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         path->keep_locks = 1;
 939
 940         while (1) {
 941                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 942                 if (ret != 0)
 943                         goto none;
 944                 if (min_key.objectid != ino)
 945                         goto none;
 946                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 947                         goto none;
 948
 949                 leaf = path->nodes[0];
 950                 extent = btrfs_item_ptr(leaf, path->slots[0],
 951                                         struct btrfs_file_extent_item);
 952
 953                 type = btrfs_file_extent_type(leaf, extent);
 954                 if (type == BTRFS_FILE_EXTENT_REG &&
 955                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 956                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 957                         *off = min_key.offset;
 958                         btrfs_free_path(path);
 959                         return 0;
 960                 }
 961
 962                 if (min_key.offset == (u64)-1)
 963                         goto none;
 964
 965                 min_key.offset++;
 966                 btrfs_release_path(path);
 967         }
 968 none:
 969         btrfs_free_path(path);
 970         return -ENOENT;
 971 }
 972
 973 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 974 {
 975         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 976         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 977         struct extent_map *em;
 978         u64 len = PAGE_CACHE_SIZE;
 979
 980         /*
 981          * hopefully we have this extent in the tree already, try without
 982          * the full extent lock
 983          */
 984         read_lock(&em_tree->lock);
 985         em = lookup_extent_mapping(em_tree, start, len);
 986         read_unlock(&em_tree->lock);
 987
 988         if (!em) {
 989                 struct extent_state *cached = NULL;
 990                 u64 end = start + len - 1;
 991
 992                 /* get the big lock and read metadata off disk */
 993                 lock_extent_bits(io_tree, start, end, 0, &cached);
 994                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 995                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 996
 997                 if (IS_ERR(em))
 998                         return NULL;
 999         }
1000
1001         return em;
1002 }
1003
1004 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1005 {
1006         struct extent_map *next;
1007         bool ret = true;
1008
1009         /* this is the last extent */
1010         if (em->start + em->len >= i_size_read(inode))
1011                 return false;
1012
1013         next = defrag_lookup_extent(inode, em->start + em->len);
1014         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE ||
1015             (em->block_start + em->block_len == next->block_start))
1016                 ret = false;
1017
1018         free_extent_map(next);
1019         return ret;
1020 }
1021
1022 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
1023                                u64 *last_len, u64 *skip, u64 *defrag_end,
1024                                int compress)
1025 {
1026         struct extent_map *em;
1027         int ret = 1;
1028         bool next_mergeable = true;
1029
1030         /*
1031          * make sure that once we start defragging an extent, we keep on
1032          * defragging it
1033          */
1034         if (start < *defrag_end)
1035                 return 1;
1036
1037         *skip = 0;
1038
1039         em = defrag_lookup_extent(inode, start);
1040         if (!em)
1041                 return 0;
1042
1043         /* this will cover holes, and inline extents */
1044         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1045                 ret = 0;
1046                 goto out;
1047         }
1048
1049         next_mergeable = defrag_check_next_extent(inode, em);
1050
1051         /*
1052          * we hit a real extent, if it is big or the next extent is not a
1053          * real extent, don't bother defragging it
1054          */
1055         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1056             (em->len >= thresh || !next_mergeable))
1057                 ret = 0;
1058 out:
1059         /*
1060          * last_len ends up being a counter of how many bytes we've defragged.
1061          * every time we choose not to defrag an extent, we reset *last_len
1062          * so that the next tiny extent will force a defrag.
1063          *
1064          * The end result of this is that tiny extents before a single big
1065          * extent will force at least part of that big extent to be defragged.
1066          */
1067         if (ret) {
1068                 *defrag_end = extent_map_end(em);
1069         } else {
1070                 *last_len = 0;
1071                 *skip = extent_map_end(em);
1072                 *defrag_end = 0;
1073         }
1074
1075         free_extent_map(em);
1076         return ret;
1077 }
1078
1079 /*
1080  * it doesn't do much good to defrag one or two pages
1081  * at a time.  This pulls in a nice chunk of pages
1082  * to COW and defrag.
1083  *
1084  * It also makes sure the delalloc code has enough
1085  * dirty data to avoid making new small extents as part
1086  * of the defrag
1087  *
1088  * It's a good idea to start RA on this range
1089  * before calling this.
1090  */
1091 static int cluster_pages_for_defrag(struct inode *inode,
1092                                     struct page **pages,
1093                                     unsigned long start_index,
1094                                     unsigned long num_pages)
1095 {
1096         unsigned long file_end;
1097         u64 isize = i_size_read(inode);
1098         u64 page_start;
1099         u64 page_end;
1100         u64 page_cnt;
1101         int ret;
1102         int i;
1103         int i_done;
1104         struct btrfs_ordered_extent *ordered;
1105         struct extent_state *cached_state = NULL;
1106         struct extent_io_tree *tree;
1107         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1108
1109         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1110         if (!isize || start_index > file_end)
1111                 return 0;
1112
1113         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1114
1115         ret = btrfs_delalloc_reserve_space(inode,
1116                                            page_cnt << PAGE_CACHE_SHIFT);
1117         if (ret)
1118                 return ret;
1119         i_done = 0;
1120         tree = &BTRFS_I(inode)->io_tree;
1121
1122         /* step one, lock all the pages */
1123         for (i = 0; i < page_cnt; i++) {
1124                 struct page *page;
1125 again:
1126                 page = find_or_create_page(inode->i_mapping,
1127                                            start_index + i, mask);
1128                 if (!page)
1129                         break;
1130
1131                 page_start = page_offset(page);
1132                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1133                 while (1) {
1134                         lock_extent_bits(tree, page_start, page_end,
1135                                          0, &cached_state);
1136                         ordered = btrfs_lookup_ordered_extent(inode,
1137                                                               page_start);
1138                         unlock_extent_cached(tree, page_start, page_end,
1139                                              &cached_state, GFP_NOFS);
1140                         if (!ordered)
1141                                 break;
1142
1143                         unlock_page(page);
1144                         btrfs_start_ordered_extent(inode, ordered, 1);
1145                         btrfs_put_ordered_extent(ordered);
1146                         lock_page(page);
1147                         /*
1148                          * we unlocked the page above, so we need check if
1149                          * it was released or not.
1150                          */
1151                         if (page->mapping != inode->i_mapping) {
1152                                 unlock_page(page);
1153                                 page_cache_release(page);
1154                                 goto again;
1155                         }
1156                 }
1157
1158                 if (!PageUptodate(page)) {
1159                         btrfs_readpage(NULL, page);
1160                         lock_page(page);
1161                         if (!PageUptodate(page)) {
1162                                 unlock_page(page);
1163                                 page_cache_release(page);
1164                                 ret = -EIO;
1165                                 break;
1166                         }
1167                 }
1168
1169                 if (page->mapping != inode->i_mapping) {
1170                         unlock_page(page);
1171                         page_cache_release(page);
1172                         goto again;
1173                 }
1174
1175                 pages[i] = page;
1176                 i_done++;
1177         }
1178         if (!i_done || ret)
1179                 goto out;
1180
1181         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1182                 goto out;
1183
1184         /*
1185          * so now we have a nice long stream of locked
1186          * and up to date pages, lets wait on them
1187          */
1188         for (i = 0; i < i_done; i++)
1189                 wait_on_page_writeback(pages[i]);
1190
1191         page_start = page_offset(pages[0]);
1192         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1193
1194         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1195                          page_start, page_end - 1, 0, &cached_state);
1196         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1197                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1198                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1199                           &cached_state, GFP_NOFS);
1200
1201         if (i_done != page_cnt) {
1202                 spin_lock(&BTRFS_I(inode)->lock);
1203                 BTRFS_I(inode)->outstanding_extents++;
1204                 spin_unlock(&BTRFS_I(inode)->lock);
1205                 btrfs_delalloc_release_space(inode,
1206                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1207         }
1208
1209
1210         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1211                           &cached_state, GFP_NOFS);
1212
1213         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1214                              page_start, page_end - 1, &cached_state,
1215                              GFP_NOFS);
1216
1217         for (i = 0; i < i_done; i++) {
1218                 clear_page_dirty_for_io(pages[i]);
1219                 ClearPageChecked(pages[i]);
1220                 set_page_extent_mapped(pages[i]);
1221                 set_page_dirty(pages[i]);
1222                 unlock_page(pages[i]);
1223                 page_cache_release(pages[i]);
1224         }
1225         return i_done;
1226 out:
1227         for (i = 0; i < i_done; i++) {
1228                 unlock_page(pages[i]);
1229                 page_cache_release(pages[i]);
1230         }
1231         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1232         return ret;
1233
1234 }
1235
1236 int btrfs_defrag_file(struct inode *inode, struct file *file,
1237                       struct btrfs_ioctl_defrag_range_args *range,
1238                       u64 newer_than, unsigned long max_to_defrag)
1239 {
1240         struct btrfs_root *root = BTRFS_I(inode)->root;
1241         struct file_ra_state *ra = NULL;
1242         unsigned long last_index;
1243         u64 isize = i_size_read(inode);
1244         u64 last_len = 0;
1245         u64 skip = 0;
1246         u64 defrag_end = 0;
1247         u64 newer_off = range->start;
1248         unsigned long i;
1249         unsigned long ra_index = 0;
1250         int ret;
1251         int defrag_count = 0;
1252         int compress_type = BTRFS_COMPRESS_ZLIB;
1253         int extent_thresh = range->extent_thresh;
1254         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1255         unsigned long cluster = max_cluster;
1256         u64 new_align = ~((u64)128 * 1024 - 1);
1257         struct page **pages = NULL;
1258
1259         if (isize == 0)
1260                 return 0;
1261
1262         if (range->start >= isize)
1263                 return -EINVAL;
1264
1265         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1266                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1267                         return -EINVAL;
1268                 if (range->compress_type)
1269                         compress_type = range->compress_type;
1270         }
1271
1272         if (extent_thresh == 0)
1273                 extent_thresh = 256 * 1024;
1274
1275         /*
1276          * if we were not given a file, allocate a readahead
1277          * context
1278          */
1279         if (!file) {
1280                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1281                 if (!ra)
1282                         return -ENOMEM;
1283                 file_ra_state_init(ra, inode->i_mapping);
1284         } else {
1285                 ra = &file->f_ra;
1286         }
1287
1288         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1289                         GFP_NOFS);
1290         if (!pages) {
1291                 ret = -ENOMEM;
1292                 goto out_ra;
1293         }
1294
1295         /* find the last page to defrag */
1296         if (range->start + range->len > range->start) {
1297                 last_index = min_t(u64, isize - 1,
1298                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1299         } else {
1300                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1301         }
1302
1303         if (newer_than) {
1304                 ret = find_new_extents(root, inode, newer_than,
1305                                        &newer_off, 64 * 1024);
1306                 if (!ret) {
1307                         range->start = newer_off;
1308                         /*
1309                          * we always align our defrag to help keep
1310                          * the extents in the file evenly spaced
1311                          */
1312                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1313                 } else
1314                         goto out_ra;
1315         } else {
1316                 i = range->start >> PAGE_CACHE_SHIFT;
1317         }
1318         if (!max_to_defrag)
1319                 max_to_defrag = last_index + 1;
1320
1321         /*
1322          * make writeback starts from i, so the defrag range can be
1323          * written sequentially.
