfs_insert_empty_item(trans, root, path,
3617                                                               &new_key, size);
3618                                 if (ret) {
3619                                         btrfs_abort_transaction(trans, root,
3620                                                                 ret);
3621                                         btrfs_end_transaction(trans, root);
3622                                         goto out;
3623                                 }
3624
3625                                 if (skip) {
3626                                         u32 start =
3627                                           btrfs_file_extent_calc_inline_size(0);
3628                                         memmove(buf+start, buf+start+skip,
3629                                                 datal);
3630                                 }
3631
3632                                 leaf = path->nodes[0];
3633                                 slot = path->slots[0];
3634                                 write_extent_buffer(leaf, buf,
3635                                             btrfs_item_ptr_offset(leaf, slot),
3636                                             size);
3637                                 inode_add_bytes(inode, datal);
3638                         }
3639
3640                         /* If we have an implicit hole (NO_HOLES feature). */
3641                         if (drop_start < new_key.offset)
3642                                 clone_update_extent_map(inode, trans,
3643                                                 NULL, drop_start,
3644                                                 new_key.offset - drop_start);
3645
3646                         clone_update_extent_map(inode, trans, path, 0, 0);
3647
3648                         btrfs_mark_buffer_dirty(leaf);
3649                         btrfs_release_path(path);
3650
3651                         last_dest_end = ALIGN(new_key.offset + datal,
3652                                               root->sectorsize);
3653                         ret = clone_finish_inode_update(trans, inode,
3654                                                         last_dest_end,
3655                                                         destoff, olen);
3656                         if (ret)
3657                                 goto out;
3658                         if (new_key.offset + datal >= destoff + len)
3659                                 break;
3660                 }
3661                 btrfs_release_path(path);
3662                 key.offset = next_key_min_offset;
3663         }
3664         ret = 0;
3665
3666         if (last_dest_end < destoff + len) {
3667                 /*
3668                  * We have an implicit hole (NO_HOLES feature is enabled) that
3669                  * fully or partially overlaps our cloning range at its end.
3670                  */
3671                 btrfs_release_path(path);
3672
3673                 /*
3674                  * 1 - remove extent(s)
3675                  * 1 - inode update
3676                  */
3677                 trans = btrfs_start_transaction(root, 2);
3678                 if (IS_ERR(trans)) {
3679                         ret = PTR_ERR(trans);
3680                         goto out;
3681                 }
3682                 ret = btrfs_drop_extents(trans, root, inode,
3683                                          last_dest_end, destoff + len, 1);
3684                 if (ret) {
3685                         if (ret != -EOPNOTSUPP)
3686                                 btrfs_abort_transaction(trans, root, ret);
3687                         btrfs_end_transaction(trans, root);
3688                         goto out;
3689                 }
3690                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3691                                         destoff + len - last_dest_end);
3692                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3693                                                 destoff, olen);
3694         }
3695
3696 out:
3697         btrfs_free_path(path);
3698         vfree(buf);
3699         return ret;
3700 }
3701
3702 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3703                                        u64 off, u64 olen, u64 destoff)
3704 {
3705         struct inode *inode = file_inode(file);
3706         struct btrfs_root *root = BTRFS_I(inode)->root;
3707         struct fd src_file;
3708         struct inode *src;
3709         int ret;
3710         u64 len = olen;
3711         u64 bs = root->fs_info->sb->s_blocksize;
3712         int same_inode = 0;
3713
3714         /*
3715          * TODO:
3716          * - split compressed inline extents.  annoying: we need to
3717          *   decompress into destination's address_space (the file offset
3718          *   may change, so source mapping won't do), then recompress (or
3719          *   otherwise reinsert) a subrange.
3720          *
3721          * - split destination inode's inline extents.  The inline extents can
3722          *   be either compressed or non-compressed.
3723          */
3724
3725         /* the destination must be opened for writing */
3726         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3727                 return -EINVAL;
3728
3729         if (btrfs_root_readonly(root))
3730                 return -EROFS;
3731
3732         ret = mnt_want_write_file(file);
3733         if (ret)
3734                 return ret;
3735
3736         src_file = fdget(srcfd);
3737         if (!src_file.file) {
3738                 ret = -EBADF;
3739                 goto out_drop_write;
3740         }
3741
3742         ret = -EXDEV;
3743         if (src_file.file->f_path.mnt != file->f_path.mnt)
3744                 goto out_fput;
3745
3746         src = file_inode(src_file.file);
3747
3748         ret = -EINVAL;
3749         if (src == inode)
3750                 same_inode = 1;
3751
3752         /* the src must be open for reading */
3753         if (!(src_file.file->f_mode & FMODE_READ))
3754                 goto out_fput;
3755
3756         /* don't make the dst file partly checksummed */
3757         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3758             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3759                 goto out_fput;
3760
3761         ret = -EISDIR;
3762         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3763                 goto out_fput;
3764
3765         ret = -EXDEV;
3766         if (src->i_sb != inode->i_sb)
3767                 goto out_fput;
3768
3769         if (!same_inode) {
3770                 if (inode < src) {
3771                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3772                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3773                 } else {
3774                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3775                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3776                 }
3777         } else {
3778                 mutex_lock(&src->i_mutex);
3779         }
3780
3781         /* determine range to clone */
3782         ret = -EINVAL;
3783         if (off + len > src->i_size || off + len < off)
3784                 goto out_unlock;
3785         if (len == 0)
3786                 olen = len = src->i_size - off;
3787         /* if we extend to eof, continue to block boundary */
3788         if (off + len == src->i_size)
3789                 len = ALIGN(src->i_size, bs) - off;
3790
3791         if (len == 0) {
3792                 ret = 0;
3793                 goto out_unlock;
3794         }
3795
3796         /* verify the end result is block aligned */
3797         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3798             !IS_ALIGNED(destoff, bs))
3799                 goto out_unlock;
3800
3801         /* verify if ranges are overlapped within the same file */
3802         if (same_inode) {
3803                 if (destoff + len > off && destoff < off + len)
3804                         goto out_unlock;
3805         }
3806
3807         if (destoff > inode->i_size) {
3808                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3809                 if (ret)
3810                         goto out_unlock;
3811         }
3812
3813         /*
3814          * Lock the target range too. Right after we replace the file extent
3815          * items in the fs tree (which now point to the cloned data), we might
3816          * have a worker replace them with extent items relative to a write
3817          * operation that was issued before this clone operation (i.e. confront
3818          * with inode.c:btrfs_finish_ordered_io).
3819          */
3820         if (same_inode) {
3821                 u64 lock_start = min_t(u64, off, destoff);
3822                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3823
3824                 lock_extent_range(src, lock_start, lock_len);
3825         } else {
3826                 lock_extent_range(src, off, len);
3827                 lock_extent_range(inode, destoff, len);
3828         }
3829
3830         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3831
3832         if (same_inode) {
3833                 u64 lock_start = min_t(u64, off, destoff);
3834                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3835
3836                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3837         } else {
3838                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3839                 unlock_extent(&BTRFS_I(inode)->io_tree, destoff,
3840                               destoff + len - 1);
3841         }
3842         /*
3843          * Truncate page cache pages so that future reads will see the cloned
3844          * data immediately and not the previous data.
