fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62
  63 #ifdef CONFIG_64BIT
  64 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  65  * structures are incorrect, as the timespec structure from userspace
  66  * is 4 bytes too small. We define these alternatives here to teach
  67  * the kernel about the 32-bit struct packing.
  68  */
  69 struct btrfs_ioctl_timespec_32 {
  70         __u64 sec;
  71         __u32 nsec;
  72 } __attribute__ ((__packed__));
  73
  74 struct btrfs_ioctl_received_subvol_args_32 {
  75         char    uuid[BTRFS_UUID_SIZE];  /* in */
  76         __u64   stransid;               /* in */
  77         __u64   rtransid;               /* out */
  78         struct btrfs_ioctl_timespec_32 stime; /* in */
  79         struct btrfs_ioctl_timespec_32 rtime; /* out */
  80         __u64   flags;                  /* in */
  81         __u64   reserved[16];           /* in */
  82 } __attribute__ ((__packed__));
  83
  84 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  85                                 struct btrfs_ioctl_received_subvol_args_32)
  86 #endif
  87
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  91
  92 /* Mask out flags that are inappropriate for the given type of inode. */
  93 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  94 {
  95         if (S_ISDIR(mode))
  96                 return flags;
  97         else if (S_ISREG(mode))
  98                 return flags & ~FS_DIRSYNC_FL;
  99         else
 100                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 101 }
 102
 103 /*
 104  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 105  */
 106 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 126                 iflags |= FS_COMPR_FL;
 127         else if (flags & BTRFS_INODE_NOCOMPRESS)
 128                 iflags |= FS_NOCOMP_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_update_iflags(struct inode *inode)
 137 {
 138         struct btrfs_inode *ip = BTRFS_I(inode);
 139         unsigned int new_fl = 0;
 140
 141         if (ip->flags & BTRFS_INODE_SYNC)
 142                 new_fl |= S_SYNC;
 143         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 144                 new_fl |= S_IMMUTABLE;
 145         if (ip->flags & BTRFS_INODE_APPEND)
 146                 new_fl |= S_APPEND;
 147         if (ip->flags & BTRFS_INODE_NOATIME)
 148                 new_fl |= S_NOATIME;
 149         if (ip->flags & BTRFS_INODE_DIRSYNC)
 150                 new_fl |= S_DIRSYNC;
 151
 152         set_mask_bits(&inode->i_flags,
 153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 154                       new_fl);
 155 }
 156
 157 /*
 158  * Inherit flags from the parent inode.
 159  *
 160  * Currently only the compression flags and the cow flags are inherited.
 161  */
 162 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 163 {
 164         unsigned int flags;
 165
 166         if (!dir)
 167                 return;
 168
 169         flags = BTRFS_I(dir)->flags;
 170
 171         if (flags & BTRFS_INODE_NOCOMPRESS) {
 172                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 173                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 174         } else if (flags & BTRFS_INODE_COMPRESS) {
 175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 177         }
 178
 179         if (flags & BTRFS_INODE_NODATACOW) {
 180                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 181                 if (S_ISREG(inode->i_mode))
 182                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 183         }
 184
 185         btrfs_update_iflags(inode);
 186 }
 187
 188 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 189 {
 190         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 191         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 192
 193         if (copy_to_user(arg, &flags, sizeof(flags)))
 194                 return -EFAULT;
 195         return 0;
 196 }
 197
 198 static int check_flags(unsigned int flags)
 199 {
 200         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 201                       FS_NOATIME_FL | FS_NODUMP_FL | \
 202                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 203                       FS_NOCOMP_FL | FS_COMPR_FL |
 204                       FS_NOCOW_FL))
 205                 return -EOPNOTSUPP;
 206
 207         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 208                 return -EINVAL;
 209
 210         return 0;
 211 }
 212
 213 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 214 {
 215         struct inode *inode = file_inode(file);
 216         struct btrfs_inode *ip = BTRFS_I(inode);
 217         struct btrfs_root *root = ip->root;
 218         struct btrfs_trans_handle *trans;
 219         unsigned int flags, oldflags;
 220         int ret;
 221         u64 ip_oldflags;
 222         unsigned int i_oldflags;
 223         umode_t mode;
 224
 225         if (!inode_owner_or_capable(inode))
 226                 return -EPERM;
 227
 228         if (btrfs_root_readonly(root))
 229                 return -EROFS;
 230
 231         if (copy_from_user(&flags, arg, sizeof(flags)))
 232                 return -EFAULT;
 233
 234         ret = check_flags(flags);
 235         if (ret)
 236                 return ret;
 237
 238         ret = mnt_want_write_file(file);
 239         if (ret)
 240                 return ret;