1324          */
1325         if (i < inode->i_mapping->writeback_index)
1326                 inode->i_mapping->writeback_index = i;
1327
1328         while (i <= last_index && defrag_count < max_to_defrag &&
1329                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1330                 PAGE_CACHE_SHIFT)) {
1331                 /*
1332                  * make sure we stop running if someone unmounts
1333                  * the FS
1334                  */
1335                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1336                         break;
1337
1338                 if (btrfs_defrag_cancelled(root->fs_info)) {
1339                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1340                         ret = -EAGAIN;
1341                         break;
1342                 }
1343
1344                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1345                                          extent_thresh, &last_len, &skip,
1346                                          &defrag_end, range->flags &
1347                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1348                         unsigned long next;
1349                         /*
1350                          * the should_defrag function tells us how much to skip
1351                          * bump our counter by the suggested amount
1352                          */
1353                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1354                         i = max(i + 1, next);
1355                         continue;
1356                 }
1357
1358                 if (!newer_than) {
1359                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1360                                    PAGE_CACHE_SHIFT) - i;
1361                         cluster = min(cluster, max_cluster);
1362                 } else {
1363                         cluster = max_cluster;
1364                 }
1365
1366                 if (i + cluster > ra_index) {
1367                         ra_index = max(i, ra_index);
1368                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1369                                        cluster);
1370                         ra_index += max_cluster;
1371                 }
1372
1373                 mutex_lock(&inode->i_mutex);
1374                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1375                         BTRFS_I(inode)->force_compress = compress_type;
1376                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1377                 if (ret < 0) {
1378                         mutex_unlock(&inode->i_mutex);
1379                         goto out_ra;
1380                 }
1381
1382                 defrag_count += ret;
1383                 balance_dirty_pages_ratelimited(inode->i_mapping);
1384                 mutex_unlock(&inode->i_mutex);
1385
1386                 if (newer_than) {
1387                         if (newer_off == (u64)-1)
1388                                 break;
1389
1390                         if (ret > 0)
1391                                 i += ret;
1392
1393                         newer_off = max(newer_off + 1,
1394                                         (u64)i << PAGE_CACHE_SHIFT);
1395
1396                         ret = find_new_extents(root, inode,
1397                                                newer_than, &newer_off,
1398                                                64 * 1024);
1399                         if (!ret) {
1400                                 range->start = newer_off;
1401                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1402                         } else {
1403                                 break;
1404                         }
1405                 } else {
1406                         if (ret > 0) {
1407                                 i += ret;
1408                                 last_len += ret << PAGE_CACHE_SHIFT;
1409                         } else {
1410                                 i++;
1411                                 last_len = 0;
1412                         }
1413                 }
1414         }
1415
1416         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1417                 filemap_flush(inode->i_mapping);
1418                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1419                              &BTRFS_I(inode)->runtime_flags))
1420                         filemap_flush(inode->i_mapping);
1421         }
1422
1423         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1424                 /* the filemap_flush will queue IO into the worker threads, but
1425                  * we have to make sure the IO is actually started and that
1426                  * ordered extents get created before we return
1427                  */
1428                 atomic_inc(&root->fs_info->async_submit_draining);
1429                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1430                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1431                         wait_event(root->fs_info->async_submit_wait,
1432                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1433                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1434                 }
1435                 atomic_dec(&root->fs_info->async_submit_draining);
1436         }
1437
1438         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1439                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1440         }
1441
1442         ret = defrag_count;
1443
1444 out_ra:
1445         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1446                 mutex_lock(&inode->i_mutex);
1447                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1448                 mutex_unlock(&inode->i_mutex);
1449         }
1450         if (!file)
1451                 kfree(ra);
1452         kfree(pages);
1453         return ret;
1454 }
1455
1456 static noinline int btrfs_ioctl_resize(struct file *file,
1457                                         void __user *arg)
1458 {
1459         u64 new_size;
1460         u64 old_size;
1461         u64 devid = 1;
1462         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1463         struct btrfs_ioctl_vol_args *vol_args;
1464         struct btrfs_trans_handle *trans;
1465         struct btrfs_device *device = NULL;
1466         char *sizestr;
1467         char *devstr = NULL;
1468         int ret = 0;
1469         int mod = 0;
1470
1471         if (!capable(CAP_SYS_ADMIN))
1472                 return -EPERM;
1473
1474         ret = mnt_want_write_file(file);
1475         if (ret)
1476                 return ret;
1477
1478         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1479                         1)) {
1480                 mnt_drop_write_file(file);
1481                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1482         }
1483
1484         mutex_lock(&root->fs_info->volume_mutex);
1485         vol_args = memdup_user(arg, sizeof(*vol_args));
1486         if (IS_ERR(vol_args)) {
1487                 ret = PTR_ERR(vol_args);
1488                 goto out;
1489         }
1490
1491         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1492
1493         sizestr = vol_args->name;
1494         devstr = strchr(sizestr, ':');
1495         if (devstr) {
1496                 char *end;
1497                 sizestr = devstr + 1;
1498                 *devstr = '\0';
1499                 devstr = vol_args->name;
1500                 devid = simple_strtoull(devstr, &end, 10);
1501                 if (!devid) {
1502                         ret = -EINVAL;
1503                         goto out_free;
1504                 }
1505                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1506         }
1507
1508         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1509         if (!device) {
1510                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1511                        devid);
1512                 ret = -ENODEV;
1513                 goto out_free;
1514         }
1515
1516         if (!device->writeable) {
1517                 btrfs_info(root->fs_info,
1518                            "resizer unable to apply on readonly device %llu",
1519                        devid);
1520                 ret = -EPERM;
1521                 goto out_free;
1522         }
1523
1524         if (!strcmp(sizestr, "max"))
1525                 new_size = device->bdev->bd_inode->i_size;
1526         else {
1527                 if (sizestr[0] == '-') {
1528                         mod = -1;
1529                         sizestr++;
1530                 } else if (sizestr[0] == '+') {
1531                         mod = 1;
1532                         sizestr++;
1533                 }
1534                 new_size = memparse(sizestr, NULL);
1535                 if (new_size == 0) {
1536                         ret = -EINVAL;
1537                         goto out_free;
1538                 }
1539         }
1540
1541         if (device->is_tgtdev_for_dev_replace) {
1542                 ret = -EPERM;
1543                 goto out_free;
1544         }
1545
1546         old_size = device->total_bytes;
1547
1548         if (mod < 0) {
1549                 if (new_size > old_size) {
1550                         ret = -EINVAL;
1551                         goto out_free;
1552                 }
1553                 new_size = old_size - new_size;
1554         } else if (mod > 0) {
1555                 if (new_size > ULLONG_MAX - old_size) {
1556                         ret = -EINVAL;
1557                         goto out_free;
1558                 }
1559                 new_size = old_size + new_size;
1560         }
1561
1562         if (new_size < 256 * 1024 * 1024) {
1563                 ret = -EINVAL;
1564                 goto out_free;
1565         }
1566         if (new_size > device->bdev->bd_inode->i_size) {
1567                 ret = -EFBIG;
1568                 goto out_free;
1569         }
1570
1571         do_div(new_size, root->sectorsize);
1572         new_size *= root->sectorsize;
1573
1574         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1575                       rcu_str_deref(device->name), new_size);
1576
1577         if (new_size > old_size) {
1578                 trans = btrfs_start_transaction(root, 0);
1579                 if (IS_ERR(trans)) {
1580                         ret = PTR_ERR(trans);
1581                         goto out_free;
1582                 }
1583                 ret = btrfs_grow_device(trans, device, new_size);
1584                 btrfs_commit_transaction(trans, root);
1585         } else if (new_size < old_size) {
1586                 ret = btrfs_shrink_device(device, new_size);
1587         } /* equal, nothing need to do */
1588
1589 out_free:
1590         kfree(vol_args);
1591 out:
1592         mutex_unlock(&root->fs_info->volume_mutex);
1593         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1594         mnt_drop_write_file(file);
1595         return ret;
1596 }
1597
1598 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1599                                 char *name, unsigned long fd, int subvol,
1600                                 u64 *transid, bool readonly,
1601                                 struct btrfs_qgroup_inherit *inherit)
1602 {
1603         int namelen;
1604         int ret = 0;
1605
1606         ret = mnt_want_write_file(file);
1607         if (ret)
1608                 goto out;
1609
1610         namelen = strlen(name);
1611         if (strchr(name, '/')) {
1612                 ret = -EINVAL;
1613                 goto out_drop_write;
1614         }
1615
1616         if (name[0] == '.' &&
1617            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1618                 ret = -EEXIST;
1619                 goto out_drop_write;
1620         }
1621
1622         if (subvol) {
1623                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1624                                      NULL, transid, readonly, inherit);
1625         } else {
1626                 struct fd src = fdget(fd);
1627                 struct inode *src_inode;
1628                 if (!src.file) {
1629                         ret = -EINVAL;
1630                         goto out_drop_write;
1631                 }
1632
1633                 src_inode = file_inode(src.file);
1634                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1635                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1636                                    "Snapshot src from another FS");
1637                         ret = -EXDEV;
1638                 } else if (!inode_owner_or_capable(src_inode)) {
1639                         /*
1640                          * Subvolume creation is not restricted, but snapshots
1641                          * are limited to own subvolumes only
1642                          */
1643                         ret = -EPERM;
1644                 } else {
1645                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1646                                              BTRFS_I(src_inode)->root,
1647                                              transid, readonly, inherit);
1648                 }
1649                 fdput(src);
1650         }
1651 out_drop_write:
1652         mnt_drop_write_file(file);
1653 out:
1654         return ret;
1655 }
1656
1657 static noinline int btrfs_ioctl_snap_create(struct file *file,
1658                                             void __user *arg, int subvol)
1659 {
1660         struct btrfs_ioctl_vol_args *vol_args;
1661         int ret;
1662
1663         vol_args = memdup_user(arg, sizeof(*vol_args));
1664         if (IS_ERR(vol_args))
1665                 return PTR_ERR(vol_args);
1666         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1667
1668         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1669                                               vol_args->fd, subvol,
1670                                               NULL, false, NULL);
1671
1672         kfree(vol_args);
1673         return ret;
1674 }
1675
1676 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1677                                                void __user *arg, int subvol)
1678 {
1679         struct btrfs_ioctl_vol_args_v2 *vol_args;
1680         int ret;
1681         u64 transid = 0;
1682         u64 *ptr = NULL;
1683         bool readonly = false;
1684         struct btrfs_qgroup_inherit *inherit = NULL;
1685
1686         vol_args = memdup_user(arg, sizeof(*vol_args));
1687         if (IS_ERR(vol_args))
1688                 return PTR_ERR(vol_args);
1689         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1690
1691         if (vol_args->flags &
1692             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1693               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1694                 ret = -EOPNOTSUPP;
1695                 goto out;
1696         }
1697
1698         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1699                 ptr = &transid;
1700         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1701                 readonly = true;
1702         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1703                 if (vol_args->size > PAGE_CACHE_SIZE) {
1704                         ret = -EINVAL;
1705                         goto out;
1706                 }
1707                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1708                 if (IS_ERR(inherit)) {
1709                         ret = PTR_ERR(inherit);
1710                         goto out;
1711                 }
1712         }
1713
1714         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1715                                               vol_args->fd, subvol, ptr,
1716                                               readonly, inherit);
1717
1718         if (ret == 0 && ptr &&
1719             copy_to_user(arg +
1720                          offsetof(struct btrfs_ioctl_vol_args_v2,
1721                                   transid), ptr, sizeof(*ptr)))
1722                 ret = -EFAULT;
1723 out:
1724         kfree(vol_args);
1725         kfree(inherit);
1726         return ret;
1727 }
1728
1729 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1730                                                 void __user *arg)
1731 {
1732         struct inode *inode = file_inode(file);
1733         struct btrfs_root *root = BTRFS_I(inode)->root;
1734         int ret = 0;
1735         u64 flags = 0;
1736
1737         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1738                 return -EINVAL;
1739
1740         down_read(&root->fs_info->subvol_sem);
1741         if (btrfs_root_readonly(root))
1742                 flags |= BTRFS_SUBVOL_RDONLY;
1743         up_read(&root->fs_info->subvol_sem);
1744
1745         if (copy_to_user(arg, &flags, sizeof(flags)))
1746                 ret = -EFAULT;
1747
1748         return ret;
1749 }
1750
1751 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1752                                               void __user *arg)
1753 {
1754         struct inode *inode = file_inode(file);
1755         struct btrfs_root *root = BTRFS_I(inode)->root;
1756         struct btrfs_trans_handle *trans;
1757         u64 root_flags;
1758         u64 flags;
1759         int ret = 0;
1760
1761         if (!inode_owner_or_capable(inode))
1762                 return -EPERM;
1763
1764         ret = mnt_want_write_file(file);
1765         if (ret)
1766                 goto out;
1767
1768         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1769                 ret = -EINVAL;
1770                 goto out_drop_write;
1771         }
1772
1773         if (copy_from_user(&flags, arg, sizeof(flags))) {
1774                 ret = -EFAULT;
1775                 goto out_drop_write;
1776         }
1777
1778         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1779                 ret = -EINVAL;
1780                 goto out_drop_write;
1781         }
1782
1783         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1784                 ret = -EOPNOTSUPP;
1785                 goto out_drop_write;
1786         }
1787
1788         down_write(&root->fs_info->subvol_sem);
1789
1790         /* nothing to do */
1791         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1792                 goto out_drop_sem;
1793
1794         root_flags = btrfs_root_flags(&root->root_item);
1795         if (flags & BTRFS_SUBVOL_RDONLY) {
1796                 btrfs_set_root_flags(&root->root_item,
1797                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1798         } else {
1799                 /*
1800                  * Block RO -> RW transition if this subvolume is involved in
1801                  * send
1802                  */
1803                 spin_lock(&root->root_item_lock);
1804                 if (root->send_in_progress == 0) {
1805                         btrfs_set_root_flags(&root->root_item,
1806                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1807                         spin_unlock(&root->root_item_lock);
1808                 } else {
1809                         spin_unlock(&root->root_item_lock);
1810                         btrfs_warn(root->fs_info,
1811                         "Attempt to set subvolume %llu read-write during send",
1812                                         root->root_key.objectid);
1813                         ret = -EPERM;
1814                         goto out_drop_sem;
1815                 }
1816         }
1817
1818         trans = btrfs_start_transaction(root, 1);
1819         if (IS_ERR(trans)) {
1820                 ret = PTR_ERR(trans);
1821                 goto out_reset;
1822         }
1823
1824         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1825                                 &root->root_key, &root->root_item);
1826
1827         btrfs_commit_transaction(trans, root);
1828 out_reset:
1829         if (ret)
1830                 btrfs_set_root_flags(&root->root_item, root_flags);
1831 out_drop_sem:
1832         up_write(&root->fs_info->subvol_sem);
1833 out_drop_write:
1834         mnt_drop_write_file(file);
1835 out:
1836         return ret;
1837 }
1838
1839 /*
1840  * helper to check if the subvolume references other subvolumes
1841  */
1842 static noinline int may_destroy_subvol(struct btrfs_root *root)
1843 {
1844         struct btrfs_path *path;
1845         struct btrfs_dir_item *di;
1846         struct btrfs_key key;
1847         u64 dir_id;
1848         int ret;
1849
1850         path = btrfs_alloc_path();
1851         if (!path)
1852                 return -ENOMEM;
1853
1854         /* Make sure this root isn't set as the default subvol */
1855         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1856         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1857                                    dir_id, "default", 7, 0);
1858         if (di && !IS_ERR(di)) {
1859                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1860                 if (key.objectid == root->root_key.objectid) {
1861                         ret = -ENOTEMPTY;
1862                         goto out;
1863                 }
1864                 btrfs_release_path(path);
1865         }
1866
1867         key.objectid = root->root_key.objectid;
1868         key.type = BTRFS_ROOT_REF_KEY;
1869         key.offset = (u64)-1;
1870
1871         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1872                                 &key, path, 0, 0);
1873         if (ret < 0)
1874                 goto out;
1875         BUG_ON(ret == 0);
1876
1877         ret = 0;
1878         if (path->slots[0] > 0) {
1879                 path->slots[0]--;
1880                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1881                 if (key.objectid == root->root_key.objectid &&
1882                     key.type == BTRFS_ROOT_REF_KEY)
1883                         ret = -ENOTEMPTY;
1884         }
1885 out:
1886         btrfs_free_path(path);
1887         return ret;
1888 }
1889
1890 static noinline int key_in_sk(struct btrfs_key *key,
1891                               struct btrfs_ioctl_search_key *sk)
1892 {
1893         struct btrfs_key test;
1894         int ret;
1895
1896         test.objectid = sk->min_objectid;
1897         test.type = sk->min_type;
1898         test.offset = sk->min_offset;
1899
1900         ret = btrfs_comp_cpu_keys(key, &test);
1901         if (ret < 0)
1902                 return 0;
1903
1904         test.objectid = sk->max_objectid;
1905         test.type = sk->max_type;
1906         test.offset = sk->max_offset;
1907
1908         ret = btrfs_comp_cpu_keys(key, &test);
1909         if (ret > 0)
1910                 return 0;
1911         return 1;
1912 }
1913
1914 static noinline int copy_to_sk(struct btrfs_root *root,
1915                                struct btrfs_path *path,
1916                                struct btrfs_key *key,
1917                                struct btrfs_ioctl_search_key *sk,
1918                                char *buf,
1919                                unsigned long *sk_offset,
1920                                int *num_found)
1921 {
1922         u64 found_transid;
1923         struct extent_buffer *leaf;
1924         struct btrfs_ioctl_search_header sh;
1925         unsigned long item_off;
1926         unsigned long item_len;
1927         int nritems;
1928         int i;
1929         int slot;
1930         int ret = 0;
1931
1932         leaf = path->nodes[0];
1933         slot = path->slots[0];
1934         nritems = btrfs_header_nritems(leaf);
1935
1936         if (btrfs_header_generation(leaf) > sk->max_transid) {
1937                 i = nritems;
1938                 goto advance_key;
1939         }
1940         found_transid = btrfs_header_generation(leaf);
1941
1942         for (i = slot; i < nritems; i++) {
1943                 item_off = btrfs_item_ptr_offset(leaf, i);
1944                 item_len = btrfs_item_size_nr(leaf, i);
1945
1946                 btrfs_item_key_to_cpu(leaf, key, i);
1947                 if (!key_in_sk(key, sk))
1948                         continue;
1949
1950                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1951                         item_len = 0;
1952
1953                 if (sizeof(sh) + item_len + *sk_offset >
1954                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1955                         ret = 1;
1956                         goto overflow;
1957                 }
1958
1959                 sh.objectid = key->objectid;
1960                 sh.offset = key->offset;
1961                 sh.type = key->type;
1962                 sh.len = item_len;
1963                 sh.transid = found_transid;
1964
1965                 /* copy search result header */
1966                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1967                 *sk_offset += sizeof(sh);
1968
1969                 if (item_len) {
1970                         char *p = buf + *sk_offset;
1971                         /* copy the item */
1972                         read_extent_buffer(leaf, p,
1973                                            item_off, item_len);
1974                         *sk_offset += item_len;
1975                 }
1976                 (*num_found)++;
1977
1978                 if (*num_found >= sk->nr_items)
1979                         break;
1980         }
1981 advance_key:
1982         ret = 0;
1983         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1984                 key->offset++;
1985         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1986                 key->offset = 0;
1987                 key->type++;
1988         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1989                 key->offset = 0;
1990                 key->type = 0;
1991                 key->objectid++;
1992         } else
1993                 ret = 1;
1994 overflow:
1995         return ret;
1996 }
1997
1998 static noinline int search_ioctl(struct inode *inode,
1999                                  struct btrfs_ioctl_search_args *args)
2000 {
2001         struct btrfs_root *root;
2002         struct btrfs_key key;
2003         struct btrfs_path *path;
2004         struct btrfs_ioctl_search_key *sk = &args->key;
2005         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2006         int ret;
2007         int num_found = 0;
2008         unsigned long sk_offset = 0;
2009
2010         path = btrfs_alloc_path();
2011         if (!path)
2012                 return -ENOMEM;
2013
2014         if (sk->tree_id == 0) {
2015                 /* search the root of the inode that was passed */
2016                 root = BTRFS_I(inode)->root;
2017         } else {
2018                 key.objectid = sk->tree_id;
2019                 key.type = BTRFS_ROOT_ITEM_KEY;
2020                 key.offset = (u64)-1;
2021                 root = btrfs_read_fs_root_no_name(info, &key);
2022                 if (IS_ERR(root)) {
2023                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2024                                sk->tree_id);
2025                         btrfs_free_path(path);
2026                         return -ENOENT;
2027                 }
2028         }
2029
2030         key.objectid = sk->min_objectid;
2031         key.type = sk->min_type;
2032         key.offset = sk->min_offset;
2033
2034         path->keep_locks = 1;
2035
2036         while (1) {
2037                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2038                 if (ret != 0) {
2039                         if (ret > 0)
2040                                 ret = 0;
2041                         goto err;
2042                 }
2043                 ret = copy_to_sk(root, path, &key, sk, args->buf,
2044                                  &sk_offset, &num_found);
2045                 btrfs_release_path(path);
2046                 if (ret || num_found >= sk->nr_items)
2047                         break;
2048
2049         }
2050         ret = 0;
2051 err:
2052         sk->nr_items = num_found;
2053         btrfs_free_path(path);
2054         return ret;
2055 }
2056
2057 static noinline int btrfs_ioctl_tree_search(struct file *file,
2058                                            void __user *argp)
2059 {
2060          struct btrfs_ioctl_search_args *args;
2061          struct inode *inode;
2062          int ret;
2063
2064         if (!capable(CAP_SYS_ADMIN))
2065                 return -EPERM;
2066
2067         args = memdup_user(argp, sizeof(*args));
2068         if (IS_ERR(args))
2069                 return PTR_ERR(args);
2070
2071         inode = file_inode(file);
2072         ret = search_ioctl(inode, args);
2073         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2074                 ret = -EFAULT;
2075         kfree(args);
2076         return ret;
2077 }
2078
2079 /*
2080  * Search INODE_REFs to identify path name of 'dirid' directory
2081  * in a 'tree_id' tree. and sets path name to 'name'.