3845          */
3846         truncate_inode_pages_range(&inode->i_data, destoff,
3847                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3848 out_unlock:
3849         if (!same_inode) {
3850                 if (inode < src) {
3851                         mutex_unlock(&src->i_mutex);
3852                         mutex_unlock(&inode->i_mutex);
3853                 } else {
3854                         mutex_unlock(&inode->i_mutex);
3855                         mutex_unlock(&src->i_mutex);
3856                 }
3857         } else {
3858                 mutex_unlock(&src->i_mutex);
3859         }
3860 out_fput:
3861         fdput(src_file);
3862 out_drop_write:
3863         mnt_drop_write_file(file);
3864         return ret;
3865 }
3866
3867 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3868 {
3869         struct btrfs_ioctl_clone_range_args args;
3870
3871         if (copy_from_user(&args, argp, sizeof(args)))
3872                 return -EFAULT;
3873         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3874                                  args.src_length, args.dest_offset);
3875 }
3876
3877 /*
3878  * there are many ways the trans_start and trans_end ioctls can lead
3879  * to deadlocks.  They should only be used by applications that
3880  * basically own the machine, and have a very in depth understanding
3881  * of all the possible deadlocks and enospc problems.
3882  */
3883 static long btrfs_ioctl_trans_start(struct file *file)
3884 {
3885         struct inode *inode = file_inode(file);
3886         struct btrfs_root *root = BTRFS_I(inode)->root;
3887         struct btrfs_trans_handle *trans;
3888         int ret;
3889
3890         ret = -EPERM;
3891         if (!capable(CAP_SYS_ADMIN))
3892                 goto out;
3893
3894         ret = -EINPROGRESS;
3895         if (file->private_data)
3896                 goto out;
3897
3898         ret = -EROFS;
3899         if (btrfs_root_readonly(root))
3900                 goto out;
3901
3902         ret = mnt_want_write_file(file);
3903         if (ret)
3904                 goto out;
3905
3906         atomic_inc(&root->fs_info->open_ioctl_trans);
3907
3908         ret = -ENOMEM;
3909         trans = btrfs_start_ioctl_transaction(root);
3910         if (IS_ERR(trans))
3911                 goto out_drop;
3912
3913         file->private_data = trans;
3914         return 0;
3915
3916 out_drop:
3917         atomic_dec(&root->fs_info->open_ioctl_trans);
3918         mnt_drop_write_file(file);
3919 out:
3920         return ret;
3921 }
3922
3923 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3924 {
3925         struct inode *inode = file_inode(file);
3926         struct btrfs_root *root = BTRFS_I(inode)->root;
3927         struct btrfs_root *new_root;
3928         struct btrfs_dir_item *di;
3929         struct btrfs_trans_handle *trans;
3930         struct btrfs_path *path;
3931         struct btrfs_key location;
3932         struct btrfs_disk_key disk_key;
3933         u64 objectid = 0;
3934         u64 dir_id;
3935         int ret;
3936
3937         if (!capable(CAP_SYS_ADMIN))
3938                 return -EPERM;
3939
3940         ret = mnt_want_write_file(file);
3941         if (ret)
3942                 return ret;
3943
3944         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3945                 ret = -EFAULT;
3946                 goto out;
3947         }
3948
3949         if (!objectid)
3950                 objectid = BTRFS_FS_TREE_OBJECTID;
3951
3952         location.objectid = objectid;
3953         location.type = BTRFS_ROOT_ITEM_KEY;
3954         location.offset = (u64)-1;
3955
3956         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3957         if (IS_ERR(new_root)) {
3958                 ret = PTR_ERR(new_root);
3959                 goto out;
3960         }
3961
3962         path = btrfs_alloc_path();
3963         if (!path) {
3964                 ret = -ENOMEM;
3965                 goto out;
3966         }
3967         path->leave_spinning = 1;
3968
3969         trans = btrfs_start_transaction(root, 1);
3970         if (IS_ERR(trans)) {
3971                 btrfs_free_path(path);
3972                 ret = PTR_ERR(trans);
3973                 goto out;
3974         }
3975
3976         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3977         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3978                                    dir_id, "default", 7, 1);
3979         if (IS_ERR_OR_NULL(di)) {
3980                 btrfs_free_path(path);
3981                 btrfs_end_transaction(trans, root);
3982                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3983                            "item, this isn't going to work");
3984                 ret = -ENOENT;
3985                 goto out;
3986         }
3987
3988         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3989         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3990         btrfs_mark_buffer_dirty(path->nodes[0]);
3991         btrfs_free_path(path);
3992
3993         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3994         btrfs_end_transaction(trans, root);
3995 out:
3996         mnt_drop_write_file(file);
3997         return ret;
3998 }
3999
4000 void btrfs_get_block_group_info(struct list_head *groups_list,
4001                                 struct btrfs_ioctl_space_info *space)
4002 {
4003         struct btrfs_block_group_cache *block_group;
4004
4005         space->total_bytes = 0;
4006         space->used_bytes = 0;
4007         space->flags = 0;
4008         list_for_each_entry(block_group, groups_list, list) {
4009                 space->flags = block_group->flags;
4010                 space->total_bytes += block_group->key.offset;
4011                 space->used_bytes +=
4012                         btrfs_block_group_used(&block_group->item);
4013         }
4014 }
4015
4016 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4017 {
4018         struct btrfs_ioctl_space_args space_args;
4019         struct btrfs_ioctl_space_info space;
4020         struct btrfs_ioctl_space_info *dest;
4021         struct btrfs_ioctl_space_info *dest_orig;
4022         struct btrfs_ioctl_space_info __user *user_dest;
4023         struct btrfs_space_info *info;
4024         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4025                        BTRFS_BLOCK_GROUP_SYSTEM,
4026                        BTRFS_BLOCK_GROUP_METADATA,
4027                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4028         int num_types = 4;
4029         int alloc_size;
4030         int ret = 0;
4031         u64 slot_count = 0;
4032         int i, c;
4033
4034         if (copy_from_user(&space_args,
4035                            (struct btrfs_ioctl_space_args __user *)arg,
4036                            sizeof(space_args)))
4037                 return -EFAULT;
4038
4039         for (i = 0; i < num_types; i++) {
4040                 struct btrfs_space_info *tmp;
4041
4042                 info = NULL;
4043                 rcu_read_lock();
4044                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4045                                         list) {
4046                         if (tmp->flags == types[i]) {
4047                                 info = tmp;
4048                                 break;
4049                         }
4050                 }
4051                 rcu_read_unlock();
4052
4053                 if (!info)
4054                         continue;
4055
4056                 down_read(&info->groups_sem);
4057                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4058                         if (!list_empty(&info->block_groups[c]))
4059                                 slot_count++;
4060                 }
4061                 up_read(&info->groups_sem);
4062         }
4063
4064         /*
4065          * Global block reserve, exported as a space_info
4066          */
4067         slot_count++;
4068
4069         /* space_slots == 0 means they are asking for a count */
4070         if (space_args.space_slots == 0) {
4071                 space_args.total_spaces = slot_count;
4072                 goto out;
4073         }
4074
4075         slot_count = min_t(u64, space_args.space_slots, slot_count);
4076
4077         alloc_size = sizeof(*dest) * slot_count;
4078
4079         /* we generally have at most 6 or so space infos, one for each raid
4080          * level.  So, a whole page should be more than enough for everyone
4081          */
4082         if (alloc_size > PAGE_CACHE_SIZE)
4083                 return -ENOMEM;
4084
4085         space_args.total_spaces = 0;
4086         dest = kmalloc(alloc_size, GFP_NOFS);
4087         if (!dest)
4088                 return -ENOMEM;
4089         dest_orig = dest;
4090
4091         /* now we have a buffer to copy into */
4092         for (i = 0; i < num_types; i++) {
4093                 struct btrfs_space_info *tmp;
4094
4095                 if (!slot_count)
4096                         break;
4097
4098                 info = NULL;
4099                 rcu_read_lock();
4100                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4101                                         list) {
4102                         if (tmp->flags == types[i]) {
4103                                 info = tmp;
4104                                 break;
4105                         }
4106                 }
4107                 rcu_read_unlock();
4108
4109                 if (!info)
4110                         continue;
4111                 down_read(&info->groups_sem);
4112                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4113                         if (!list_empty(&info->block_groups[c])) {
4114                                 btrfs_get_block_group_info(
4115                                         &info->block_groups[c], &space);
4116                                 memcpy(dest, &space, sizeof(space));
4117                                 dest++;
4118                                 space_args.total_spaces++;
4119                                 slot_count--;
4120                         }
4121                         if (!slot_count)
4122                                 break;
4123                 }
4124                 up_read(&info->groups_sem);
4125         }
4126
4127         /*
4128          * Add global block reserve
4129          */
4130         if (slot_count) {
4131                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4132
4133                 spin_lock(&block_rsv->lock);
4134                 space.total_bytes = block_rsv->size;
4135                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4136                 spin_unlock(&block_rsv->lock);
4137                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4138                 memcpy(dest, &space, sizeof(space));
4139                 space_args.total_spaces++;
4140         }
4141
4142         user_dest = (struct btrfs_ioctl_space_info __user *)
4143                 (arg + sizeof(struct btrfs_ioctl_space_args));
4144
4145         if (copy_to_user(user_dest, dest_orig, alloc_size))
4146                 ret = -EFAULT;
4147
4148         kfree(dest_orig);
4149 out:
4150         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4151                 ret = -EFAULT;
4152
4153         return ret;
4154 }
4155
4156 /*
4157  * there are many ways the trans_start and trans_end ioctls can lead
4158  * to deadlocks.  They should only be used by applications that
4159  * basically own the machine, and have a very in depth understanding
4160  * of all the possible deadlocks and enospc problems.
