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