2082  */
2083 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2084                                 u64 tree_id, u64 dirid, char *name)
2085 {
2086         struct btrfs_root *root;
2087         struct btrfs_key key;
2088         char *ptr;
2089         int ret = -1;
2090         int slot;
2091         int len;
2092         int total_len = 0;
2093         struct btrfs_inode_ref *iref;
2094         struct extent_buffer *l;
2095         struct btrfs_path *path;
2096
2097         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2098                 name[0]='\0';
2099                 return 0;
2100         }
2101
2102         path = btrfs_alloc_path();
2103         if (!path)
2104                 return -ENOMEM;
2105
2106         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2107
2108         key.objectid = tree_id;
2109         key.type = BTRFS_ROOT_ITEM_KEY;
2110         key.offset = (u64)-1;
2111         root = btrfs_read_fs_root_no_name(info, &key);
2112         if (IS_ERR(root)) {
2113                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2114                 ret = -ENOENT;
2115                 goto out;
2116         }
2117
2118         key.objectid = dirid;
2119         key.type = BTRFS_INODE_REF_KEY;
2120         key.offset = (u64)-1;
2121
2122         while (1) {
2123                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2124                 if (ret < 0)
2125                         goto out;
2126                 else if (ret > 0) {
2127                         ret = btrfs_previous_item(root, path, dirid,
2128                                                   BTRFS_INODE_REF_KEY);
2129                         if (ret < 0)
2130                                 goto out;
2131                         else if (ret > 0) {
2132                                 ret = -ENOENT;
2133                                 goto out;
2134                         }
2135                 }
2136
2137                 l = path->nodes[0];
2138                 slot = path->slots[0];
2139                 btrfs_item_key_to_cpu(l, &key, slot);
2140
2141                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2142                 len = btrfs_inode_ref_name_len(l, iref);
2143                 ptr -= len + 1;
2144                 total_len += len + 1;
2145                 if (ptr < name) {
2146                         ret = -ENAMETOOLONG;
2147                         goto out;
2148                 }
2149
2150                 *(ptr + len) = '/';
2151                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2152
2153                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2154                         break;
2155
2156                 btrfs_release_path(path);
2157                 key.objectid = key.offset;
2158                 key.offset = (u64)-1;
2159                 dirid = key.objectid;
2160         }
2161         memmove(name, ptr, total_len);
2162         name[total_len] = '\0';
2163         ret = 0;
2164 out:
2165         btrfs_free_path(path);
2166         return ret;
2167 }
2168
2169 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2170                                            void __user *argp)
2171 {
2172          struct btrfs_ioctl_ino_lookup_args *args;
2173          struct inode *inode;
2174          int ret;
2175
2176         if (!capable(CAP_SYS_ADMIN))
2177                 return -EPERM;
2178
2179         args = memdup_user(argp, sizeof(*args));
2180         if (IS_ERR(args))
2181                 return PTR_ERR(args);
2182
2183         inode = file_inode(file);
2184
2185         if (args->treeid == 0)
2186                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2187
2188         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2189                                         args->treeid, args->objectid,
2190                                         args->name);
2191
2192         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2193                 ret = -EFAULT;
2194
2195         kfree(args);
2196         return ret;
2197 }
2198
2199 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2200                                              void __user *arg)
2201 {
2202         struct dentry *parent = file->f_path.dentry;
2203         struct dentry *dentry;
2204         struct inode *dir = parent->d_inode;
2205         struct inode *inode;
2206         struct btrfs_root *root = BTRFS_I(dir)->root;
2207         struct btrfs_root *dest = NULL;
2208         struct btrfs_ioctl_vol_args *vol_args;
2209         struct btrfs_trans_handle *trans;
2210         struct btrfs_block_rsv block_rsv;
2211         u64 qgroup_reserved;
2212         int namelen;
2213         int ret;
2214         int err = 0;
2215
2216         vol_args = memdup_user(arg, sizeof(*vol_args));
2217         if (IS_ERR(vol_args))
2218                 return PTR_ERR(vol_args);
2219
2220         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2221         namelen = strlen(vol_args->name);
2222         if (strchr(vol_args->name, '/') ||
2223             strncmp(vol_args->name, "..", namelen) == 0) {
2224                 err = -EINVAL;
2225                 goto out;
2226         }
2227
2228         err = mnt_want_write_file(file);
2229         if (err)
2230                 goto out;
2231
2232         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2233         if (err == -EINTR)
2234                 goto out_drop_write;
2235         dentry = lookup_one_len(vol_args->name, parent, namelen);
2236         if (IS_ERR(dentry)) {
2237                 err = PTR_ERR(dentry);
2238                 goto out_unlock_dir;
2239         }
2240
2241         if (!dentry->d_inode) {
2242                 err = -ENOENT;
2243                 goto out_dput;
2244         }
2245
2246         inode = dentry->d_inode;
2247         dest = BTRFS_I(inode)->root;
2248         if (!capable(CAP_SYS_ADMIN)) {
2249                 /*
2250                  * Regular user.  Only allow this with a special mount
2251                  * option, when the user has write+exec access to the
2252                  * subvol root, and when rmdir(2) would have been
2253                  * allowed.
2254                  *
2255                  * Note that this is _not_ check that the subvol is
2256                  * empty or doesn't contain data that we wouldn't
2257                  * otherwise be able to delete.
2258                  *
2259                  * Users who want to delete empty subvols should try
2260                  * rmdir(2).
2261                  */
2262                 err = -EPERM;
2263                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2264                         goto out_dput;
2265
2266                 /*
2267                  * Do not allow deletion if the parent dir is the same
2268                  * as the dir to be deleted.  That means the ioctl
2269                  * must be called on the dentry referencing the root
2270                  * of the subvol, not a random directory contained
2271                  * within it.
2272                  */
2273                 err = -EINVAL;
2274                 if (root == dest)
2275                         goto out_dput;
2276
2277                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2278                 if (err)
2279                         goto out_dput;
2280         }
2281
2282         /* check if subvolume may be deleted by a user */
2283         err = btrfs_may_delete(dir, dentry, 1);
2284         if (err)
2285                 goto out_dput;
2286
2287         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2288                 err = -EINVAL;
2289                 goto out_dput;
2290         }
2291
2292         mutex_lock(&inode->i_mutex);
2293         err = d_invalidate(dentry);
2294         if (err)
2295                 goto out_unlock;
2296
2297         down_write(&root->fs_info->subvol_sem);
2298
2299         err = may_destroy_subvol(dest);
2300         if (err)
2301                 goto out_up_write;
2302
2303         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2304         /*
2305          * One for dir inode, two for dir entries, two for root
2306          * ref/backref.
2307          */
2308         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2309                                                5, &qgroup_reserved, true);
2310         if (err)
2311                 goto out_up_write;
2312
2313         trans = btrfs_start_transaction(root, 0);
2314         if (IS_ERR(trans)) {
2315                 err = PTR_ERR(trans);
2316                 goto out_release;
2317         }
2318         trans->block_rsv = &block_rsv;
2319         trans->bytes_reserved = block_rsv.size;
2320
2321         ret = btrfs_unlink_subvol(trans, root, dir,
2322                                 dest->root_key.objectid,
2323                                 dentry->d_name.name,
2324                                 dentry->d_name.len);
2325         if (ret) {
2326                 err = ret;
2327                 btrfs_abort_transaction(trans, root, ret);
2328                 goto out_end_trans;
2329         }
2330
2331         btrfs_record_root_in_trans(trans, dest);
2332
2333         memset(&dest->root_item.drop_progress, 0,
2334                 sizeof(dest->root_item.drop_progress));
2335         dest->root_item.drop_level = 0;
2336         btrfs_set_root_refs(&dest->root_item, 0);
2337
2338         if (!xchg(&dest->orphan_item_inserted, 1)) {
2339                 ret = btrfs_insert_orphan_item(trans,
2340                                         root->fs_info->tree_root,
2341                                         dest->root_key.objectid);
2342                 if (ret) {
2343                         btrfs_abort_transaction(trans, root, ret);
2344                         err = ret;
2345                         goto out_end_trans;
2346                 }
2347         }
2348
2349         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2350                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2351                                   dest->root_key.objectid);
2352         if (ret && ret != -ENOENT) {
2353                 btrfs_abort_transaction(trans, root, ret);
2354                 err = ret;
2355                 goto out_end_trans;
2356         }
2357         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2358                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2359                                           dest->root_item.received_uuid,
2360                                           BTRFS_UUID_KEY_RECEIVED_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         }
2368
2369 out_end_trans:
2370         trans->block_rsv = NULL;
2371         trans->bytes_reserved = 0;
2372         ret = btrfs_end_transaction(trans, root);
2373         if (ret && !err)
2374                 err = ret;
2375         inode->i_flags |= S_DEAD;
2376 out_release:
2377         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2378 out_up_write:
2379         up_write(&root->fs_info->subvol_sem);
2380 out_unlock:
2381         mutex_unlock(&inode->i_mutex);
2382         if (!err) {
2383                 shrink_dcache_sb(root->fs_info->sb);
2384                 btrfs_invalidate_inodes(dest);
2385                 d_delete(dentry);
2386
2387                 /* the last ref */
2388                 if (dest->cache_inode) {
2389                         iput(dest->cache_inode);
2390                         dest->cache_inode = NULL;
2391                 }
2392         }
2393 out_dput:
2394         dput(dentry);
2395 out_unlock_dir:
2396         mutex_unlock(&dir->i_mutex);
2397 out_drop_write:
2398         mnt_drop_write_file(file);
2399 out:
2400         kfree(vol_args);
2401         return err;
2402 }
2403
2404 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2405 {
2406         struct inode *inode = file_inode(file);
2407         struct btrfs_root *root = BTRFS_I(inode)->root;
2408         struct btrfs_ioctl_defrag_range_args *range;
2409         int ret;
2410
2411         ret = mnt_want_write_file(file);
2412         if (ret)
2413                 return ret;
2414
2415         if (btrfs_root_readonly(root)) {
2416                 ret = -EROFS;
2417                 goto out;
2418         }
2419
2420         switch (inode->i_mode & S_IFMT) {
2421         case S_IFDIR:
2422                 if (!capable(CAP_SYS_ADMIN)) {
2423                         ret = -EPERM;
2424                         goto out;
2425                 }
2426                 ret = btrfs_defrag_root(root);
2427                 if (ret)
2428                         goto out;
2429                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2430                 break;
2431         case S_IFREG:
2432                 if (!(file->f_mode & FMODE_WRITE)) {
2433                         ret = -EINVAL;
2434                         goto out;
2435                 }
2436
2437                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2438                 if (!range) {
2439                         ret = -ENOMEM;
2440                         goto out;
2441                 }
2442
2443                 if (argp) {
2444                         if (copy_from_user(range, argp,
2445                                            sizeof(*range))) {
2446                                 ret = -EFAULT;
2447                                 kfree(range);
2448                                 goto out;
2449                         }
2450                         /* compression requires us to start the IO */
2451                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2452                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2453                                 range->extent_thresh = (u32)-1;
2454                         }
2455                 } else {
2456                         /* the rest are all set to zero by kzalloc */
2457                         range->len = (u64)-1;
2458                 }
2459                 ret = btrfs_defrag_file(file_inode(file), file,
2460                                         range, 0, 0);
2461                 if (ret > 0)
2462                         ret = 0;
2463                 kfree(range);
2464                 break;
2465         default:
2466                 ret = -EINVAL;
2467         }
2468 out:
2469         mnt_drop_write_file(file);
2470         return ret;
2471 }
2472
2473 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2474 {
2475         struct btrfs_ioctl_vol_args *vol_args;
2476         int ret;
2477
2478         if (!capable(CAP_SYS_ADMIN))
2479                 return -EPERM;
2480
2481         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2482                         1)) {
2483                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2484         }
2485
2486         mutex_lock(&root->fs_info->volume_mutex);
2487         vol_args = memdup_user(arg, sizeof(*vol_args));
2488         if (IS_ERR(vol_args)) {
2489                 ret = PTR_ERR(vol_args);
2490                 goto out;
2491         }
2492
2493         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2494         ret = btrfs_init_new_device(root, vol_args->name);
2495
2496         kfree(vol_args);
2497 out:
2498         mutex_unlock(&root->fs_info->volume_mutex);
2499         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2500         return ret;
2501 }
2502
2503 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2504 {
2505         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2506         struct btrfs_ioctl_vol_args *vol_args;
2507         int ret;
2508
2509         if (!capable(CAP_SYS_ADMIN))
2510                 return -EPERM;
2511
2512         ret = mnt_want_write_file(file);
2513         if (ret)
2514                 return ret;
2515
2516         vol_args = memdup_user(arg, sizeof(*vol_args));
2517         if (IS_ERR(vol_args)) {
2518                 ret = PTR_ERR(vol_args);