4161  */
4162 long btrfs_ioctl_trans_end(struct file *file)
4163 {
4164         struct inode *inode = file_inode(file);
4165         struct btrfs_root *root = BTRFS_I(inode)->root;
4166         struct btrfs_trans_handle *trans;
4167
4168         trans = file->private_data;
4169         if (!trans)
4170                 return -EINVAL;
4171         file->private_data = NULL;
4172
4173         btrfs_end_transaction(trans, root);
4174
4175         atomic_dec(&root->fs_info->open_ioctl_trans);
4176
4177         mnt_drop_write_file(file);
4178         return 0;
4179 }
4180
4181 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4182                                             void __user *argp)
4183 {
4184         struct btrfs_trans_handle *trans;
4185         u64 transid;
4186         int ret;
4187
4188         trans = btrfs_attach_transaction_barrier(root);
4189         if (IS_ERR(trans)) {
4190                 if (PTR_ERR(trans) != -ENOENT)
4191                         return PTR_ERR(trans);
4192
4193                 /* No running transaction, don't bother */
4194                 transid = root->fs_info->last_trans_committed;
4195                 goto out;
4196         }
4197         transid = trans->transid;
4198         ret = btrfs_commit_transaction_async(trans, root, 0);
4199         if (ret) {
4200                 btrfs_end_transaction(trans, root);
4201                 return ret;
4202         }
4203 out:
4204         if (argp)
4205                 if (copy_to_user(argp, &transid, sizeof(transid)))
4206                         return -EFAULT;
4207         return 0;
4208 }
4209
4210 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4211                                            void __user *argp)
4212 {
4213         u64 transid;
4214
4215         if (argp) {
4216                 if (copy_from_user(&transid, argp, sizeof(transid)))
4217                         return -EFAULT;
4218         } else {
4219                 transid = 0;  /* current trans */
4220         }
4221         return btrfs_wait_for_commit(root, transid);
4222 }
4223
4224 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4225 {
4226         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4227         struct btrfs_ioctl_scrub_args *sa;
4228         int ret;
4229
4230         if (!capable(CAP_SYS_ADMIN))
4231                 return -EPERM;
4232
4233         sa = memdup_user(arg, sizeof(*sa));
4234         if (IS_ERR(sa))
4235                 return PTR_ERR(sa);
4236
4237         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4238                 ret = mnt_want_write_file(file);
4239                 if (ret)
4240                         goto out;
4241         }
4242
4243         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4244                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4245                               0);
4246
4247         if (copy_to_user(arg, sa, sizeof(*sa)))
4248                 ret = -EFAULT;
4249
4250         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4251                 mnt_drop_write_file(file);
4252 out:
4253         kfree(sa);
4254         return ret;
4255 }
4256
4257 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4258 {
4259         if (!capable(CAP_SYS_ADMIN))
4260                 return -EPERM;
4261
4262         return btrfs_scrub_cancel(root->fs_info);
4263 }
4264
4265 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4266                                        void __user *arg)
4267 {
4268         struct btrfs_ioctl_scrub_args *sa;
4269         int ret;
4270
4271         if (!capable(CAP_SYS_ADMIN))
4272                 return -EPERM;
4273
4274         sa = memdup_user(arg, sizeof(*sa));
4275         if (IS_ERR(sa))
4276                 return PTR_ERR(sa);
4277
4278         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4279
4280         if (copy_to_user(arg, sa, sizeof(*sa)))
4281                 ret = -EFAULT;
4282
4283         kfree(sa);
4284         return ret;
4285 }
4286
4287 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4288                                       void __user *arg)
4289 {
4290         struct btrfs_ioctl_get_dev_stats *sa;
4291         int ret;
4292
4293         sa = memdup_user(arg, sizeof(*sa));
4294         if (IS_ERR(sa))
4295                 return PTR_ERR(sa);
4296
4297         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4298                 kfree(sa);
4299                 return -EPERM;
4300         }
4301
4302         ret = btrfs_get_dev_stats(root, sa);
4303
4304         if (copy_to_user(arg, sa, sizeof(*sa)))
4305                 ret = -EFAULT;
4306
4307         kfree(sa);
4308         return ret;
4309 }
4310
4311 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4312 {
4313         struct btrfs_ioctl_dev_replace_args *p;
4314         int ret;
4315
4316         if (!capable(CAP_SYS_ADMIN))
4317                 return -EPERM;
4318
4319         p = memdup_user(arg, sizeof(*p));
4320         if (IS_ERR(p))
4321                 return PTR_ERR(p);
4322
4323         switch (p->cmd) {
4324         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4325                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4326                         ret = -EROFS;
4327                         goto out;
4328                 }
4329                 if (atomic_xchg(
4330                         &root->fs_info->mutually_exclusive_operation_running,
4331                         1)) {
4332                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4333                 } else {
4334                         ret = btrfs_dev_replace_start(root, p);
4335                         atomic_set(
4336                          &root->fs_info->mutually_exclusive_operation_running,
4337                          0);
4338                 }
4339                 break;
4340         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4341                 btrfs_dev_replace_status(root->fs_info, p);
4342                 ret = 0;
4343                 break;
4344         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4345                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4346                 break;
4347         default:
4348                 ret = -EINVAL;
4349                 break;
4350         }
4351
4352         if (copy_to_user(arg, p, sizeof(*p)))
4353                 ret = -EFAULT;
4354 out:
4355         kfree(p);
4356         return ret;
4357 }
4358
4359 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4360 {
4361         int ret = 0;
4362         int i;
4363         u64 rel_ptr;
4364         int size;
4365         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4366         struct inode_fs_paths *ipath = NULL;
4367         struct btrfs_path *path;
4368
4369         if (!capable(CAP_DAC_READ_SEARCH))
4370                 return -EPERM;
4371
4372         path = btrfs_alloc_path();
4373         if (!path) {
4374                 ret = -ENOMEM;
4375                 goto out;
4376         }
4377
4378         ipa = memdup_user(arg, sizeof(*ipa));
4379         if (IS_ERR(ipa)) {
4380                 ret = PTR_ERR(ipa);
4381                 ipa = NULL;
4382                 goto out;
4383         }
4384
4385         size = min_t(u32, ipa->size, 4096);
4386         ipath = init_ipath(size, root, path);
4387         if (IS_ERR(ipath)) {
4388                 ret = PTR_ERR(ipath);
4389                 ipath = NULL;
4390                 goto out;
4391         }
4392
4393         ret = paths_from_inode(ipa->inum, ipath);
4394         if (ret < 0)
4395                 goto out;
4396
4397         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4398                 rel_ptr = ipath->fspath->val[i] -
4399                           (u64)(unsigned long)ipath->fspath->val;
4400                 ipath->fspath->val[i] = rel_ptr;
4401         }
4402
4403         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4404                            (void *)(unsigned long)ipath->fspath, size);
4405         if (ret) {
4406                 ret = -EFAULT;
4407                 goto out;
4408         }
4409
4410 out:
4411         btrfs_free_path(path);
4412         free_ipath(ipath);
4413         kfree(ipa);
4414
4415         return ret;
4416 }
4417
4418 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4419 {
