fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62
  63 #ifdef CONFIG_64BIT
  64 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  65  * structures are incorrect, as the timespec structure from userspace
  66  * is 4 bytes too small. We define these alternatives here to teach
  67  * the kernel about the 32-bit struct packing.
  68  */
  69 struct btrfs_ioctl_timespec_32 {
  70         __u64 sec;
  71         __u32 nsec;
  72 } __attribute__ ((__packed__));
  73
  74 struct btrfs_ioctl_received_subvol_args_32 {
  75         char    uuid[BTRFS_UUID_SIZE];  /* in */
  76         __u64   stransid;               /* in */
  77         __u64   rtransid;               /* out */
  78         struct btrfs_ioctl_timespec_32 stime; /* in */
  79         struct btrfs_ioctl_timespec_32 rtime; /* out */
  80         __u64   flags;                  /* in */
  81         __u64   reserved[16];           /* in */
  82 } __attribute__ ((__packed__));
  83
  84 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  85                                 struct btrfs_ioctl_received_subvol_args_32)
  86 #endif
  87
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  91
  92 /* Mask out flags that are inappropriate for the given type of inode. */
  93 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  94 {
  95         if (S_ISDIR(mode))
  96                 return flags;
  97         else if (S_ISREG(mode))
  98                 return flags & ~FS_DIRSYNC_FL;
  99         else
 100                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 101 }
 102
 103 /*
 104  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 105  */
 106 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 126                 iflags |= FS_COMPR_FL;
 127         else if (flags & BTRFS_INODE_NOCOMPRESS)
 128                 iflags |= FS_NOCOMP_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_update_iflags(struct inode *inode)
 137 {
 138         struct btrfs_inode *ip = BTRFS_I(inode);
 139         unsigned int new_fl = 0;
 140
 141         if (ip->flags & BTRFS_INODE_SYNC)
 142                 new_fl |= S_SYNC;
 143         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 144                 new_fl |= S_IMMUTABLE;
 145         if (ip->flags & BTRFS_INODE_APPEND)
 146                 new_fl |= S_APPEND;
 147         if (ip->flags & BTRFS_INODE_NOATIME)
 148                 new_fl |= S_NOATIME;
 149         if (ip->flags & BTRFS_INODE_DIRSYNC)
 150                 new_fl |= S_DIRSYNC;
 151
 152         set_mask_bits(&inode->i_flags,
 153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 154                       new_fl);
 155 }
 156
 157 /*
 158  * Inherit flags from the parent inode.
 159  *
 160  * Currently only the compression flags and the cow flags are inherited.
 161  */
 162 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 163 {
 164         unsigned int flags;
 165
 166         if (!dir)
 167                 return;
 168
 169         flags = BTRFS_I(dir)->flags;
 170
 171         if (flags & BTRFS_INODE_NOCOMPRESS) {
 172                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 173                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 174         } else if (flags & BTRFS_INODE_COMPRESS) {
 175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 177         }
 178
 179         if (flags & BTRFS_INODE_NODATACOW) {
 180                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 181                 if (S_ISREG(inode->i_mode))
 182                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 183         }
 184
 185         btrfs_update_iflags(inode);
 186 }
 187
 188 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 189 {
 190         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 191         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 192
 193         if (copy_to_user(arg, &flags, sizeof(flags)))
 194                 return -EFAULT;
 195         return 0;
 196 }
 197
 198 static int check_flags(unsigned int flags)
 199 {
 200         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 201                       FS_NOATIME_FL | FS_NODUMP_FL | \
 202                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 203                       FS_NOCOMP_FL | FS_COMPR_FL |
 204                       FS_NOCOW_FL))
 205                 return -EOPNOTSUPP;
 206
 207         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 208                 return -EINVAL;
 209
 210         return 0;
 211 }
 212
 213 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 214 {
 215         struct inode *inode = file_inode(file);
 216         struct btrfs_inode *ip = BTRFS_I(inode);
 217         struct btrfs_root *root = ip->root;
 218         struct btrfs_trans_handle *trans;
 219         unsigned int flags, oldflags;
 220         int ret;
 221         u64 ip_oldflags;
 222         unsigned int i_oldflags;
 223         umode_t mode;
 224
 225         if (!inode_owner_or_capable(inode))
 226                 return -EPERM;
 227
 228         if (btrfs_root_readonly(root))
 229                 return -EROFS;
 230
 231         if (copy_from_user(&flags, arg, sizeof(flags)))
 232                 return -EFAULT;
 233
 234         ret = check_flags(flags);
 235         if (ret)
 236                 return ret;
 237
 238         ret = mnt_want_write_file(file);
 239         if (ret)
 240                 return ret;
 241
 242         mutex_lock(&inode->i_mutex);
 243
 244         ip_oldflags = ip->flags;
 245         i_oldflags = inode->i_flags;
 246         mode = inode->i_mode;
 247
 248         flags = btrfs_mask_flags(inode->i_mode, flags);
 249         oldflags = btrfs_flags_to_ioctl(ip->flags);
 250         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 251                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 252                         ret = -EPERM;
 253                         goto out_unlock;
 254                 }
 255         }
 256
 257         if (flags & FS_SYNC_FL)
 258                 ip->flags |= BTRFS_INODE_SYNC;
 259         else
 260                 ip->flags &= ~BTRFS_INODE_SYNC;
 261         if (flags & FS_IMMUTABLE_FL)
 262                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 263         else
 264                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 265         if (flags & FS_APPEND_FL)
 266                 ip->flags |= BTRFS_INODE_APPEND;
 267         else
 268                 ip->flags &= ~BTRFS_INODE_APPEND;
 269         if (flags & FS_NODUMP_FL)
 270                 ip->flags |= BTRFS_INODE_NODUMP;
 271         else
 272                 ip->flags &= ~BTRFS_INODE_NODUMP;
 273         if (flags & FS_NOATIME_FL)
 274                 ip->flags |= BTRFS_INODE_NOATIME;
 275         else
 276                 ip->flags &= ~BTRFS_INODE_NOATIME;
 277         if (flags & FS_DIRSYNC_FL)
 278                 ip->flags |= BTRFS_INODE_DIRSYNC;
 279         else
 280                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 281         if (flags & FS_NOCOW_FL) {
 282                 if (S_ISREG(mode)) {
 283                         /*
 284                          * It's safe to turn csums off here, no extents exist.
 285                          * Otherwise we want the flag to reflect the real COW
 286                          * status of the file and will not set it.
 287                          */
 288                         if (inode->i_size == 0)
 289                                 ip->flags |= BTRFS_INODE_NODATACOW
 290                                            | BTRFS_INODE_NODATASUM;
 291                 } else {
 292                         ip->flags |= BTRFS_INODE_NODATACOW;
 293                 }
 294         } else {
 295                 /*
 296                  * Revert back under same assuptions as above
 297                  */
 298                 if (S_ISREG(mode)) {
 299                         if (inode->i_size == 0)
 300                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 301                                              | BTRFS_INODE_NODATASUM);
 302                 } else {
 303                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 304                 }
 305         }
 306
 307         /*
 308          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 309          * flag may be changed automatically if compression code won't make
 310          * things smaller.
 311          */
 312         if (flags & FS_NOCOMP_FL) {
 313                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 314                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 315
 316                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 317                 if (ret && ret != -ENODATA)
 318                         goto out_drop;
 319         } else if (flags & FS_COMPR_FL) {
 320                 const char *comp;
 321
 322                 ip->flags |= BTRFS_INODE_COMPRESS;
 323                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 324
 325                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 326                         comp = "lzo";
 327                 else
 328                         comp = "zlib";
 329                 ret = btrfs_set_prop(inode, "btrfs.compression",
 330                                      comp, strlen(comp), 0);
 331                 if (ret)
 332                         goto out_drop;
 333
 334         } else {
 335                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 336                 if (ret && ret != -ENODATA)
 337                         goto out_drop;
 338                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 339         }
 340
 341         trans = btrfs_start_transaction(root, 1);
 342         if (IS_ERR(trans)) {
 343                 ret = PTR_ERR(trans);
 344                 goto out_drop;
 345         }
 346
 347         btrfs_update_iflags(inode);
 348         inode_inc_iversion(inode);
 349         inode->i_ctime = CURRENT_TIME;
 350         ret = btrfs_update_inode(trans, root, inode);
 351
 352         btrfs_end_transaction(trans, root);
 353  out_drop:
 354         if (ret) {
 355                 ip->flags = ip_oldflags;
 356                 inode->i_flags = i_oldflags;
 357         }
 358
 359  out_unlock:
 360         mutex_unlock(&inode->i_mutex);
 361         mnt_drop_write_file(file);
 362         return ret;
 363 }
 364
 365 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 366 {
 367         struct inode *inode = file_inode(file);
 368
 369         return put_user(inode->i_generation, arg);
 370 }
 371
 372 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 373 {
 374         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 375         struct btrfs_device *device;
 376         struct request_queue *q;
 377         struct fstrim_range range;
 378         u64 minlen = ULLONG_MAX;
 379         u64 num_devices = 0;
 380         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 381         int ret;
 382
 383         if (!capable(CAP_SYS_ADMIN))
 384                 return -EPERM;
 385
 386         rcu_read_lock();
 387         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 388                                 dev_list) {
 389                 if (!device->bdev)
 390                         continue;
 391                 q = bdev_get_queue(device->bdev);
 392                 if (blk_queue_discard(q)) {
 393                         num_devices++;
 394                         minlen = min((u64)q->limits.discard_granularity,
 395                                      minlen);
 396                 }
 397         }
 398         rcu_read_unlock();
 399
 400         if (!num_devices)
 401                 return -EOPNOTSUPP;
 402         if (copy_from_user(&range, arg, sizeof(range)))
 403                 return -EFAULT;
 404         if (range.start > total_bytes ||
 405             range.len < fs_info->sb->s_blocksize)
 406                 return -EINVAL;
 407
 408         range.len = min(range.len, total_bytes - range.start);
 409         range.minlen = max(range.minlen, minlen);
 410         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 411         if (ret < 0)
 412                 return ret;
 413
 414         if (copy_to_user(arg, &range, sizeof(range)))
 415                 return -EFAULT;
 416
 417         return 0;
 418 }
 419
 420 int btrfs_is_empty_uuid(u8 *uuid)
 421 {
 422         int i;
 423
 424         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 425                 if (uuid[i])
 426                         return 0;
 427         }
 428         return 1;
 429 }
 430
 431 static noinline int create_subvol(struct inode *dir,
 432                                   struct dentry *dentry,
 433                                   char *name, int namelen,
 434                                   u64 *async_transid,
 435                                   struct btrfs_qgroup_inherit *inherit)
 436 {
 437         struct btrfs_trans_handle *trans;
 438         struct btrfs_key key;
 439         struct btrfs_root_item root_item;
 440         struct btrfs_inode_item *inode_item;
 441         struct extent_buffer *leaf;
 442         struct btrfs_root *root = BTRFS_I(dir)->root;
 443         struct btrfs_root *new_root;
 444         struct btrfs_block_rsv block_rsv;
 445         struct timespec cur_time = CURRENT_TIME;
 446         struct inode *inode;
 447         int ret;
 448         int err;
 449         u64 objectid;
 450         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 451         u64 index = 0;
 452         u64 qgroup_reserved;
 453         uuid_le new_uuid;
 454
 455         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 456         if (ret)
 457                 return ret;
 458
 459         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 460         /*
 461          * The same as the snapshot creation, please see the comment
 462          * of create_snapshot().
 463          */
 464         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 465                                                8, &qgroup_reserved, false);
 466         if (ret)
 467                 return ret;
 468
 469         trans = btrfs_start_transaction(root, 0);
 470         if (IS_ERR(trans)) {
 471                 ret = PTR_ERR(trans);
 472                 btrfs_subvolume_release_metadata(root, &block_rsv,
 473                                                  qgroup_reserved);
 474                 return ret;
 475         }
 476         trans->block_rsv = &block_rsv;
 477         trans->bytes_reserved = block_rsv.size;
 478
 479         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 480         if (ret)
 481                 goto fail;
 482
 483         leaf = btrfs_alloc_free_block(trans, root, root->nodesize,
 484                                       0, objectid, NULL, 0, 0, 0);
 485         if (IS_ERR(leaf)) {
 486                 ret = PTR_ERR(leaf);
 487                 goto fail;
 488         }
 489
 490         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 491         btrfs_set_header_bytenr(leaf, leaf->start);
 492         btrfs_set_header_generation(leaf, trans->transid);
 493         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 494         btrfs_set_header_owner(leaf, objectid);
 495
 496         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 497                             BTRFS_FSID_SIZE);
 498         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 499                             btrfs_header_chunk_tree_uuid(leaf),
 500                             BTRFS_UUID_SIZE);
 501         btrfs_mark_buffer_dirty(leaf);
 502
 503         memset(&root_item, 0, sizeof(root_item));
 504
 505         inode_item = &root_item.inode;
 506         btrfs_set_stack_inode_generation(inode_item, 1);
 507         btrfs_set_stack_inode_size(inode_item, 3);
 508         btrfs_set_stack_inode_nlink(inode_item, 1);
 509         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 510         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 511
 512         btrfs_set_root_flags(&root_item, 0);
 513         btrfs_set_root_limit(&root_item, 0);
 514         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 515
 516         btrfs_set_root_bytenr(&root_item, leaf->start);
 517         btrfs_set_root_generation(&root_item, trans->transid);
 518         btrfs_set_root_level(&root_item, 0);
 519         btrfs_set_root_refs(&root_item, 1);
 520         btrfs_set_root_used(&root_item, leaf->len);
 521         btrfs_set_root_last_snapshot(&root_item, 0);
 522
 523         btrfs_set_root_generation_v2(&root_item,
 524                         btrfs_root_generation(&root_item));
 525         uuid_le_gen(&new_uuid);
 526         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 527         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 528         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 529         root_item.ctime = root_item.otime;
 530         btrfs_set_root_ctransid(&root_item, trans->transid);
 531         btrfs_set_root_otransid(&root_item, trans->transid);
 532
 533         btrfs_tree_unlock(leaf);
 534         free_extent_buffer(leaf);
 535         leaf = NULL;
 536
 537         btrfs_set_root_dirid(&root_item, new_dirid);
 538
 539         key.objectid = objectid;
 540         key.offset = 0;
 541         key.type = BTRFS_ROOT_ITEM_KEY;
 542         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 543                                 &root_item);
 544         if (ret)
 545                 goto fail;
 546
 547         key.offset = (u64)-1;
 548         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 549         if (IS_ERR(new_root)) {
 550                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 551                 ret = PTR_ERR(new_root);
 552                 goto fail;
 553         }
 554
 555         btrfs_record_root_in_trans(trans, new_root);
 556
 557         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 558         if (ret) {
 559                 /* We potentially lose an unused inode item here */
 560                 btrfs_abort_transaction(trans, root, ret);
 561                 goto fail;
 562         }
 563
 564         /*
 565          * insert the directory item
 566          */
 567         ret = btrfs_set_inode_index(dir, &index);
 568         if (ret) {
 569                 btrfs_abort_transaction(trans, root, ret);
 570                 goto fail;
 571         }
 572
 573         ret = btrfs_insert_dir_item(trans, root,
 574                                     name, namelen, dir, &key,
 575                                     BTRFS_FT_DIR, index);
 576         if (ret) {
 577                 btrfs_abort_transaction(trans, root, ret);
 578                 goto fail;
 579         }
 580
 581         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 582         ret = btrfs_update_inode(trans, root, dir);
 583         BUG_ON(ret);
 584
 585         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 586                                  objectid, root->root_key.objectid,
 587                                  btrfs_ino(dir), index, name, namelen);
 588         BUG_ON(ret);
 589
 590         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 591                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 592                                   objectid);
 593         if (ret)
 594                 btrfs_abort_transaction(trans, root, ret);
 595
 596 fail:
 597         trans->block_rsv = NULL;
 598         trans->bytes_reserved = 0;
 599         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 600
 601         if (async_transid) {
 602                 *async_transid = trans->transid;
 603                 err = btrfs_commit_transaction_async(trans, root, 1);
 604                 if (err)
 605                         err = btrfs_commit_transaction(trans, root);
 606         } else {
 607                 err = btrfs_commit_transaction(trans, root);
 608         }
 609         if (err && !ret)
 610                 ret = err;
 611
 612         if (!ret) {
 613                 inode = btrfs_lookup_dentry(dir, dentry);
 614                 if (IS_ERR(inode))
 615                         return PTR_ERR(inode);
 616                 d_instantiate(dentry, inode);
 617         }
 618         return ret;
 619 }
 620
 621 static void btrfs_wait_nocow_write(struct btrfs_root *root)
 622 {
 623         s64 writers;
 624         DEFINE_WAIT(wait);
 625
 626         do {
 627                 prepare_to_wait(&root->subv_writers->wait, &wait,
 628                                 TASK_UNINTERRUPTIBLE);
 629
 630                 writers = percpu_counter_sum(&root->subv_writers->counter);
 631                 if (writers)
 632                         schedule();
 633
 634                 finish_wait(&root->subv_writers->wait, &wait);
 635         } while (writers);
 636 }
 637
 638 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 639                            struct dentry *dentry, char *name, int namelen,
 640                            u64 *async_transid, bool readonly,
 641                            struct btrfs_qgroup_inherit *inherit)
 642 {
 643         struct inode *inode;
 644         struct btrfs_pending_snapshot *pending_snapshot;
 645         struct btrfs_trans_handle *trans;
 646         int ret;
 647
 648         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 649                 return -EINVAL;
 650
 651         atomic_inc(&root->will_be_snapshoted);
 652         smp_mb__after_atomic();
 653         btrfs_wait_nocow_write(root);
 654
 655         ret = btrfs_start_delalloc_inodes(root, 0);
 656         if (ret)
 657                 goto out;
 658
 659         btrfs_wait_ordered_extents(root, -1);
 660
 661         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 662         if (!pending_snapshot) {
 663                 ret = -ENOMEM;
 664                 goto out;
 665         }
 666
 667         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 668                              BTRFS_BLOCK_RSV_TEMP);
 669         /*
 670          * 1 - parent dir inode
 671          * 2 - dir entries
 672          * 1 - root item
 673          * 2 - root ref/backref
 674          * 1 - root of snapshot
 675          * 1 - UUID item
 676          */
 677         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 678                                         &pending_snapshot->block_rsv, 8,
 679                                         &pending_snapshot->qgroup_reserved,
 680                                         false);
 681         if (ret)
 682                 goto free;
 683
 684         pending_snapshot->dentry = dentry;
 685         pending_snapshot->root = root;
 686         pending_snapshot->readonly = readonly;
 687         pending_snapshot->dir = dir;
 688         pending_snapshot->inherit = inherit;
 689
 690         trans = btrfs_start_transaction(root, 0);
 691         if (IS_ERR(trans)) {
 692                 ret = PTR_ERR(trans);
 693                 goto fail;
 694         }
 695
 696         spin_lock(&root->fs_info->trans_lock);
 697         list_add(&pending_snapshot->list,
 698                  &trans->transaction->pending_snapshots);
 699         spin_unlock(&root->fs_info->trans_lock);
 700         if (async_transid) {
 701                 *async_transid = trans->transid;
 702                 ret = btrfs_commit_transaction_async(trans,
 703                                      root->fs_info->extent_root, 1);
 704                 if (ret)
 705                         ret = btrfs_commit_transaction(trans, root);
 706         } else {
 707                 ret = btrfs_commit_transaction(trans,
 708                                                root->fs_info->extent_root);
 709         }
 710         if (ret)
 711                 goto fail;
 712
 713         ret = pending_snapshot->error;
 714         if (ret)
 715                 goto fail;
 716
 717         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 718         if (IS_ERR(inode)) {
 719                 ret = PTR_ERR(inode);
 720                 goto fail;
 721         }
 722
 723         d_instantiate(dentry, inode);
 724         ret = 0;
 725 fail:
 726         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 727                                          &pending_snapshot->block_rsv,
 728                                          pending_snapshot->qgroup_reserved);
 729 free:
 730         kfree(pending_snapshot);
 731 out:
 732         atomic_dec(&root->will_be_snapshoted);
 733         return ret;
 734 }
 735
 736 /*  copy of check_sticky in fs/namei.c()
 737 * It's inline, so penalty for filesystems that don't use sticky bit is
 738 * minimal.