 241
 242         mutex_lock(&inode->i_mutex);
 243
 244         ip_oldflags = ip->flags;
 245         i_oldflags = inode->i_flags;
 246         mode = inode->i_mode;
 247
 248         flags = btrfs_mask_flags(inode->i_mode, flags);
 249         oldflags = btrfs_flags_to_ioctl(ip->flags);
 250         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 251                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 252                         ret = -EPERM;
 253                         goto out_unlock;
 254                 }
 255         }
 256
 257         if (flags & FS_SYNC_FL)
 258                 ip->flags |= BTRFS_INODE_SYNC;
 259         else
 260                 ip->flags &= ~BTRFS_INODE_SYNC;
 261         if (flags & FS_IMMUTABLE_FL)
 262                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 263         else
 264                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 265         if (flags & FS_APPEND_FL)
 266                 ip->flags |= BTRFS_INODE_APPEND;
 267         else
 268                 ip->flags &= ~BTRFS_INODE_APPEND;
 269         if (flags & FS_NODUMP_FL)
 270                 ip->flags |= BTRFS_INODE_NODUMP;
 271         else
 272                 ip->flags &= ~BTRFS_INODE_NODUMP;
 273         if (flags & FS_NOATIME_FL)
 274                 ip->flags |= BTRFS_INODE_NOATIME;
 275         else
 276                 ip->flags &= ~BTRFS_INODE_NOATIME;
 277         if (flags & FS_DIRSYNC_FL)
 278                 ip->flags |= BTRFS_INODE_DIRSYNC;
 279         else
 280                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 281         if (flags & FS_NOCOW_FL) {
 282                 if (S_ISREG(mode)) {
 283                         /*
 284                          * It's safe to turn csums off here, no extents exist.
 285                          * Otherwise we want the flag to reflect the real COW
 286                          * status of the file and will not set it.
 287                          */
 288                         if (inode->i_size == 0)
 289                                 ip->flags |= BTRFS_INODE_NODATACOW
 290                                            | BTRFS_INODE_NODATASUM;
 291                 } else {
 292                         ip->flags |= BTRFS_INODE_NODATACOW;
 293                 }
 294         } else {
 295                 /*
 296                  * Revert back under same assuptions as above
 297                  */
 298                 if (S_ISREG(mode)) {
 299                         if (inode->i_size == 0)
 300                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 301                                              | BTRFS_INODE_NODATASUM);
 302                 } else {
 303                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 304                 }
 305         }
 306
 307         /*
 308          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 309          * flag may be changed automatically if compression code won't make
 310          * things smaller.
 311          */
 312         if (flags & FS_NOCOMP_FL) {
 313                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 314                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 315
 316                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 317                 if (ret && ret != -ENODATA)
 318                         goto out_drop;
 319         } else if (flags & FS_COMPR_FL) {
 320                 const char *comp;
 321
 322                 ip->flags |= BTRFS_INODE_COMPRESS;
 323                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 324
 325                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 326                         comp = "lzo";
 327                 else
 328                         comp = "zlib";
 329                 ret = btrfs_set_prop(inode, "btrfs.compression",
 330                                      comp, strlen(comp), 0);
 331                 if (ret)
 332                         goto out_drop;
 333
 334         } else {
 335                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 336                 if (ret && ret != -ENODATA)
 337                         goto out_drop;
 338                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 339         }
 340
 341         trans = btrfs_start_transaction(root, 1);
 342         if (IS_ERR(trans)) {
 343                 ret = PTR_ERR(trans);
 344                 goto out_drop;
 345         }
 346
 347         btrfs_update_iflags(inode);
 348         inode_inc_iversion(inode);
 349         inode->i_ctime = CURRENT_TIME;
 350         ret = btrfs_update_inode(trans, root, inode);
 351
 352         btrfs_end_transaction(trans, root);
 353  out_drop:
 354         if (ret) {
 355                 ip->flags = ip_oldflags;
 356                 inode->i_flags = i_oldflags;
 357         }
 358
 359  out_unlock:
 360         mutex_unlock(&inode->i_mutex);
 361         mnt_drop_write_file(file);
 362         return ret;
 363 }
 364
 365 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 366 {
 367         struct inode *inode = file_inode(file);
 368
 369         return put_user(inode->i_generation, arg);
 370 }
 371
 372 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 373 {
 374         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 375         struct btrfs_device *device;
 376         struct request_queue *q;