2519                 goto out;
2520         }
2521
2522         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2523
2524         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2525                         1)) {
2526                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2527                 goto out;
2528         }
2529
2530         mutex_lock(&root->fs_info->volume_mutex);
2531         ret = btrfs_rm_device(root, vol_args->name);
2532         mutex_unlock(&root->fs_info->volume_mutex);
2533         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2534
2535 out:
2536         kfree(vol_args);
2537         mnt_drop_write_file(file);
2538         return ret;
2539 }
2540
2541 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2542 {
2543         struct btrfs_ioctl_fs_info_args *fi_args;
2544         struct btrfs_device *device;
2545         struct btrfs_device *next;
2546         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2547         int ret = 0;
2548
2549         if (!capable(CAP_SYS_ADMIN))
2550                 return -EPERM;
2551
2552         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2553         if (!fi_args)
2554                 return -ENOMEM;
2555
2556         mutex_lock(&fs_devices->device_list_mutex);
2557         fi_args->num_devices = fs_devices->num_devices;
2558         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2559
2560         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2561                 if (device->devid > fi_args->max_id)
2562                         fi_args->max_id = device->devid;
2563         }
2564         mutex_unlock(&fs_devices->device_list_mutex);
2565
2566         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2567                 ret = -EFAULT;
2568
2569         kfree(fi_args);
2570         return ret;
2571 }
2572
2573 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2574 {
2575         struct btrfs_ioctl_dev_info_args *di_args;
2576         struct btrfs_device *dev;
2577         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2578         int ret = 0;
2579         char *s_uuid = NULL;
2580
2581         if (!capable(CAP_SYS_ADMIN))
2582                 return -EPERM;
2583
2584         di_args = memdup_user(arg, sizeof(*di_args));
2585         if (IS_ERR(di_args))
2586                 return PTR_ERR(di_args);
2587
2588         if (!btrfs_is_empty_uuid(di_args->uuid))
2589                 s_uuid = di_args->uuid;
2590
2591         mutex_lock(&fs_devices->device_list_mutex);
2592         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2593
2594         if (!dev) {
2595                 ret = -ENODEV;
2596                 goto out;
2597         }
2598
2599         di_args->devid = dev->devid;
2600         di_args->bytes_used = dev->bytes_used;
2601         di_args->total_bytes = dev->total_bytes;
2602         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2603         if (dev->name) {
2604                 struct rcu_string *name;
2605
2606                 rcu_read_lock();
2607                 name = rcu_dereference(dev->name);
2608                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2609                 rcu_read_unlock();
2610                 di_args->path[sizeof(di_args->path) - 1] = 0;
2611         } else {
2612                 di_args->path[0] = '\0';
2613         }
2614
2615 out:
2616         mutex_unlock(&fs_devices->device_list_mutex);
2617         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2618                 ret = -EFAULT;
2619
2620         kfree(di_args);
2621         return ret;
2622 }
2623
2624 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2625 {
2626         struct page *page;
2627         pgoff_t index;
2628         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2629
2630         index = off >> PAGE_CACHE_SHIFT;
2631
2632         page = grab_cache_page(inode->i_mapping, index);
2633         if (!page)
2634                 return NULL;
2635
2636         if (!PageUptodate(page)) {
2637                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2638                                                  0))
2639                         return NULL;
2640                 lock_page(page);
2641                 if (!PageUptodate(page)) {
2642                         unlock_page(page);
2643                         page_cache_release(page);
2644                         return NULL;
2645                 }
2646         }
2647         unlock_page(page);
2648
2649         return page;
2650 }
2651
2652 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2653 {
2654         /* do any pending delalloc/csum calc on src, one way or
2655            another, and lock file content */
2656         while (1) {
2657                 struct btrfs_ordered_extent *ordered;
2658                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2659                 ordered = btrfs_lookup_first_ordered_extent(inode,
2660                                                             off + len - 1);
2661                 if (!ordered &&
2662                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2663                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2664                         break;
2665                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2666                 if (ordered)
2667                         btrfs_put_ordered_extent(ordered);
2668                 btrfs_wait_ordered_range(inode, off, len);
2669         }
2670 }
2671
2672 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2673                                 struct inode *inode2, u64 loff2, u64 len)
2674 {
2675         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2676         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2677
2678         mutex_unlock(&inode1->i_mutex);
2679         mutex_unlock(&inode2->i_mutex);
2680 }
2681
2682 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2683                               struct inode *inode2, u64 loff2, u64 len)
2684 {
2685         if (inode1 < inode2) {
2686                 swap(inode1, inode2);
2687                 swap(loff1, loff2);
2688         }
2689
2690         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2691         lock_extent_range(inode1, loff1, len);
2692         if (inode1 != inode2) {
2693                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2694                 lock_extent_range(inode2, loff2, len);
2695         }
2696 }
2697
2698 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2699                           u64 dst_loff, u64 len)
2700 {
2701         int ret = 0;
2702         struct page *src_page, *dst_page;
2703         unsigned int cmp_len = PAGE_CACHE_SIZE;
2704         void *addr, *dst_addr;
2705
2706         while (len) {
2707                 if (len < PAGE_CACHE_SIZE)
2708                         cmp_len = len;
2709
2710                 src_page = extent_same_get_page(src, loff);
2711                 if (!src_page)
2712                         return -EINVAL;
2713                 dst_page = extent_same_get_page(dst, dst_loff);
2714                 if (!dst_page) {
2715                         page_cache_release(src_page);
2716                         return -EINVAL;
2717                 }
2718                 addr = kmap_atomic(src_page);
2719                 dst_addr = kmap_atomic(dst_page);
2720
2721                 flush_dcache_page(src_page);
2722                 flush_dcache_page(dst_page);
2723
2724                 if (memcmp(addr, dst_addr, cmp_len))
2725                         ret = BTRFS_SAME_DATA_DIFFERS;
2726
2727                 kunmap_atomic(addr);
2728                 kunmap_atomic(dst_addr);
2729                 page_cache_release(src_page);
2730                 page_cache_release(dst_page);
2731
2732                 if (ret)
2733                         break;
2734
2735                 loff += cmp_len;
2736                 dst_loff += cmp_len;
2737                 len -= cmp_len;
2738         }
2739
2740         return ret;
2741 }
2742
2743 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2744 {
2745         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2746
2747         if (off + len > inode->i_size || off + len < off)
2748                 return -EINVAL;
2749         /* Check that we are block aligned - btrfs_clone() requires this */
2750         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2751                 return -EINVAL;
2752
2753         return 0;
2754 }
2755
2756 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2757                              struct inode *dst, u64 dst_loff)
2758 {
2759         int ret;
2760
2761         /*
2762          * btrfs_clone() can't handle extents in the same file
2763          * yet. Once that works, we can drop this check and replace it
2764          * with a check for the same inode, but overlapping extents.
2765          */
2766         if (src == dst)
2767                 return -EINVAL;
2768
2769         btrfs_double_lock(src, loff, dst, dst_loff, len);
2770
2771         ret = extent_same_check_offsets(src, loff, len);
2772         if (ret)
2773                 goto out_unlock;
2774
2775         ret = extent_same_check_offsets(dst, dst_loff, len);
2776         if (ret)
2777                 goto out_unlock;
2778
2779         /* don't make the dst file partly checksummed */
2780         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2781             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2782                 ret = -EINVAL;
2783                 goto out_unlock;
2784         }
2785
2786         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2787         if (ret == 0)
2788                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2789
2790 out_unlock:
2791         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2792
2793         return ret;
2794 }
2795
2796 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2797
2798 static long btrfs_ioctl_file_extent_same(struct file *file,
2799                                          void __user *argp)
2800 {
2801         struct btrfs_ioctl_same_args tmp;
2802         struct btrfs_ioctl_same_args *same;
2803         struct btrfs_ioctl_same_extent_info *info;
2804         struct inode *src = file->f_dentry->d_inode;
2805         struct file *dst_file = NULL;
2806         struct inode *dst;
2807         u64 off;
2808         u64 len;
2809         int i;
2810         int ret;
2811         unsigned long size;
2812         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2813         bool is_admin = capable(CAP_SYS_ADMIN);
2814
2815         if (!(file->f_mode & FMODE_READ))
2816                 return -EINVAL;
2817
2818         ret = mnt_want_write_file(file);
2819         if (ret)
2820                 return ret;
2821
2822         if (copy_from_user(&tmp,
2823                            (struct btrfs_ioctl_same_args __user *)argp,
2824                            sizeof(tmp))) {
2825                 ret = -EFAULT;
2826                 goto out;
2827         }
2828
2829         size = sizeof(tmp) +
2830                 tmp.dest_count * sizeof(struct btrfs_ioctl_same_extent_info);
2831
2832         same = memdup_user((struct btrfs_ioctl_same_args __user *)argp, size);
2833
2834         if (IS_ERR(same)) {
2835                 ret = PTR_ERR(same);
2836                 goto out;
2837         }
2838
2839         off = same->logical_offset;
2840         len = same->length;
2841
2842         /*
2843          * Limit the total length we will dedupe for each operation.
2844          * This is intended to bound the total time spent in this
2845          * ioctl to something sane.
2846          */
2847         if (len > BTRFS_MAX_DEDUPE_LEN)
2848                 len = BTRFS_MAX_DEDUPE_LEN;
2849
2850         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2851                 /*
2852                  * Btrfs does not support blocksize < page_size. As a
2853                  * result, btrfs_cmp_data() won't correctly handle
2854                  * this situation without an update.
2855                  */
2856                 ret = -EINVAL;
2857                 goto out;
2858         }
2859
2860         ret = -EISDIR;
2861         if (S_ISDIR(src->i_mode))
2862                 goto out;
2863
2864         ret = -EACCES;
2865         if (!S_ISREG(src->i_mode))
2866                 goto out;
2867
2868         /* pre-format output fields to sane values */
2869         for (i = 0; i < same->dest_count; i++) {
2870                 same->info[i].bytes_deduped = 0ULL;
2871                 same->info[i].status = 0;
2872         }
2873
2874         ret = 0;
2875         for (i = 0; i < same->dest_count; i++) {
2876                 info = &same->info[i];
2877
2878                 dst_file = fget(info->fd);
2879                 if (!dst_file) {
2880                         info->status = -EBADF;
2881                         goto next;
2882                 }
2883
2884                 if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2885                         info->status = -EINVAL;
2886                         goto next;
2887                 }
2888
2889                 info->status = -EXDEV;
2890                 if (file->f_path.mnt != dst_file->f_path.mnt)
2891                         goto next;
2892
2893                 dst = dst_file->f_dentry->d_inode;
2894                 if (src->i_sb != dst->i_sb)
2895                         goto next;
2896
2897                 if (S_ISDIR(dst->i_mode)) {
2898                         info->status = -EISDIR;
2899                         goto next;
2900                 }
2901
2902                 if (!S_ISREG(dst->i_mode)) {
2903                         info->status = -EACCES;
2904                         goto next;
2905                 }
2906
2907                 info->status = btrfs_extent_same(src, off, len, dst,
2908                                                 info->logical_offset);
2909                 if (info->status == 0)
2910                         info->bytes_deduped += len;
2911
2912 next:
2913                 if (dst_file)
2914                         fput(dst_file);
2915         }
2916
2917         ret = copy_to_user(argp, same, size);
2918         if (ret)
2919                 ret = -EFAULT;
2920
2921 out:
2922         mnt_drop_write_file(file);
2923         return ret;
2924 }
2925
2926 /**
2927  * btrfs_clone() - clone a range from inode file to another
2928  *
2929  * @src: Inode to clone from
2930  * @inode: Inode to clone to
2931  * @off: Offset within source to start clone from
2932  * @olen: Original length, passed by user, of range to clone
2933  * @olen_aligned: Block-aligned value of olen, extent_same uses
2934  *               identical values here
2935  * @destoff: Offset within @inode to start clone
2936  */
2937 static int btrfs_clone(struct inode *src, struct inode *inode,
2938                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2939 {
2940         struct btrfs_root *root = BTRFS_I(inode)->root;
2941         struct btrfs_path *path = NULL;
2942         struct extent_buffer *leaf;
2943         struct btrfs_trans_handle *trans;
2944         char *buf = NULL;
2945         struct btrfs_key key;
2946         u32 nritems;
2947         int slot;
2948         int ret;
2949         u64 len = olen_aligned;
2950
2951         ret = -ENOMEM;
2952         buf = vmalloc(btrfs_level_size(root, 0));
2953         if (!buf)
2954                 return ret;
2955
2956         path = btrfs_alloc_path();
2957         if (!path) {
2958                 vfree(buf);
2959                 return ret;
2960         }
2961
2962         path->reada = 2;
2963         /* clone data */
2964         key.objectid = btrfs_ino(src);
2965         key.type = BTRFS_EXTENT_DATA_KEY;
2966         key.offset = 0;
2967
2968         while (1) {
2969                 /*
2970                  * note the key will change type as we walk through the
2971                  * tree.