 739 */
 740 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 741 {
 742         kuid_t fsuid = current_fsuid();
 743
 744         if (!(dir->i_mode & S_ISVTX))
 745                 return 0;
 746         if (uid_eq(inode->i_uid, fsuid))
 747                 return 0;
 748         if (uid_eq(dir->i_uid, fsuid))
 749                 return 0;
 750         return !capable(CAP_FOWNER);
 751 }
 752
 753 /*  copy of may_delete in fs/namei.c()
 754  *      Check whether we can remove a link victim from directory dir, check
 755  *  whether the type of victim is right.
 756  *  1. We can't do it if dir is read-only (done in permission())
 757  *  2. We should have write and exec permissions on dir
 758  *  3. We can't remove anything from append-only dir
 759  *  4. We can't do anything with immutable dir (done in permission())
 760  *  5. If the sticky bit on dir is set we should either
 761  *      a. be owner of dir, or
 762  *      b. be owner of victim, or
 763  *      c. have CAP_FOWNER capability
 764  *  6. If the victim is append-only or immutable we can't do antyhing with
 765  *     links pointing to it.
 766  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 767  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 768  *  9. We can't remove a root or mountpoint.
 769  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 770  *     nfs_async_unlink().
 771  */
 772
 773 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 774 {
 775         int error;
 776
 777         if (!victim->d_inode)
 778                 return -ENOENT;
 779
 780         BUG_ON(victim->d_parent->d_inode != dir);
 781         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 782
 783         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 784         if (error)
 785                 return error;
 786         if (IS_APPEND(dir))
 787                 return -EPERM;
 788         if (btrfs_check_sticky(dir, victim->d_inode)||
 789                 IS_APPEND(victim->d_inode)||
 790             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 791                 return -EPERM;
 792         if (isdir) {
 793                 if (!S_ISDIR(victim->d_inode->i_mode))
 794                         return -ENOTDIR;
 795                 if (IS_ROOT(victim))
 796                         return -EBUSY;
 797         } else if (S_ISDIR(victim->d_inode->i_mode))
 798                 return -EISDIR;
 799         if (IS_DEADDIR(dir))
 800                 return -ENOENT;
 801         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 802                 return -EBUSY;
 803         return 0;
 804 }
 805
 806 /* copy of may_create in fs/namei.c() */
 807 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 808 {
 809         if (child->d_inode)
 810                 return -EEXIST;
 811         if (IS_DEADDIR(dir))
 812                 return -ENOENT;
 813         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 814 }
 815
 816 /*
 817  * Create a new subvolume below @parent.  This is largely modeled after
 818  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 819  * inside this filesystem so it's quite a bit simpler.
 820  */
 821 static noinline int btrfs_mksubvol(struct path *parent,
 822                                    char *name, int namelen,
 823                                    struct btrfs_root *snap_src,
 824                                    u64 *async_transid, bool readonly,
 825                                    struct btrfs_qgroup_inherit *inherit)
 826 {
 827         struct inode *dir  = parent->dentry->d_inode;
 828         struct dentry *dentry;
 829         int error;
 830
 831         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 832         if (error == -EINTR)
 833                 return error;
 834
 835         dentry = lookup_one_len(name, parent->dentry, namelen);
 836         error = PTR_ERR(dentry);
 837         if (IS_ERR(dentry))
 838                 goto out_unlock;
 839
 840         error = -EEXIST;
 841         if (dentry->d_inode)
 842                 goto out_dput;
 843
 844         error = btrfs_may_create(dir, dentry);
 845         if (error)
 846                 goto out_dput;
 847
 848         /*
 849          * even if this name doesn't exist, we may get hash collisions.
 850          * check for them now when we can safely fail
 851          */
 852         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 853                                                dir->i_ino, name,
 854                                                namelen);
 855         if (error)
 856                 goto out_dput;
 857
 858         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 859
 860         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 861                 goto out_up_read;
 862
 863         if (snap_src) {
 864                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 865                                         async_transid, readonly, inherit);
 866         } else {
 867                 error = create_subvol(dir, dentry, name, namelen,
 868                                       async_transid, inherit);
 869         }
 870         if (!error)
 871                 fsnotify_mkdir(dir, dentry);
 872 out_up_read:
 873         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 874 out_dput:
 875         dput(dentry);
 876 out_unlock:
 877         mutex_unlock(&dir->i_mutex);
 878         return error;
 879 }
 880
 881 /*
 882  * When we're defragging a range, we don't want to kick it off again
 883  * if it is really just waiting for delalloc to send it down.
 884  * If we find a nice big extent or delalloc range for the bytes in the
 885  * file you want to defrag, we return 0 to let you know to skip this
 886  * part of the file
 887  */
 888 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 889 {
 890         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 891         struct extent_map *em = NULL;
 892         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 893         u64 end;
 894
 895         read_lock(&em_tree->lock);
 896         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 897         read_unlock(&em_tree->lock);
 898
 899         if (em) {
 900                 end = extent_map_end(em);
 901                 free_extent_map(em);
 902                 if (end - offset > thresh)
 903                         return 0;
 904         }
 905         /* if we already have a nice delalloc here, just stop */
 906         thresh /= 2;
 907         end = count_range_bits(io_tree, &offset, offset + thresh,
 908                                thresh, EXTENT_DELALLOC, 1);
 909         if (end >= thresh)
 910                 return 0;
 911         return 1;
 912 }
 913
 914 /*
 915  * helper function to walk through a file and find extents
 916  * newer than a specific transid, and smaller than thresh.
 917  *
 918  * This is used by the defragging code to find new and small
 919  * extents
 920  */
 921 static int find_new_extents(struct btrfs_root *root,
 922                             struct inode *inode, u64 newer_than,
 923                             u64 *off, u32 thresh)
 924 {
 925         struct btrfs_path *path;
 926         struct btrfs_key min_key;
 927         struct extent_buffer *leaf;
 928         struct btrfs_file_extent_item *extent;
 929         int type;
 930         int ret;
 931         u64 ino = btrfs_ino(inode);
 932
 933         path = btrfs_alloc_path();
 934         if (!path)
 935                 return -ENOMEM;
 936
 937         min_key.objectid = ino;
 938         min_key.type = BTRFS_EXTENT_DATA_KEY;
 939         min_key.offset = *off;
 940
 941         while (1) {
 942                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 943                 if (ret != 0)
 944                         goto none;
 945 process_slot:
 946                 if (min_key.objectid != ino)
 947                         goto none;
 948                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 949                         goto none;
 950
 951                 leaf = path->nodes[0];
 952                 extent = btrfs_item_ptr(leaf, path->slots[0],
 953                                         struct btrfs_file_extent_item);
 954
 955                 type = btrfs_file_extent_type(leaf, extent);
 956                 if (type == BTRFS_FILE_EXTENT_REG &&
 957                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 958                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 959                         *off = min_key.offset;
 960                         btrfs_free_path(path);
 961                         return 0;
 962                 }
 963
 964                 path->slots[0]++;
 965                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 966                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 967                         goto process_slot;
 968                 }
 969
 970                 if (min_key.offset == (u64)-1)
 971                         goto none;
 972
 973                 min_key.offset++;
 974                 btrfs_release_path(path);
 975         }
 976 none:
 977         btrfs_free_path(path);
 978         return -ENOENT;
 979 }
 980
 981 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 982 {
 983         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 984         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 985         struct extent_map *em;
 986         u64 len = PAGE_CACHE_SIZE;
 987
 988         /*
 989          * hopefully we have this extent in the tree already, try without
 990          * the full extent lock
 991          */
 992         read_lock(&em_tree->lock);
 993         em = lookup_extent_mapping(em_tree, start, len);
 994         read_unlock(&em_tree->lock);
 995
 996         if (!em) {
 997                 struct extent_state *cached = NULL;
 998                 u64 end = start + len - 1;
 999
1000                 /* get the big lock and read metadata off disk */
1001                 lock_extent_bits(io_tree, start, end, 0, &cached);
1002                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1003                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1004
1005                 if (IS_ERR(em))
1006                         return NULL;
1007         }
1008
1009         return em;
1010 }
1011
1012 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1013 {
1014         struct extent_map *next;
1015         bool ret = true;
1016
1017         /* this is the last extent */
1018         if (em->start + em->len >= i_size_read(inode))
1019                 return false;
1020
1021         next = defrag_lookup_extent(inode, em->start + em->len);
1022         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1023                 ret = false;
1024         else if ((em->block_start + em->block_len == next->block_start) &&
1025                  (em->block_len > 128 * 1024 && next->block_len > 128 * 1024))
1026                 ret = false;
1027
1028         free_extent_map(next);
1029         return ret;
1030 }
1031
1032 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1033                                u64 *last_len, u64 *skip, u64 *defrag_end,
1034                                int compress)
1035 {
1036         struct extent_map *em;
1037         int ret = 1;
1038         bool next_mergeable = true;
1039
1040         /*
1041          * make sure that once we start defragging an extent, we keep on
1042          * defragging it
1043          */
1044         if (start < *defrag_end)
1045                 return 1;
1046
1047         *skip = 0;
1048
1049         em = defrag_lookup_extent(inode, start);
1050         if (!em)
1051                 return 0;
1052
1053         /* this will cover holes, and inline extents */
1054         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1055                 ret = 0;
1056                 goto out;
1057         }
1058
1059         next_mergeable = defrag_check_next_extent(inode, em);
1060         /*
1061          * we hit a real extent, if it is big or the next extent is not a
1062          * real extent, don't bother defragging it
1063          */
1064         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1065             (em->len >= thresh || !next_mergeable))
1066                 ret = 0;
1067 out:
1068         /*
1069          * last_len ends up being a counter of how many bytes we've defragged.
1070          * every time we choose not to defrag an extent, we reset *last_len
1071          * so that the next tiny extent will force a defrag.
1072          *
1073          * The end result of this is that tiny extents before a single big
1074          * extent will force at least part of that big extent to be defragged.
1075          */
1076         if (ret) {
1077                 *defrag_end = extent_map_end(em);
1078         } else {
1079                 *last_len = 0;
1080                 *skip = extent_map_end(em);
1081                 *defrag_end = 0;
1082         }
1083
1084         free_extent_map(em);
1085         return ret;
1086 }
1087
1088 /*
1089  * it doesn't do much good to defrag one or two pages
1090  * at a time.  This pulls in a nice chunk of pages
1091  * to COW and defrag.
1092  *
1093  * It also makes sure the delalloc code has enough
1094  * dirty data to avoid making new small extents as part
1095  * of the defrag
1096  *
1097  * It's a good idea to start RA on this range
1098  * before calling this.
1099  */
1100 static int cluster_pages_for_defrag(struct inode *inode,
1101                                     struct page **pages,
1102                                     unsigned long start_index,
1103                                     unsigned long num_pages)
1104 {
1105         unsigned long file_end;
1106         u64 isize = i_size_read(inode);
1107         u64 page_start;
1108         u64 page_end;
1109         u64 page_cnt;
1110         int ret;
1111         int i;
1112         int i_done;
1113         struct btrfs_ordered_extent *ordered;
1114         struct extent_state *cached_state = NULL;
1115         struct extent_io_tree *tree;
1116         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1117
1118         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1119         if (!isize || start_index > file_end)
1120                 return 0;
1121
1122         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1123
1124         ret = btrfs_delalloc_reserve_space(inode,
1125                                            page_cnt << PAGE_CACHE_SHIFT);
1126         if (ret)
1127                 return ret;
1128         i_done = 0;
1129         tree = &BTRFS_I(inode)->io_tree;
1130
1131         /* step one, lock all the pages */
1132         for (i = 0; i < page_cnt; i++) {
1133                 struct page *page;
1134 again:
1135                 page = find_or_create_page(inode->i_mapping,
1136                                            start_index + i, mask);
1137                 if (!page)
1138                         break;
1139
1140                 page_start = page_offset(page);
1141                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1142                 while (1) {
1143                         lock_extent_bits(tree, page_start, page_end,
1144                                          0, &cached_state);
1145                         ordered = btrfs_lookup_ordered_extent(inode,
1146                                                               page_start);
1147                         unlock_extent_cached(tree, page_start, page_end,
1148                                              &cached_state, GFP_NOFS);
1149                         if (!ordered)
1150                                 break;
1151
1152                         unlock_page(page);
1153                         btrfs_start_ordered_extent(inode, ordered, 1);
1154                         btrfs_put_ordered_extent(ordered);
1155                         lock_page(page);
1156                         /*
1157                          * we unlocked the page above, so we need check if
1158                          * it was released or not.
1159                          */
1160                         if (page->mapping != inode->i_mapping) {
1161                                 unlock_page(page);
1162                                 page_cache_release(page);
1163                                 goto again;
1164                         }
1165                 }
1166
1167                 if (!PageUptodate(page)) {
1168                         btrfs_readpage(NULL, page);
1169                         lock_page(page);
1170                         if (!PageUptodate(page)) {
1171                                 unlock_page(page);
1172                                 page_cache_release(page);
1173                                 ret = -EIO;
1174                                 break;
1175                         }
1176                 }
1177
1178                 if (page->mapping != inode->i_mapping) {
1179                         unlock_page(page);
1180                         page_cache_release(page);
1181                         goto again;
1182                 }
1183
1184                 pages[i] = page;
1185                 i_done++;
1186         }
1187         if (!i_done || ret)
1188                 goto out;
1189
1190         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1191                 goto out;
1192
1193         /*
1194          * so now we have a nice long stream of locked
1195          * and up to date pages, lets wait on them
1196          */
1197         for (i = 0; i < i_done; i++)
1198                 wait_on_page_writeback(pages[i]);
1199
1200         page_start = page_offset(pages[0]);
1201         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1202
1203         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1204                          page_start, page_end - 1, 0, &cached_state);
1205         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1206                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1207                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1208                           &cached_state, GFP_NOFS);
1209
1210         if (i_done != page_cnt) {
1211                 spin_lock(&BTRFS_I(inode)->lock);
1212                 BTRFS_I(inode)->outstanding_extents++;
1213                 spin_unlock(&BTRFS_I(inode)->lock);
1214                 btrfs_delalloc_release_space(inode,
1215                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1216         }
1217
1218
1219         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1220                           &cached_state, GFP_NOFS);
1221
1222         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1223                              page_start, page_end - 1, &cached_state,
1224                              GFP_NOFS);
1225
1226         for (i = 0; i < i_done; i++) {
1227                 clear_page_dirty_for_io(pages[i]);
1228                 ClearPageChecked(pages[i]);
1229                 set_page_extent_mapped(pages[i]);
1230                 set_page_dirty(pages[i]);
1231                 unlock_page(pages[i]);
1232                 page_cache_release(pages[i]);
1233         }
1234         return i_done;
1235 out:
1236         for (i = 0; i < i_done; i++) {
1237                 unlock_page(pages[i]);
1238                 page_cache_release(pages[i]);
1239         }
1240         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1241         return ret;
1242
1243 }
1244
1245 int btrfs_defrag_file(struct inode *inode, struct file *file,
1246                       struct btrfs_ioctl_defrag_range_args *range,
1247                       u64 newer_than, unsigned long max_to_defrag)
1248 {
1249         struct btrfs_root *root = BTRFS_I(inode)->root;
1250         struct file_ra_state *ra = NULL;
1251         unsigned long last_index;
1252         u64 isize = i_size_read(inode);
1253         u64 last_len = 0;
1254         u64 skip = 0;
1255         u64 defrag_end = 0;
1256         u64 newer_off = range->start;
1257         unsigned long i;
1258         unsigned long ra_index = 0;
1259         int ret;
1260         int defrag_count = 0;
1261         int compress_type = BTRFS_COMPRESS_ZLIB;
1262         u32 extent_thresh = range->extent_thresh;
1263         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1264         unsigned long cluster = max_cluster;
1265         u64 new_align = ~((u64)128 * 1024 - 1);
1266         struct page **pages = NULL;
1267
1268         if (isize == 0)
1269                 return 0;
1270
1271         if (range->start >= isize)
1272                 return -EINVAL;
1273
1274         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1275                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1276                         return -EINVAL;
1277                 if (range->compress_type)
1278                         compress_type = range->compress_type;
1279         }
1280
1281         if (extent_thresh == 0)
1282                 extent_thresh = 256 * 1024;
1283
1284         /*
1285          * if we were not given a file, allocate a readahead
1286          * context
1287          */
1288         if (!file) {
1289                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1290                 if (!ra)
1291                         return -ENOMEM;
1292                 file_ra_state_init(ra, inode->i_mapping);
1293         } else {
1294                 ra = &file->f_ra;
1295         }
1296
1297         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1298                         GFP_NOFS);
1299         if (!pages) {
1300                 ret = -ENOMEM;
1301                 goto out_ra;
1302         }
1303
1304         /* find the last page to defrag */
1305         if (range->start + range->len > range->start) {
1306                 last_index = min_t(u64, isize - 1,
1307                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1308         } else {
1309                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1310         }
1311
1312         if (newer_than) {
1313                 ret = find_new_extents(root, inode, newer_than,
1314                                        &newer_off, 64 * 1024);
1315                 if (!ret) {
1316                         range->start = newer_off;
1317                         /*
1318                          * we always align our defrag to help keep
1319                          * the extents in the file evenly spaced
1320                          */
1321                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1322                 } else
1323                         goto out_ra;
1324         } else {
1325                 i = range->start >> PAGE_CACHE_SHIFT;
1326         }
1327         if (!max_to_defrag)
1328                 max_to_defrag = last_index + 1;
1329
1330         /*
1331          * make writeback starts from i, so the defrag range can be
1332          * written sequentially.