 377         struct fstrim_range range;
 378         u64 minlen = ULLONG_MAX;
 379         u64 num_devices = 0;
 380         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 381         int ret;
 382
 383         if (!capable(CAP_SYS_ADMIN))
 384                 return -EPERM;
 385
 386         rcu_read_lock();
 387         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 388                                 dev_list) {
 389                 if (!device->bdev)
 390                         continue;
 391                 q = bdev_get_queue(device->bdev);
 392                 if (blk_queue_discard(q)) {
 393                         num_devices++;
 394                         minlen = min((u64)q->limits.discard_granularity,
 395                                      minlen);
 396                 }
 397         }
 398         rcu_read_unlock();
 399
 400         if (!num_devices)
 401                 return -EOPNOTSUPP;
 402         if (copy_from_user(&range, arg, sizeof(range)))
 403                 return -EFAULT;
 404         if (range.start > total_bytes ||
 405             range.len < fs_info->sb->s_blocksize)
 406                 return -EINVAL;
 407
 408         range.len = min(range.len, total_bytes - range.start);
 409         range.minlen = max(range.minlen, minlen);
 410         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 411         if (ret < 0)
 412                 return ret;
 413
 414         if (copy_to_user(arg, &range, sizeof(range)))
 415                 return -EFAULT;
 416
 417         return 0;
 418 }
 419
 420 int btrfs_is_empty_uuid(u8 *uuid)
 421 {
 422         int i;
 423
 424         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 425                 if (uuid[i])
 426                         return 0;
 427         }
 428         return 1;
 429 }
 430
 431 static noinline int create_subvol(struct inode *dir,
 432                                   struct dentry *dentry,
 433                                   char *name, int namelen,
 434                                   u64 *async_transid,
 435                                   struct btrfs_qgroup_inherit *inherit)
 436 {
 437         struct btrfs_trans_handle *trans;
 438         struct btrfs_key key;
 439         struct btrfs_root_item root_item;
 440         struct btrfs_inode_item *inode_item;
 441         struct extent_buffer *leaf;
 442         struct btrfs_root *root = BTRFS_I(dir)->root;
 443         struct btrfs_root *new_root;
 444         struct btrfs_block_rsv block_rsv;
 445         struct timespec cur_time = CURRENT_TIME;
 446         struct inode *inode;
 447         int ret;
 448         int err;
 449         u64 objectid;
 450         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 451         u64 index = 0;
 452         u64 qgroup_reserved;
 453         uuid_le new_uuid;
 454
 455         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 456         if (ret)
 457                 return ret;
 458
 459         /*
 460          * Don't create subvolume whose level is not zero. Or qgroup will be
 461          * screwed up since it assume subvolme qgroup's level to be 0.
 462          */
 463         if (btrfs_qgroup_level(objectid))
 464                 return -ENOSPC;
 465
 466         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 467         /*
 468          * The same as the snapshot creation, please see the comment
 469          * of create_snapshot().
 470          */
 471         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 472                                                8, &qgroup_reserved, false);
 473         if (ret)
 474                 return ret;
 475
 476         trans = btrfs_start_transaction(root, 0);
 477         if (IS_ERR(trans)) {
 478                 ret = PTR_ERR(trans);
 479                 btrfs_subvolume_release_metadata(root, &block_rsv,
 480                                                  qgroup_reserved);
 481                 return ret;
 482         }
 483         trans->block_rsv = &block_rsv;
 484         trans->bytes_reserved = block_rsv.size;
 485
 486         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 487         if (ret)
 488                 goto fail;
 489
 490         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 491         if (IS_ERR(leaf)) {
 492                 ret = PTR_ERR(leaf);
 493                 goto fail;
 494         }
 495
 496         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 497         btrfs_set_header_bytenr(leaf, leaf->start);
 498         btrfs_set_header_generation(leaf, trans->transid);
 499         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 500         btrfs_set_header_owner(leaf, objectid);
 501
 502         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 503                             BTRFS_FSID_SIZE);
 504         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 505                             btrfs_header_chunk_tree_uuid(leaf),
 506                             BTRFS_UUID_SIZE);
 507         btrfs_mark_buffer_dirty(leaf);
 508
 509         memset(&root_item, 0, sizeof(root_item));
 510
 511         inode_item = &root_item.inode;
 512         btrfs_set_stack_inode_generation(inode_item, 1);
 513         btrfs_set_stack_inode_size(inode_item, 3);
 514         btrfs_set_stack_inode_nlink(inode_item, 1);
 515         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 516         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 517