4420         struct btrfs_data_container *inodes = ctx;
4421         const size_t c = 3 * sizeof(u64);
4422
4423         if (inodes->bytes_left >= c) {
4424                 inodes->bytes_left -= c;
4425                 inodes->val[inodes->elem_cnt] = inum;
4426                 inodes->val[inodes->elem_cnt + 1] = offset;
4427                 inodes->val[inodes->elem_cnt + 2] = root;
4428                 inodes->elem_cnt += 3;
4429         } else {
4430                 inodes->bytes_missing += c - inodes->bytes_left;
4431                 inodes->bytes_left = 0;
4432                 inodes->elem_missed += 3;
4433         }
4434
4435         return 0;
4436 }
4437
4438 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4439                                         void __user *arg)
4440 {
4441         int ret = 0;
4442         int size;
4443         struct btrfs_ioctl_logical_ino_args *loi;
4444         struct btrfs_data_container *inodes = NULL;
4445         struct btrfs_path *path = NULL;
4446
4447         if (!capable(CAP_SYS_ADMIN))
4448                 return -EPERM;
4449
4450         loi = memdup_user(arg, sizeof(*loi));
4451         if (IS_ERR(loi)) {
4452                 ret = PTR_ERR(loi);
4453                 loi = NULL;
4454                 goto out;
4455         }
4456
4457         path = btrfs_alloc_path();
4458         if (!path) {
4459                 ret = -ENOMEM;
4460                 goto out;
4461         }
4462
4463         size = min_t(u32, loi->size, 64 * 1024);
4464         inodes = init_data_container(size);
4465         if (IS_ERR(inodes)) {
4466                 ret = PTR_ERR(inodes);
4467                 inodes = NULL;
4468                 goto out;
4469         }
4470
4471         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4472                                           build_ino_list, inodes);
4473         if (ret == -EINVAL)
4474                 ret = -ENOENT;
4475         if (ret < 0)
4476                 goto out;
4477
4478         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4479                            (void *)(unsigned long)inodes, size);
4480         if (ret)
4481                 ret = -EFAULT;
4482
4483 out:
4484         btrfs_free_path(path);
4485         vfree(inodes);
4486         kfree(loi);
4487
4488         return ret;
4489 }
4490
4491 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4492                                struct btrfs_ioctl_balance_args *bargs)
4493 {
4494         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4495
4496         bargs->flags = bctl->flags;
4497
4498         if (atomic_read(&fs_info->balance_running))
4499                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4500         if (atomic_read(&fs_info->balance_pause_req))
4501                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4502         if (atomic_read(&fs_info->balance_cancel_req))
4503                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4504
4505         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4506         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4507         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4508
4509         if (lock) {
4510                 spin_lock(&fs_info->balance_lock);
4511                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4512                 spin_unlock(&fs_info->balance_lock);
4513         } else {
4514                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4515         }
4516 }
4517
4518 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4519 {
4520         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4521         struct btrfs_fs_info *fs_info = root->fs_info;
4522         struct btrfs_ioctl_balance_args *bargs;
4523         struct btrfs_balance_control *bctl;
4524         bool need_unlock; /* for mut. excl. ops lock */
4525         int ret;
4526
4527         if (!capable(CAP_SYS_ADMIN))
4528                 return -EPERM;
4529
4530         ret = mnt_want_write_file(file);
4531         if (ret)
4532                 return ret;
4533
4534 again:
4535         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4536                 mutex_lock(&fs_info->volume_mutex);
4537                 mutex_lock(&fs_info->balance_mutex);
4538                 need_unlock = true;
4539                 goto locked;
4540         }
4541
4542         /*
4543          * mut. excl. ops lock is locked.  Three possibilites:
4544          *   (1) some other op is running
4545          *   (2) balance is running
4546          *   (3) balance is paused -- special case (think resume)
4547          */
4548         mutex_lock(&fs_info->balance_mutex);
4549         if (fs_info->balance_ctl) {
4550                 /* this is either (2) or (3) */
4551                 if (!atomic_read(&fs_info->balance_running)) {
4552                         mutex_unlock(&fs_info->balance_mutex);
4553                         if (!mutex_trylock(&fs_info->volume_mutex))
4554                                 goto again;
4555                         mutex_lock(&fs_info->balance_mutex);
4556
4557                         if (fs_info->balance_ctl &&
4558                             !atomic_read(&fs_info->balance_running)) {
4559                                 /* this is (3) */
4560                                 need_unlock = false;
4561                                 goto locked;
4562                         }
4563
4564                         mutex_unlock(&fs_info->balance_mutex);
4565                         mutex_unlock(&fs_info->volume_mutex);
4566                         goto again;
4567                 } else {
4568                         /* this is (2) */
4569                         mutex_unlock(&fs_info->balance_mutex);
4570                         ret = -EINPROGRESS;
4571                         goto out;
4572                 }
4573         } else {
4574                 /* this is (1) */
4575                 mutex_unlock(&fs_info->balance_mutex);
4576                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4577                 goto out;
4578         }
4579
4580 locked:
4581         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4582
4583         if (arg) {
4584                 bargs = memdup_user(arg, sizeof(*bargs));
4585                 if (IS_ERR(bargs)) {
4586                         ret = PTR_ERR(bargs);
4587                         goto out_unlock;
4588                 }
4589
4590                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4591                         if (!fs_info->balance_ctl) {
4592                                 ret = -ENOTCONN;
4593                                 goto out_bargs;
4594                         }
4595
4596                         bctl = fs_info->balance_ctl;
4597                         spin_lock(&fs_info->balance_lock);
4598                         bctl->flags |= BTRFS_BALANCE_RESUME;
4599                         spin_unlock(&fs_info->balance_lock);
4600
4601                         goto do_balance;
4602                 }
4603         } else {
4604                 bargs = NULL;
4605         }
4606
4607         if (fs_info->balance_ctl) {
4608                 ret = -EINPROGRESS;
4609                 goto out_bargs;
4610         }
4611
4612         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4613         if (!bctl) {
4614                 ret = -ENOMEM;
4615                 goto out_bargs;
4616         }
4617
4618         bctl->fs_info = fs_info;
4619         if (arg) {
4620                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4621                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4622                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4623
4624                 bctl->flags = bargs->flags;
4625         } else {
4626                 /* balance everything - no filters */
4627                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4628         }
4629
4630 do_balance:
4631         /*
4632          * Ownership of bctl and mutually_exclusive_operation_running
4633          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4634          * or, if restriper was paused all the way until unmount, in
4635          * free_fs_info.  mutually_exclusive_operation_running is
4636          * cleared in __cancel_balance.