1333          */
1334         if (i < inode->i_mapping->writeback_index)
1335                 inode->i_mapping->writeback_index = i;
1336
1337         while (i <= last_index && defrag_count < max_to_defrag &&
1338                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1339                 /*
1340                  * make sure we stop running if someone unmounts
1341                  * the FS
1342                  */
1343                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1344                         break;
1345
1346                 if (btrfs_defrag_cancelled(root->fs_info)) {
1347                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1348                         ret = -EAGAIN;
1349                         break;
1350                 }
1351
1352                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1353                                          extent_thresh, &last_len, &skip,
1354                                          &defrag_end, range->flags &
1355                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1356                         unsigned long next;
1357                         /*
1358                          * the should_defrag function tells us how much to skip
1359                          * bump our counter by the suggested amount
1360                          */
1361                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1362                         i = max(i + 1, next);
1363                         continue;
1364                 }
1365
1366                 if (!newer_than) {
1367                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1368                                    PAGE_CACHE_SHIFT) - i;
1369                         cluster = min(cluster, max_cluster);
1370                 } else {
1371                         cluster = max_cluster;
1372                 }
1373
1374                 if (i + cluster > ra_index) {
1375                         ra_index = max(i, ra_index);
1376                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1377                                        cluster);
1378                         ra_index += max_cluster;
1379                 }
1380
1381                 mutex_lock(&inode->i_mutex);
1382                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1383                         BTRFS_I(inode)->force_compress = compress_type;
1384                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1385                 if (ret < 0) {
1386                         mutex_unlock(&inode->i_mutex);
1387                         goto out_ra;
1388                 }
1389
1390                 defrag_count += ret;
1391                 balance_dirty_pages_ratelimited(inode->i_mapping);
1392                 mutex_unlock(&inode->i_mutex);
1393
1394                 if (newer_than) {
1395                         if (newer_off == (u64)-1)
1396                                 break;
1397
1398                         if (ret > 0)
1399                                 i += ret;
1400
1401                         newer_off = max(newer_off + 1,
1402                                         (u64)i << PAGE_CACHE_SHIFT);
1403
1404                         ret = find_new_extents(root, inode,
1405                                                newer_than, &newer_off,
1406                                                64 * 1024);
1407                         if (!ret) {
1408                                 range->start = newer_off;
1409                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1410                         } else {
1411                                 break;
1412                         }
1413                 } else {
1414                         if (ret > 0) {
1415                                 i += ret;
1416                                 last_len += ret << PAGE_CACHE_SHIFT;
1417                         } else {
1418                                 i++;
1419                                 last_len = 0;
1420                         }
1421                 }
1422         }
1423
1424         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1425                 filemap_flush(inode->i_mapping);
1426                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1427                              &BTRFS_I(inode)->runtime_flags))
1428                         filemap_flush(inode->i_mapping);
1429         }
1430
1431         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1432                 /* the filemap_flush will queue IO into the worker threads, but
1433                  * we have to make sure the IO is actually started and that
1434                  * ordered extents get created before we return
1435                  */
1436                 atomic_inc(&root->fs_info->async_submit_draining);
1437                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1438                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1439                         wait_event(root->fs_info->async_submit_wait,
1440                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1441                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1442                 }
1443                 atomic_dec(&root->fs_info->async_submit_draining);
1444         }
1445
1446         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1447                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1448         }
1449
1450         ret = defrag_count;
1451
1452 out_ra:
1453         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1454                 mutex_lock(&inode->i_mutex);
1455                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1456                 mutex_unlock(&inode->i_mutex);
1457         }
1458         if (!file)
1459                 kfree(ra);
1460         kfree(pages);
1461         return ret;
1462 }
1463
1464 static noinline int btrfs_ioctl_resize(struct file *file,
1465                                         void __user *arg)
1466 {
1467         u64 new_size;
1468         u64 old_size;
1469         u64 devid = 1;
1470         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1471         struct btrfs_ioctl_vol_args *vol_args;
1472         struct btrfs_trans_handle *trans;
1473         struct btrfs_device *device = NULL;
1474         char *sizestr;
1475         char *retptr;
1476         char *devstr = NULL;
1477         int ret = 0;
1478         int mod = 0;
1479
1480         if (!capable(CAP_SYS_ADMIN))
1481                 return -EPERM;
1482
1483         ret = mnt_want_write_file(file);
1484         if (ret)
1485                 return ret;
1486
1487         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1488                         1)) {
1489                 mnt_drop_write_file(file);
1490                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1491         }
1492
1493         mutex_lock(&root->fs_info->volume_mutex);
1494         vol_args = memdup_user(arg, sizeof(*vol_args));
1495         if (IS_ERR(vol_args)) {
1496                 ret = PTR_ERR(vol_args);
1497                 goto out;
1498         }
1499
1500         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1501
1502         sizestr = vol_args->name;
1503         devstr = strchr(sizestr, ':');
1504         if (devstr) {
1505                 sizestr = devstr + 1;
1506                 *devstr = '\0';
1507                 devstr = vol_args->name;
1508                 ret = kstrtoull(devstr, 10, &devid);
1509                 if (ret)
1510                         goto out_free;
1511                 if (!devid) {
1512                         ret = -EINVAL;
1513                         goto out_free;
1514                 }
1515                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1516         }
1517
1518         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1519         if (!device) {
1520                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1521                        devid);
1522                 ret = -ENODEV;
1523                 goto out_free;
1524         }
1525
1526         if (!device->writeable) {
1527                 btrfs_info(root->fs_info,
1528                            "resizer unable to apply on readonly device %llu",
1529                        devid);
1530                 ret = -EPERM;
1531                 goto out_free;
1532         }
1533
1534         if (!strcmp(sizestr, "max"))
1535                 new_size = device->bdev->bd_inode->i_size;
1536         else {
1537                 if (sizestr[0] == '-') {
1538                         mod = -1;
1539                         sizestr++;
1540                 } else if (sizestr[0] == '+') {
1541                         mod = 1;
1542                         sizestr++;
1543                 }
1544                 new_size = memparse(sizestr, &retptr);
1545                 if (*retptr != '\0' || new_size == 0) {
1546                         ret = -EINVAL;
1547                         goto out_free;
1548                 }
1549         }
1550
1551         if (device->is_tgtdev_for_dev_replace) {
1552                 ret = -EPERM;
1553                 goto out_free;
1554         }
1555
1556         old_size = btrfs_device_get_total_bytes(device);
1557
1558         if (mod < 0) {
1559                 if (new_size > old_size) {
1560                         ret = -EINVAL;
1561                         goto out_free;
1562                 }
1563                 new_size = old_size - new_size;
1564         } else if (mod > 0) {
1565                 if (new_size > ULLONG_MAX - old_size) {
1566                         ret = -ERANGE;
1567                         goto out_free;
1568                 }
1569                 new_size = old_size + new_size;
1570         }
1571
1572         if (new_size < 256 * 1024 * 1024) {
1573                 ret = -EINVAL;
1574                 goto out_free;
1575         }
1576         if (new_size > device->bdev->bd_inode->i_size) {
1577                 ret = -EFBIG;
1578                 goto out_free;
1579         }
1580
1581         do_div(new_size, root->sectorsize);
1582         new_size *= root->sectorsize;
1583
1584         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1585                       rcu_str_deref(device->name), new_size);
1586
1587         if (new_size > old_size) {
1588                 trans = btrfs_start_transaction(root, 0);
1589                 if (IS_ERR(trans)) {
1590                         ret = PTR_ERR(trans);
1591                         goto out_free;
1592                 }
1593                 ret = btrfs_grow_device(trans, device, new_size);
1594                 btrfs_commit_transaction(trans, root);
1595         } else if (new_size < old_size) {
1596                 ret = btrfs_shrink_device(device, new_size);
1597         } /* equal, nothing need to do */
1598
1599 out_free:
1600         kfree(vol_args);
1601 out:
1602         mutex_unlock(&root->fs_info->volume_mutex);
1603         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1604         mnt_drop_write_file(file);
1605         return ret;
1606 }
1607
1608 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1609                                 char *name, unsigned long fd, int subvol,
1610                                 u64 *transid, bool readonly,
1611                                 struct btrfs_qgroup_inherit *inherit)
1612 {
1613         int namelen;
1614         int ret = 0;
1615
1616         ret = mnt_want_write_file(file);
1617         if (ret)
1618                 goto out;
1619
1620         namelen = strlen(name);
1621         if (strchr(name, '/')) {
1622                 ret = -EINVAL;
1623                 goto out_drop_write;
1624         }
1625
1626         if (name[0] == '.' &&
1627            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1628                 ret = -EEXIST;
1629                 goto out_drop_write;
1630         }
1631
1632         if (subvol) {
1633                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1634                                      NULL, transid, readonly, inherit);
1635         } else {
1636                 struct fd src = fdget(fd);
1637                 struct inode *src_inode;
1638                 if (!src.file) {
1639                         ret = -EINVAL;
1640                         goto out_drop_write;
1641                 }
1642
1643                 src_inode = file_inode(src.file);
1644                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1645                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1646                                    "Snapshot src from another FS");
1647                         ret = -EXDEV;
1648                 } else if (!inode_owner_or_capable(src_inode)) {
1649                         /*
1650                          * Subvolume creation is not restricted, but snapshots
1651                          * are limited to own subvolumes only
1652                          */
1653                         ret = -EPERM;
1654                 } else {
1655                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1656                                              BTRFS_I(src_inode)->root,
1657                                              transid, readonly, inherit);
1658                 }
1659                 fdput(src);
1660         }
1661 out_drop_write:
1662         mnt_drop_write_file(file);
1663 out:
1664         return ret;
1665 }
1666
1667 static noinline int btrfs_ioctl_snap_create(struct file *file,
1668                                             void __user *arg, int subvol)
1669 {
1670         struct btrfs_ioctl_vol_args *vol_args;
1671         int ret;
1672
1673         vol_args = memdup_user(arg, sizeof(*vol_args));
1674         if (IS_ERR(vol_args))
1675                 return PTR_ERR(vol_args);
1676         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1677
1678         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1679                                               vol_args->fd, subvol,
1680                                               NULL, false, NULL);
1681
1682         kfree(vol_args);
1683         return ret;
1684 }
1685
1686 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1687                                                void __user *arg, int subvol)
1688 {
1689         struct btrfs_ioctl_vol_args_v2 *vol_args;
1690         int ret;
1691         u64 transid = 0;
1692         u64 *ptr = NULL;
1693         bool readonly = false;
1694         struct btrfs_qgroup_inherit *inherit = NULL;
1695
1696         vol_args = memdup_user(arg, sizeof(*vol_args));
1697         if (IS_ERR(vol_args))
1698                 return PTR_ERR(vol_args);
1699         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1700
1701         if (vol_args->flags &
1702             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1703               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1704                 ret = -EOPNOTSUPP;
1705                 goto free_args;
1706         }
1707
1708         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1709                 ptr = &transid;
1710         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1711                 readonly = true;
1712         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1713                 if (vol_args->size > PAGE_CACHE_SIZE) {
1714                         ret = -EINVAL;
1715                         goto free_args;
1716                 }
1717                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1718                 if (IS_ERR(inherit)) {
1719                         ret = PTR_ERR(inherit);
1720                         goto free_args;
1721                 }
1722         }
1723
1724         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1725                                               vol_args->fd, subvol, ptr,
1726                                               readonly, inherit);
1727         if (ret)
1728                 goto free_inherit;
1729
1730         if (ptr && copy_to_user(arg +
1731                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1732                                         transid),
1733                                 ptr, sizeof(*ptr)))
1734                 ret = -EFAULT;
1735
1736 free_inherit:
1737         kfree(inherit);
1738 free_args:
1739         kfree(vol_args);
1740         return ret;
1741 }
1742
1743 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1744                                                 void __user *arg)
1745 {
1746         struct inode *inode = file_inode(file);
1747         struct btrfs_root *root = BTRFS_I(inode)->root;
1748         int ret = 0;
1749         u64 flags = 0;
1750
1751         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1752                 return -EINVAL;
1753
1754         down_read(&root->fs_info->subvol_sem);
1755         if (btrfs_root_readonly(root))
1756                 flags |= BTRFS_SUBVOL_RDONLY;
1757         up_read(&root->fs_info->subvol_sem);
1758
1759         if (copy_to_user(arg, &flags, sizeof(flags)))
1760                 ret = -EFAULT;
1761
1762         return ret;
1763 }
1764
1765 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1766                                               void __user *arg)
1767 {
1768         struct inode *inode = file_inode(file);
1769         struct btrfs_root *root = BTRFS_I(inode)->root;
1770         struct btrfs_trans_handle *trans;
1771         u64 root_flags;
1772         u64 flags;
1773         int ret = 0;
1774
1775         if (!inode_owner_or_capable(inode))
1776                 return -EPERM;
1777
1778         ret = mnt_want_write_file(file);
1779         if (ret)
1780                 goto out;
1781
1782         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1783                 ret = -EINVAL;
1784                 goto out_drop_write;
1785         }
1786
1787         if (copy_from_user(&flags, arg, sizeof(flags))) {
1788                 ret = -EFAULT;
1789                 goto out_drop_write;
1790         }
1791
1792         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1793                 ret = -EINVAL;
1794                 goto out_drop_write;
1795         }
1796
1797         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1798                 ret = -EOPNOTSUPP;
1799                 goto out_drop_write;
1800         }
1801
1802         down_write(&root->fs_info->subvol_sem);
1803
1804         /* nothing to do */
1805         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1806                 goto out_drop_sem;
1807
1808         root_flags = btrfs_root_flags(&root->root_item);
1809         if (flags & BTRFS_SUBVOL_RDONLY) {
1810                 btrfs_set_root_flags(&root->root_item,
1811                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1812         } else {
1813                 /*
1814                  * Block RO -> RW transition if this subvolume is involved in
1815                  * send
1816                  */
1817                 spin_lock(&root->root_item_lock);
1818                 if (root->send_in_progress == 0) {
1819                         btrfs_set_root_flags(&root->root_item,
1820                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1821                         spin_unlock(&root->root_item_lock);
1822                 } else {
1823                         spin_unlock(&root->root_item_lock);
1824                         btrfs_warn(root->fs_info,
1825                         "Attempt to set subvolume %llu read-write during send",
1826                                         root->root_key.objectid);
1827                         ret = -EPERM;
1828                         goto out_drop_sem;
1829                 }
1830         }
1831
1832         trans = btrfs_start_transaction(root, 1);
1833         if (IS_ERR(trans)) {
1834                 ret = PTR_ERR(trans);
1835                 goto out_reset;
1836         }
1837
1838         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1839                                 &root->root_key, &root->root_item);
1840
1841         btrfs_commit_transaction(trans, root);
1842 out_reset:
1843         if (ret)
1844                 btrfs_set_root_flags(&root->root_item, root_flags);
1845 out_drop_sem:
1846         up_write(&root->fs_info->subvol_sem);
1847 out_drop_write:
1848         mnt_drop_write_file(file);
1849 out:
1850         return ret;
1851 }
1852
1853 /*
1854  * helper to check if the subvolume references other subvolumes
1855  */
1856 static noinline int may_destroy_subvol(struct btrfs_root *root)
1857 {
1858         struct btrfs_path *path;
1859         struct btrfs_dir_item *di;
1860         struct btrfs_key key;
1861         u64 dir_id;
1862         int ret;
1863
1864         path = btrfs_alloc_path();
1865         if (!path)
1866                 return -ENOMEM;
1867
1868         /* Make sure this root isn't set as the default subvol */
1869         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1870         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1871                                    dir_id, "default", 7, 0);
1872         if (di && !IS_ERR(di)) {
1873                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1874                 if (key.objectid == root->root_key.objectid) {
1875                         ret = -EPERM;
1876                         btrfs_err(root->fs_info, "deleting default subvolume "
1877                                   "%llu is not allowed", key.objectid);
1878                         goto out;
1879                 }
1880                 btrfs_release_path(path);
1881         }
1882
1883         key.objectid = root->root_key.objectid;
1884         key.type = BTRFS_ROOT_REF_KEY;
1885         key.offset = (u64)-1;
1886
1887         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1888                                 &key, path, 0, 0);
1889         if (ret < 0)
1890                 goto out;
1891         BUG_ON(ret == 0);
1892
1893         ret = 0;
1894         if (path->slots[0] > 0) {
1895                 path->slots[0]--;
1896                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1897                 if (key.objectid == root->root_key.objectid &&
1898                     key.type == BTRFS_ROOT_REF_KEY)
1899                         ret = -ENOTEMPTY;
1900         }
1901 out:
1902         btrfs_free_path(path);
1903         return ret;
1904 }
1905
1906 static noinline int key_in_sk(struct btrfs_key *key,
1907                               struct btrfs_ioctl_search_key *sk)
1908 {
1909         struct btrfs_key test;
1910         int ret;
1911
1912         test.objectid = sk->min_objectid;
1913         test.type = sk->min_type;
1914         test.offset = sk->min_offset;
1915
1916         ret = btrfs_comp_cpu_keys(key, &test);
1917         if (ret < 0)
1918                 return 0;
1919
1920         test.objectid = sk->max_objectid;
1921         test.type = sk->max_type;
1922         test.offset = sk->max_offset;
1923
1924         ret = btrfs_comp_cpu_keys(key, &test);
1925         if (ret > 0)
1926                 return 0;
1927         return 1;
1928 }
1929
1930 static noinline int copy_to_sk(struct btrfs_root *root,
1931                                struct btrfs_path *path,
1932                                struct btrfs_key *key,
1933                                struct btrfs_ioctl_search_key *sk,
1934                                size_t *buf_size,
1935                                char __user *ubuf,
1936                                unsigned long *sk_offset,
1937                                int *num_found)
1938 {
1939         u64 found_transid;
1940         struct extent_buffer *leaf;
1941         struct btrfs_ioctl_search_header sh;
1942         unsigned long item_off;
1943         unsigned long item_len;
1944         int nritems;
1945         int i;
1946         int slot;
1947         int ret = 0;
1948
1949         leaf = path->nodes[0];
1950         slot = path->slots[0];
1951         nritems = btrfs_header_nritems(leaf);
1952
1953         if (btrfs_header_generation(leaf) > sk->max_transid) {
1954                 i = nritems;
1955                 goto advance_key;
1956         }
1957         found_transid = btrfs_header_generation(leaf);
1958
1959         for (i = slot; i < nritems; i++) {
1960                 item_off = btrfs_item_ptr_offset(leaf, i);
1961                 item_len = btrfs_item_size_nr(leaf, i);
1962
1963                 btrfs_item_key_to_cpu(leaf, key, i);
1964                 if (!key_in_sk(key, sk))
1965                         continue;
1966
1967                 if (sizeof(sh) + item_len > *buf_size) {
1968                         if (*num_found) {
1969                                 ret = 1;
1970                                 goto out;
1971                         }
1972
1973                         /*
1974                          * return one empty item back for v1, which does not
1975                          * handle -EOVERFLOW
1976                          */
1977
1978                         *buf_size = sizeof(sh) + item_len;
1979                         item_len = 0;
1980                         ret = -EOVERFLOW;
1981                 }
1982
1983                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1984                         ret = 1;
1985                         goto out;
1986                 }
1987
1988                 sh.objectid = key->objectid;
1989                 sh.offset = key->offset;
1990                 sh.type = key->type;
1991                 sh.len = item_len;
1992                 sh.transid = found_transid;
1993
1994                 /* copy search result header */
1995                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1996                         ret = -EFAULT;
1997                         goto out;
1998                 }
1999
2000                 *sk_offset += sizeof(sh);
2001
2002                 if (item_len) {
2003                         char __user *up = ubuf + *sk_offset;
2004                         /* copy the item */
2005                         if (read_extent_buffer_to_user(leaf, up,
2006                                                        item_off, item_len)) {
2007                                 ret = -EFAULT;
2008                                 goto out;
2009                         }
2010
2011                         *sk_offset += item_len;
2012                 }
2013                 (*num_found)++;
2014
2015                 if (ret) /* -EOVERFLOW from above */
2016                         goto out;
2017
2018                 if (*num_found >= sk->nr_items) {
2019                         ret = 1;
2020                         goto out;
2021                 }
2022         }
2023 advance_key:
2024         ret = 0;
2025         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2026                 key->offset++;
2027         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2028                 key->offset = 0;
2029                 key->type++;
2030         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2031                 key->offset = 0;
2032                 key->type = 0;
2033                 key->objectid++;
2034         } else
2035                 ret = 1;
2036 out:
2037         /*
2038          *  0: all items from this leaf copied, continue with next
2039          *  1: * more items can be copied, but unused buffer is too small
2040          *     * all items were found
2041          *     Either way, it will stops the loop which iterates to the next
2042          *     leaf
2043          *  -EOVERFLOW: item was to large for buffer
2044          *  -EFAULT: could not copy extent buffer back to userspace
2045          */
2046         return ret;
2047 }
2048
2049 static noinline int search_ioctl(struct inode *inode,
2050                                  struct btrfs_ioctl_search_key *sk,
2051                                  size_t *buf_size,
2052                                  char __user *ubuf)
2053 {
2054         struct btrfs_root *root;
2055         struct btrfs_key key;
2056         struct btrfs_path *path;
2057         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2058         int ret;
2059         int num_found = 0;
2060         unsigned long sk_offset = 0;
2061
2062         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2063                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2064                 return -EOVERFLOW;
2065         }
2066
2067         path = btrfs_alloc_path();
2068         if (!path)
2069                 return -ENOMEM;
2070
2071         if (sk->tree_id == 0) {
2072                 /* search the root of the inode that was passed */
2073                 root = BTRFS_I(inode)->root;
2074         } else {
2075                 key.objectid = sk->tree_id;
2076                 key.type = BTRFS_ROOT_ITEM_KEY;
2077                 key.offset = (u64)-1;
2078                 root = btrfs_read_fs_root_no_name(info, &key);
2079                 if (IS_ERR(root)) {
2080                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2081                                sk->tree_id);
2082                         btrfs_free_path(path);
2083                         return -ENOENT;
2084                 }
2085         }
2086
2087         key.objectid = sk->min_objectid;
2088         key.type = sk->min_type;
2089         key.offset = sk->min_offset;
2090
2091         while (1) {
2092                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2093                 if (ret != 0) {
2094                         if (ret > 0)
2095                                 ret = 0;
2096                         goto err;
2097                 }
2098                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2099                                  &sk_offset, &num_found);
2100                 btrfs_release_path(path);
2101                 if (ret)
2102                         break;
2103
2104         }
2105         if (ret > 0)
2106                 ret = 0;
2107 err:
2108         sk->nr_items = num_found;
2109         btrfs_free_path(path);
2110         return ret;
2111 }
2112
2113 static noinline int btrfs_ioctl_tree_search(struct file *file,
2114                                            void __user *argp)
2115 {
2116         struct btrfs_ioctl_search_args __user *uargs;
2117         struct btrfs_ioctl_search_key sk;
2118         struct inode *inode;
2119         int ret;
2120         size_t buf_size;
2121
2122         if (!capable(CAP_SYS_ADMIN))
2123                 return -EPERM;
2124
2125         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2126
2127         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2128                 return -EFAULT;
2129
2130         buf_size = sizeof(uargs->buf);
2131
2132         inode = file_inode(file);
2133         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2134
2135         /*
2136          * In the origin implementation an overflow is handled by returning a
2137          * search header with a len of zero, so reset ret.