 518         btrfs_set_root_flags(&root_item, 0);
 519         btrfs_set_root_limit(&root_item, 0);
 520         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 521
 522         btrfs_set_root_bytenr(&root_item, leaf->start);
 523         btrfs_set_root_generation(&root_item, trans->transid);
 524         btrfs_set_root_level(&root_item, 0);
 525         btrfs_set_root_refs(&root_item, 1);
 526         btrfs_set_root_used(&root_item, leaf->len);
 527         btrfs_set_root_last_snapshot(&root_item, 0);
 528
 529         btrfs_set_root_generation_v2(&root_item,
 530                         btrfs_root_generation(&root_item));
 531         uuid_le_gen(&new_uuid);
 532         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 533         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 534         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 535         root_item.ctime = root_item.otime;
 536         btrfs_set_root_ctransid(&root_item, trans->transid);
 537         btrfs_set_root_otransid(&root_item, trans->transid);
 538
 539         btrfs_tree_unlock(leaf);
 540         free_extent_buffer(leaf);
 541         leaf = NULL;
 542
 543         btrfs_set_root_dirid(&root_item, new_dirid);
 544
 545         key.objectid = objectid;
 546         key.offset = 0;
 547         key.type = BTRFS_ROOT_ITEM_KEY;
 548         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 549                                 &root_item);
 550         if (ret)
 551                 goto fail;
 552
 553         key.offset = (u64)-1;
 554         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 555         if (IS_ERR(new_root)) {
 556                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 557                 ret = PTR_ERR(new_root);
 558                 goto fail;
 559         }
 560
 561         btrfs_record_root_in_trans(trans, new_root);
 562
 563         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 564         if (ret) {
 565                 /* We potentially lose an unused inode item here */
 566                 btrfs_abort_transaction(trans, root, ret);
 567                 goto fail;
 568         }
 569
 570         /*
 571          * insert the directory item
 572          */
 573         ret = btrfs_set_inode_index(dir, &index);
 574         if (ret) {
 575                 btrfs_abort_transaction(trans, root, ret);
 576                 goto fail;
 577         }
 578
 579         ret = btrfs_insert_dir_item(trans, root,
 580                                     name, namelen, dir, &key,
 581                                     BTRFS_FT_DIR, index);
 582         if (ret) {
 583                 btrfs_abort_transaction(trans, root, ret);
 584                 goto fail;
 585         }
 586
 587         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 588         ret = btrfs_update_inode(trans, root, dir);
 589         BUG_ON(ret);
 590
 591         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 592                                  objectid, root->root_key.objectid,
 593                                  btrfs_ino(dir), index, name, namelen);
 594         BUG_ON(ret);
 595
 596         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 597                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 598                                   objectid);
 599         if (ret)
 600                 btrfs_abort_transaction(trans, root, ret);
 601
 602 fail:
 603         trans->block_rsv = NULL;
 604         trans->bytes_reserved = 0;
 605         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 606
 607         if (async_transid) {
 608                 *async_transid = trans->transid;
 609                 err = btrfs_commit_transaction_async(trans, root, 1);
 610                 if (err)
 611                         err = btrfs_commit_transaction(trans, root);
 612         } else {
 613                 err = btrfs_commit_transaction(trans, root);
 614         }
 615         if (err && !ret)
 616                 ret = err;
 617
 618         if (!ret) {
 619                 inode = btrfs_lookup_dentry(dir, dentry);
 620                 if (IS_ERR(inode))
 621                         return PTR_ERR(inode);
 622                 d_instantiate(dentry, inode);
 623         }
 624         return ret;
 625 }
 626
 627 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 628 {
 629         s64 writers;
 630         DEFINE_WAIT(wait);
 631
 632         do {
 633                 prepare_to_wait(&root->subv_writers->wait, &wait,
 634                                 TASK_UNINTERRUPTIBLE);
 635
 636                 writers = percpu_counter_sum(&root->subv_writers->counter);
 637                 if (writers)
 638                         schedule();
 639
 640                 finish_wait(&root->subv_writers->wait, &wait);
 641         } while (writers);
 642 }
 643
 644 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 645                            struct dentry *dentry, char *name, int namelen,
 646                            u64 *async_transid, bool readonly,
 647                            struct btrfs_qgroup_inherit *inherit)
 648 {
 649         struct inode *inode;
 650         struct btrfs_pending_snapshot *pending_snapshot;
 651         struct btrfs_trans_handle *trans;
 652         int ret;
 653
 654         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 655                 return -EINVAL;