2972                  */
2973                 path->leave_spinning = 1;
2974                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2975                                 0, 0);
2976                 if (ret < 0)
2977                         goto out;
2978
2979                 nritems = btrfs_header_nritems(path->nodes[0]);
2980 process_slot:
2981                 if (path->slots[0] >= nritems) {
2982                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2983                         if (ret < 0)
2984                                 goto out;
2985                         if (ret > 0)
2986                                 break;
2987                         nritems = btrfs_header_nritems(path->nodes[0]);
2988                 }
2989                 leaf = path->nodes[0];
2990                 slot = path->slots[0];
2991
2992                 btrfs_item_key_to_cpu(leaf, &key, slot);
2993                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2994                     key.objectid != btrfs_ino(src))
2995                         break;
2996
2997                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2998                         struct btrfs_file_extent_item *extent;
2999                         int type;
3000                         u32 size;
3001                         struct btrfs_key new_key;
3002                         u64 disko = 0, diskl = 0;
3003                         u64 datao = 0, datal = 0;
3004                         u8 comp;
3005                         u64 endoff;
3006
3007                         extent = btrfs_item_ptr(leaf, slot,
3008                                                 struct btrfs_file_extent_item);
3009                         comp = btrfs_file_extent_compression(leaf, extent);
3010                         type = btrfs_file_extent_type(leaf, extent);
3011                         if (type == BTRFS_FILE_EXTENT_REG ||
3012                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3013                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3014                                                                       extent);
3015                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3016                                                                  extent);
3017                                 datao = btrfs_file_extent_offset(leaf, extent);
3018                                 datal = btrfs_file_extent_num_bytes(leaf,
3019                                                                     extent);
3020                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3021                                 /* take upper bound, may be compressed */
3022                                 datal = btrfs_file_extent_ram_bytes(leaf,
3023                                                                     extent);
3024                         }
3025
3026                         if (key.offset + datal <= off ||
3027                             key.offset >= off + len - 1) {
3028                                 path->slots[0]++;
3029                                 goto process_slot;
3030                         }
3031
3032                         size = btrfs_item_size_nr(leaf, slot);
3033                         read_extent_buffer(leaf, buf,
3034                                            btrfs_item_ptr_offset(leaf, slot),
3035                                            size);
3036
3037                         btrfs_release_path(path);
3038                         path->leave_spinning = 0;
3039
3040                         memcpy(&new_key, &key, sizeof(new_key));
3041                         new_key.objectid = btrfs_ino(inode);
3042                         if (off <= key.offset)
3043                                 new_key.offset = key.offset + destoff - off;
3044                         else
3045                                 new_key.offset = destoff;
3046
3047                         /*
3048                          * 1 - adjusting old extent (we may have to split it)
3049                          * 1 - add new extent
3050                          * 1 - inode update
3051                          */
3052                         trans = btrfs_start_transaction(root, 3);
3053                         if (IS_ERR(trans)) {
3054                                 ret = PTR_ERR(trans);
3055                                 goto out;
3056                         }
3057
3058                         if (type == BTRFS_FILE_EXTENT_REG ||
3059                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3060                                 /*
3061                                  *    a  | --- range to clone ---|  b
3062                                  * | ------------- extent ------------- |
3063                                  */
3064
3065                                 /* substract range b */
3066                                 if (key.offset + datal > off + len)
3067                                         datal = off + len - key.offset;
3068
3069                                 /* substract range a */
3070                                 if (off > key.offset) {
3071                                         datao += off - key.offset;
3072                                         datal -= off - key.offset;
3073                                 }
3074
3075                                 ret = btrfs_drop_extents(trans, root, inode,
3076                                                          new_key.offset,
3077                                                          new_key.offset + datal,
3078                                                          1);
3079                                 if (ret) {
3080                                         btrfs_abort_transaction(trans, root,
3081                                                                 ret);
3082                                         btrfs_end_transaction(trans, root);
3083                                         goto out;
3084                                 }
3085
3086                                 ret = btrfs_insert_empty_item(trans, root, path,
3087                                                               &new_key, size);
3088                                 if (ret) {
3089                                         btrfs_abort_transaction(trans, root,
3090                                                                 ret);
3091                                         btrfs_end_transaction(trans, root);
3092                                         goto out;
3093                                 }
3094
3095                                 leaf = path->nodes[0];
3096                                 slot = path->slots[0];
3097                                 write_extent_buffer(leaf, buf,
3098                                             btrfs_item_ptr_offset(leaf, slot),
3099                                             size);
3100
3101                                 extent = btrfs_item_ptr(leaf, slot,
3102                                                 struct btrfs_file_extent_item);
3103
3104                                 /* disko == 0 means it's a hole */
3105                                 if (!disko)
3106                                         datao = 0;
3107
3108                                 btrfs_set_file_extent_offset(leaf, extent,
3109                                                              datao);
3110                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3111                                                                 datal);
3112                                 if (disko) {
3113                                         inode_add_bytes(inode, datal);
3114                                         ret = btrfs_inc_extent_ref(trans, root,
3115                                                         disko, diskl, 0,
3116                                                         root->root_key.objectid,
3117                                                         btrfs_ino(inode),
3118                                                         new_key.offset - datao,
3119                                                         0);
3120                                         if (ret) {
3121                                                 btrfs_abort_transaction(trans,
3122                                                                         root,
3123                                                                         ret);
3124                                                 btrfs_end_transaction(trans,
3125                                                                       root);
3126                                                 goto out;
3127
3128                                         }
3129                                 }
3130                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3131                                 u64 skip = 0;
3132                                 u64 trim = 0;
3133                                 if (off > key.offset) {
3134                                         skip = off - key.offset;
3135                                         new_key.offset += skip;
3136                                 }
3137
3138                                 if (key.offset + datal > off + len)
3139                                         trim = key.offset + datal - (off + len);
3140
3141                                 if (comp && (skip || trim)) {
3142                                         ret = -EINVAL;
3143                                         btrfs_end_transaction(trans, root);
3144                                         goto out;
3145                                 }
3146                                 size -= skip + trim;
3147                                 datal -= skip + trim;
3148
3149                                 ret = btrfs_drop_extents(trans, root, inode,
3150                                                          new_key.offset,
3151                                                          new_key.offset + datal,
3152                                                          1);
3153                                 if (ret) {
3154                                         btrfs_abort_transaction(trans, root,
3155                                                                 ret);
3156                                         btrfs_end_transaction(trans, root);
3157                                         goto out;
3158                                 }
3159
3160                                 ret = btrfs_insert_empty_item(trans, root, path,
3161                                                               &new_key, size);
3162                                 if (ret) {
3163                                         btrfs_abort_transaction(trans, root,
3164                                                                 ret);
3165                                         btrfs_end_transaction(trans, root);
3166                                         goto out;
3167                                 }
3168
3169                                 if (skip) {
3170                                         u32 start =
3171                                           btrfs_file_extent_calc_inline_size(0);
3172                                         memmove(buf+start, buf+start+skip,
3173                                                 datal);
3174                                 }
3175
3176                                 leaf = path->nodes[0];
3177                                 slot = path->slots[0];
3178                                 write_extent_buffer(leaf, buf,
3179                                             btrfs_item_ptr_offset(leaf, slot),
3180                                             size);
3181                                 inode_add_bytes(inode, datal);
3182                         }
3183
3184                         btrfs_mark_buffer_dirty(leaf);
3185                         btrfs_release_path(path);
3186
3187                         inode_inc_iversion(inode);
3188                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3189
3190                         /*
3191                          * we round up to the block size at eof when
3192                          * determining which extents to clone above,
3193                          * but shouldn't round up the file size
3194                          */
3195                         endoff = new_key.offset + datal;
3196                         if (endoff > destoff+olen)
3197                                 endoff = destoff+olen;
3198                         if (endoff > inode->i_size)
3199                                 btrfs_i_size_write(inode, endoff);
3200
3201                         ret = btrfs_update_inode(trans, root, inode);
3202                         if (ret) {
3203                                 btrfs_abort_transaction(trans, root, ret);
3204                                 btrfs_end_transaction(trans, root);
3205                                 goto out;
3206                         }
3207                         ret = btrfs_end_transaction(trans, root);
3208                 }
3209                 btrfs_release_path(path);
3210                 key.offset++;
3211         }
3212         ret = 0;
3213
3214 out:
3215         btrfs_release_path(path);
3216         btrfs_free_path(path);
3217         vfree(buf);
3218         return ret;
3219 }
3220
3221 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3222                                        u64 off, u64 olen, u64 destoff)
3223 {
3224         struct inode *inode = file_inode(file);
3225         struct btrfs_root *root = BTRFS_I(inode)->root;
3226         struct fd src_file;
3227         struct inode *src;
3228         int ret;
3229         u64 len = olen;
3230         u64 bs = root->fs_info->sb->s_blocksize;
3231         int same_inode = 0;
3232
3233         /*
3234          * TODO:
3235          * - split compressed inline extents.  annoying: we need to
3236          *   decompress into destination's address_space (the file offset
3237          *   may change, so source mapping won't do), then recompress (or
3238          *   otherwise reinsert) a subrange.
3239          * - allow ranges within the same file to be cloned (provided
3240          *   they don't overlap)?
3241          */
3242
3243         /* the destination must be opened for writing */
3244         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3245                 return -EINVAL;
3246
3247         if (btrfs_root_readonly(root))
3248                 return -EROFS;
3249
3250         ret = mnt_want_write_file(file);
3251         if (ret)
3252                 return ret;
3253
3254         src_file = fdget(srcfd);
3255         if (!src_file.file) {
3256                 ret = -EBADF;
3257                 goto out_drop_write;
3258         }
3259
3260         ret = -EXDEV;
3261         if (src_file.file->f_path.mnt != file->f_path.mnt)
3262                 goto out_fput;
3263
3264         src = file_inode(src_file.file);
3265
3266         ret = -EINVAL;
3267         if (src == inode)
3268                 same_inode = 1;
3269
3270         /* the src must be open for reading */
3271         if (!(src_file.file->f_mode & FMODE_READ))
3272                 goto out_fput;
3273
3274         /* don't make the dst file partly checksummed */
3275         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3276             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3277                 goto out_fput;
3278
3279         ret = -EISDIR;
3280         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3281                 goto out_fput;
3282
3283         ret = -EXDEV;
3284         if (src->i_sb != inode->i_sb)
3285                 goto out_fput;
3286
3287         if (!same_inode) {
3288                 if (inode < src) {
3289                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3290                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3291                 } else {
3292                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3293                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3294                 }
3295         } else {
3296                 mutex_lock(&src->i_mutex);
3297         }
3298
3299         /* determine range to clone */
3300         ret = -EINVAL;
3301         if (off + len > src->i_size || off + len < off)
3302                 goto out_unlock;
3303         if (len == 0)
3304                 olen = len = src->i_size - off;
3305         /* if we extend to eof, continue to block boundary */
3306         if (off + len == src->i_size)
3307                 len = ALIGN(src->i_size, bs) - off;
3308
3309         /* verify the end result is block aligned */
3310         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3311             !IS_ALIGNED(destoff, bs))
3312                 goto out_unlock;
3313
3314         /* verify if ranges are overlapped within the same file */
3315         if (same_inode) {
3316                 if (destoff + len > off && destoff < off + len)
3317                         goto out_unlock;
3318         }
3319
3320         if (destoff > inode->i_size) {
3321                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3322                 if (ret)
3323                         goto out_unlock;
3324         }
3325
3326         /* truncate page cache pages from target inode range */
3327         truncate_inode_pages_range(&inode->i_data, destoff,
3328                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3329
3330         lock_extent_range(src, off, len);
3331
3332         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3333
3334         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3335 out_unlock:
3336         if (!same_inode) {
3337                 if (inode < src) {
3338                         mutex_unlock(&src->i_mutex);
3339                         mutex_unlock(&inode->i_mutex);
3340                 } else {
3341                         mutex_unlock(&inode->i_mutex);
3342                         mutex_unlock(&src->i_mutex);
3343                 }
3344         } else {
3345                 mutex_unlock(&src->i_mutex);
3346         }
3347 out_fput:
3348         fdput(src_file);
3349 out_drop_write:
3350         mnt_drop_write_file(file);
3351         return ret;
3352 }
3353
3354 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3355 {
3356         struct btrfs_ioctl_clone_range_args args;
3357
3358         if (copy_from_user(&args, argp, sizeof(args)))
3359                 return -EFAULT;
3360         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3361                                  args.src_length, args.dest_offset);
3362 }
3363
3364 /*
3365  * there are many ways the trans_start and trans_end ioctls can lead
3366  * to deadlocks.  They should only be used by applications that
3367  * basically own the machine, and have a very in depth understanding
3368  * of all the possible deadlocks and enospc problems.