4637          */
4638         need_unlock = false;
4639
4640         ret = btrfs_balance(bctl, bargs);
4641
4642         if (arg) {
4643                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4644                         ret = -EFAULT;
4645         }
4646
4647 out_bargs:
4648         kfree(bargs);
4649 out_unlock:
4650         mutex_unlock(&fs_info->balance_mutex);
4651         mutex_unlock(&fs_info->volume_mutex);
4652         if (need_unlock)
4653                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4654 out:
4655         mnt_drop_write_file(file);
4656         return ret;
4657 }
4658
4659 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4660 {
4661         if (!capable(CAP_SYS_ADMIN))
4662                 return -EPERM;
4663
4664         switch (cmd) {
4665         case BTRFS_BALANCE_CTL_PAUSE:
4666                 return btrfs_pause_balance(root->fs_info);
4667         case BTRFS_BALANCE_CTL_CANCEL:
4668                 return btrfs_cancel_balance(root->fs_info);
4669         }
4670
4671         return -EINVAL;
4672 }
4673
4674 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4675                                          void __user *arg)
4676 {
4677         struct btrfs_fs_info *fs_info = root->fs_info;
4678         struct btrfs_ioctl_balance_args *bargs;
4679         int ret = 0;
4680
4681         if (!capable(CAP_SYS_ADMIN))
4682                 return -EPERM;
4683
4684         mutex_lock(&fs_info->balance_mutex);
4685         if (!fs_info->balance_ctl) {
4686                 ret = -ENOTCONN;
4687                 goto out;
4688         }
4689
4690         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4691         if (!bargs) {
4692                 ret = -ENOMEM;
4693                 goto out;
4694         }
4695
4696         update_ioctl_balance_args(fs_info, 1, bargs);
4697
4698         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4699                 ret = -EFAULT;
4700
4701         kfree(bargs);
4702 out:
4703         mutex_unlock(&fs_info->balance_mutex);
4704         return ret;
4705 }
4706
4707 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4708 {
4709         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4710         struct btrfs_ioctl_quota_ctl_args *sa;
4711         struct btrfs_trans_handle *trans = NULL;
4712         int ret;
4713         int err;
4714
4715         if (!capable(CAP_SYS_ADMIN))
4716                 return -EPERM;
4717
4718         ret = mnt_want_write_file(file);
4719         if (ret)
4720                 return ret;
4721
4722         sa = memdup_user(arg, sizeof(*sa));
4723         if (IS_ERR(sa)) {
4724                 ret = PTR_ERR(sa);
4725                 goto drop_write;
4726         }
4727
4728         down_write(&root->fs_info->subvol_sem);
4729         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4730         if (IS_ERR(trans)) {
4731                 ret = PTR_ERR(trans);
4732                 goto out;
4733         }
4734
4735         switch (sa->cmd) {
4736         case BTRFS_QUOTA_CTL_ENABLE:
4737                 ret = btrfs_quota_enable(trans, root->fs_info);
4738                 break;
4739         case BTRFS_QUOTA_CTL_DISABLE:
4740                 ret = btrfs_quota_disable(trans, root->fs_info);
4741                 break;
4742         default:
4743                 ret = -EINVAL;
4744                 break;
4745         }
4746
4747         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4748         if (err && !ret)
4749                 ret = err;
4750 out:
4751         kfree(sa);
4752         up_write(&root->fs_info->subvol_sem);
4753 drop_write:
4754         mnt_drop_write_file(file);
4755         return ret;
4756 }
4757
4758 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4759 {
4760         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4761         struct btrfs_ioctl_qgroup_assign_args *sa;
4762         struct btrfs_trans_handle *trans;
4763         int ret;
4764         int err;
4765
4766         if (!capable(CAP_SYS_ADMIN))
4767                 return -EPERM;
4768
4769         ret = mnt_want_write_file(file);
4770         if (ret)
4771                 return ret;
4772
4773         sa = memdup_user(arg, sizeof(*sa));
4774         if (IS_ERR(sa)) {
4775                 ret = PTR_ERR(sa);
4776                 goto drop_write;
4777         }
4778
4779         trans = btrfs_join_transaction(root);
4780         if (IS_ERR(trans)) {
4781                 ret = PTR_ERR(trans);
4782                 goto out;
4783         }
4784
4785         /* FIXME: check if the IDs really exist */
4786         if (sa->assign) {
4787                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4788                                                 sa->src, sa->dst);
4789         } else {
4790                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4791                                                 sa->src, sa->dst);
4792         }
4793
4794         /* update qgroup status and info */
4795         err = btrfs_run_qgroups(trans, root->fs_info);
4796         if (err < 0)
4797                 btrfs_error(root->fs_info, ret,
4798                             "failed to update qgroup status and info\n");
4799         err = btrfs_end_transaction(trans, root);
4800         if (err && !ret)
4801                 ret = err;
4802
4803 out:
4804         kfree(sa);
4805 drop_write:
4806         mnt_drop_write_file(file);
4807         return ret;
4808 }
4809
4810 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4811 {
4812         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4813         struct btrfs_ioctl_qgroup_create_args *sa;
4814         struct btrfs_trans_handle *trans;
4815         int ret;
4816         int err;
4817
4818         if (!capable(CAP_SYS_ADMIN))
4819                 return -EPERM;
4820
4821         ret = mnt_want_write_file(file);
4822         if (ret)
4823                 return ret;
4824
4825         sa = memdup_user(arg, sizeof(*sa));
4826         if (IS_ERR(sa)) {
4827                 ret = PTR_ERR(sa);
4828                 goto drop_write;
4829         }
4830
4831         if (!sa->qgroupid) {
4832                 ret = -EINVAL;
4833                 goto out;
4834         }
4835
4836         trans = btrfs_join_transaction(root);
4837         if (IS_ERR(trans)) {
4838                 ret = PTR_ERR(trans);
4839                 goto out;
4840         }
4841
4842         /* FIXME: check if the IDs really exist */
4843         if (sa->create) {
4844                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4845         } else {
4846                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4847         }
4848
4849         err = btrfs_end_transaction(trans, root);
4850         if (err && !ret)
4851                 ret = err;
4852
4853 out:
4854         kfree(sa);
4855 drop_write:
4856         mnt_drop_write_file(file);
4857         return ret;
4858 }
4859
4860 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4861 {
4862         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4863         struct btrfs_ioctl_qgroup_limit_args *sa;
4864         struct btrfs_trans_handle *trans;
4865         int ret;
4866         int err;
4867         u64 qgroupid;
4868
4869         if (!capable(CAP_SYS_ADMIN))
4870                 return -EPERM;
4871
4872         ret = mnt_want_write_file(file);
4873         if (ret)
4874                 return ret;
4875
4876         sa = memdup_user(arg, sizeof(*sa));
4877         if (IS_ERR(sa)) {
4878                 ret = PTR_ERR(sa);
4879                 goto drop_write;
4880         }
4881
4882         trans = btrfs_join_transaction(root);
4883         if (IS_ERR(trans)) {
4884                 ret = PTR_ERR(trans);
4885                 goto out;
4886         }
4887
4888         qgroupid = sa->qgroupid;
4889         if (!qgroupid) {
4890                 /* take the current subvol as qgroup */
4891                 qgroupid = root->root_key.objectid;
4892         }
4893
4894         /* FIXME: check if the IDs really exist */
4895         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4896
4897         err = btrfs_end_transaction(trans, root);