2138          */
2139         if (ret == -EOVERFLOW)
2140                 ret = 0;
2141
2142         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2143                 ret = -EFAULT;
2144         return ret;
2145 }
2146
2147 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2148                                                void __user *argp)
2149 {
2150         struct btrfs_ioctl_search_args_v2 __user *uarg;
2151         struct btrfs_ioctl_search_args_v2 args;
2152         struct inode *inode;
2153         int ret;
2154         size_t buf_size;
2155         const size_t buf_limit = 16 * 1024 * 1024;
2156
2157         if (!capable(CAP_SYS_ADMIN))
2158                 return -EPERM;
2159
2160         /* copy search header and buffer size */
2161         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2162         if (copy_from_user(&args, uarg, sizeof(args)))
2163                 return -EFAULT;
2164
2165         buf_size = args.buf_size;
2166
2167         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2168                 return -EOVERFLOW;
2169
2170         /* limit result size to 16MB */
2171         if (buf_size > buf_limit)
2172                 buf_size = buf_limit;
2173
2174         inode = file_inode(file);
2175         ret = search_ioctl(inode, &args.key, &buf_size,
2176                            (char *)(&uarg->buf[0]));
2177         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2178                 ret = -EFAULT;
2179         else if (ret == -EOVERFLOW &&
2180                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2181                 ret = -EFAULT;
2182
2183         return ret;
2184 }
2185
2186 /*
2187  * Search INODE_REFs to identify path name of 'dirid' directory
2188  * in a 'tree_id' tree. and sets path name to 'name'.
2189  */
2190 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2191                                 u64 tree_id, u64 dirid, char *name)
2192 {
2193         struct btrfs_root *root;
2194         struct btrfs_key key;
2195         char *ptr;
2196         int ret = -1;
2197         int slot;
2198         int len;
2199         int total_len = 0;
2200         struct btrfs_inode_ref *iref;
2201         struct extent_buffer *l;
2202         struct btrfs_path *path;
2203
2204         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2205                 name[0]='\0';
2206                 return 0;
2207         }
2208
2209         path = btrfs_alloc_path();
2210         if (!path)
2211                 return -ENOMEM;
2212
2213         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2214
2215         key.objectid = tree_id;
2216         key.type = BTRFS_ROOT_ITEM_KEY;
2217         key.offset = (u64)-1;
2218         root = btrfs_read_fs_root_no_name(info, &key);
2219         if (IS_ERR(root)) {
2220                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2221                 ret = -ENOENT;
2222                 goto out;
2223         }
2224
2225         key.objectid = dirid;
2226         key.type = BTRFS_INODE_REF_KEY;
2227         key.offset = (u64)-1;
2228
2229         while (1) {
2230                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2231                 if (ret < 0)
2232                         goto out;
2233                 else if (ret > 0) {
2234                         ret = btrfs_previous_item(root, path, dirid,
2235                                                   BTRFS_INODE_REF_KEY);
2236                         if (ret < 0)
2237                                 goto out;
2238                         else if (ret > 0) {
2239                                 ret = -ENOENT;
2240                                 goto out;
2241                         }
2242                 }
2243
2244                 l = path->nodes[0];
2245                 slot = path->slots[0];
2246                 btrfs_item_key_to_cpu(l, &key, slot);
2247
2248                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2249                 len = btrfs_inode_ref_name_len(l, iref);
2250                 ptr -= len + 1;
2251                 total_len += len + 1;
2252                 if (ptr < name) {
2253                         ret = -ENAMETOOLONG;
2254                         goto out;
2255                 }
2256
2257                 *(ptr + len) = '/';
2258                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2259
2260                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2261                         break;
2262
2263                 btrfs_release_path(path);
2264                 key.objectid = key.offset;
2265                 key.offset = (u64)-1;
2266                 dirid = key.objectid;
2267         }
2268         memmove(name, ptr, total_len);
2269         name[total_len] = '\0';
2270         ret = 0;
2271 out:
2272         btrfs_free_path(path);
2273         return ret;
2274 }
2275
2276 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2277                                            void __user *argp)
2278 {
2279          struct btrfs_ioctl_ino_lookup_args *args;
2280          struct inode *inode;
2281          int ret;
2282
2283         if (!capable(CAP_SYS_ADMIN))
2284                 return -EPERM;
2285
2286         args = memdup_user(argp, sizeof(*args));
2287         if (IS_ERR(args))
2288                 return PTR_ERR(args);
2289
2290         inode = file_inode(file);
2291
2292         if (args->treeid == 0)
2293                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2294
2295         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2296                                         args->treeid, args->objectid,
2297                                         args->name);
2298
2299         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2300                 ret = -EFAULT;
2301
2302         kfree(args);
2303         return ret;
2304 }
2305
2306 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2307                                              void __user *arg)
2308 {
2309         struct dentry *parent = file->f_path.dentry;
2310         struct dentry *dentry;
2311         struct inode *dir = parent->d_inode;
2312         struct inode *inode;
2313         struct btrfs_root *root = BTRFS_I(dir)->root;
2314         struct btrfs_root *dest = NULL;
2315         struct btrfs_ioctl_vol_args *vol_args;
2316         struct btrfs_trans_handle *trans;
2317         struct btrfs_block_rsv block_rsv;
2318         u64 root_flags;
2319         u64 qgroup_reserved;
2320         int namelen;
2321         int ret;
2322         int err = 0;
2323
2324         vol_args = memdup_user(arg, sizeof(*vol_args));
2325         if (IS_ERR(vol_args))
2326                 return PTR_ERR(vol_args);
2327
2328         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2329         namelen = strlen(vol_args->name);
2330         if (strchr(vol_args->name, '/') ||
2331             strncmp(vol_args->name, "..", namelen) == 0) {
2332                 err = -EINVAL;
2333                 goto out;
2334         }
2335
2336         err = mnt_want_write_file(file);
2337         if (err)
2338                 goto out;
2339
2340
2341         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2342         if (err == -EINTR)
2343                 goto out_drop_write;
2344         dentry = lookup_one_len(vol_args->name, parent, namelen);
2345         if (IS_ERR(dentry)) {
2346                 err = PTR_ERR(dentry);
2347                 goto out_unlock_dir;
2348         }
2349
2350         if (!dentry->d_inode) {
2351                 err = -ENOENT;
2352                 goto out_dput;
2353         }
2354
2355         inode = dentry->d_inode;
2356         dest = BTRFS_I(inode)->root;
2357         if (!capable(CAP_SYS_ADMIN)) {
2358                 /*
2359                  * Regular user.  Only allow this with a special mount
2360                  * option, when the user has write+exec access to the
2361                  * subvol root, and when rmdir(2) would have been
2362                  * allowed.
2363                  *
2364                  * Note that this is _not_ check that the subvol is
2365                  * empty or doesn't contain data that we wouldn't
2366                  * otherwise be able to delete.
2367                  *
2368                  * Users who want to delete empty subvols should try
2369                  * rmdir(2).
2370                  */
2371                 err = -EPERM;
2372                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2373                         goto out_dput;
2374
2375                 /*
2376                  * Do not allow deletion if the parent dir is the same
2377                  * as the dir to be deleted.  That means the ioctl
2378                  * must be called on the dentry referencing the root
2379                  * of the subvol, not a random directory contained
2380                  * within it.
2381                  */
2382                 err = -EINVAL;
2383                 if (root == dest)
2384                         goto out_dput;
2385
2386                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2387                 if (err)
2388                         goto out_dput;
2389         }
2390
2391         /* check if subvolume may be deleted by a user */
2392         err = btrfs_may_delete(dir, dentry, 1);
2393         if (err)
2394                 goto out_dput;
2395
2396         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2397                 err = -EINVAL;
2398                 goto out_dput;
2399         }
2400
2401         mutex_lock(&inode->i_mutex);
2402
2403         /*
2404          * Don't allow to delete a subvolume with send in progress. This is
2405          * inside the i_mutex so the error handling that has to drop the bit
2406          * again is not run concurrently.
2407          */
2408         spin_lock(&dest->root_item_lock);
2409         root_flags = btrfs_root_flags(&dest->root_item);
2410         if (dest->send_in_progress == 0) {
2411                 btrfs_set_root_flags(&dest->root_item,
2412                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2413                 spin_unlock(&dest->root_item_lock);
2414         } else {
2415                 spin_unlock(&dest->root_item_lock);
2416                 btrfs_warn(root->fs_info,
2417                         "Attempt to delete subvolume %llu during send",
2418                         dest->root_key.objectid);
2419                 err = -EPERM;
2420                 goto out_dput;
2421         }
2422
2423         err = d_invalidate(dentry);
2424         if (err)
2425                 goto out_unlock;
2426
2427         down_write(&root->fs_info->subvol_sem);
2428
2429         err = may_destroy_subvol(dest);
2430         if (err)
2431                 goto out_up_write;
2432
2433         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2434         /*
2435          * One for dir inode, two for dir entries, two for root
2436          * ref/backref.
2437          */
2438         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2439                                                5, &qgroup_reserved, true);
2440         if (err)
2441                 goto out_up_write;
2442
2443         trans = btrfs_start_transaction(root, 0);
2444         if (IS_ERR(trans)) {
2445                 err = PTR_ERR(trans);
2446                 goto out_release;
2447         }
2448         trans->block_rsv = &block_rsv;
2449         trans->bytes_reserved = block_rsv.size;
2450
2451         ret = btrfs_unlink_subvol(trans, root, dir,
2452                                 dest->root_key.objectid,
2453                                 dentry->d_name.name,
2454                                 dentry->d_name.len);
2455         if (ret) {
2456                 err = ret;
2457                 btrfs_abort_transaction(trans, root, ret);
2458                 goto out_end_trans;
2459         }
2460
2461         btrfs_record_root_in_trans(trans, dest);
2462
2463         memset(&dest->root_item.drop_progress, 0,
2464                 sizeof(dest->root_item.drop_progress));
2465         dest->root_item.drop_level = 0;
2466         btrfs_set_root_refs(&dest->root_item, 0);
2467
2468         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2469                 ret = btrfs_insert_orphan_item(trans,
2470                                         root->fs_info->tree_root,
2471                                         dest->root_key.objectid);
2472                 if (ret) {
2473                         btrfs_abort_transaction(trans, root, ret);
2474                         err = ret;
2475                         goto out_end_trans;
2476                 }
2477         }
2478
2479         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2480                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2481                                   dest->root_key.objectid);
2482         if (ret && ret != -ENOENT) {
2483                 btrfs_abort_transaction(trans, root, ret);
2484                 err = ret;
2485                 goto out_end_trans;
2486         }
2487         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2488                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2489                                           dest->root_item.received_uuid,
2490                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2491                                           dest->root_key.objectid);
2492                 if (ret && ret != -ENOENT) {
2493                         btrfs_abort_transaction(trans, root, ret);
2494                         err = ret;
2495                         goto out_end_trans;
2496                 }
2497         }
2498
2499 out_end_trans:
2500         trans->block_rsv = NULL;
2501         trans->bytes_reserved = 0;
2502         ret = btrfs_end_transaction(trans, root);
2503         if (ret && !err)
2504                 err = ret;
2505         inode->i_flags |= S_DEAD;
2506 out_release:
2507         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2508 out_up_write:
2509         up_write(&root->fs_info->subvol_sem);
2510 out_unlock:
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         mutex_unlock(&inode->i_mutex);
2519         if (!err) {
2520                 shrink_dcache_sb(root->fs_info->sb);
2521                 btrfs_invalidate_inodes(dest);
2522                 d_delete(dentry);
2523                 ASSERT(dest->send_in_progress == 0);
2524
2525                 /* the last ref */
2526                 if (dest->ino_cache_inode) {
2527                         iput(dest->ino_cache_inode);
2528                         dest->ino_cache_inode = NULL;
2529                 }
2530         }
2531 out_dput:
2532         dput(dentry);
2533 out_unlock_dir:
2534         mutex_unlock(&dir->i_mutex);
2535 out_drop_write:
2536         mnt_drop_write_file(file);
2537 out:
2538         kfree(vol_args);
2539         return err;
2540 }
2541
2542 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2543 {
2544         struct inode *inode = file_inode(file);
2545         struct btrfs_root *root = BTRFS_I(inode)->root;
2546         struct btrfs_ioctl_defrag_range_args *range;
2547         int ret;
2548
2549         ret = mnt_want_write_file(file);
2550         if (ret)
2551                 return ret;
2552
2553         if (btrfs_root_readonly(root)) {
2554                 ret = -EROFS;
2555                 goto out;
2556         }
2557
2558         switch (inode->i_mode & S_IFMT) {
2559         case S_IFDIR:
2560                 if (!capable(CAP_SYS_ADMIN)) {
2561                         ret = -EPERM;
2562                         goto out;
2563                 }
2564                 ret = btrfs_defrag_root(root);
2565                 if (ret)
2566                         goto out;
2567                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2568                 break;
2569         case S_IFREG:
2570                 if (!(file->f_mode & FMODE_WRITE)) {
2571                         ret = -EINVAL;
2572                         goto out;
2573                 }
2574
2575                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2576                 if (!range) {
2577                         ret = -ENOMEM;
2578                         goto out;
2579                 }
2580
2581                 if (argp) {
2582                         if (copy_from_user(range, argp,
2583                                            sizeof(*range))) {
2584                                 ret = -EFAULT;
2585                                 kfree(range);
2586                                 goto out;
2587                         }
2588                         /* compression requires us to start the IO */
2589                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2590                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2591                                 range->extent_thresh = (u32)-1;
2592                         }
2593                 } else {
2594                         /* the rest are all set to zero by kzalloc */
2595                         range->len = (u64)-1;
2596                 }
2597                 ret = btrfs_defrag_file(file_inode(file), file,
2598                                         range, 0, 0);
2599                 if (ret > 0)
2600                         ret = 0;
2601                 kfree(range);
2602                 break;
2603         default:
2604                 ret = -EINVAL;
2605         }
2606 out:
2607         mnt_drop_write_file(file);
2608         return ret;
2609 }
2610
2611 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2612 {
2613         struct btrfs_ioctl_vol_args *vol_args;
2614         int ret;
2615
2616         if (!capable(CAP_SYS_ADMIN))
2617                 return -EPERM;
2618
2619         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2620                         1)) {
2621                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2622         }
2623
2624         mutex_lock(&root->fs_info->volume_mutex);
2625         vol_args = memdup_user(arg, sizeof(*vol_args));
2626         if (IS_ERR(vol_args)) {
2627                 ret = PTR_ERR(vol_args);
2628                 goto out;
2629         }
2630
2631         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2632         ret = btrfs_init_new_device(root, vol_args->name);
2633
2634         if (!ret)
2635                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2636
2637         kfree(vol_args);
2638 out:
2639         mutex_unlock(&root->fs_info->volume_mutex);
2640         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2641         return ret;
2642 }
2643
2644 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2645 {
2646         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2647         struct btrfs_ioctl_vol_args *vol_args;
2648         int ret;
2649
2650         if (!capable(CAP_SYS_ADMIN))
2651                 return -EPERM;
2652
2653         ret = mnt_want_write_file(file);
2654         if (ret)
2655                 return ret;
2656
2657         vol_args = memdup_user(arg, sizeof(*vol_args));
2658         if (IS_ERR(vol_args)) {
2659                 ret = PTR_ERR(vol_args);
2660                 goto err_drop;
2661         }
2662
2663         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2664
2665         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2666                         1)) {
2667                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2668                 goto out;
2669         }
2670
2671         mutex_lock(&root->fs_info->volume_mutex);
2672         ret = btrfs_rm_device(root, vol_args->name);
2673         mutex_unlock(&root->fs_info->volume_mutex);
2674         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2675
2676         if (!ret)
2677                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2678
2679 out:
2680         kfree(vol_args);
2681 err_drop:
2682         mnt_drop_write_file(file);
2683         return ret;
2684 }
2685
2686 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2687 {
2688         struct btrfs_ioctl_fs_info_args *fi_args;
2689         struct btrfs_device *device;
2690         struct btrfs_device *next;
2691         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2692         int ret = 0;
2693
2694         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2695         if (!fi_args)
2696                 return -ENOMEM;
2697
2698         mutex_lock(&fs_devices->device_list_mutex);
2699         fi_args->num_devices = fs_devices->num_devices;
2700         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2701
2702         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2703                 if (device->devid > fi_args->max_id)
2704                         fi_args->max_id = device->devid;
2705         }
2706         mutex_unlock(&fs_devices->device_list_mutex);
2707
2708         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2709         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2710         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2711
2712         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2713                 ret = -EFAULT;
2714
2715         kfree(fi_args);
2716         return ret;
2717 }
2718
2719 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2720 {
2721         struct btrfs_ioctl_dev_info_args *di_args;
2722         struct btrfs_device *dev;
2723         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2724         int ret = 0;
2725         char *s_uuid = NULL;
2726
2727         di_args = memdup_user(arg, sizeof(*di_args));
2728         if (IS_ERR(di_args))
2729                 return PTR_ERR(di_args);
2730
2731         if (!btrfs_is_empty_uuid(di_args->uuid))
2732                 s_uuid = di_args->uuid;
2733
2734         mutex_lock(&fs_devices->device_list_mutex);
2735         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2736
2737         if (!dev) {
2738                 ret = -ENODEV;
2739                 goto out;
2740         }
2741
2742         di_args->devid = dev->devid;
2743         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2744         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2745         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2746         if (dev->name) {
2747                 struct rcu_string *name;
2748
2749                 rcu_read_lock();
2750                 name = rcu_dereference(dev->name);
2751                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2752                 rcu_read_unlock();
2753                 di_args->path[sizeof(di_args->path) - 1] = 0;
2754         } else {
2755                 di_args->path[0] = '\0';
2756         }
2757
2758 out:
2759         mutex_unlock(&fs_devices->device_list_mutex);
2760         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2761                 ret = -EFAULT;
2762
2763         kfree(di_args);
2764         return ret;
2765 }
2766
2767 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2768 {
2769         struct page *page;
2770         pgoff_t index;
2771         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2772
2773         index = off >> PAGE_CACHE_SHIFT;
2774
2775         page = grab_cache_page(inode->i_mapping, index);
2776         if (!page)
2777                 return NULL;
2778
2779         if (!PageUptodate(page)) {
2780                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2781                                                  0))
2782                         return NULL;
2783                 lock_page(page);
2784                 if (!PageUptodate(page)) {
2785                         unlock_page(page);
2786                         page_cache_release(page);
2787                         return NULL;
2788                 }
2789         }
2790         unlock_page(page);
2791
2792         return page;
2793 }
2794
2795 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2796 {
2797         /* do any pending delalloc/csum calc on src, one way or
2798            another, and lock file content */
2799         while (1) {
2800                 struct btrfs_ordered_extent *ordered;
2801                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2802                 ordered = btrfs_lookup_first_ordered_extent(inode,
2803                                                             off + len - 1);
2804                 if ((!ordered ||
2805                      ordered->file_offset + ordered->len <= off ||
2806                      ordered->file_offset >= off + len) &&
2807                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2808                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2809                         if (ordered)
2810                                 btrfs_put_ordered_extent(ordered);
2811                         break;
2812                 }
2813                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2814                 if (ordered)
2815                         btrfs_put_ordered_extent(ordered);
2816                 btrfs_wait_ordered_range(inode, off, len);
2817         }
2818 }
2819
2820 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2821                                 struct inode *inode2, u64 loff2, u64 len)
2822 {
2823         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2824         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2825
2826         mutex_unlock(&inode1->i_mutex);
2827         mutex_unlock(&inode2->i_mutex);
2828 }
2829
2830 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2831                               struct inode *inode2, u64 loff2, u64 len)
2832 {
2833         if (inode1 < inode2) {
2834                 swap(inode1, inode2);
2835                 swap(loff1, loff2);
2836         }
2837
2838         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2839         lock_extent_range(inode1, loff1, len);
2840         if (inode1 != inode2) {
2841                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2842                 lock_extent_range(inode2, loff2, len);
2843         }
2844 }
2845
2846 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2847                           u64 dst_loff, u64 len)
2848 {
2849         int ret = 0;
2850         struct page *src_page, *dst_page;
2851         unsigned int cmp_len = PAGE_CACHE_SIZE;
2852         void *addr, *dst_addr;
2853
2854         while (len) {
2855                 if (len < PAGE_CACHE_SIZE)
2856                         cmp_len = len;
2857
2858                 src_page = extent_same_get_page(src, loff);
2859                 if (!src_page)
2860                         return -EINVAL;
2861                 dst_page = extent_same_get_page(dst, dst_loff);
2862                 if (!dst_page) {
2863                         page_cache_release(src_page);
2864                         return -EINVAL;
2865                 }
2866                 addr = kmap_atomic(src_page);
2867                 dst_addr = kmap_atomic(dst_page);
2868
2869                 flush_dcache_page(src_page);
2870                 flush_dcache_page(dst_page);
2871
2872                 if (memcmp(addr, dst_addr, cmp_len))
2873                         ret = BTRFS_SAME_DATA_DIFFERS;
2874
2875                 kunmap_atomic(addr);
2876                 kunmap_atomic(dst_addr);
2877                 page_cache_release(src_page);
2878                 page_cache_release(dst_page);
2879
2880                 if (ret)
2881                         break;
2882
2883                 loff += cmp_len;
2884                 dst_loff += cmp_len;
2885                 len -= cmp_len;
2886         }
2887
2888         return ret;
2889 }
2890
2891 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2892 {
2893         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2894
2895         if (off + len > inode->i_size || off + len < off)
2896                 return -EINVAL;
2897         /* Check that we are block aligned - btrfs_clone() requires this */
2898         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2899                 return -EINVAL;
2900
2901         return 0;
2902 }
2903
2904 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2905                              struct inode *dst, u64 dst_loff)
2906 {
2907         int ret;
2908
2909         /*
2910          * btrfs_clone() can't handle extents in the same file
2911          * yet. Once that works, we can drop this check and replace it
2912          * with a check for the same inode, but overlapping extents.