3369  */
3370 static long btrfs_ioctl_trans_start(struct file *file)
3371 {
3372         struct inode *inode = file_inode(file);
3373         struct btrfs_root *root = BTRFS_I(inode)->root;
3374         struct btrfs_trans_handle *trans;
3375         int ret;
3376
3377         ret = -EPERM;
3378         if (!capable(CAP_SYS_ADMIN))
3379                 goto out;
3380
3381         ret = -EINPROGRESS;
3382         if (file->private_data)
3383                 goto out;
3384
3385         ret = -EROFS;
3386         if (btrfs_root_readonly(root))
3387                 goto out;
3388
3389         ret = mnt_want_write_file(file);
3390         if (ret)
3391                 goto out;
3392
3393         atomic_inc(&root->fs_info->open_ioctl_trans);
3394
3395         ret = -ENOMEM;
3396         trans = btrfs_start_ioctl_transaction(root);
3397         if (IS_ERR(trans))
3398                 goto out_drop;
3399
3400         file->private_data = trans;
3401         return 0;
3402
3403 out_drop:
3404         atomic_dec(&root->fs_info->open_ioctl_trans);
3405         mnt_drop_write_file(file);
3406 out:
3407         return ret;
3408 }
3409
3410 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3411 {
3412         struct inode *inode = file_inode(file);
3413         struct btrfs_root *root = BTRFS_I(inode)->root;
3414         struct btrfs_root *new_root;
3415         struct btrfs_dir_item *di;
3416         struct btrfs_trans_handle *trans;
3417         struct btrfs_path *path;
3418         struct btrfs_key location;
3419         struct btrfs_disk_key disk_key;
3420         u64 objectid = 0;
3421         u64 dir_id;
3422         int ret;
3423
3424         if (!capable(CAP_SYS_ADMIN))
3425                 return -EPERM;
3426
3427         ret = mnt_want_write_file(file);
3428         if (ret)
3429                 return ret;
3430
3431         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3432                 ret = -EFAULT;
3433                 goto out;
3434         }
3435
3436         if (!objectid)
3437                 objectid = BTRFS_FS_TREE_OBJECTID;
3438
3439         location.objectid = objectid;
3440         location.type = BTRFS_ROOT_ITEM_KEY;
3441         location.offset = (u64)-1;
3442
3443         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3444         if (IS_ERR(new_root)) {
3445                 ret = PTR_ERR(new_root);
3446                 goto out;
3447         }
3448
3449         path = btrfs_alloc_path();
3450         if (!path) {
3451                 ret = -ENOMEM;
3452                 goto out;
3453         }
3454         path->leave_spinning = 1;
3455
3456         trans = btrfs_start_transaction(root, 1);
3457         if (IS_ERR(trans)) {
3458                 btrfs_free_path(path);
3459                 ret = PTR_ERR(trans);
3460                 goto out;
3461         }
3462
3463         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3464         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3465                                    dir_id, "default", 7, 1);
3466         if (IS_ERR_OR_NULL(di)) {
3467                 btrfs_free_path(path);
3468                 btrfs_end_transaction(trans, root);
3469                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3470                            "item, this isn't going to work");
3471                 ret = -ENOENT;
3472                 goto out;
3473         }
3474
3475         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3476         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3477         btrfs_mark_buffer_dirty(path->nodes[0]);
3478         btrfs_free_path(path);
3479
3480         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3481         btrfs_end_transaction(trans, root);
3482 out:
3483         mnt_drop_write_file(file);
3484         return ret;
3485 }
3486
3487 void btrfs_get_block_group_info(struct list_head *groups_list,
3488                                 struct btrfs_ioctl_space_info *space)
3489 {
3490         struct btrfs_block_group_cache *block_group;
3491
3492         space->total_bytes = 0;
3493         space->used_bytes = 0;
3494         space->flags = 0;
3495         list_for_each_entry(block_group, groups_list, list) {
3496                 space->flags = block_group->flags;
3497                 space->total_bytes += block_group->key.offset;
3498                 space->used_bytes +=
3499                         btrfs_block_group_used(&block_group->item);
3500         }
3501 }
3502
3503 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3504 {
3505         struct btrfs_ioctl_space_args space_args;
3506         struct btrfs_ioctl_space_info space;
3507         struct btrfs_ioctl_space_info *dest;
3508         struct btrfs_ioctl_space_info *dest_orig;
3509         struct btrfs_ioctl_space_info __user *user_dest;
3510         struct btrfs_space_info *info;
3511         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3512                        BTRFS_BLOCK_GROUP_SYSTEM,
3513                        BTRFS_BLOCK_GROUP_METADATA,
3514                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3515         int num_types = 4;
3516         int alloc_size;
3517         int ret = 0;
3518         u64 slot_count = 0;
3519         int i, c;
3520
3521         if (copy_from_user(&space_args,
3522                            (struct btrfs_ioctl_space_args __user *)arg,
3523                            sizeof(space_args)))
3524                 return -EFAULT;
3525
3526         for (i = 0; i < num_types; i++) {
3527                 struct btrfs_space_info *tmp;
3528
3529                 info = NULL;
3530                 rcu_read_lock();
3531                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3532                                         list) {
3533                         if (tmp->flags == types[i]) {
3534                                 info = tmp;
3535                                 break;
3536                         }
3537                 }
3538                 rcu_read_unlock();
3539
3540                 if (!info)
3541                         continue;
3542
3543                 down_read(&info->groups_sem);
3544                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3545                         if (!list_empty(&info->block_groups[c]))
3546                                 slot_count++;
3547                 }
3548                 up_read(&info->groups_sem);
3549         }
3550
3551         /* space_slots == 0 means they are asking for a count */
3552         if (space_args.space_slots == 0) {
3553                 space_args.total_spaces = slot_count;
3554                 goto out;
3555         }
3556
3557         slot_count = min_t(u64, space_args.space_slots, slot_count);
3558
3559         alloc_size = sizeof(*dest) * slot_count;
3560
3561         /* we generally have at most 6 or so space infos, one for each raid
3562          * level.  So, a whole page should be more than enough for everyone
3563          */
3564         if (alloc_size > PAGE_CACHE_SIZE)
3565                 return -ENOMEM;
3566
3567         space_args.total_spaces = 0;
3568         dest = kmalloc(alloc_size, GFP_NOFS);
3569         if (!dest)
3570                 return -ENOMEM;
3571         dest_orig = dest;
3572
3573         /* now we have a buffer to copy into */
3574         for (i = 0; i < num_types; i++) {
3575                 struct btrfs_space_info *tmp;
3576
3577                 if (!slot_count)
3578                         break;
3579
3580                 info = NULL;
3581                 rcu_read_lock();
3582                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3583                                         list) {
3584                         if (tmp->flags == types[i]) {
3585                                 info = tmp;
3586                                 break;
3587                         }
3588                 }
3589                 rcu_read_unlock();
3590
3591                 if (!info)
3592                         continue;
3593                 down_read(&info->groups_sem);
3594                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3595                         if (!list_empty(&info->block_groups[c])) {
3596                                 btrfs_get_block_group_info(
3597                                         &info->block_groups[c], &space);
3598                                 memcpy(dest, &space, sizeof(space));
3599                                 dest++;
3600                                 space_args.total_spaces++;
3601                                 slot_count--;
3602                         }
3603                         if (!slot_count)
3604                                 break;
3605                 }
3606                 up_read(&info->groups_sem);
3607         }
3608
3609         user_dest = (struct btrfs_ioctl_space_info __user *)
3610                 (arg + sizeof(struct btrfs_ioctl_space_args));
3611
3612         if (copy_to_user(user_dest, dest_orig, alloc_size))
3613                 ret = -EFAULT;
3614
3615         kfree(dest_orig);
3616 out:
3617         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3618                 ret = -EFAULT;
3619
3620         return ret;
3621 }
3622
3623 /*
3624  * there are many ways the trans_start and trans_end ioctls can lead
3625  * to deadlocks.  They should only be used by applications that
3626  * basically own the machine, and have a very in depth understanding
3627  * of all the possible deadlocks and enospc problems.
3628  */
3629 long btrfs_ioctl_trans_end(struct file *file)
3630 {
3631         struct inode *inode = file_inode(file);
3632         struct btrfs_root *root = BTRFS_I(inode)->root;
3633         struct btrfs_trans_handle *trans;
3634
3635         trans = file->private_data;
3636         if (!trans)
3637                 return -EINVAL;
3638         file->private_data = NULL;
3639
3640         btrfs_end_transaction(trans, root);
3641
3642         atomic_dec(&root->fs_info->open_ioctl_trans);
3643
3644         mnt_drop_write_file(file);
3645         return 0;
3646 }
3647
3648 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3649                                             void __user *argp)
3650 {
3651         struct btrfs_trans_handle *trans;
3652         u64 transid;
3653         int ret;
3654
3655         trans = btrfs_attach_transaction_barrier(root);
3656         if (IS_ERR(trans)) {
3657                 if (PTR_ERR(trans) != -ENOENT)
3658                         return PTR_ERR(trans);
3659
3660                 /* No running transaction, don't bother */
3661                 transid = root->fs_info->last_trans_committed;
3662                 goto out;
3663         }
3664         transid = trans->transid;
3665         ret = btrfs_commit_transaction_async(trans, root, 0);
3666         if (ret) {
3667                 btrfs_end_transaction(trans, root);
3668                 return ret;
3669         }
3670 out:
3671         if (argp)
3672                 if (copy_to_user(argp, &transid, sizeof(transid)))
3673                         return -EFAULT;
3674         return 0;
3675 }
3676
3677 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3678                                            void __user *argp)
3679 {
3680         u64 transid;
3681
3682         if (argp) {
3683                 if (copy_from_user(&transid, argp, sizeof(transid)))
3684                         return -EFAULT;
3685         } else {
3686                 transid = 0;  /* current trans */
3687         }
3688         return btrfs_wait_for_commit(root, transid);
3689 }
3690
3691 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3692 {
3693         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3694         struct btrfs_ioctl_scrub_args *sa;
3695         int ret;
3696
3697         if (!capable(CAP_SYS_ADMIN))
3698                 return -EPERM;
3699
3700         sa = memdup_user(arg, sizeof(*sa));
3701         if (IS_ERR(sa))
3702                 return PTR_ERR(sa);
3703
3704         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3705                 ret = mnt_want_write_file(file);
3706                 if (ret)
3707                         goto out;
3708         }
3709
3710         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3711                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3712                               0);
3713
3714         if (copy_to_user(arg, sa, sizeof(*sa)))
3715                 ret = -EFAULT;
3716
3717         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3718                 mnt_drop_write_file(file);
3719 out:
3720         kfree(sa);
3721         return ret;
3722 }
3723
3724 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3725 {
3726         if (!capable(CAP_SYS_ADMIN))
3727                 return -EPERM;
3728
3729         return btrfs_scrub_cancel(root->fs_info);
3730 }
3731
3732 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3733                                        void __user *arg)
3734 {
3735         struct btrfs_ioctl_scrub_args *sa;
3736         int ret;
3737
3738         if (!capable(CAP_SYS_ADMIN))
3739                 return -EPERM;
3740
3741         sa = memdup_user(arg, sizeof(*sa));
3742         if (IS_ERR(sa))
3743                 return PTR_ERR(sa);
3744
3745         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3746
3747         if (copy_to_user(arg, sa, sizeof(*sa)))
3748                 ret = -EFAULT;
3749
3750         kfree(sa);
3751         return ret;
3752 }
3753
3754 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3755                                       void __user *arg)
3756 {
3757         struct btrfs_ioctl_get_dev_stats *sa;
3758         int ret;
3759
3760         sa = memdup_user(arg, sizeof(*sa));
3761         if (IS_ERR(sa))
3762                 return PTR_ERR(sa);
3763
3764         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3765                 kfree(sa);
3766                 return -EPERM;
3767         }
3768
3769         ret = btrfs_get_dev_stats(root, sa);
3770
3771         if (copy_to_user(arg, sa, sizeof(*sa)))
3772                 ret = -EFAULT;
3773
3774         kfree(sa);
3775         return ret;
3776 }
3777
3778 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3779 {
3780         struct btrfs_ioctl_dev_replace_args *p;
3781         int ret;
3782
3783         if (!capable(CAP_SYS_ADMIN))
3784                 return -EPERM;
3785
3786         p = memdup_user(arg, sizeof(*p));
3787         if (IS_ERR(p))
3788                 return PTR_ERR(p);
3789
3790         switch (p->cmd) {
3791         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3792                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
3793                         ret = -EROFS;
3794                         goto out;
3795                 }
3796                 if (atomic_xchg(
3797                         &root->fs_info->mutually_exclusive_operation_running,
3798                         1)) {
3799                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3800                 } else {
3801                         ret = btrfs_dev_replace_start(root, p);
3802                         atomic_set(
3803                          &root->fs_info->mutually_exclusive_operation_running,
3804                          0);
3805                 }
3806                 break;
3807         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3808                 btrfs_dev_replace_status(root->fs_info, p);
3809                 ret = 0;
3810                 break;
3811         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3812                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3813                 break;
3814         default:
3815                 ret = -EINVAL;
3816                 break;
3817         }
3818
3819         if (copy_to_user(arg, p, sizeof(*p)))
3820                 ret = -EFAULT;
3821 out:
3822         kfree(p);
3823         return ret;
3824 }
3825
3826 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3827 {
3828         int ret = 0;
3829         int i;
3830         u64 rel_ptr;
3831         int size;
3832         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3833         struct inode_fs_paths *ipath = NULL;
3834         struct btrfs_path *path;
3835
3836         if (!capable(CAP_DAC_READ_SEARCH))
3837                 return -EPERM;
3838
3839         path = btrfs_alloc_path();
3840         if (!path) {
3841                 ret = -ENOMEM;
3842                 goto out;
3843         }
3844
3845         ipa = memdup_user(arg, sizeof(*ipa));
3846         if (IS_ERR(ipa)) {
3847                 ret = PTR_ERR(ipa);
3848                 ipa = NULL;
3849                 goto out;
3850         }
3851
3852         size = min_t(u32, ipa->size, 4096);
3853         ipath = init_ipath(size, root, path);
3854         if (IS_ERR(ipath)) {
3855                 ret = PTR_ERR(ipath);
3856                 ipath = NULL;
3857                 goto out;
3858         }
3859
3860         ret = paths_from_inode(ipa->inum, ipath);
3861         if (ret < 0)
3862                 goto out;
3863
3864         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3865                 rel_ptr = ipath->fspath->val[i] -
3866                           (u64)(unsigned long)ipath->fspath->val;
3867                 ipath->fspath->val[i] = rel_ptr;
3868         }
3869
3870         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3871                            (void *)(unsigned long)ipath->fspath, size);
3872         if (ret) {
3873                 ret = -EFAULT;
3874                 goto out;
3875         }
3876
3877 out:
3878         btrfs_free_path(path);
3879         free_ipath(ipath);
3880         kfree(ipa);
3881
3882         return ret;
3883 }
3884
3885 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3886 {
3887         struct btrfs_data_container *inodes = ctx;
3888         const size_t c = 3 * sizeof(u64);
3889
3890         if (inodes->bytes_left >= c) {
3891                 inodes->bytes_left -= c;
3892                 inodes->val[inodes->elem_cnt] = inum;
3893                 inodes->val[inodes->elem_cnt + 1] = offset;
3894                 inodes->val[inodes->elem_cnt + 2] = root;
3895                 inodes->elem_cnt += 3;
3896         } else {
3897                 inodes->bytes_missing += c - inodes->bytes_left;
3898                 inodes->bytes_left = 0;
3899                 inodes->elem_missed += 3;
3900         }
3901
3902         return 0;
3903 }
3904
3905 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3906                                         void __user *arg)
3907 {
3908         int ret = 0;
3909         int size;
3910         struct btrfs_ioctl_logical_ino_args *loi;
3911         struct btrfs_data_container *inodes = NULL;
3912         struct btrfs_path *path = NULL;
3913
3914         if (!capable(CAP_SYS_ADMIN))
3915                 return -EPERM;
3916
3917         loi = memdup_user(arg, sizeof(*loi));
3918         if (IS_ERR(loi)) {
3919                 ret = PTR_ERR(loi);
3920                 loi = NULL;
3921                 goto out;
3922         }
3923
3924         path = btrfs_alloc_path();
3925         if (!path) {
3926                 ret = -ENOMEM;
3927                 goto out;
3928         }
3929
3930         size = min_t(u32, loi->size, 64 * 1024);
3931         inodes = init_data_container(size);
3932         if (IS_ERR(inodes)) {
3933                 ret = PTR_ERR(inodes);
3934                 inodes = NULL;
3935                 goto out;
3936         }
3937
3938         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3939                                           build_ino_list, inodes);
3940         if (ret == -EINVAL)
3941                 ret = -ENOENT;
3942         if (ret < 0)
3943                 goto out;
3944
3945         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3946                            (void *)(unsigned long)inodes, size);
3947         if (ret)
3948                 ret = -EFAULT;
3949
3950 out:
3951         btrfs_free_path(path);
3952         vfree(inodes);
3953         kfree(loi);
3954
3955         return ret;
3956 }
3957
3958 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3959                                struct btrfs_ioctl_balance_args *bargs)
3960 {
3961         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3962
3963         bargs->flags = bctl->flags;
3964
3965         if (atomic_read(&fs_info->balance_running))
3966                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3967         if (atomic_read(&fs_info->balance_pause_req))
3968                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3969         if (atomic_read(&fs_info->balance_cancel_req))
3970                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3971
3972         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3973         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3974         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3975
3976         if (lock) {
3977                 spin_lock(&fs_info->balance_lock);
3978                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3979                 spin_unlock(&fs_info->balance_lock);
3980         } else {
3981                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3982         }
3983 }
3984
3985 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3986 {
3987         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3988         struct btrfs_fs_info *fs_info = root->fs_info;
3989         struct btrfs_ioctl_balance_args *bargs;
3990         struct btrfs_balance_control *bctl;
3991         bool need_unlock; /* for mut. excl. ops lock */
3992         int ret;
3993
3994         if (!capable(CAP_SYS_ADMIN))
3995                 return -EPERM;
3996
3997         ret = mnt_want_write_file(file);
3998         if (ret)
3999                 return ret;
4000
4001 again:
4002         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4003                 mutex_lock(&fs_info->volume_mutex);
4004                 mutex_lock(&fs_info->balance_mutex);
4005                 need_unlock = true;
4006                 goto locked;
4007         }
4008
4009         /*
4010          * mut. excl. ops lock is locked.  Three possibilites:
4011          *   (1) some other op is running
4012          *   (2) balance is running
4013          *   (3) balance is paused -- special case (think resume)
4014          */
4015         mutex_lock(&fs_info->balance_mutex);
4016         if (fs_info->balance_ctl) {
4017                 /* this is either (2) or (3) */
4018                 if (!atomic_read(&fs_info->balance_running)) {
4019                         mutex_unlock(&fs_info->balance_mutex);
4020                         if (!mutex_trylock(&fs_info->volume_mutex))
4021                                 goto again;
4022                         mutex_lock(&fs_info->balance_mutex);
4023
4024                         if (fs_info->balance_ctl &&
4025                             !atomic_read(&fs_info->balance_running)) {
4026                                 /* this is (3) */
4027                                 need_unlock = false;
4028                                 goto locked;
4029                         }
4030
4031                         mutex_unlock(&fs_info->balance_mutex);
4032                         mutex_unlock(&fs_info->volume_mutex);
4033                         goto again;
4034                 } else {
4035                         /* this is (2) */
4036                         mutex_unlock(&fs_info->balance_mutex);
4037                         ret = -EINPROGRESS;
4038                         goto out;
4039                 }
4040         } else {
4041                 /* this is (1) */
4042                 mutex_unlock(&fs_info->balance_mutex);
4043                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4044                 goto out;
4045         }
4046
4047 locked:
4048         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4049
4050         if (arg) {
4051                 bargs = memdup_user(arg, sizeof(*bargs));
4052                 if (IS_ERR(bargs)) {
4053                         ret = PTR_ERR(bargs);
4054                         goto out_unlock;
4055                 }
4056
4057                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4058                         if (!fs_info->balance_ctl) {
4059                                 ret = -ENOTCONN;
4060                                 goto out_bargs;
4061                         }
4062
4063                         bctl = fs_info->balance_ctl;
4064                         spin_lock(&fs_info->balance_lock);
4065                         bctl->flags |= BTRFS_BALANCE_RESUME;
4066                         spin_unlock(&fs_info->balance_lock);
4067
4068                         goto do_balance;
4069                 }
4070         } else {
4071                 bargs = NULL;
4072         }
4073
4074         if (fs_info->balance_ctl) {
4075                 ret = -EINPROGRESS;
4076                 goto out_bargs;
4077         }
4078
4079         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4080         if (!bctl) {
4081                 ret = -ENOMEM;
4082                 goto out_bargs;
4083         }
4084
4085         bctl->fs_info = fs_info;
4086         if (arg) {
4087                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4088                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4089                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4090
4091                 bctl->flags = bargs->flags;
4092         } else {
4093                 /* balance everything - no filters */
4094                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4095         }
4096
4097 do_balance:
4098         /*
4099          * Ownership of bctl and mutually_exclusive_operation_running
4100          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4101          * or, if restriper was paused all the way until unmount, in
4102          * free_fs_info.  mutually_exclusive_operation_running is
4103          * cleared in __cancel_balance.