4898         if (err && !ret)
4899                 ret = err;
4900
4901 out:
4902         kfree(sa);
4903 drop_write:
4904         mnt_drop_write_file(file);
4905         return ret;
4906 }
4907
4908 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4909 {
4910         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4911         struct btrfs_ioctl_quota_rescan_args *qsa;
4912         int ret;
4913
4914         if (!capable(CAP_SYS_ADMIN))
4915                 return -EPERM;
4916
4917         ret = mnt_want_write_file(file);
4918         if (ret)
4919                 return ret;
4920
4921         qsa = memdup_user(arg, sizeof(*qsa));
4922         if (IS_ERR(qsa)) {
4923                 ret = PTR_ERR(qsa);
4924                 goto drop_write;
4925         }
4926
4927         if (qsa->flags) {
4928                 ret = -EINVAL;
4929                 goto out;
4930         }
4931
4932         ret = btrfs_qgroup_rescan(root->fs_info);
4933
4934 out:
4935         kfree(qsa);
4936 drop_write:
4937         mnt_drop_write_file(file);
4938         return ret;
4939 }
4940
4941 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4942 {
4943         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4944         struct btrfs_ioctl_quota_rescan_args *qsa;
4945         int ret = 0;
4946
4947         if (!capable(CAP_SYS_ADMIN))
4948                 return -EPERM;
4949
4950         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4951         if (!qsa)
4952                 return -ENOMEM;
4953
4954         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4955                 qsa->flags = 1;
4956                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4957         }
4958
4959         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4960                 ret = -EFAULT;
4961
4962         kfree(qsa);
4963         return ret;
4964 }
4965
4966 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4967 {
4968         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4969
4970         if (!capable(CAP_SYS_ADMIN))
4971                 return -EPERM;
4972
4973         return btrfs_qgroup_wait_for_completion(root->fs_info);
4974 }
4975
4976 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4977                                             struct btrfs_ioctl_received_subvol_args *sa)
4978 {
4979         struct inode *inode = file_inode(file);
4980         struct btrfs_root *root = BTRFS_I(inode)->root;
4981         struct btrfs_root_item *root_item = &root->root_item;
4982         struct btrfs_trans_handle *trans;
4983         struct timespec ct = CURRENT_TIME;
4984         int ret = 0;
4985         int received_uuid_changed;
4986
4987         if (!inode_owner_or_capable(inode))
4988                 return -EPERM;
4989
4990         ret = mnt_want_write_file(file);
4991         if (ret < 0)
4992                 return ret;
4993
4994         down_write(&root->fs_info->subvol_sem);
4995
4996         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4997                 ret = -EINVAL;
4998                 goto out;
4999         }
5000
5001         if (btrfs_root_readonly(root)) {
5002                 ret = -EROFS;
5003                 goto out;
5004         }
5005
5006         /*
5007          * 1 - root item
5008          * 2 - uuid items (received uuid + subvol uuid)
5009          */
5010         trans = btrfs_start_transaction(root, 3);
5011         if (IS_ERR(trans)) {
5012                 ret = PTR_ERR(trans);
5013                 trans = NULL;
5014                 goto out;
5015         }
5016
5017         sa->rtransid = trans->transid;
5018         sa->rtime.sec = ct.tv_sec;
5019         sa->rtime.nsec = ct.tv_nsec;
5020
5021         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5022                                        BTRFS_UUID_SIZE);
5023         if (received_uuid_changed &&
5024             !btrfs_is_empty_uuid(root_item->received_uuid))
5025                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5026                                     root_item->received_uuid,
5027                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5028                                     root->root_key.objectid);
5029         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5030         btrfs_set_root_stransid(root_item, sa->stransid);
5031         btrfs_set_root_rtransid(root_item, sa->rtransid);
5032         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5033         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5034         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5035         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5036
5037         ret = btrfs_update_root(trans, root->fs_info->tree_root,
5038                                 &root->root_key, &root->root_item);
5039         if (ret < 0) {
5040                 btrfs_end_transaction(trans, root);
5041                 goto out;
5042         }
5043         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5044                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5045                                           sa->uuid,
5046                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5047                                           root->root_key.objectid);
5048                 if (ret < 0 && ret != -EEXIST) {
5049                         btrfs_abort_transaction(trans, root, ret);
5050                         goto out;
5051                 }
5052         }
5053         ret = btrfs_commit_transaction(trans, root);
5054         if (ret < 0) {
5055                 btrfs_abort_transaction(trans, root, ret);
5056                 goto out;
5057         }
5058
5059 out:
5060         up_write(&root->fs_info->subvol_sem);
5061         mnt_drop_write_file(file);
5062         return ret;
5063 }
5064
5065 #ifdef CONFIG_64BIT
5066 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5067                                                 void __user *arg)
5068 {
5069         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5070         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5071         int ret = 0;
5072
5073         args32 = memdup_user(arg, sizeof(*args32));
5074         if (IS_ERR(args32)) {
5075                 ret = PTR_ERR(args32);
5076                 args32 = NULL;
5077                 goto out;
5078         }
5079
5080         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
5081         if (!args64) {
5082                 ret = -ENOMEM;
5083                 goto out;
5084         }
5085
5086         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5087         args64->stransid = args32->stransid;
5088         args64->rtransid = args32->rtransid;
5089         args64->stime.sec = args32->stime.sec;
5090         args64->stime.nsec = args32->stime.nsec;
5091         args64->rtime.sec = args32->rtime.sec;
5092         args64->rtime.nsec = args32->rtime.nsec;
5093         args64->flags = args32->flags;
5094
5095         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5096         if (ret)
5097                 goto out;
5098
5099         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5100         args32->stransid = args64->stransid;
5101         args32->rtransid = args64->rtransid;
5102         args32->stime.sec = args64->stime.sec;
5103         args32->stime.nsec = args64->stime.nsec;
5104         args32->rtime.sec = args64->rtime.sec;
5105         args32->rtime.nsec = args64->rtime.nsec;
5106         args32->flags = args64->flags;
5107
5108         ret = copy_to_user(arg, args32, sizeof(*args32));
5109         if (ret)
5110                 ret = -EFAULT;
5111
5112 out:
5113         kfree(args32);
5114         kfree(args64);
5115         return ret;
5116 }
5117 #endif
5118
5119 static long btrfs_ioctl_set_received_subvol(struct file *file,
5120                                             void __user *arg)
5121 {
5122         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5123         int ret = 0;
5124
5125         sa = memdup_user(arg, sizeof(*sa));
5126         if (IS_ERR(sa)) {
5127                 ret = PTR_ERR(sa);
5128                 sa = NULL;