2913          */
2914         if (src == dst)
2915                 return -EINVAL;
2916
2917         btrfs_double_lock(src, loff, dst, dst_loff, len);
2918
2919         ret = extent_same_check_offsets(src, loff, len);
2920         if (ret)
2921                 goto out_unlock;
2922
2923         ret = extent_same_check_offsets(dst, dst_loff, len);
2924         if (ret)
2925                 goto out_unlock;
2926
2927         /* don't make the dst file partly checksummed */
2928         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2929             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2930                 ret = -EINVAL;
2931                 goto out_unlock;
2932         }
2933
2934         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2935         if (ret == 0)
2936                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2937
2938 out_unlock:
2939         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2940
2941         return ret;
2942 }
2943
2944 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2945
2946 static long btrfs_ioctl_file_extent_same(struct file *file,
2947                         struct btrfs_ioctl_same_args __user *argp)
2948 {
2949         struct btrfs_ioctl_same_args *same;
2950         struct btrfs_ioctl_same_extent_info *info;
2951         struct inode *src = file_inode(file);
2952         u64 off;
2953         u64 len;
2954         int i;
2955         int ret;
2956         unsigned long size;
2957         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2958         bool is_admin = capable(CAP_SYS_ADMIN);
2959         u16 count;
2960
2961         if (!(file->f_mode & FMODE_READ))
2962                 return -EINVAL;
2963
2964         ret = mnt_want_write_file(file);
2965         if (ret)
2966                 return ret;
2967
2968         if (get_user(count, &argp->dest_count)) {
2969                 ret = -EFAULT;
2970                 goto out;
2971         }
2972
2973         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2974
2975         same = memdup_user(argp, size);
2976
2977         if (IS_ERR(same)) {
2978                 ret = PTR_ERR(same);
2979                 goto out;
2980         }
2981
2982         off = same->logical_offset;
2983         len = same->length;
2984
2985         /*
2986          * Limit the total length we will dedupe for each operation.
2987          * This is intended to bound the total time spent in this
2988          * ioctl to something sane.
2989          */
2990         if (len > BTRFS_MAX_DEDUPE_LEN)
2991                 len = BTRFS_MAX_DEDUPE_LEN;
2992
2993         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2994                 /*
2995                  * Btrfs does not support blocksize < page_size. As a
2996                  * result, btrfs_cmp_data() won't correctly handle
2997                  * this situation without an update.
2998                  */
2999                 ret = -EINVAL;
3000                 goto out;
3001         }
3002
3003         ret = -EISDIR;
3004         if (S_ISDIR(src->i_mode))
3005                 goto out;
3006
3007         ret = -EACCES;
3008         if (!S_ISREG(src->i_mode))
3009                 goto out;
3010
3011         /* pre-format output fields to sane values */
3012         for (i = 0; i < count; i++) {
3013                 same->info[i].bytes_deduped = 0ULL;
3014                 same->info[i].status = 0;
3015         }
3016
3017         for (i = 0, info = same->info; i < count; i++, info++) {
3018                 struct inode *dst;
3019                 struct fd dst_file = fdget(info->fd);
3020                 if (!dst_file.file) {
3021                         info->status = -EBADF;
3022                         continue;
3023                 }
3024                 dst = file_inode(dst_file.file);
3025
3026                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3027                         info->status = -EINVAL;
3028                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3029                         info->status = -EXDEV;
3030                 } else if (S_ISDIR(dst->i_mode)) {
3031                         info->status = -EISDIR;
3032                 } else if (!S_ISREG(dst->i_mode)) {
3033                         info->status = -EACCES;
3034                 } else {
3035                         info->status = btrfs_extent_same(src, off, len, dst,
3036                                                         info->logical_offset);
3037                         if (info->status == 0)
3038                                 info->bytes_deduped += len;
3039                 }
3040                 fdput(dst_file);
3041         }
3042
3043         ret = copy_to_user(argp, same, size);
3044         if (ret)
3045                 ret = -EFAULT;
3046
3047 out:
3048         mnt_drop_write_file(file);
3049         return ret;
3050 }
3051
3052 /* Helper to check and see if this root currently has a ref on the given disk
3053  * bytenr.  If it does then we need to update the quota for this root.  This
3054  * doesn't do anything if quotas aren't enabled.
3055  */
3056 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3057                      u64 disko)
3058 {
3059         struct seq_list tree_mod_seq_elem = {};
3060         struct ulist *roots;
3061         struct ulist_iterator uiter;
3062         struct ulist_node *root_node = NULL;
3063         int ret;
3064
3065         if (!root->fs_info->quota_enabled)
3066                 return 1;
3067
3068         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3069         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3070                                    tree_mod_seq_elem.seq, &roots);
3071         if (ret < 0)
3072                 goto out;
3073         ret = 0;
3074         ULIST_ITER_INIT(&uiter);
3075         while ((root_node = ulist_next(roots, &uiter))) {
3076                 if (root_node->val == root->objectid) {
3077                         ret = 1;
3078                         break;
3079                 }
3080         }
3081         ulist_free(roots);
3082 out:
3083         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3084         return ret;
3085 }
3086
3087 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3088                                      struct inode *inode,
3089                                      u64 endoff,
3090                                      const u64 destoff,
3091                                      const u64 olen)
3092 {
3093         struct btrfs_root *root = BTRFS_I(inode)->root;
3094         int ret;
3095
3096         inode_inc_iversion(inode);
3097         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3098         /*
3099          * We round up to the block size at eof when determining which
3100          * extents to clone above, but shouldn't round up the file size.
3101          */
3102         if (endoff > destoff + olen)
3103                 endoff = destoff + olen;
3104         if (endoff > inode->i_size)
3105                 btrfs_i_size_write(inode, endoff);
3106
3107         ret = btrfs_update_inode(trans, root, inode);
3108         if (ret) {
3109                 btrfs_abort_transaction(trans, root, ret);
3110                 btrfs_end_transaction(trans, root);
3111                 goto out;
3112         }
3113         ret = btrfs_end_transaction(trans, root);
3114 out:
3115         return ret;
3116 }
3117
3118 static void clone_update_extent_map(struct inode *inode,
3119                                     const struct btrfs_trans_handle *trans,
3120                                     const struct btrfs_path *path,
3121                                     const u64 hole_offset,
3122                                     const u64 hole_len)
3123 {
3124         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3125         struct extent_map *em;
3126         int ret;
3127
3128         em = alloc_extent_map();
3129         if (!em) {
3130                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3131                         &BTRFS_I(inode)->runtime_flags);
3132                 return;
3133         }
3134
3135         if (path) {
3136                 struct btrfs_file_extent_item *fi;
3137
3138                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3139                                     struct btrfs_file_extent_item);
3140                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3141                 em->generation = -1;
3142                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3143                     BTRFS_FILE_EXTENT_INLINE)
3144                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3145                                 &BTRFS_I(inode)->runtime_flags);
3146         } else {
3147                 em->start = hole_offset;
3148                 em->len = hole_len;
3149                 em->ram_bytes = em->len;
3150                 em->orig_start = hole_offset;
3151                 em->block_start = EXTENT_MAP_HOLE;
3152                 em->block_len = 0;
3153                 em->orig_block_len = 0;
3154                 em->compress_type = BTRFS_COMPRESS_NONE;
3155                 em->generation = trans->transid;
3156         }
3157
3158         while (1) {
3159                 write_lock(&em_tree->lock);
3160                 ret = add_extent_mapping(em_tree, em, 1);
3161                 write_unlock(&em_tree->lock);
3162                 if (ret != -EEXIST) {
3163                         free_extent_map(em);
3164                         break;
3165                 }
3166                 btrfs_drop_extent_cache(inode, em->start,
3167                                         em->start + em->len - 1, 0);
3168         }
3169
3170         if (unlikely(ret))
3171                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3172                         &BTRFS_I(inode)->runtime_flags);
3173 }
3174
3175 /**
3176  * btrfs_clone() - clone a range from inode file to another
3177  *
3178  * @src: Inode to clone from
3179  * @inode: Inode to clone to
3180  * @off: Offset within source to start clone from
3181  * @olen: Original length, passed by user, of range to clone
3182  * @olen_aligned: Block-aligned value of olen, extent_same uses
3183  *               identical values here
3184  * @destoff: Offset within @inode to start clone
3185  */
3186 static int btrfs_clone(struct inode *src, struct inode *inode,
3187                        const u64 off, const u64 olen, const u64 olen_aligned,
3188                        const u64 destoff)
3189 {
3190         struct btrfs_root *root = BTRFS_I(inode)->root;
3191         struct btrfs_path *path = NULL;
3192         struct extent_buffer *leaf;
3193         struct btrfs_trans_handle *trans;
3194         char *buf = NULL;
3195         struct btrfs_key key;
3196         u32 nritems;
3197         int slot;
3198         int ret;
3199         int no_quota;
3200         const u64 len = olen_aligned;
3201         u64 last_disko = 0;
3202         u64 last_dest_end = destoff;
3203
3204         ret = -ENOMEM;
3205         buf = vmalloc(root->nodesize);
3206         if (!buf)
3207                 return ret;
3208
3209         path = btrfs_alloc_path();
3210         if (!path) {
3211                 vfree(buf);
3212                 return ret;
3213         }
3214
3215         path->reada = 2;
3216         /* clone data */
3217         key.objectid = btrfs_ino(src);
3218         key.type = BTRFS_EXTENT_DATA_KEY;
3219         key.offset = off;
3220
3221         while (1) {
3222                 /*
3223                  * note the key will change type as we walk through the
3224                  * tree.
3225                  */
3226                 path->leave_spinning = 1;
3227                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3228                                 0, 0);
3229                 if (ret < 0)
3230                         goto out;
3231                 /*
3232                  * First search, if no extent item that starts at offset off was
3233                  * found but the previous item is an extent item, it's possible
3234                  * it might overlap our target range, therefore process it.
3235                  */
3236                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3237                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3238                                               path->slots[0] - 1);
3239                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3240                                 path->slots[0]--;
3241                 }
3242
3243                 nritems = btrfs_header_nritems(path->nodes[0]);
3244 process_slot:
3245                 no_quota = 1;
3246                 if (path->slots[0] >= nritems) {
3247                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3248                         if (ret < 0)
3249                                 goto out;
3250                         if (ret > 0)
3251                                 break;
3252                         nritems = btrfs_header_nritems(path->nodes[0]);
3253                 }
3254                 leaf = path->nodes[0];
3255                 slot = path->slots[0];
3256
3257                 btrfs_item_key_to_cpu(leaf, &key, slot);
3258                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3259                     key.objectid != btrfs_ino(src))
3260                         break;
3261
3262                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3263                         struct btrfs_file_extent_item *extent;
3264                         int type;
3265                         u32 size;
3266                         struct btrfs_key new_key;
3267                         u64 disko = 0, diskl = 0;
3268                         u64 datao = 0, datal = 0;
3269                         u8 comp;
3270                         u64 drop_start;
3271
3272                         extent = btrfs_item_ptr(leaf, slot,
3273                                                 struct btrfs_file_extent_item);
3274                         comp = btrfs_file_extent_compression(leaf, extent);
3275                         type = btrfs_file_extent_type(leaf, extent);
3276                         if (type == BTRFS_FILE_EXTENT_REG ||
3277                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3278                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3279                                                                       extent);
3280                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3281                                                                  extent);
3282                                 datao = btrfs_file_extent_offset(leaf, extent);
3283                                 datal = btrfs_file_extent_num_bytes(leaf,
3284                                                                     extent);
3285                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3286                                 /* take upper bound, may be compressed */
3287                                 datal = btrfs_file_extent_ram_bytes(leaf,
3288                                                                     extent);
3289                         }
3290
3291                         /*
3292                          * The first search might have left us at an extent
3293                          * item that ends before our target range's start, can
3294                          * happen if we have holes and NO_HOLES feature enabled.
3295                          */
3296                         if (key.offset + datal <= off) {
3297                                 path->slots[0]++;
3298                                 goto process_slot;
3299                         } else if (key.offset >= off + len) {
3300                                 break;
3301                         }
3302
3303                         size = btrfs_item_size_nr(leaf, slot);
3304                         read_extent_buffer(leaf, buf,
3305                                            btrfs_item_ptr_offset(leaf, slot),
3306                                            size);
3307
3308                         btrfs_release_path(path);
3309                         path->leave_spinning = 0;
3310
3311                         memcpy(&new_key, &key, sizeof(new_key));
3312                         new_key.objectid = btrfs_ino(inode);
3313                         if (off <= key.offset)
3314                                 new_key.offset = key.offset + destoff - off;
3315                         else
3316                                 new_key.offset = destoff;
3317
3318                         /*
3319                          * Deal with a hole that doesn't have an extent item
3320                          * that represents it (NO_HOLES feature enabled).
3321                          * This hole is either in the middle of the cloning
3322                          * range or at the beginning (fully overlaps it or
3323                          * partially overlaps it).
3324                          */
3325                         if (new_key.offset != last_dest_end)
3326                                 drop_start = last_dest_end;
3327                         else
3328                                 drop_start = new_key.offset;
3329
3330                         /*
3331                          * 1 - adjusting old extent (we may have to split it)
3332                          * 1 - add new extent
3333                          * 1 - inode update
3334                          */
3335                         trans = btrfs_start_transaction(root, 3);
3336                         if (IS_ERR(trans)) {
3337                                 ret = PTR_ERR(trans);
3338                                 goto out;
3339                         }
3340
3341                         if (type == BTRFS_FILE_EXTENT_REG ||
3342                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3343                                 /*
3344                                  *    a  | --- range to clone ---|  b
3345                                  * | ------------- extent ------------- |
3346                                  */
3347
3348                                 /* subtract range b */
3349                                 if (key.offset + datal > off + len)
3350                                         datal = off + len - key.offset;
3351
3352                                 /* subtract range a */
3353                                 if (off > key.offset) {
3354                                         datao += off - key.offset;
3355                                         datal -= off - key.offset;
3356                                 }
3357
3358                                 ret = btrfs_drop_extents(trans, root, inode,
3359                                                          drop_start,
3360                                                          new_key.offset + datal,
3361                                                          1);
3362                                 if (ret) {
3363                                         if (ret != -EOPNOTSUPP)
3364                                                 btrfs_abort_transaction(trans,
3365                                                                 root, ret);
3366                                         btrfs_end_transaction(trans, root);
3367                                         goto out;
3368                                 }
3369
3370                                 ret = btrfs_insert_empty_item(trans, root, path,
3371                                                               &new_key, size);
3372                                 if (ret) {
3373                                         btrfs_abort_transaction(trans, root,
3374                                                                 ret);
3375                                         btrfs_end_transaction(trans, root);
3376                                         goto out;
3377                                 }
3378
3379                                 leaf = path->nodes[0];
3380                                 slot = path->slots[0];
3381                                 write_extent_buffer(leaf, buf,
3382                                             btrfs_item_ptr_offset(leaf, slot),
3383                                             size);
3384
3385                                 extent = btrfs_item_ptr(leaf, slot,
3386                                                 struct btrfs_file_extent_item);
3387
3388                                 /* disko == 0 means it's a hole */
3389                                 if (!disko)
3390                                         datao = 0;
3391
3392                                 btrfs_set_file_extent_offset(leaf, extent,
3393                                                              datao);
3394                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3395                                                                 datal);
3396
3397                                 /*
3398                                  * We need to look up the roots that point at
3399                                  * this bytenr and see if the new root does.  If
3400                                  * it does not we need to make sure we update
3401                                  * quotas appropriately.