4104          */
4105         need_unlock = false;
4106
4107         ret = btrfs_balance(bctl, bargs);
4108
4109         if (arg) {
4110                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4111                         ret = -EFAULT;
4112         }
4113
4114 out_bargs:
4115         kfree(bargs);
4116 out_unlock:
4117         mutex_unlock(&fs_info->balance_mutex);
4118         mutex_unlock(&fs_info->volume_mutex);
4119         if (need_unlock)
4120                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4121 out:
4122         mnt_drop_write_file(file);
4123         return ret;
4124 }
4125
4126 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4127 {
4128         if (!capable(CAP_SYS_ADMIN))
4129                 return -EPERM;
4130
4131         switch (cmd) {
4132         case BTRFS_BALANCE_CTL_PAUSE:
4133                 return btrfs_pause_balance(root->fs_info);
4134         case BTRFS_BALANCE_CTL_CANCEL:
4135                 return btrfs_cancel_balance(root->fs_info);
4136         }
4137
4138         return -EINVAL;
4139 }
4140
4141 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4142                                          void __user *arg)
4143 {
4144         struct btrfs_fs_info *fs_info = root->fs_info;
4145         struct btrfs_ioctl_balance_args *bargs;
4146         int ret = 0;
4147
4148         if (!capable(CAP_SYS_ADMIN))
4149                 return -EPERM;
4150
4151         mutex_lock(&fs_info->balance_mutex);
4152         if (!fs_info->balance_ctl) {
4153                 ret = -ENOTCONN;
4154                 goto out;
4155         }
4156
4157         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4158         if (!bargs) {
4159                 ret = -ENOMEM;
4160                 goto out;
4161         }
4162
4163         update_ioctl_balance_args(fs_info, 1, bargs);
4164
4165         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4166                 ret = -EFAULT;
4167
4168         kfree(bargs);
4169 out:
4170         mutex_unlock(&fs_info->balance_mutex);
4171         return ret;
4172 }
4173
4174 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4175 {
4176         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4177         struct btrfs_ioctl_quota_ctl_args *sa;
4178         struct btrfs_trans_handle *trans = NULL;
4179         int ret;
4180         int err;
4181
4182         if (!capable(CAP_SYS_ADMIN))
4183                 return -EPERM;
4184
4185         ret = mnt_want_write_file(file);
4186         if (ret)
4187                 return ret;
4188
4189         sa = memdup_user(arg, sizeof(*sa));
4190         if (IS_ERR(sa)) {
4191                 ret = PTR_ERR(sa);
4192                 goto drop_write;
4193         }
4194
4195         down_write(&root->fs_info->subvol_sem);
4196         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4197         if (IS_ERR(trans)) {
4198                 ret = PTR_ERR(trans);
4199                 goto out;
4200         }
4201
4202         switch (sa->cmd) {
4203         case BTRFS_QUOTA_CTL_ENABLE:
4204                 ret = btrfs_quota_enable(trans, root->fs_info);
4205                 break;
4206         case BTRFS_QUOTA_CTL_DISABLE:
4207                 ret = btrfs_quota_disable(trans, root->fs_info);
4208                 break;
4209         default:
4210                 ret = -EINVAL;
4211                 break;
4212         }
4213
4214         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4215         if (err && !ret)
4216                 ret = err;
4217 out:
4218         kfree(sa);
4219         up_write(&root->fs_info->subvol_sem);
4220 drop_write:
4221         mnt_drop_write_file(file);
4222         return ret;
4223 }
4224
4225 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4226 {
4227         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4228         struct btrfs_ioctl_qgroup_assign_args *sa;
4229         struct btrfs_trans_handle *trans;
4230         int ret;
4231         int err;
4232
4233         if (!capable(CAP_SYS_ADMIN))
4234                 return -EPERM;
4235
4236         ret = mnt_want_write_file(file);
4237         if (ret)
4238                 return ret;
4239
4240         sa = memdup_user(arg, sizeof(*sa));
4241         if (IS_ERR(sa)) {
4242                 ret = PTR_ERR(sa);
4243                 goto drop_write;
4244         }
4245
4246         trans = btrfs_join_transaction(root);
4247         if (IS_ERR(trans)) {
4248                 ret = PTR_ERR(trans);
4249                 goto out;
4250         }
4251
4252         /* FIXME: check if the IDs really exist */
4253         if (sa->assign) {
4254                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4255                                                 sa->src, sa->dst);
4256         } else {
4257                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4258                                                 sa->src, sa->dst);
4259         }
4260
4261         err = btrfs_end_transaction(trans, root);
4262         if (err && !ret)
4263                 ret = err;
4264
4265 out:
4266         kfree(sa);
4267 drop_write:
4268         mnt_drop_write_file(file);
4269         return ret;
4270 }
4271
4272 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4273 {
4274         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4275         struct btrfs_ioctl_qgroup_create_args *sa;
4276         struct btrfs_trans_handle *trans;
4277         int ret;
4278         int err;
4279
4280         if (!capable(CAP_SYS_ADMIN))
4281                 return -EPERM;
4282
4283         ret = mnt_want_write_file(file);
4284         if (ret)
4285                 return ret;
4286
4287         sa = memdup_user(arg, sizeof(*sa));
4288         if (IS_ERR(sa)) {
4289                 ret = PTR_ERR(sa);
4290                 goto drop_write;
4291         }
4292
4293         if (!sa->qgroupid) {
4294                 ret = -EINVAL;
4295                 goto out;
4296         }
4297
4298         trans = btrfs_join_transaction(root);
4299         if (IS_ERR(trans)) {
4300                 ret = PTR_ERR(trans);
4301                 goto out;
4302         }
4303
4304         /* FIXME: check if the IDs really exist */
4305         if (sa->create) {
4306                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4307                                           NULL);
4308         } else {
4309                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4310         }
4311
4312         err = btrfs_end_transaction(trans, root);
4313         if (err && !ret)
4314                 ret = err;
4315
4316 out:
4317         kfree(sa);
4318 drop_write:
4319         mnt_drop_write_file(file);
4320         return ret;
4321 }
4322
4323 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4324 {
4325         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4326         struct btrfs_ioctl_qgroup_limit_args *sa;
4327         struct btrfs_trans_handle *trans;
4328         int ret;
4329         int err;
4330         u64 qgroupid;
4331
4332         if (!capable(CAP_SYS_ADMIN))
4333                 return -EPERM;
4334
4335         ret = mnt_want_write_file(file);
4336         if (ret)
4337                 return ret;
4338
4339         sa = memdup_user(arg, sizeof(*sa));
4340         if (IS_ERR(sa)) {
4341                 ret = PTR_ERR(sa);
4342                 goto drop_write;
4343         }
4344
4345         trans = btrfs_join_transaction(root);
4346         if (IS_ERR(trans)) {
4347                 ret = PTR_ERR(trans);
4348                 goto out;
4349         }
4350
4351         qgroupid = sa->qgroupid;
4352         if (!qgroupid) {
4353                 /* take the current subvol as qgroup */
4354                 qgroupid = root->root_key.objectid;
4355         }
4356
4357         /* FIXME: check if the IDs really exist */
4358         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4359
4360         err = btrfs_end_transaction(trans, root);
4361         if (err && !ret)
4362                 ret = err;
4363
4364 out:
4365         kfree(sa);
4366 drop_write:
4367         mnt_drop_write_file(file);
4368         return ret;
4369 }
4370
4371 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4372 {
4373         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4374         struct btrfs_ioctl_quota_rescan_args *qsa;
4375         int ret;
4376
4377         if (!capable(CAP_SYS_ADMIN))
4378                 return -EPERM;
4379
4380         ret = mnt_want_write_file(file);
4381         if (ret)
4382                 return ret;
4383
4384         qsa = memdup_user(arg, sizeof(*qsa));
4385         if (IS_ERR(qsa)) {
4386                 ret = PTR_ERR(qsa);
4387                 goto drop_write;
4388         }
4389
4390         if (qsa->flags) {
4391                 ret = -EINVAL;
4392                 goto out;
4393         }
4394
4395         ret = btrfs_qgroup_rescan(root->fs_info);
4396
4397 out:
4398         kfree(qsa);
4399 drop_write:
4400         mnt_drop_write_file(file);
4401         return ret;
4402 }
4403
4404 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4405 {
4406         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4407         struct btrfs_ioctl_quota_rescan_args *qsa;
4408         int ret = 0;
4409
4410         if (!capable(CAP_SYS_ADMIN))
4411                 return -EPERM;
4412
4413         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4414         if (!qsa)
4415                 return -ENOMEM;
4416
4417         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4418                 qsa->flags = 1;
4419                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4420         }
4421
4422         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4423                 ret = -EFAULT;
4424
4425         kfree(qsa);
4426         return ret;
4427 }
4428
4429 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4430 {
4431         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4432
4433         if (!capable(CAP_SYS_ADMIN))
4434                 return -EPERM;
4435
4436         return btrfs_qgroup_wait_for_completion(root->fs_info);
4437 }
4438
4439 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4440                                             struct btrfs_ioctl_received_subvol_args *sa)
4441 {
4442         struct inode *inode = file_inode(file);
4443         struct btrfs_root *root = BTRFS_I(inode)->root;
4444         struct btrfs_root_item *root_item = &root->root_item;
4445         struct btrfs_trans_handle *trans;
4446         struct timespec ct = CURRENT_TIME;
4447         int ret = 0;
4448         int received_uuid_changed;
4449
4450         if (!inode_owner_or_capable(inode))
4451                 return -EPERM;
4452
4453         ret = mnt_want_write_file(file);
4454         if (ret < 0)
4455                 return ret;
4456
4457         down_write(&root->fs_info->subvol_sem);
4458
4459         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4460                 ret = -EINVAL;
4461                 goto out;
4462         }
4463
4464         if (btrfs_root_readonly(root)) {
4465                 ret = -EROFS;
4466                 goto out;
4467         }
4468
4469         /*
4470          * 1 - root item
4471          * 2 - uuid items (received uuid + subvol uuid)
4472          */
4473         trans = btrfs_start_transaction(root, 3);
4474         if (IS_ERR(trans)) {
4475                 ret = PTR_ERR(trans);
4476                 trans = NULL;
4477                 goto out;
4478         }
4479
4480         sa->rtransid = trans->transid;
4481         sa->rtime.sec = ct.tv_sec;
4482         sa->rtime.nsec = ct.tv_nsec;
4483
4484         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4485                                        BTRFS_UUID_SIZE);
4486         if (received_uuid_changed &&
4487             !btrfs_is_empty_uuid(root_item->received_uuid))
4488                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4489                                     root_item->received_uuid,
4490                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4491                                     root->root_key.objectid);
4492         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4493         btrfs_set_root_stransid(root_item, sa->stransid);
4494         btrfs_set_root_rtransid(root_item, sa->rtransid);
4495         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4496         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4497         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4498         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4499
4500         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4501                                 &root->root_key, &root->root_item);
4502         if (ret < 0) {
4503                 btrfs_end_transaction(trans, root);
4504                 goto out;
4505         }
4506         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4507                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4508                                           sa->uuid,
4509                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4510                                           root->root_key.objectid);
4511                 if (ret < 0 && ret != -EEXIST) {
4512                         btrfs_abort_transaction(trans, root, ret);
4513                         goto out;
4514                 }
4515         }
4516         ret = btrfs_commit_transaction(trans, root);
4517         if (ret < 0) {
4518                 btrfs_abort_transaction(trans, root, ret);
4519                 goto out;
4520         }
4521
4522 out:
4523         up_write(&root->fs_info->subvol_sem);
4524         mnt_drop_write_file(file);
4525         return ret;
4526 }
4527
4528 #ifdef CONFIG_64BIT
4529 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4530                                                 void __user *arg)
4531 {
4532         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4533         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4534         int ret = 0;
4535
4536         args32 = memdup_user(arg, sizeof(*args32));
4537         if (IS_ERR(args32)) {
4538                 ret = PTR_ERR(args32);
4539                 args32 = NULL;
4540                 goto out;
4541         }
4542
4543         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4544         if (IS_ERR(args64)) {
4545                 ret = PTR_ERR(args64);
4546                 args64 = NULL;
4547                 goto out;
4548         }
4549
4550         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4551         args64->stransid = args32->stransid;
4552         args64->rtransid = args32->rtransid;
4553         args64->stime.sec = args32->stime.sec;
4554         args64->stime.nsec = args32->stime.nsec;
4555         args64->rtime.sec = args32->rtime.sec;
4556         args64->rtime.nsec = args32->rtime.nsec;
4557         args64->flags = args32->flags;
4558
4559         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4560         if (ret)
4561                 goto out;
4562
4563         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4564         args32->stransid = args64->stransid;
4565         args32->rtransid = args64->rtransid;
4566         args32->stime.sec = args64->stime.sec;
4567         args32->stime.nsec = args64->stime.nsec;
4568         args32->rtime.sec = args64->rtime.sec;
4569         args32->rtime.nsec = args64->rtime.nsec;
4570         args32->flags = args64->flags;
4571
4572         ret = copy_to_user(arg, args32, sizeof(*args32));
4573         if (ret)
4574                 ret = -EFAULT;
4575
4576 out:
4577         kfree(args32);
4578         kfree(args64);
4579         return ret;
4580 }
4581 #endif
4582
4583 static long btrfs_ioctl_set_received_subvol(struct file *file,
4584                                             void __user *arg)
4585 {
4586         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4587         int ret = 0;
4588
4589         sa = memdup_user(arg, sizeof(*sa));