5129                 goto out;
5130         }
5131
5132         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5133
5134         if (ret)
5135                 goto out;
5136
5137         ret = copy_to_user(arg, sa, sizeof(*sa));
5138         if (ret)
5139                 ret = -EFAULT;
5140
5141 out:
5142         kfree(sa);
5143         return ret;
5144 }
5145
5146 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5147 {
5148         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5149         size_t len;
5150         int ret;
5151         char label[BTRFS_LABEL_SIZE];
5152
5153         spin_lock(&root->fs_info->super_lock);
5154         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5155         spin_unlock(&root->fs_info->super_lock);
5156
5157         len = strnlen(label, BTRFS_LABEL_SIZE);
5158
5159         if (len == BTRFS_LABEL_SIZE) {
5160                 btrfs_warn(root->fs_info,
5161                         "label is too long, return the first %zu bytes", --len);
5162         }
5163
5164         ret = copy_to_user(arg, label, len);
5165
5166         return ret ? -EFAULT : 0;
5167 }
5168
5169 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5170 {
5171         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5172         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5173         struct btrfs_trans_handle *trans;
5174         char label[BTRFS_LABEL_SIZE];
5175         int ret;
5176
5177         if (!capable(CAP_SYS_ADMIN))
5178                 return -EPERM;
5179
5180         if (copy_from_user(label, arg, sizeof(label)))
5181                 return -EFAULT;
5182
5183         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5184                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5185                        BTRFS_LABEL_SIZE - 1);
5186                 return -EINVAL;
5187         }
5188
5189         ret = mnt_want_write_file(file);
5190         if (ret)
5191                 return ret;
5192
5193         trans = btrfs_start_transaction(root, 0);
5194         if (IS_ERR(trans)) {
5195                 ret = PTR_ERR(trans);
5196                 goto out_unlock;
5197         }
5198
5199         spin_lock(&root->fs_info->super_lock);
5200         strcpy(super_block->label, label);
5201         spin_unlock(&root->fs_info->super_lock);
5202         ret = btrfs_commit_transaction(trans, root);
5203
5204 out_unlock:
5205         mnt_drop_write_file(file);
5206         return ret;
5207 }
5208
5209 #define INIT_FEATURE_FLAGS(suffix) \
5210         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5211           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5212           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5213
5214 static int btrfs_ioctl_get_supported_features(struct file *file,
5215                                               void __user *arg)
5216 {
5217         static struct btrfs_ioctl_feature_flags features[3] = {
5218                 INIT_FEATURE_FLAGS(SUPP),
5219                 INIT_FEATURE_FLAGS(SAFE_SET),
5220                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5221         };
5222
5223         if (copy_to_user(arg, &features, sizeof(features)))
5224                 return -EFAULT;
5225
5226         return 0;
5227 }
5228
5229 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5230 {
5231         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5232         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5233         struct btrfs_ioctl_feature_flags features;
5234
5235         features.compat_flags = btrfs_super_compat_flags(super_block);
5236         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5237         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5238
5239         if (copy_to_user(arg, &features, sizeof(features)))
5240                 return -EFAULT;
5241
5242         return 0;
5243 }
5244
5245 static int check_feature_bits(struct btrfs_root *root,
5246                               enum btrfs_feature_set set,
5247                               u64 change_mask, u64 flags, u64 supported_flags,
5248                               u64 safe_set, u64 safe_clear)
5249 {
5250         const char *type = btrfs_feature_set_names[set];
5251         char *names;
5252         u64 disallowed, unsupported;
5253         u64 set_mask = flags & change_mask;
5254         u64 clear_mask = ~flags & change_mask;
5255
5256         unsupported = set_mask & ~supported_flags;
5257         if (unsupported) {
5258                 names = btrfs_printable_features(set, unsupported);
5259                 if (names) {
5260                         btrfs_warn(root->fs_info,
5261                            "this kernel does not support the %s feature bit%s",
5262                            names, strchr(names, ',') ? "s" : "");
5263                         kfree(names);
5264                 } else
5265                         btrfs_warn(root->fs_info,
5266                            "this kernel does not support %s bits 0x%llx",
5267                            type, unsupported);
5268                 return -EOPNOTSUPP;
5269         }
5270
5271         disallowed = set_mask & ~safe_set;
5272         if (disallowed) {
5273                 names = btrfs_printable_features(set, disallowed);
5274                 if (names) {
5275                         btrfs_warn(root->fs_info,
5276                            "can't set the %s feature bit%s while mounted",
5277                            names, strchr(names, ',') ? "s" : "");
5278                         kfree(names);
5279                 } else
5280                         btrfs_warn(root->fs_info,
5281                            "can't set %s bits 0x%llx while mounted",
5282                            type, disallowed);
5283                 return -EPERM;
5284         }
5285
5286         disallowed = clear_mask & ~safe_clear;
5287         if (disallowed) {
5288                 names = btrfs_printable_features(set, disallowed);
5289                 if (names) {
5290                         btrfs_warn(root->fs_info,
5291                            "can't clear the %s feature bit%s while mounted",
5292                            names, strchr(names, ',') ? "s" : "");
5293                         kfree(names);
5294                 } else
5295                         btrfs_warn(root->fs_info,
5296                            "can't clear %s bits 0x%llx while mounted",
5297                            type, disallowed);
5298                 return -EPERM;
5299         }
5300
5301         return 0;
5302 }
5303
5304 #define check_feature(root, change_mask, flags, mask_base)      \
5305 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5306                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5307                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5308                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5309
5310 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5311 {
5312         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5313         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5314         struct btrfs_ioctl_feature_flags flags[2];
5315         struct btrfs_trans_handle *trans;
5316         u64 newflags;
5317         int ret;
5318
5319         if (!capable(CAP_SYS_ADMIN))
5320                 return -EPERM;
5321
5322         if (copy_from_user(flags, arg, sizeof(flags)))
5323                 return -EFAULT;
5324
5325         /* Nothing to do */
5326         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5327             !flags[0].incompat_flags)
5328                 return 0;
5329
5330         ret = check_feature(root, flags[0].compat_flags,
5331                             flags[1].compat_flags, COMPAT);
5332         if (ret)
5333                 return ret;
5334
5335         ret = check_feature(root, flags[0].compat_ro_flags,
5336                             flags[1].compat_ro_flags, COMPAT_RO);
5337         if (ret)
5338                 return ret;
5339
5340         ret = check_feature(root, flags[0].incompat_flags,
5341                             flags[1].incompat_flags, INCOMPAT);
5342         if (ret)
5343                 return ret;