3402                                  */
3403                                 if (disko && root != BTRFS_I(src)->root &&
3404                                     disko != last_disko) {
3405                                         no_quota = check_ref(trans, root,
3406                                                              disko);
3407                                         if (no_quota < 0) {
3408                                                 btrfs_abort_transaction(trans,
3409                                                                         root,
3410                                                                         ret);
3411                                                 btrfs_end_transaction(trans,
3412                                                                       root);
3413                                                 ret = no_quota;
3414                                                 goto out;
3415                                         }
3416                                 }
3417
3418                                 if (disko) {
3419                                         inode_add_bytes(inode, datal);
3420                                         ret = btrfs_inc_extent_ref(trans, root,
3421                                                         disko, diskl, 0,
3422                                                         root->root_key.objectid,
3423                                                         btrfs_ino(inode),
3424                                                         new_key.offset - datao,
3425                                                         no_quota);
3426                                         if (ret) {
3427                                                 btrfs_abort_transaction(trans,
3428                                                                         root,
3429                                                                         ret);
3430                                                 btrfs_end_transaction(trans,
3431                                                                       root);
3432                                                 goto out;
3433
3434                                         }
3435                                 }
3436                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3437                                 u64 skip = 0;
3438                                 u64 trim = 0;
3439                                 u64 aligned_end = 0;
3440
3441                                 if (off > key.offset) {
3442                                         skip = off - key.offset;
3443                                         new_key.offset += skip;
3444                                 }
3445
3446                                 if (key.offset + datal > off + len)
3447                                         trim = key.offset + datal - (off + len);
3448
3449                                 if (comp && (skip || trim)) {
3450                                         ret = -EINVAL;
3451                                         btrfs_end_transaction(trans, root);
3452                                         goto out;
3453                                 }
3454                                 size -= skip + trim;
3455                                 datal -= skip + trim;
3456
3457                                 aligned_end = ALIGN(new_key.offset + datal,
3458                                                     root->sectorsize);
3459                                 ret = btrfs_drop_extents(trans, root, inode,
3460                                                          drop_start,
3461                                                          aligned_end,
3462                                                          1);
3463                                 if (ret) {
3464                                         if (ret != -EOPNOTSUPP)
3465                                                 btrfs_abort_transaction(trans,
3466                                                         root, ret);
3467                                         btrfs_end_transaction(trans, root);
3468                                         goto out;
3469                                 }
3470
3471                                 ret = btrfs_insert_empty_item(trans, root, path,
3472                                                               &new_key, size);
3473                                 if (ret) {
3474                                         btrfs_abort_transaction(trans, root,
3475                                                                 ret);
3476                                         btrfs_end_transaction(trans, root);
3477                                         goto out;
3478                                 }
3479
3480                                 if (skip) {
3481                                         u32 start =
3482                                           btrfs_file_extent_calc_inline_size(0);
3483                                         memmove(buf+start, buf+start+skip,
3484                                                 datal);
3485                                 }
3486
3487                                 leaf = path->nodes[0];
3488                                 slot = path->slots[0];
3489                                 write_extent_buffer(leaf, buf,
3490                                             btrfs_item_ptr_offset(leaf, slot),
3491                                             size);
3492                                 inode_add_bytes(inode, datal);
3493                         }
3494
3495                         /* If we have an implicit hole (NO_HOLES feature). */
3496                         if (drop_start < new_key.offset)
3497                                 clone_update_extent_map(inode, trans,
3498                                                 NULL, drop_start,
3499                                                 new_key.offset - drop_start);
3500
3501                         clone_update_extent_map(inode, trans, path, 0, 0);
3502
3503                         btrfs_mark_buffer_dirty(leaf);
3504                         btrfs_release_path(path);
3505
3506                         last_dest_end = ALIGN(new_key.offset + datal,
3507                                               root->sectorsize);
3508                         ret = clone_finish_inode_update(trans, inode,
3509                                                         last_dest_end,
3510                                                         destoff, olen);
3511                         if (ret)
3512                                 goto out;
3513                         if (new_key.offset + datal >= destoff + len)
3514                                 break;
3515                 }
3516                 btrfs_release_path(path);
3517                 key.offset++;
3518         }
3519         ret = 0;
3520
3521         if (last_dest_end < destoff + len) {
3522                 /*
3523                  * We have an implicit hole (NO_HOLES feature is enabled) that
3524                  * fully or partially overlaps our cloning range at its end.
3525                  */
3526                 btrfs_release_path(path);
3527
3528                 /*
3529                  * 1 - remove extent(s)
3530                  * 1 - inode update
3531                  */
3532                 trans = btrfs_start_transaction(root, 2);
3533                 if (IS_ERR(trans)) {
3534                         ret = PTR_ERR(trans);
3535                         goto out;
3536                 }
3537                 ret = btrfs_drop_extents(trans, root, inode,
3538                                          last_dest_end, destoff + len, 1);
3539                 if (ret) {
3540                         if (ret != -EOPNOTSUPP)
3541                                 btrfs_abort_transaction(trans, root, ret);
3542                         btrfs_end_transaction(trans, root);
3543                         goto out;
3544                 }
3545                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3546                                         destoff + len - last_dest_end);
3547                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3548                                                 destoff, olen);
3549         }
3550
3551 out:
3552         btrfs_free_path(path);
3553         vfree(buf);
3554         return ret;
3555 }
3556
3557 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3558                                        u64 off, u64 olen, u64 destoff)
3559 {
3560         struct inode *inode = file_inode(file);
3561         struct btrfs_root *root = BTRFS_I(inode)->root;
3562         struct fd src_file;
3563         struct inode *src;
3564         int ret;
3565         u64 len = olen;
3566         u64 bs = root->fs_info->sb->s_blocksize;
3567         int same_inode = 0;
3568
3569         /*
3570          * TODO:
3571          * - split compressed inline extents.  annoying: we need to
3572          *   decompress into destination's address_space (the file offset
3573          *   may change, so source mapping won't do), then recompress (or
3574          *   otherwise reinsert) a subrange.
3575          *
3576          * - split destination inode's inline extents.  The inline extents can
3577          *   be either compressed or non-compressed.
3578          */
3579
3580         /* the destination must be opened for writing */
3581         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3582                 return -EINVAL;
3583
3584         if (btrfs_root_readonly(root))
3585                 return -EROFS;
3586
3587         ret = mnt_want_write_file(file);
3588         if (ret)
3589                 return ret;
3590
3591         src_file = fdget(srcfd);
3592         if (!src_file.file) {
3593                 ret = -EBADF;
3594                 goto out_drop_write;
3595         }
3596
3597         ret = -EXDEV;
3598         if (src_file.file->f_path.mnt != file->f_path.mnt)
3599                 goto out_fput;
3600
3601         src = file_inode(src_file.file);
3602
3603         ret = -EINVAL;
3604         if (src == inode)
3605                 same_inode = 1;
3606
3607         /* the src must be open for reading */
3608         if (!(src_file.file->f_mode & FMODE_READ))
3609                 goto out_fput;
3610
3611         /* don't make the dst file partly checksummed */
3612         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3613             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3614                 goto out_fput;
3615
3616         ret = -EISDIR;
3617         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3618                 goto out_fput;
3619
3620         ret = -EXDEV;
3621         if (src->i_sb != inode->i_sb)
3622                 goto out_fput;
3623
3624         if (!same_inode) {
3625                 if (inode < src) {
3626                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3627                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3628                 } else {
3629                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3630                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3631                 }
3632         } else {
3633                 mutex_lock(&src->i_mutex);
3634         }
3635
3636         /* determine range to clone */
3637         ret = -EINVAL;
3638         if (off + len > src->i_size || off + len < off)
3639                 goto out_unlock;
3640         if (len == 0)
3641                 olen = len = src->i_size - off;
3642         /* if we extend to eof, continue to block boundary */
3643         if (off + len == src->i_size)
3644                 len = ALIGN(src->i_size, bs) - off;
3645
3646         /* verify the end result is block aligned */
3647         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3648             !IS_ALIGNED(destoff, bs))
3649                 goto out_unlock;
3650
3651         /* verify if ranges are overlapped within the same file */
3652         if (same_inode) {
3653                 if (destoff + len > off && destoff < off + len)
3654                         goto out_unlock;
3655         }
3656
3657         if (destoff > inode->i_size) {
3658                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3659                 if (ret)
3660                         goto out_unlock;
3661         }
3662
3663         /*
3664          * Lock the target range too. Right after we replace the file extent
3665          * items in the fs tree (which now point to the cloned data), we might
3666          * have a worker replace them with extent items relative to a write
3667          * operation that was issued before this clone operation (i.e. confront
3668          * with inode.c:btrfs_finish_ordered_io).
3669          */
3670         if (same_inode) {
3671                 u64 lock_start = min_t(u64, off, destoff);
3672                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3673
3674                 lock_extent_range(src, lock_start, lock_len);
3675         } else {
3676                 lock_extent_range(src, off, len);
3677                 lock_extent_range(inode, destoff, len);
3678         }
3679
3680         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3681
3682         if (same_inode) {
3683                 u64 lock_start = min_t(u64, off, destoff);
3684                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3685
3686                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3687         } else {
3688                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3689                 unlock_extent(&BTRFS_I(inode)->io_tree, destoff,
3690                               destoff + len - 1);
3691         }
3692         /*
3693          * Truncate page cache pages so that future reads will see the cloned
3694          * data immediately and not the previous data.
3695          */
3696         truncate_inode_pages_range(&inode->i_data, destoff,
3697                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3698 out_unlock:
3699         if (!same_inode) {
3700                 if (inode < src) {
3701                         mutex_unlock(&src->i_mutex);
3702                         mutex_unlock(&inode->i_mutex);
3703                 } else {
3704                         mutex_unlock(&inode->i_mutex);
3705                         mutex_unlock(&src->i_mutex);
3706                 }
3707         } else {
3708                 mutex_unlock(&src->i_mutex);
3709         }
3710 out_fput:
3711         fdput(src_file);
3712 out_drop_write:
3713         mnt_drop_write_file(file);
3714         return ret;
3715 }
3716
3717 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3718 {
3719         struct btrfs_ioctl_clone_range_args args;
3720
3721         if (copy_from_user(&args, argp, sizeof(args)))
3722                 return -EFAULT;
3723         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3724                                  args.src_length, args.dest_offset);
3725 }
3726
3727 /*
3728  * there are many ways the trans_start and trans_end ioctls can lead
3729  * to deadlocks.  They should only be used by applications that
3730  * basically own the machine, and have a very in depth understanding
3731  * of all the possible deadlocks and enospc problems.
3732  */
3733 static long btrfs_ioctl_trans_start(struct file *file)
3734 {
3735         struct inode *inode = file_inode(file);
3736         struct btrfs_root *root = BTRFS_I(inode)->root;
3737         struct btrfs_trans_handle *trans;
3738         int ret;
3739
3740         ret = -EPERM;
3741         if (!capable(CAP_SYS_ADMIN))
3742                 goto out;
3743
3744         ret = -EINPROGRESS;
3745         if (file->private_data)
3746                 goto out;
3747
3748         ret = -EROFS;
3749         if (btrfs_root_readonly(root))
3750                 goto out;
3751
3752         ret = mnt_want_write_file(file);
3753         if (ret)
3754                 goto out;
3755
3756         atomic_inc(&root->fs_info->open_ioctl_trans);
3757
3758         ret = -ENOMEM;
3759         trans = btrfs_start_ioctl_transaction(root);
3760         if (IS_ERR(trans))
3761                 goto out_drop;
3762
3763         file->private_data = trans;
3764         return 0;
3765
3766 out_drop:
3767         atomic_dec(&root->fs_info->open_ioctl_trans);
3768         mnt_drop_write_file(file);
3769 out:
3770         return ret;
3771 }
3772
3773 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3774 {
3775         struct inode *inode = file_inode(file);
3776         struct btrfs_root *root = BTRFS_I(inode)->root;
3777         struct btrfs_root *new_root;
3778         struct btrfs_dir_item *di;
3779         struct btrfs_trans_handle *trans;
3780         struct btrfs_path *path;
3781         struct btrfs_key location;
3782         struct btrfs_disk_key disk_key;
3783         u64 objectid = 0;
3784         u64 dir_id;
3785         int ret;
3786
3787         if (!capable(CAP_SYS_ADMIN))
3788                 return -EPERM;
3789
3790         ret = mnt_want_write_file(file);
3791         if (ret)
3792                 return ret;
3793
3794         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3795                 ret = -EFAULT;
3796                 goto out;
3797         }
3798
3799         if (!objectid)
3800                 objectid = BTRFS_FS_TREE_OBJECTID;
3801
3802         location.objectid = objectid;
3803         location.type = BTRFS_ROOT_ITEM_KEY;
3804         location.offset = (u64)-1;
3805
3806         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3807         if (IS_ERR(new_root)) {
3808                 ret = PTR_ERR(new_root);
3809                 goto out;
3810         }
3811
3812         path = btrfs_alloc_path();
3813         if (!path) {
3814                 ret = -ENOMEM;
3815                 goto out;
3816         }
3817         path->leave_spinning = 1;
3818
3819         trans = btrfs_start_transaction(root, 1);
3820         if (IS_ERR(trans)) {
3821                 btrfs_free_path(path);
3822                 ret = PTR_ERR(trans);
3823                 goto out;
3824         }
3825
3826         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3827         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3828                                    dir_id, "default", 7, 1);
3829         if (IS_ERR_OR_NULL(di)) {
3830                 btrfs_free_path(path);
3831                 btrfs_end_transaction(trans, root);
3832                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3833                            "item, this isn't going to work");
3834                 ret = -ENOENT;
3835                 goto out;
3836         }
3837
3838         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3839         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3840         btrfs_mark_buffer_dirty(path->nodes[0]);
3841         btrfs_free_path(path);
3842
3843         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3844         btrfs_end_transaction(trans, root);
3845 out:
3846         mnt_drop_write_file(file);
3847         return ret;
3848 }
3849
3850 void btrfs_get_block_group_info(struct list_head *groups_list,
3851                                 struct btrfs_ioctl_space_info *space)
3852 {
3853         struct btrfs_block_group_cache *block_group;
3854
3855         space->total_bytes = 0;
3856         space->used_bytes = 0;
3857         space->flags = 0;
3858         list_for_each_entry(block_group, groups_list, list) {
3859                 space->flags = block_group->flags;
3860                 space->total_bytes += block_group->key.offset;
3861                 space->used_bytes +=
3862                         btrfs_block_group_used(&block_group->item);
3863         }
3864 }
3865
3866 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3867 {
3868         struct btrfs_ioctl_space_args space_args;
3869         struct btrfs_ioctl_space_info space;
3870         struct btrfs_ioctl_space_info *dest;
3871         struct btrfs_ioctl_space_info *dest_orig;
3872         struct btrfs_ioctl_space_info __user *user_dest;
3873         struct btrfs_space_info *info;
3874         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3875                        BTRFS_BLOCK_GROUP_SYSTEM,
3876                        BTRFS_BLOCK_GROUP_METADATA,
3877                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3878         int num_types = 4;
3879         int alloc_size;
3880         int ret = 0;
3881         u64 slot_count = 0;
3882         int i, c;
3883
3884         if (copy_from_user(&space_args,
3885                            (struct btrfs_ioctl_space_args __user *)arg,
3886                            sizeof(space_args)))
3887                 return -EFAULT;
3888
3889         for (i = 0; i < num_types; i++) {
3890                 struct btrfs_space_info *tmp;
3891
3892                 info = NULL;
3893                 rcu_read_lock();
3894                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3895                                         list) {
3896                         if (tmp->flags == types[i]) {
3897                                 info = tmp;
3898                                 break;
3899                         }
3900                 }
3901                 rcu_read_unlock();
3902
3903                 if (!info)
3904                         continue;
3905
3906                 down_read(&info->groups_sem);
3907                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3908                         if (!list_empty(&info->block_groups[c]))
3909                                 slot_count++;
3910                 }
3911                 up_read(&info->groups_sem);
3912         }
3913
3914         /*
3915          * Global block reserve, exported as a space_info
3916          */
3917         slot_count++;
3918
3919         /* space_slots == 0 means they are asking for a count */
3920         if (space_args.space_slots == 0) {
3921                 space_args.total_spaces = slot_count;
3922                 goto out;
3923         }
3924
3925         slot_count = min_t(u64, space_args.space_slots, slot_count);
3926
3927         alloc_size = sizeof(*dest) * slot_count;
3928
3929         /* we generally have at most 6 or so space infos, one for each raid
3930          * level.  So, a whole page should be more than enough for everyone
3931          */
3932         if (alloc_size > PAGE_CACHE_SIZE)
3933                 return -ENOMEM;
3934
3935         space_args.total_spaces = 0;
3936         dest = kmalloc(alloc_size, GFP_NOFS);
3937         if (!dest)
3938                 return -ENOMEM;
3939         dest_orig = dest;
3940
3941         /* now we have a buffer to copy into */
3942         for (i = 0; i < num_types; i++) {
3943                 struct btrfs_space_info *tmp;
3944
3945                 if (!slot_count)
3946                         break;
3947
3948                 info = NULL;
3949                 rcu_read_lock();
3950                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3951                                         list) {
3952                         if (tmp->flags == types[i]) {
3953                                 info = tmp;
3954                                 break;
3955                         }
3956                 }
3957                 rcu_read_unlock();
3958
3959                 if (!info)
3960                         continue;
3961                 down_read(&info->groups_sem);
3962                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3963                         if (!list_empty(&info->block_groups[c])) {
3964                                 btrfs_get_block_group_info(
3965                                         &info->block_groups[c], &space);
3966                                 memcpy(dest, &space, sizeof(space));
3967                                 dest++;
3968                                 space_args.total_spaces++;
3969                                 slot_count--;
3970                         }
3971                         if (!slot_count)
3972                                 break;
3973                 }
3974                 up_read(&info->groups_sem);
3975         }
3976
3977         /*
3978          * Add global block reserve
3979          */
3980         if (slot_count) {
3981                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3982
3983                 spin_lock(&block_rsv->lock);
3984                 space.total_bytes = block_rsv->size;
3985                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3986                 spin_unlock(&block_rsv->lock);
3987                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3988                 memcpy(dest, &space, sizeof(space));
3989                 space_args.total_spaces++;
3990         }
3991
3992         user_dest = (struct btrfs_ioctl_space_info __user *)
3993                 (arg + sizeof(struct btrfs_ioctl_space_args));
3994
3995         if (copy_to_user(user_dest, dest_orig, alloc_size))
3996                 ret = -EFAULT;
3997
3998         kfree(dest_orig);
3999 out:
4000         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4001                 ret = -EFAULT;
4002
4003         return ret;
4004 }
4005
4006 /*
4007  * there are many ways the trans_start and trans_end ioctls can lead
4008  * to deadlocks.  They should only be used by applications that
4009  * basically own the machine, and have a very in depth understanding
4010  * of all the possible deadlocks and enospc problems.