4590         if (IS_ERR(sa)) {
4591                 ret = PTR_ERR(sa);
4592                 sa = NULL;
4593                 goto out;
4594         }
4595
4596         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4597
4598         if (ret)
4599                 goto out;
4600
4601         ret = copy_to_user(arg, sa, sizeof(*sa));
4602         if (ret)
4603                 ret = -EFAULT;
4604
4605 out:
4606         kfree(sa);
4607         return ret;
4608 }
4609
4610 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4611 {
4612         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4613         size_t len;
4614         int ret;
4615         char label[BTRFS_LABEL_SIZE];
4616
4617         spin_lock(&root->fs_info->super_lock);
4618         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4619         spin_unlock(&root->fs_info->super_lock);
4620
4621         len = strnlen(label, BTRFS_LABEL_SIZE);
4622
4623         if (len == BTRFS_LABEL_SIZE) {
4624                 btrfs_warn(root->fs_info,
4625                         "label is too long, return the first %zu bytes", --len);
4626         }
4627
4628         ret = copy_to_user(arg, label, len);
4629
4630         return ret ? -EFAULT : 0;
4631 }
4632
4633 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4634 {
4635         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4636         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4637         struct btrfs_trans_handle *trans;
4638         char label[BTRFS_LABEL_SIZE];
4639         int ret;
4640
4641         if (!capable(CAP_SYS_ADMIN))
4642                 return -EPERM;
4643
4644         if (copy_from_user(label, arg, sizeof(label)))
4645                 return -EFAULT;
4646
4647         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4648                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
4649                        BTRFS_LABEL_SIZE - 1);
4650                 return -EINVAL;
4651         }
4652
4653         ret = mnt_want_write_file(file);
4654         if (ret)
4655                 return ret;
4656
4657         trans = btrfs_start_transaction(root, 0);
4658         if (IS_ERR(trans)) {
4659                 ret = PTR_ERR(trans);
4660                 goto out_unlock;
4661         }
4662
4663         spin_lock(&root->fs_info->super_lock);
4664         strcpy(super_block->label, label);
4665         spin_unlock(&root->fs_info->super_lock);
4666         ret = btrfs_commit_transaction(trans, root);
4667
4668 out_unlock:
4669         mnt_drop_write_file(file);
4670         return ret;
4671 }
4672
4673 #define INIT_FEATURE_FLAGS(suffix) \
4674         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4675           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4676           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4677
4678 static int btrfs_ioctl_get_supported_features(struct file *file,
4679                                               void __user *arg)
4680 {
4681         static struct btrfs_ioctl_feature_flags features[3] = {
4682                 INIT_FEATURE_FLAGS(SUPP),
4683                 INIT_FEATURE_FLAGS(SAFE_SET),
4684                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4685         };
4686
4687         if (copy_to_user(arg, &features, sizeof(features)))
4688                 return -EFAULT;
4689
4690         return 0;
4691 }
4692
4693 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
4694 {
4695         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4696         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4697         struct btrfs_ioctl_feature_flags features;
4698
4699         features.compat_flags = btrfs_super_compat_flags(super_block);
4700         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4701         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4702
4703         if (copy_to_user(arg, &features, sizeof(features)))
4704                 return -EFAULT;
4705
4706         return 0;
4707 }
4708
4709 static int check_feature_bits(struct btrfs_root *root,
4710                               enum btrfs_feature_set set,
4711                               u64 change_mask, u64 flags, u64 supported_flags,
4712                               u64 safe_set, u64 safe_clear)
4713 {
4714         const char *type = btrfs_feature_set_names[set];
4715         char *names;
4716         u64 disallowed, unsupported;
4717         u64 set_mask = flags & change_mask;
4718         u64 clear_mask = ~flags & change_mask;
4719
4720         unsupported = set_mask & ~supported_flags;
4721         if (unsupported) {
4722                 names = btrfs_printable_features(set, unsupported);
4723                 if (names) {
4724                         btrfs_warn(root->fs_info,
4725                            "this kernel does not support the %s feature bit%s",
4726                            names, strchr(names, ',') ? "s" : "");
4727                         kfree(names);
4728                 } else
4729                         btrfs_warn(root->fs_info,
4730                            "this kernel does not support %s bits 0x%llx",
4731                            type, unsupported);
4732                 return -EOPNOTSUPP;
4733         }
4734
4735         disallowed = set_mask & ~safe_set;
4736         if (disallowed) {
4737                 names = btrfs_printable_features(set, disallowed);
4738                 if (names) {
4739                         btrfs_warn(root->fs_info,
4740                            "can't set the %s feature bit%s while mounted",
4741                            names, strchr(names, ',') ? "s" : "");
4742                         kfree(names);
4743                 } else
4744                         btrfs_warn(root->fs_info,
4745                            "can't set %s bits 0x%llx while mounted",
4746                            type, disallowed);
4747                 return -EPERM;
4748         }
4749
4750         disallowed = clear_mask & ~safe_clear;
4751         if (disallowed) {
4752                 names = btrfs_printable_features(set, disallowed);
4753                 if (names) {
4754                         btrfs_warn(root->fs_info,
4755                            "can't clear the %s feature bit%s while mounted",
4756                            names, strchr(names, ',') ? "s" : "");
4757                         kfree(names);
4758                 } else
4759                         btrfs_warn(root->fs_info,
4760                            "can't clear %s bits 0x%llx while mounted",
4761                            type, disallowed);
4762                 return -EPERM;
4763         }
4764
4765         return 0;
4766 }
4767
4768 #define check_feature(root, change_mask, flags, mask_base)      \
4769 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
4770                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4771                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4772                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4773
4774 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4775 {
4776         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4777         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4778         struct btrfs_ioctl_feature_flags flags[2];
4779         struct btrfs_trans_handle *trans;
4780         u64 newflags;
4781         int ret;
4782
4783         if (!capable(CAP_SYS_ADMIN))
4784                 return -EPERM;
4785
4786         if (copy_from_user(flags, arg, sizeof(flags)))
4787                 return -EFAULT;
4788
4789         /* Nothing to do */
4790         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4791             !flags[0].incompat_flags)
4792                 return 0;
4793
4794         ret = check_feature(root, flags[0].compat_flags,
4795                             flags[1].compat_flags, COMPAT);
4796         if (ret)
4797                 return ret;
4798
4799         ret = check_feature(root, flags[0].compat_ro_flags,
4800                             flags[1].compat_ro_flags, COMPAT_RO);
4801         if (ret)
4802                 return ret;
4803
4804         ret = check_feature(root, flags[0].incompat_flags,
4805                             flags[1].incompat_flags, INCOMPAT);
4806         if (ret)
4807                 return ret;
4808
4809         trans = btrfs_start_transaction(root, 0);
4810         if (IS_ERR(trans))
4811                 return PTR_ERR(trans);
4812
4813         spin_lock(&root->fs_info->super_lock);
4814         newflags = btrfs_super_compat_flags(super_block);
4815         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4816         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4817         btrfs_set_super_compat_flags(super_block, newflags);
4818
4819         newflags = btrfs_super_compat_ro_flags(super_block);
4820         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4821         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4822         btrfs_set_super_compat_ro_flags(super_block, newflags);
4823
4824         newflags = btrfs_super_incompat_flags(super_block);
4825         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4826         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4827         btrfs_set_super_incompat_flags(super_block, newflags);
4828         spin_unlock(&root->fs_info->super_lock);
4829
4830         return btrfs_commit_transaction(trans, root);
4831 }
4832
4833 long btrfs_ioctl(struct file *file, unsigned int
4834                 cmd, unsigned long arg)
4835 {
4836         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4837         void __user *argp = (void __user *)arg;
4838
4839         switch (cmd) {
4840         case FS_IOC_GETFLAGS:
4841                 return btrfs_ioctl_getflags(file, argp);
4842         case FS_IOC_SETFLAGS:
4843                 return btrfs_ioctl_setflags(file, argp);
4844         case FS_IOC_GETVERSION:
4845                 return btrfs_ioctl_getversion(file, argp);
4846         case FITRIM:
4847                 return btrfs_ioctl_fitrim(file, argp);
4848         case BTRFS_IOC_SNAP_CREATE:
4849                 return btrfs_ioctl_snap_create(file, argp, 0);
4850         case BTRFS_IOC_SNAP_CREATE_V2:
4851                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4852         case BTRFS_IOC_SUBVOL_CREATE:
4853                 return btrfs_ioctl_snap_create(file, argp, 1);
4854         case BTRFS_IOC_SUBVOL_CREATE_V2:
4855                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4856         case BTRFS_IOC_SNAP_DESTROY:
4857                 return btrfs_ioctl_snap_destroy(file, argp);
4858         case BTRFS_IOC_SUBVOL_GETFLAGS:
4859                 return btrfs_ioctl_subvol_getflags(file, argp);
4860         case BTRFS_IOC_SUBVOL_SETFLAGS:
4861                 return btrfs_ioctl_subvol_setflags(file, argp);
4862         case BTRFS_IOC_DEFAULT_SUBVOL:
4863                 return btrfs_ioctl_default_subvol(file, argp);
4864         case BTRFS_IOC_DEFRAG:
4865                 return btrfs_ioctl_defrag(file, NULL);
4866         case BTRFS_IOC_DEFRAG_RANGE:
4867                 return btrfs_ioctl_defrag(file, argp);
4868         case BTRFS_IOC_RESIZE:
4869                 return btrfs_ioctl_resize(file, argp);
4870         case BTRFS_IOC_ADD_DEV:
4871                 return btrfs_ioctl_add_dev(root, argp);
4872         case BTRFS_IOC_RM_DEV:
4873                 return btrfs_ioctl_rm_dev(file, argp);
4874         case BTRFS_IOC_FS_INFO:
4875                 return btrfs_ioctl_fs_info(root, argp);
4876         case BTRFS_IOC_DEV_INFO:
4877                 return btrfs_ioctl_dev_info(root, argp);
4878         case BTRFS_IOC_BALANCE:
4879                 return btrfs_ioctl_balance(file, NULL);
4880         case BTRFS_IOC_CLONE:
4881                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4882         case BTRFS_IOC_CLONE_RANGE:
4883                 return btrfs_ioctl_clone_range(file, argp);
4884         case BTRFS_IOC_TRANS_START:
4885                 return btrfs_ioctl_trans_start(file);
4886         case BTRFS_IOC_TRANS_END:
4887                 return btrfs_ioctl_trans_end(file);
4888         case BTRFS_IOC_TREE_SEARCH:
4889                 return btrfs_ioctl_tree_search(file, argp);
4890         case BTRFS_IOC_INO_LOOKUP:
4891                 return btrfs_ioctl_ino_lookup(file, argp);
4892         case BTRFS_IOC_INO_PATHS:
4893                 return btrfs_ioctl_ino_to_path(root, argp);
4894         case BTRFS_IOC_LOGICAL_INO:
4895                 return btrfs_ioctl_logical_to_ino(root, argp);
4896         case BTRFS_IOC_SPACE_INFO:
4897                 return btrfs_ioctl_space_info(root, argp);
4898         case BTRFS_IOC_SYNC: {
4899                 int ret;
4900
4901                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
4902                 if (ret)
4903                         return ret;
4904                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
4905                 return ret;
4906         }
4907         case BTRFS_IOC_START_SYNC:
4908                 return btrfs_ioctl_start_sync(root, argp);
4909         case BTRFS_IOC_WAIT_SYNC:
4910                 return btrfs_ioctl_wait_sync(root, argp);
4911         case BTRFS_IOC_SCRUB:
4912                 return btrfs_ioctl_scrub(file, argp);
4913         case BTRFS_IOC_SCRUB_CANCEL:
4914                 return btrfs_ioctl_scrub_cancel(root, argp);
4915         case BTRFS_IOC_SCRUB_PROGRESS:
4916                 return btrfs_ioctl_scrub_progress(root, argp);
4917         case BTRFS_IOC_BALANCE_V2:
4918                 return btrfs_ioctl_balance(file, argp);
4919         case BTRFS_IOC_BALANCE_CTL:
4920                 return btrfs_ioctl_balance_ctl(root, arg);
4921         case BTRFS_IOC_BALANCE_PROGRESS:
4922                 return btrfs_ioctl_balance_progress(root, argp);
4923         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4924                 return btrfs_ioctl_set_received_subvol(file, argp);
4925 #ifdef CONFIG_64BIT
4926         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4927                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4928 #endif
4929         case BTRFS_IOC_SEND:
4930                 return btrfs_ioctl_send(file, argp);
4931         case BTRFS_IOC_GET_DEV_STATS:
4932                 return btrfs_ioctl_get_dev_stats(root, argp);
4933         case BTRFS_IOC_QUOTA_CTL:
4934                 return btrfs_ioctl_quota_ctl(file, argp);
4935         case BTRFS_IOC_QGROUP_ASSIGN:
4936                 return btrfs_ioctl_qgroup_assign(file, argp);
4937         case BTRFS_IOC_QGROUP_CREATE:
4938                 return btrfs_ioctl_qgroup_create(file, argp);
4939         case BTRFS_IOC_QGROUP_LIMIT:
4940                 return btrfs_ioctl_qgroup_limit(file, argp);
4941         case BTRFS_IOC_QUOTA_RESCAN:
4942                 return btrfs_ioctl_quota_rescan(file, argp);
4943         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4944                 return btrfs_ioctl_quota_rescan_status(file, argp);
4945         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4946                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4947         case BTRFS_IOC_DEV_REPLACE:
4948                 return btrfs_ioctl_dev_replace(root, argp);
4949         case BTRFS_IOC_GET_FSLABEL:
4950                 return btrfs_ioctl_get_fslabel(file, argp);
4951         case BTRFS_IOC_SET_FSLABEL:
4952                 return btrfs_ioctl_set_fslabel(file, argp);
4953         case BTRFS_IOC_FILE_EXTENT_SAME:
4954                 return btrfs_ioctl_file_extent_same(file, argp);
4955         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4956                 return btrfs_ioctl_get_supported_features(file, argp);
4957         case BTRFS_IOC_GET_FEATURES:
4958                 return btrfs_ioctl_get_features(file, argp);
4959         case BTRFS_IOC_SET_FEATURES:
4960                 return btrfs_ioctl_set_features(file, argp);
4961         }
4962
4963         return -ENOTTY;
4964 }