5344
5345         trans = btrfs_start_transaction(root, 0);
5346         if (IS_ERR(trans))
5347                 return PTR_ERR(trans);
5348
5349         spin_lock(&root->fs_info->super_lock);
5350         newflags = btrfs_super_compat_flags(super_block);
5351         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5352         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5353         btrfs_set_super_compat_flags(super_block, newflags);
5354
5355         newflags = btrfs_super_compat_ro_flags(super_block);
5356         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5357         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5358         btrfs_set_super_compat_ro_flags(super_block, newflags);
5359
5360         newflags = btrfs_super_incompat_flags(super_block);
5361         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5362         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5363         btrfs_set_super_incompat_flags(super_block, newflags);
5364         spin_unlock(&root->fs_info->super_lock);
5365
5366         return btrfs_commit_transaction(trans, root);
5367 }
5368
5369 long btrfs_ioctl(struct file *file, unsigned int
5370                 cmd, unsigned long arg)
5371 {
5372         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5373         void __user *argp = (void __user *)arg;
5374
5375         switch (cmd) {
5376         case FS_IOC_GETFLAGS:
5377                 return btrfs_ioctl_getflags(file, argp);
5378         case FS_IOC_SETFLAGS:
5379                 return btrfs_ioctl_setflags(file, argp);
5380         case FS_IOC_GETVERSION:
5381                 return btrfs_ioctl_getversion(file, argp);
5382         case FITRIM:
5383                 return btrfs_ioctl_fitrim(file, argp);
5384         case BTRFS_IOC_SNAP_CREATE:
5385                 return btrfs_ioctl_snap_create(file, argp, 0);
5386         case BTRFS_IOC_SNAP_CREATE_V2:
5387                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5388         case BTRFS_IOC_SUBVOL_CREATE:
5389                 return btrfs_ioctl_snap_create(file, argp, 1);
5390         case BTRFS_IOC_SUBVOL_CREATE_V2:
5391                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5392         case BTRFS_IOC_SNAP_DESTROY:
5393                 return btrfs_ioctl_snap_destroy(file, argp);
5394         case BTRFS_IOC_SUBVOL_GETFLAGS:
5395                 return btrfs_ioctl_subvol_getflags(file, argp);
5396         case BTRFS_IOC_SUBVOL_SETFLAGS:
5397                 return btrfs_ioctl_subvol_setflags(file, argp);
5398         case BTRFS_IOC_DEFAULT_SUBVOL:
5399                 return btrfs_ioctl_default_subvol(file, argp);
5400         case BTRFS_IOC_DEFRAG:
5401                 return btrfs_ioctl_defrag(file, NULL);
5402         case BTRFS_IOC_DEFRAG_RANGE:
5403                 return btrfs_ioctl_defrag(file, argp);
5404         case BTRFS_IOC_RESIZE:
5405                 return btrfs_ioctl_resize(file, argp);
5406         case BTRFS_IOC_ADD_DEV:
5407                 return btrfs_ioctl_add_dev(root, argp);
5408         case BTRFS_IOC_RM_DEV:
5409                 return btrfs_ioctl_rm_dev(file, argp);
5410         case BTRFS_IOC_FS_INFO:
5411                 return btrfs_ioctl_fs_info(root, argp);
5412         case BTRFS_IOC_DEV_INFO:
5413                 return btrfs_ioctl_dev_info(root, argp);
5414         case BTRFS_IOC_BALANCE:
5415                 return btrfs_ioctl_balance(file, NULL);
5416         case BTRFS_IOC_CLONE:
5417                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5418         case BTRFS_IOC_CLONE_RANGE:
5419                 return btrfs_ioctl_clone_range(file, argp);
5420         case BTRFS_IOC_TRANS_START:
5421                 return btrfs_ioctl_trans_start(file);
5422         case BTRFS_IOC_TRANS_END:
5423                 return btrfs_ioctl_trans_end(file);
5424         case BTRFS_IOC_TREE_SEARCH:
5425                 return btrfs_ioctl_tree_search(file, argp);
5426         case BTRFS_IOC_TREE_SEARCH_V2:
5427                 return btrfs_ioctl_tree_search_v2(file, argp);
5428         case BTRFS_IOC_INO_LOOKUP:
5429                 return btrfs_ioctl_ino_lookup(file, argp);
5430         case BTRFS_IOC_INO_PATHS:
5431                 return btrfs_ioctl_ino_to_path(root, argp);
5432         case BTRFS_IOC_LOGICAL_INO:
5433                 return btrfs_ioctl_logical_to_ino(root, argp);
5434         case BTRFS_IOC_SPACE_INFO:
5435                 return btrfs_ioctl_space_info(root, argp);
5436         case BTRFS_IOC_SYNC: {
5437                 int ret;
5438
5439                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5440                 if (ret)
5441                         return ret;
5442                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5443                 /*
5444                  * The transaction thread may want to do more work,
5445                  * namely it pokes the cleaner ktread that will start
5446                  * processing uncleaned subvols.
5447                  */
5448                 wake_up_process(root->fs_info->transaction_kthread);
5449                 return ret;
5450         }
5451         case BTRFS_IOC_START_SYNC:
5452                 return btrfs_ioctl_start_sync(root, argp);
5453         case BTRFS_IOC_WAIT_SYNC:
5454                 return btrfs_ioctl_wait_sync(root, argp);
5455         case BTRFS_IOC_SCRUB:
5456                 return btrfs_ioctl_scrub(file, argp);
5457         case BTRFS_IOC_SCRUB_CANCEL:
5458                 return btrfs_ioctl_scrub_cancel(root, argp);
5459         case BTRFS_IOC_SCRUB_PROGRESS:
5460                 return btrfs_ioctl_scrub_progress(root, argp);
5461         case BTRFS_IOC_BALANCE_V2:
5462                 return btrfs_ioctl_balance(file, argp);
5463         case BTRFS_IOC_BALANCE_CTL:
5464                 return btrfs_ioctl_balance_ctl(root, arg);
5465         case BTRFS_IOC_BALANCE_PROGRESS:
5466                 return btrfs_ioctl_balance_progress(root, argp);
5467         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5468                 return btrfs_ioctl_set_received_subvol(file, argp);
5469 #ifdef CONFIG_64BIT
5470         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5471                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5472 #endif
5473         case BTRFS_IOC_SEND:
5474                 return btrfs_ioctl_send(file, argp);
5475         case BTRFS_IOC_GET_DEV_STATS:
5476                 return btrfs_ioctl_get_dev_stats(root, argp);
5477         case BTRFS_IOC_QUOTA_CTL:
5478                 return btrfs_ioctl_quota_ctl(file, argp);
5479         case BTRFS_IOC_QGROUP_ASSIGN:
5480                 return btrfs_ioctl_qgroup_assign(file, argp);
5481         case BTRFS_IOC_QGROUP_CREATE:
5482                 return btrfs_ioctl_qgroup_create(file, argp);
5483         case BTRFS_IOC_QGROUP_LIMIT:
5484                 return btrfs_ioctl_qgroup_limit(file, argp);
5485         case BTRFS_IOC_QUOTA_RESCAN:
5486                 return btrfs_ioctl_quota_rescan(file, argp);
5487         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5488                 return btrfs_ioctl_quota_rescan_status(file, argp);
5489         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5490                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5491         case BTRFS_IOC_DEV_REPLACE:
5492                 return btrfs_ioctl_dev_replace(root, argp);
5493         case BTRFS_IOC_GET_FSLABEL:
5494                 return btrfs_ioctl_get_fslabel(file, argp);
5495         case BTRFS_IOC_SET_FSLABEL:
5496                 return btrfs_ioctl_set_fslabel(file, argp);
5497         case BTRFS_IOC_FILE_EXTENT_SAME:
5498                 return btrfs_ioctl_file_extent_same(file, argp);
5499         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5500                 return btrfs_ioctl_get_supported_features(file, argp);
5501         case BTRFS_IOC_GET_FEATURES:
5502                 return btrfs_ioctl_get_features(file, argp);
5503         case BTRFS_IOC_SET_FEATURES:
5504                 return btrfs_ioctl_set_features(file, argp);
5505         }
5506
5507         return -ENOTTY;
5508 }