4011  */
4012 long btrfs_ioctl_trans_end(struct file *file)
4013 {
4014         struct inode *inode = file_inode(file);
4015         struct btrfs_root *root = BTRFS_I(inode)->root;
4016         struct btrfs_trans_handle *trans;
4017
4018         trans = file->private_data;
4019         if (!trans)
4020                 return -EINVAL;
4021         file->private_data = NULL;
4022
4023         btrfs_end_transaction(trans, root);
4024
4025         atomic_dec(&root->fs_info->open_ioctl_trans);
4026
4027         mnt_drop_write_file(file);
4028         return 0;
4029 }
4030
4031 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4032                                             void __user *argp)
4033 {
4034         struct btrfs_trans_handle *trans;
4035         u64 transid;
4036         int ret;
4037
4038         trans = btrfs_attach_transaction_barrier(root);
4039         if (IS_ERR(trans)) {
4040                 if (PTR_ERR(trans) != -ENOENT)
4041                         return PTR_ERR(trans);
4042
4043                 /* No running transaction, don't bother */
4044                 transid = root->fs_info->last_trans_committed;
4045                 goto out;
4046         }
4047         transid = trans->transid;
4048         ret = btrfs_commit_transaction_async(trans, root, 0);
4049         if (ret) {
4050                 btrfs_end_transaction(trans, root);
4051                 return ret;
4052         }
4053 out:
4054         if (argp)
4055                 if (copy_to_user(argp, &transid, sizeof(transid)))
4056                         return -EFAULT;
4057         return 0;
4058 }
4059
4060 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4061                                            void __user *argp)
4062 {
4063         u64 transid;
4064
4065         if (argp) {
4066                 if (copy_from_user(&transid, argp, sizeof(transid)))
4067                         return -EFAULT;
4068         } else {
4069                 transid = 0;  /* current trans */
4070         }
4071         return btrfs_wait_for_commit(root, transid);
4072 }
4073
4074 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4075 {
4076         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4077         struct btrfs_ioctl_scrub_args *sa;
4078         int ret;
4079
4080         if (!capable(CAP_SYS_ADMIN))
4081                 return -EPERM;
4082
4083         sa = memdup_user(arg, sizeof(*sa));
4084         if (IS_ERR(sa))
4085                 return PTR_ERR(sa);
4086
4087         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4088                 ret = mnt_want_write_file(file);
4089                 if (ret)
4090                         goto out;
4091         }
4092
4093         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4094                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4095                               0);
4096
4097         if (copy_to_user(arg, sa, sizeof(*sa)))
4098                 ret = -EFAULT;
4099
4100         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4101                 mnt_drop_write_file(file);
4102 out:
4103         kfree(sa);
4104         return ret;
4105 }
4106
4107 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4108 {
4109         if (!capable(CAP_SYS_ADMIN))
4110                 return -EPERM;
4111
4112         return btrfs_scrub_cancel(root->fs_info);
4113 }
4114
4115 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4116                                        void __user *arg)
4117 {
4118         struct btrfs_ioctl_scrub_args *sa;
4119         int ret;
4120
4121         if (!capable(CAP_SYS_ADMIN))
4122                 return -EPERM;
4123
4124         sa = memdup_user(arg, sizeof(*sa));
4125         if (IS_ERR(sa))
4126                 return PTR_ERR(sa);
4127
4128         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4129
4130         if (copy_to_user(arg, sa, sizeof(*sa)))
4131                 ret = -EFAULT;
4132
4133         kfree(sa);
4134         return ret;
4135 }
4136
4137 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4138                                       void __user *arg)
4139 {
4140         struct btrfs_ioctl_get_dev_stats *sa;
4141         int ret;
4142
4143         sa = memdup_user(arg, sizeof(*sa));
4144         if (IS_ERR(sa))
4145                 return PTR_ERR(sa);
4146
4147         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4148                 kfree(sa);
4149                 return -EPERM;
4150         }
4151
4152         ret = btrfs_get_dev_stats(root, sa);
4153
4154         if (copy_to_user(arg, sa, sizeof(*sa)))
4155                 ret = -EFAULT;
4156
4157         kfree(sa);
4158         return ret;
4159 }
4160
4161 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4162 {
4163         struct btrfs_ioctl_dev_replace_args *p;
4164         int ret;
4165
4166         if (!capable(CAP_SYS_ADMIN))
4167                 return -EPERM;
4168
4169         p = memdup_user(arg, sizeof(*p));
4170         if (IS_ERR(p))
4171                 return PTR_ERR(p);
4172
4173         switch (p->cmd) {
4174         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4175                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4176                         ret = -EROFS;
4177                         goto out;
4178                 }
4179                 if (atomic_xchg(
4180                         &root->fs_info->mutually_exclusive_operation_running,
4181                         1)) {
4182                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4183                 } else {
4184                         ret = btrfs_dev_replace_start(root, p);
4185                         atomic_set(
4186                          &root->fs_info->mutually_exclusive_operation_running,
4187                          0);
4188                 }
4189                 break;
4190         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4191                 btrfs_dev_replace_status(root->fs_info, p);
4192                 ret = 0;
4193                 break;
4194         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4195                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4196                 break;
4197         default:
4198                 ret = -EINVAL;
4199                 break;
4200         }
4201
4202         if (copy_to_user(arg, p, sizeof(*p)))
4203                 ret = -EFAULT;
4204 out:
4205         kfree(p);
4206         return ret;
4207 }
4208
4209 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4210 {
4211         int ret = 0;
4212         int i;
4213         u64 rel_ptr;
4214         int size;
4215         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4216         struct inode_fs_paths *ipath = NULL;
4217         struct btrfs_path *path;
4218
4219         if (!capable(CAP_DAC_READ_SEARCH))
4220                 return -EPERM;
4221
4222         path = btrfs_alloc_path();
4223         if (!path) {
4224                 ret = -ENOMEM;
4225                 goto out;
4226         }
4227
4228         ipa = memdup_user(arg, sizeof(*ipa));
4229         if (IS_ERR(ipa)) {
4230                 ret = PTR_ERR(ipa);
4231                 ipa = NULL;
4232                 goto out;
4233         }
4234
4235         size = min_t(u32, ipa->size, 4096);
4236         ipath = init_ipath(size, root, path);
4237         if (IS_ERR(ipath)) {
4238                 ret = PTR_ERR(ipath);
4239                 ipath = NULL;
4240                 goto out;
4241         }
4242
4243         ret = paths_from_inode(ipa->inum, ipath);
4244         if (ret < 0)
4245                 goto out;
4246
4247         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4248                 rel_ptr = ipath->fspath->val[i] -
4249                           (u64)(unsigned long)ipath->fspath->val;
4250                 ipath->fspath->val[i] = rel_ptr;
4251         }
4252
4253         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4254                            (void *)(unsigned long)ipath->fspath, size);
4255         if (ret) {
4256                 ret = -EFAULT;
4257                 goto out;
4258         }
4259
4260 out:
4261         btrfs_free_path(path);
4262         free_ipath(ipath);
4263         kfree(ipa);
4264
4265         return ret;
4266 }
4267
4268 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4269 {
4270         struct btrfs_data_container *inodes = ctx;
4271         const size_t c = 3 * sizeof(u64);
4272
4273         if (inodes->bytes_left >= c) {
4274                 inodes->bytes_left -= c;
4275                 inodes->val[inodes->elem_cnt] = inum;
4276                 inodes->val[inodes->elem_cnt + 1] = offset;
4277                 inodes->val[inodes->elem_cnt + 2] = root;
4278                 inodes->elem_cnt += 3;
4279         } else {
4280                 inodes->bytes_missing += c - inodes->bytes_left;
4281                 inodes->bytes_left = 0;
4282                 inodes->elem_missed += 3;
4283         }
4284
4285         return 0;
4286 }
4287
4288 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4289                                         void __user *arg)
4290 {
4291         int ret = 0;
4292         int size;
4293         struct btrfs_ioctl_logical_ino_args *loi;
4294         struct btrfs_data_container *inodes = NULL;
4295         struct btrfs_path *path = NULL;
4296
4297         if (!capable(CAP_SYS_ADMIN))
4298                 return -EPERM;
4299
4300         loi = memdup_user(arg, sizeof(*loi));
4301         if (IS_ERR(loi)) {
4302                 ret = PTR_ERR(loi);
4303                 loi = NULL;
4304                 goto out;
4305         }
4306
4307         path = btrfs_alloc_path();
4308         if (!path) {
4309                 ret = -ENOMEM;
4310                 goto out;
4311         }
4312
4313         size = min_t(u32, loi->size, 64 * 1024);
4314         inodes = init_data_container(size);
4315         if (IS_ERR(inodes)) {
4316                 ret = PTR_ERR(inodes);
4317                 inodes = NULL;
4318                 goto out;
4319         }
4320
4321         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4322                                           build_ino_list, inodes);
4323         if (ret == -EINVAL)
4324                 ret = -ENOENT;
4325         if (ret < 0)
4326                 goto out;
4327
4328         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4329                            (void *)(unsigned long)inodes, size);
4330         if (ret)
4331                 ret = -EFAULT;
4332
4333 out:
4334         btrfs_free_path(path);
4335         vfree(inodes);
4336         kfree(loi);
4337
4338         return ret;
4339 }
4340
4341 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4342                                struct btrfs_ioctl_balance_args *bargs)
4343 {
4344         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4345
4346         bargs->flags = bctl->flags;
4347
4348         if (atomic_read(&fs_info->balance_running))
4349                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4350         if (atomic_read(&fs_info->balance_pause_req))
4351                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4352         if (atomic_read(&fs_info->balance_cancel_req))
4353                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4354
4355         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4356         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4357         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4358
4359         if (lock) {
4360                 spin_lock(&fs_info->balance_lock);
4361                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4362                 spin_unlock(&fs_info->balance_lock);
4363         } else {
4364                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4365         }
4366 }
4367
4368 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4369 {
4370         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4371         struct btrfs_fs_info *fs_info = root->fs_info;
4372         struct btrfs_ioctl_balance_args *bargs;
4373         struct btrfs_balance_control *bctl;
4374         bool need_unlock; /* for mut. excl. ops lock */
4375         int ret;
4376
4377         if (!capable(CAP_SYS_ADMIN))
4378                 return -EPERM;
4379
4380         ret = mnt_want_write_file(file);
4381         if (ret)
4382                 return ret;
4383
4384 again:
4385         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4386                 mutex_lock(&fs_info->volume_mutex);
4387                 mutex_lock(&fs_info->balance_mutex);
4388                 need_unlock = true;
4389                 goto locked;
4390         }
4391
4392         /*
4393          * mut. excl. ops lock is locked.  Three possibilites:
4394          *   (1) some other op is running
4395          *   (2) balance is running
4396          *   (3) balance is paused -- special case (think resume)
4397          */
4398         mutex_lock(&fs_info->balance_mutex);
4399         if (fs_info->balance_ctl) {
4400                 /* this is either (2) or (3) */
4401                 if (!atomic_read(&fs_info->balance_running)) {
4402                         mutex_unlock(&fs_info->balance_mutex);
4403                         if (!mutex_trylock(&fs_info->volume_mutex))
4404                                 goto again;
4405                         mutex_lock(&fs_info->balance_mutex);
4406
4407                         if (fs_info->balance_ctl &&
4408                             !atomic_read(&fs_info->balance_running)) {
4409                                 /* this is (3) */
4410                                 need_unlock = false;
4411                                 goto locked;
4412                         }
4413
4414                         mutex_unlock(&fs_info->balance_mutex);
4415                         mutex_unlock(&fs_info->volume_mutex);
4416                         goto again;
4417                 } else {
4418                         /* this is (2) */
4419                         mutex_unlock(&fs_info->balance_mutex);
4420                         ret = -EINPROGRESS;
4421                         goto out;
4422                 }
4423         } else {
4424                 /* this is (1) */
4425                 mutex_unlock(&fs_info->balance_mutex);
4426                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4427                 goto out;
4428         }
4429
4430 locked:
4431         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4432
4433         if (arg) {
4434                 bargs = memdup_user(arg, sizeof(*bargs));
4435                 if (IS_ERR(bargs)) {
4436                         ret = PTR_ERR(bargs);
4437                         goto out_unlock;
4438                 }
4439
4440                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4441                         if (!fs_info->balance_ctl) {
4442                                 ret = -ENOTCONN;
4443                                 goto out_bargs;
4444                         }
4445
4446                         bctl = fs_info->balance_ctl;
4447                         spin_lock(&fs_info->balance_lock);
4448                         bctl->flags |= BTRFS_BALANCE_RESUME;
4449                         spin_unlock(&fs_info->balance_lock);
4450
4451                         goto do_balance;
4452                 }
4453         } else {
4454                 bargs = NULL;
4455         }
4456
4457         if (fs_info->balance_ctl) {
4458                 ret = -EINPROGRESS;
4459                 goto out_bargs;
4460         }
4461
4462         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4463         if (!bctl) {
4464                 ret = -ENOMEM;
4465                 goto out_bargs;
4466         }
4467
4468         bctl->fs_info = fs_info;
4469         if (arg) {
4470                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4471                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4472                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4473
4474                 bctl->flags = bargs->flags;
4475         } else {
4476                 /* balance everything - no filters */
4477                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4478         }
4479
4480 do_balance:
4481         /*
4482          * Ownership of bctl and mutually_exclusive_operation_running
4483          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4484          * or, if restriper was paused all the way until unmount, in
4485          * free_fs_info.  mutually_exclusive_operation_running is
4486          * cleared in __cancel_balance.
4487          */
4488         need_unlock = false;
4489
4490         ret = btrfs_balance(bctl, bargs);
4491
4492         if (arg) {
4493                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4494                         ret = -EFAULT;
4495         }
4496
4497 out_bargs:
4498         kfree(bargs);
4499 out_unlock:
4500         mutex_unlock(&fs_info->balance_mutex);
4501         mutex_unlock(&fs_info->volume_mutex);
4502         if (need_unlock)
4503                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4504 out:
4505         mnt_drop_write_file(file);
4506         return ret;
4507 }
4508
4509 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4510 {
4511         if (!capable(CAP_SYS_ADMIN))
4512                 return -EPERM;
4513
4514         switch (cmd) {
4515         case BTRFS_BALANCE_CTL_PAUSE:
4516                 return btrfs_pause_balance(root->fs_info);
4517         case BTRFS_BALANCE_CTL_CANCEL:
4518                 return btrfs_cancel_balance(root->fs_info);
4519         }
4520
4521         return -EINVAL;
4522 }
4523
4524 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4525                                          void __user *arg)
4526 {
4527         struct btrfs_fs_info *fs_info = root->fs_info;
4528         struct btrfs_ioctl_balance_args *bargs;
4529         int ret = 0;
4530
4531         if (!capable(CAP_SYS_ADMIN))
4532                 return -EPERM;
4533
4534         mutex_lock(&fs_info->balance_mutex);
4535         if (!fs_info->balance_ctl) {
4536                 ret = -ENOTCONN;
4537                 goto out;
4538         }
4539
4540         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4541         if (!bargs) {
4542                 ret = -ENOMEM;
4543                 goto out;
4544         }
4545
4546         update_ioctl_balance_args(fs_info, 1, bargs);
4547
4548         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4549                 ret = -EFAULT;
4550
4551         kfree(bargs);
4552 out:
4553         mutex_unlock(&fs_info->balance_mutex);
4554         return ret;
4555 }
4556
4557 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4558 {
4559         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4560         struct btrfs_ioctl_quota_ctl_args *sa;
4561         struct btrfs_trans_handle *trans = NULL;
4562         int ret;
4563         int err;
4564
4565         if (!capable(CAP_SYS_ADMIN))
4566                 return -EPERM;
4567
4568         ret = mnt_want_write_file(file);
4569         if (ret)
4570                 return ret;
4571
4572         sa = memdup_user(arg, sizeof(*sa));
4573         if (IS_ERR(sa)) {
4574                 ret = PTR_ERR(sa);
4575                 goto drop_write;
4576         }
4577
4578         down_write(&root->fs_info->subvol_sem);
4579         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4580         if (IS_ERR(trans)) {
4581                 ret = PTR_ERR(trans);
4582                 goto out;
4583         }
4584
4585         switch (sa->cmd) {
4586         case BTRFS_QUOTA_CTL_ENABLE:
4587                 ret = btrfs_quota_enable(trans, root->fs_info);
4588                 break;
4589         case BTRFS_QUOTA_CTL_DISABLE:
4590                 ret = btrfs_quota_disable(trans, root->fs_info);
4591                 break;
4592         default:
4593                 ret = -EINVAL;
4594                 break;
4595         }
4596
4597         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4598         if (err && !ret)
4599                 ret = err;
4600 out:
4601         kfree(sa);
4602         up_write(&root->fs_info->subvol_sem);
4603 drop_write:
4604         mnt_drop_write_file(file);
4605         return ret;
4606 }
4607
4608 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4609 {
4610         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4611         struct btrfs_ioctl_qgroup_assign_args *sa;
4612         struct btrfs_trans_handle *trans;
4613         int ret;
4614         int err;
4615
4616         if (!capable(CAP_SYS_ADMIN))
4617                 return -EPERM;
4618
4619         ret = mnt_want_write_file(file);
4620         if (ret)
4621                 return ret;
4622
4623         sa = memdup_user(arg, sizeof(*sa));
4624         if (IS_ERR(sa)) {
4625                 ret = PTR_ERR(sa);
4626                 goto drop_write;
4627         }
4628
4629         trans = btrfs_join_transaction(root);
4630         if (IS_ERR(trans)) {
4631                 ret = PTR_ERR(trans);
4632                 goto out;
4633         }
4634
4635         /* FIXME: check if the IDs really exist */
4636         if (sa->assign) {
4637                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4638                                                 sa->src, sa->dst);
4639         } else {
4640                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4641                                                 sa->src, sa->dst);
4642         }
4643
4644         err = btrfs_end_transaction(trans, root);
4645         if (err && !ret)
4646                 ret = err;
4647
4648 out:
4649         kfree(sa);
4650 drop_write:
4651         mnt_drop_write_file(file);
4652         return ret;
4653 }
4654
4655 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4656 {
4657         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4658         struct btrfs_ioctl_qgroup_create_args *sa;
4659         struct btrfs_trans_handle *trans;
4660         int ret;
4661         int err;
4662
4663         if (!capable(CAP_SYS_ADMIN))
4664                 return -EPERM;
4665
4666         ret = mnt_want_write_file(file);
4667         if (ret)
4668                 return ret;
4669
4670         sa = memdup_user(arg, sizeof(*sa));
4671         if (IS_ERR(sa)) {
4672                 ret = PTR_ERR(sa);
4673                 goto drop_write;
4674         }
4675
4676         if (!sa->qgroupid) {
4677                 ret = -EINVAL;
4678                 goto out;
4679         }
4680
4681         trans = btrfs_join_transaction(root);
4682         if (IS_ERR(trans)) {
4683                 ret = PTR_ERR(trans);
4684                 goto out;
4685         }
4686
4687         /* FIXME: check if the IDs really exist */
4688         if (sa->create) {
4689                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4690                                           NULL);
4691         } else {
4692                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4693         }
4694
4695         err = btrfs_end_transaction(trans, root);
4696         if (err && !ret)
4697                 ret = err;
4698
4699 out:
4700         kfree(sa);
4701 drop_write:
4702         mnt_drop_write_file(file);
4703         return ret;
4704 }
4705
4706 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4707 {
4708         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4709         struct btrfs_ioctl_qgroup_limit_args *sa;
4710         struct btrfs_trans_handle *trans;
4711         int ret;
4712         int err;
4713         u64 qgroupid;
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         trans = btrfs_join_transaction(root);
4729         if (IS_ERR(trans)) {
4730                 ret = PTR_ERR(trans);
4731                 goto out;
4732         }
4733
4734         qgroupid = sa->qgroupid;
4735         if (!qgroupid) {
4736                 /* take the current subvol as qgroup */
4737                 qgroupid = root->root_key.objectid;
4738         }
4739
4740         /* FIXME: check if the IDs really exist */
4741         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4742
4743         err = btrfs_end_transaction(trans, root);
4744         if (err && !ret)
4745                 ret = err;
4746
4747 out:
4748         kfree(sa);
4749 drop_write:
4750         mnt_drop_write_file(file);
4751         return ret;
4752 }
4753
4754 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4755 {
4756         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4757         struct btrfs_ioctl_quota_rescan_args *qsa;
4758         int ret;
4759
4760         if (!capable(CAP_SYS_ADMIN))
4761                 return -EPERM;
4762
4763         ret = mnt_want_write_file(file);
4764         if (ret)
4765                 return ret;
4766
4767         qsa = memdup_user(arg, sizeof(*qsa));
4768         if (IS_ERR(qsa)) {
4769                 ret = PTR_ERR(qsa);
4770                 goto drop_write;
4771         }
4772
4773         if (qsa->flags) {
4774                 ret = -EINVAL;
4775                 goto out;
4776         }
4777
4778         ret = btrfs_qgroup_rescan(root->fs_info);
4779
4780 out:
4781         kfree(qsa);
4782 drop_write:
4783         mnt_drop_write_file(file);
4784         return ret;
4785 }
4786
4787 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4788 {
4789         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4790         struct btrfs_ioctl_quota_rescan_args *qsa;
4791         int ret = 0;
4792
4793         if (!capable(CAP_SYS_ADMIN))
4794                 return -EPERM;
4795
4796         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4797         if (!qsa)
4798                 return -ENOMEM;
4799
4800         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4801                 qsa->flags = 1;
4802                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4803         }
4804
4805         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4806                 ret = -EFAULT;
4807
4808         kfree(qsa);
4809         return ret;
4810 }
4811
4812 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4813 {
4814         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4815
4816         if (!capable(CAP_SYS_ADMIN))
4817                 return -EPERM;
4818
4819         return btrfs_qgroup_wait_for_completion(root->fs_info);
4820 }
4821
4822 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4823                                             struct btrfs_ioctl_received_subvol_args *sa)
4824 {
4825         struct inode *inode = file_inode(file);
4826         struct btrfs_root *root = BTRFS_I(inode)->root;
4827         struct btrfs_root_item *root_item = &root->root_item;
4828         struct btrfs_trans_handle *trans;
4829         struct timespec ct = CURRENT_TIME;
4830         int ret = 0;
4831         int received_uuid_changed;
4832
4833         if (!inode_owner_or_capable(inode))
4834                 return -EPERM;
4835
4836         ret = mnt_want_write_file(file);
4837         if (ret < 0)
4838                 return ret;
4839
4840         down_write(&root->fs_info->subvol_sem);
4841
4842         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4843                 ret = -EINVAL;
4844                 goto out;
4845         }
4846
4847         if (btrfs_root_readonly(root)) {
4848                 ret = -EROFS;
4849                 goto out;
4850         }
4851
4852         /*
4853          * 1 - root item
4854          * 2 - uuid items (received uuid + subvol uuid)
4855          */
4856         trans = btrfs_start_transaction(root, 3);
4857         if (IS_ERR(trans)) {
4858                 ret = PTR_ERR(trans);
4859                 trans = NULL;
4860                 goto out;
4861         }
4862
4863         sa->rtransid = trans->transid;
4864         sa->rtime.sec = ct.tv_sec;
4865         sa->rtime.nsec = ct.tv_nsec;
4866
4867         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4868                                        BTRFS_UUID_SIZE);
4869         if (received_uuid_changed &&
4870             !btrfs_is_empty_uuid(root_item->received_uuid))
4871                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4872                                     root_item->received_uuid,
4873                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4874                                     root->root_key.objectid);
4875         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4876         btrfs_set_root_stransid(root_item, sa->stransid);
4877         btrfs_set_root_rtransid(root_item, sa->rtransid);
4878         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4879         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4880         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4881         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4882
4883         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4884                                 &root->root_key, &root->root_item);
4885         if (ret < 0) {
4886                 btrfs_end_transaction(trans, root);
4887                 goto out;
4888         }
4889         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4890                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4891                                           sa->uuid,
4892                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4893                                           root->root_key.objectid);
4894                 if (ret < 0 && ret != -EEXIST) {
4895                         btrfs_abort_transaction(trans, root, ret);
4896                         goto out;
4897                 }
4898         }
4899         ret = btrfs_commit_transaction(trans, root);
4900         if (ret < 0) {
4901                 btrfs_abort_transaction(trans, root, ret);
4902                 goto out;
4903         }
4904
4905 out:
4906         up_write(&root->fs_info->subvol_sem);
4907         mnt_drop_write_file(file);
4908         return ret;
4909 }
4910
4911 #ifdef CONFIG_64BIT
4912 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4913                                                 void __user *arg)
4914 {
4915         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4916         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4917         int ret = 0;
4918
4919         args32 = memdup_user(arg, sizeof(*args32));
4920         if (IS_ERR(args32)) {
4921                 ret = PTR_ERR(args32);
4922                 args32 = NULL;
4923                 goto out;
4924         }
4925
4926         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4927         if (!args64) {
4928                 ret = -ENOMEM;
4929                 goto out;
4930         }
4931
4932         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4933         args64->stransid = args32->stransid;
4934         args64->rtransid = args32->rtransid;
4935         args64->stime.sec = args32->stime.sec;
4936         args64->stime.nsec = args32->stime.nsec;
4937         args64->rtime.sec = args32->rtime.sec;
4938         args64->rtime.nsec = args32->rtime.nsec;
4939         args64->flags = args32->flags;
4940
4941         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4942         if (ret)
4943                 goto out;
4944
4945         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4946         args32->stransid = args64->stransid;
4947         args32->rtransid = args64->rtransid;
4948         args32->stime.sec = args64->stime.sec;
4949         args32->stime.nsec = args64->stime.nsec;
4950         args32->rtime.sec = args64->rtime.sec;
4951         args32->rtime.nsec = args64->rtime.nsec;
4952         args32->flags = args64->flags;
4953
4954         ret = copy_to_user(arg, args32, sizeof(*args32));
4955         if (ret)
4956                 ret = -EFAULT;
4957
4958 out:
4959         kfree(args32);
4960         kfree(args64);
4961         return ret;
4962 }
4963 #endif
4964
4965 static long btrfs_ioctl_set_received_subvol(struct file *file,
4966                                             void __user *arg)
4967 {
4968         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4969         int ret = 0;
4970
4971         sa = memdup_user(arg, sizeof(*sa));
4972         if (IS_ERR(sa)) {
4973                 ret = PTR_ERR(sa);
4974                 sa = NULL;
4975                 goto out;
4976         }
4977
4978         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4979
4980         if (ret)
4981                 goto out;
4982
4983         ret = copy_to_user(arg, sa, sizeof(*sa));
4984         if (ret)
4985                 ret = -EFAULT;
4986
4987 out:
4988         kfree(sa);
4989         return ret;
4990 }
4991
4992 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4993 {
4994         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4995         size_t len;
4996         int ret;
4997         char label[BTRFS_LABEL_SIZE];
4998
4999         spin_lock(&root->fs_info->super_lock);
5000         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5001         spin_unlock(&root->fs_info->super_lock);
5002
5003         len = strnlen(label, BTRFS_LABEL_SIZE);
5004
5005         if (len == BTRFS_LABEL_SIZE) {
5006                 btrfs_warn(root->fs_info,
5007                         "label is too long, return the first %zu bytes", --len);
5008         }
5009
5010         ret = copy_to_user(arg, label, len);
5011
5012         return ret ? -EFAULT : 0;
5013 }
5014
5015 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5016 {
5017         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5018         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5019         struct btrfs_trans_handle *trans;
5020         char label[BTRFS_LABEL_SIZE];
5021         int ret;
5022
5023         if (!capable(CAP_SYS_ADMIN))
5024                 return -EPERM;
5025
5026         if (copy_from_user(label, arg, sizeof(label)))
5027                 return -EFAULT;
5028
5029         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5030                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5031                        BTRFS_LABEL_SIZE - 1);
5032                 return -EINVAL;
5033         }
5034
5035         ret = mnt_want_write_file(file);
5036         if (ret)
5037                 return ret;
5038
5039         trans = btrfs_start_transaction(root, 0);
5040         if (IS_ERR(trans)) {
5041                 ret = PTR_ERR(trans);
5042                 goto out_unlock;
5043         }
5044
5045         spin_lock(&root->fs_info->super_lock);
5046         strcpy(super_block->label, label);
5047         spin_unlock(&root->fs_info->super_lock);
5048         ret = btrfs_commit_transaction(trans, root);
5049
5050 out_unlock:
5051         mnt_drop_write_file(file);
5052         return ret;
5053 }
5054
5055 #define INIT_FEATURE_FLAGS(suffix) \
5056         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5057           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5058           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5059
5060 static int btrfs_ioctl_get_supported_features(struct file *file,
5061                                               void __user *arg)
5062 {
5063         static struct btrfs_ioctl_feature_flags features[3] = {
5064                 INIT_FEATURE_FLAGS(SUPP),
5065                 INIT_FEATURE_FLAGS(SAFE_SET),
5066                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5067         };
5068
5069         if (copy_to_user(arg, &features, sizeof(features)))
5070                 return -EFAULT;
5071
5072         return 0;
5073 }
5074
5075 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5076 {
5077         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5078         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5079         struct btrfs_ioctl_feature_flags features;
5080
5081         features.compat_flags = btrfs_super_compat_flags(super_block);
5082         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5083         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5084
5085         if (copy_to_user(arg, &features, sizeof(features)))
5086                 return -EFAULT;
5087
5088         return 0;
5089 }
5090
5091 static int check_feature_bits(struct btrfs_root *root,
5092                               enum btrfs_feature_set set,
5093                               u64 change_mask, u64 flags, u64 supported_flags,
5094                               u64 safe_set, u64 safe_clear)
5095 {
5096         const char *type = btrfs_feature_set_names[set];
5097         char *names;
5098         u64 disallowed, unsupported;
5099         u64 set_mask = flags & change_mask;
5100         u64 clear_mask = ~flags & change_mask;
5101
5102         unsupported = set_mask & ~supported_flags;
5103         if (unsupported) {
5104                 names = btrfs_printable_features(set, unsupported);
5105                 if (names) {
5106                         btrfs_warn(root->fs_info,
5107                            "this kernel does not support the %s feature bit%s",
5108                            names, strchr(names, ',') ? "s" : "");
5109                         kfree(names);
5110                 } else
5111                         btrfs_warn(root->fs_info,
5112                            "this kernel does not support %s bits 0x%llx",
5113                            type, unsupported);
5114                 return -EOPNOTSUPP;
5115         }
5116
5117         disallowed = set_mask & ~safe_set;
5118         if (disallowed) {
5119                 names = btrfs_printable_features(set, disallowed);
5120                 if (names) {
5121                         btrfs_warn(root->fs_info,
5122                            "can't set the %s feature bit%s while mounted",
5123                            names, strchr(names, ',') ? "s" : "");
5124                         kfree(names);
5125                 } else
5126                         btrfs_warn(root->fs_info,
5127                            "can't set %s bits 0x%llx while mounted",
5128                            type, disallowed);
5129                 return -EPERM;
5130         }
5131
5132         disallowed = clear_mask & ~safe_clear;
5133         if (disallowed) {
5134                 names = btrfs_printable_features(set, disallowed);
5135                 if (names) {
5136                         btrfs_warn(root->fs_info,
5137                            "can't clear the %s feature bit%s while mounted",
5138                            names, strchr(names, ',') ? "s" : "");
5139                         kfree(names);
5140                 } else
5141                         btrfs_warn(root->fs_info,
5142                            "can't clear %s bits 0x%llx while mounted",
5143                            type, disallowed);
5144                 return -EPERM;
5145         }
5146
5147         return 0;
5148 }
5149
5150 #define check_feature(root, change_mask, flags, mask_base)      \
5151 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5152                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5153                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5154                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5155
5156 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5157 {
5158         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5159         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5160         struct btrfs_ioctl_feature_flags flags[2];
5161         struct btrfs_trans_handle *trans;
5162         u64 newflags;
5163         int ret;
5164
5165         if (!capable(CAP_SYS_ADMIN))
5166                 return -EPERM;
5167
5168         if (copy_from_user(flags, arg, sizeof(flags)))
5169                 return -EFAULT;
5170
5171         /* Nothing to do */
5172         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5173             !flags[0].incompat_flags)
5174                 return 0;
5175
5176         ret = check_feature(root, flags[0].compat_flags,
5177                             flags[1].compat_flags, COMPAT);
5178         if (ret)
5179                 return ret;
5180
5181         ret = check_feature(root, flags[0].compat_ro_flags,
5182                             flags[1].compat_ro_flags, COMPAT_RO);
5183         if (ret)
5184                 return ret;
5185
5186         ret = check_feature(root, flags[0].incompat_flags,
5187                             flags[1].incompat_flags, INCOMPAT);
5188         if (ret)
5189                 return ret;
5190
5191         trans = btrfs_start_transaction(root, 0);
5192         if (IS_ERR(trans))
5193                 return PTR_ERR(trans);
5194
5195         spin_lock(&root->fs_info->super_lock);
5196         newflags = btrfs_super_compat_flags(super_block);
5197         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5198         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5199         btrfs_set_super_compat_flags(super_block, newflags);
5200
5201         newflags = btrfs_super_compat_ro_flags(super_block);
5202         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5203         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5204         btrfs_set_super_compat_ro_flags(super_block, newflags);
5205
5206         newflags = btrfs_super_incompat_flags(super_block);
5207         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5208         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5209         btrfs_set_super_incompat_flags(super_block, newflags);
5210         spin_unlock(&root->fs_info->super_lock);
5211
5212         return btrfs_commit_transaction(trans, root);
5213 }
5214
5215 long btrfs_ioctl(struct file *file, unsigned int
5216                 cmd, unsigned long arg)
5217 {
5218         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5219         void __user *argp = (void __user *)arg;
5220
5221         switch (cmd) {
5222         case FS_IOC_GETFLAGS:
5223                 return btrfs_ioctl_getflags(file, argp);
5224         case FS_IOC_SETFLAGS:
5225                 return btrfs_ioctl_setflags(file, argp);
5226         case FS_IOC_GETVERSION:
5227                 return btrfs_ioctl_getversion(file, argp);
5228         case FITRIM:
5229                 return btrfs_ioctl_fitrim(file, argp);
5230         case BTRFS_IOC_SNAP_CREATE:
5231                 return btrfs_ioctl_snap_create(file, argp, 0);
5232         case BTRFS_IOC_SNAP_CREATE_V2:
5233                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5234         case BTRFS_IOC_SUBVOL_CREATE:
5235                 return btrfs_ioctl_snap_create(file, argp, 1);
5236         case BTRFS_IOC_SUBVOL_CREATE_V2:
5237                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5238         case BTRFS_IOC_SNAP_DESTROY:
5239                 return btrfs_ioctl_snap_destroy(file, argp);
5240         case BTRFS_IOC_SUBVOL_GETFLAGS:
5241                 return btrfs_ioctl_subvol_getflags(file, argp);
5242         case BTRFS_IOC_SUBVOL_SETFLAGS:
5243                 return btrfs_ioctl_subvol_setflags(file, argp);
5244         case BTRFS_IOC_DEFAULT_SUBVOL:
5245                 return btrfs_ioctl_default_subvol(file, argp);
5246         case BTRFS_IOC_DEFRAG:
5247                 return btrfs_ioctl_defrag(file, NULL);
5248         case BTRFS_IOC_DEFRAG_RANGE:
5249                 return btrfs_ioctl_defrag(file, argp);
5250         case BTRFS_IOC_RESIZE:
5251                 return btrfs_ioctl_resize(file, argp);
5252         case BTRFS_IOC_ADD_DEV:
5253                 return btrfs_ioctl_add_dev(root, argp);
5254         case BTRFS_IOC_RM_DEV:
5255                 return btrfs_ioctl_rm_dev(file, argp);
5256         case BTRFS_IOC_FS_INFO:
5257                 return btrfs_ioctl_fs_info(root, argp);
5258         case BTRFS_IOC_DEV_INFO:
5259                 return btrfs_ioctl_dev_info(root, argp);
5260         case BTRFS_IOC_BALANCE:
5261                 return btrfs_ioctl_balance(file, NULL);
5262         case BTRFS_IOC_CLONE:
5263                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5264         case BTRFS_IOC_CLONE_RANGE:
5265                 return btrfs_ioctl_clone_range(file, argp);
5266         case BTRFS_IOC_TRANS_START:
5267                 return btrfs_ioctl_trans_start(file);
5268         case BTRFS_IOC_TRANS_END:
5269                 return btrfs_ioctl_trans_end(file);
5270         case BTRFS_IOC_TREE_SEARCH:
5271                 return btrfs_ioctl_tree_search(file, argp);
5272         case BTRFS_IOC_TREE_SEARCH_V2:
5273                 return btrfs_ioctl_tree_search_v2(file, argp);
5274         case BTRFS_IOC_INO_LOOKUP:
5275                 return btrfs_ioctl_ino_lookup(file, argp);
5276         case BTRFS_IOC_INO_PATHS:
5277                 return btrfs_ioctl_ino_to_path(root, argp);
5278         case BTRFS_IOC_LOGICAL_INO:
5279                 return btrfs_ioctl_logical_to_ino(root, argp);
5280         case BTRFS_IOC_SPACE_INFO:
5281                 return btrfs_ioctl_space_info(root, argp);
5282         case BTRFS_IOC_SYNC: {
5283                 int ret;
5284
5285                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5286                 if (ret)
5287                         return ret;
5288                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
5289                 /*
5290                  * The transaction thread may want to do more work,
5291                  * namely it pokes the cleaner ktread that will start
5292                  * processing uncleaned subvols.
5293                  */
5294                 wake_up_process(root->fs_info->transaction_kthread);
5295                 return ret;
5296         }
5297         case BTRFS_IOC_START_SYNC:
5298                 return btrfs_ioctl_start_sync(root, argp);
5299         case BTRFS_IOC_WAIT_SYNC:
5300                 return btrfs_ioctl_wait_sync(root, argp);
5301         case BTRFS_IOC_SCRUB:
5302                 return btrfs_ioctl_scrub(file, argp);
5303         case BTRFS_IOC_SCRUB_CANCEL:
5304                 return btrfs_ioctl_scrub_cancel(root, argp);
5305         case BTRFS_IOC_SCRUB_PROGRESS:
5306                 return btrfs_ioctl_scrub_progress(root, argp);
5307         case BTRFS_IOC_BALANCE_V2:
5308                 return btrfs_ioctl_balance(file, argp);
5309         case BTRFS_IOC_BALANCE_CTL:
5310                 return btrfs_ioctl_balance_ctl(root, arg);
5311         case BTRFS_IOC_BALANCE_PROGRESS:
5312                 return btrfs_ioctl_balance_progress(root, argp);
5313         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5314                 return btrfs_ioctl_set_received_subvol(file, argp);
5315 #ifdef CONFIG_64BIT
5316         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5317                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5318 #endif
5319         case BTRFS_IOC_SEND:
5320                 return btrfs_ioctl_send(file, argp);
5321         case BTRFS_IOC_GET_DEV_STATS:
5322                 return btrfs_ioctl_get_dev_stats(root, argp);
5323         case BTRFS_IOC_QUOTA_CTL:
5324                 return btrfs_ioctl_quota_ctl(file, argp);
5325         case BTRFS_IOC_QGROUP_ASSIGN:
5326                 return btrfs_ioctl_qgroup_assign(file, argp);
5327         case BTRFS_IOC_QGROUP_CREATE:
5328                 return btrfs_ioctl_qgroup_create(file, argp);
5329         case BTRFS_IOC_QGROUP_LIMIT:
5330                 return btrfs_ioctl_qgroup_limit(file, argp);
5331         case BTRFS_IOC_QUOTA_RESCAN:
5332                 return btrfs_ioctl_quota_rescan(file, argp);
5333         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5334                 return btrfs_ioctl_quota_rescan_status(file, argp);
5335         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5336                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5337         case BTRFS_IOC_DEV_REPLACE:
5338                 return btrfs_ioctl_dev_replace(root, argp);
5339         case BTRFS_IOC_GET_FSLABEL:
5340                 return btrfs_ioctl_get_fslabel(file, argp);
5341         case BTRFS_IOC_SET_FSLABEL:
5342                 return btrfs_ioctl_set_fslabel(file, argp);
5343         case BTRFS_IOC_FILE_EXTENT_SAME:
5344                 return btrfs_ioctl_file_extent_same(file, argp);
5345         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5346                 return btrfs_ioctl_get_supported_features(file, argp);
5347         case BTRFS_IOC_GET_FEATURES:
5348                 return btrfs_ioctl_get_features(file, argp);
5349         case BTRFS_IOC_SET_FEATURES:
5350                 return btrfs_ioctl_set_features(file, argp);
5351         }
5352
5353         return -ENOTTY;
5354 }