 656
 657         atomic_inc(&root->will_be_snapshoted);
 658         smp_mb__after_atomic();
 659         btrfs_wait_for_no_snapshoting_writes(root);
 660
 661         ret = btrfs_start_delalloc_inodes(root, 0);
 662         if (ret)
 663                 goto out;
 664
 665         btrfs_wait_ordered_extents(root, -1);
 666
 667         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 668         if (!pending_snapshot) {
 669                 ret = -ENOMEM;
 670                 goto out;
 671         }
 672
 673         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 674                              BTRFS_BLOCK_RSV_TEMP);
 675         /*
 676          * 1 - parent dir inode
 677          * 2 - dir entries
 678          * 1 - root item
 679          * 2 - root ref/backref
 680          * 1 - root of snapshot
 681          * 1 - UUID item
 682          */
 683         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 684                                         &pending_snapshot->block_rsv, 8,
 685                                         &pending_snapshot->qgroup_reserved,
 686                                         false);
 687         if (ret)
 688                 goto free;
 689
 690         pending_snapshot->dentry = dentry;
 691         pending_snapshot->root = root;
 692         pending_snapshot->readonly = readonly;
 693         pending_snapshot->dir = dir;
 694         pending_snapshot->inherit = inherit;
 695
 696         trans = btrfs_start_transaction(root, 0);
 697         if (IS_ERR(trans)) {
 698                 ret = PTR_ERR(trans);
 699                 goto fail;
 700         }
 701
 702         spin_lock(&root->fs_info->trans_lock);
 703         list_add(&pending_snapshot->list,
 704                  &trans->transaction->pending_snapshots);
 705         spin_unlock(&root->fs_info->trans_lock);
 706         if (async_transid) {
 707                 *async_transid = trans->transid;
 708                 ret = btrfs_commit_transaction_async(trans,
 709                                      root->fs_info->extent_root, 1);
 710                 if (ret)
 711                         ret = btrfs_commit_transaction(trans, root);
 712         } else {
 713                 ret = btrfs_commit_transaction(trans,
 714                                                root->fs_info->extent_root);
 715         }
 716         if (ret)
 717                 goto fail;
 718
 719         ret = pending_snapshot->error;
 720         if (ret)
 721                 goto fail;
 722
 723         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 724         if (ret)
 725                 goto fail;
 726
 727         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 728         if (IS_ERR(inode)) {
 729                 ret = PTR_ERR(inode);
 730                 goto fail;
 731         }
 732
 733         d_instantiate(dentry, inode);
 734         ret = 0;
 735 fail:
 736         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 737                                          &pending_snapshot->block_rsv,
 738                                          pending_snapshot->qgroup_reserved);
 739 free:
 740         kfree(pending_snapshot);
 741 out:
 742         if (atomic_dec_and_test(&root->will_be_snapshoted))
 743                 wake_up_atomic_t(&root->will_be_snapshoted);
 744         return ret;
 745 }
 746
 747 /*  copy of may_delete in fs/namei.c()
 748  *      Check whether we can remove a link victim from directory dir, check
 749  *  whether the type of victim is right.
 750  *  1. We can't do it if dir is read-only (done in permission())
 751  *  2. We should have write and exec permissions on dir
 752  *  3. We can't remove anything from append-only dir
 753  *  4. We can't do anything with immutable dir (done in permission())
 754  *  5. If the sticky bit on dir is set we should either
 755  *      a. be owner of dir, or
 756  *      b. be owner of victim, or
 757  *      c. have CAP_FOWNER capability
 758  *  6. If the victim is append-only or immutable we can't do antyhing with
 759  *     links pointing to it.
 760  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 761  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 762  *  9. We can't remove a root or mountpoint.
 763  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 764  *     nfs_async_unlink().
 765  */
 766
 767 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 768 {
 769         int error;
 770
 771         if (d_really_is_negative(victim))
 772                 return -ENOENT;
 773
 774         BUG_ON(d_inode(victim->d_parent) != dir);
 775         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 776
 777         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 778         if (error)
 779                 return error;
 780         if (IS_APPEND(dir))
 781                 return -EPERM;
 782         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 783             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 784                 return -EPERM;
 785         if (isdir) {
 786                 if (!d_is_dir(victim))
 787                         return -ENOTDIR;
 788                 if (IS_ROOT(victim))
 789                         return -EBUSY;
 790         } else if (d_is_dir(victim))
 791                 return -EISDIR;
 792         if (IS_DEADDIR(dir))
 793                 return -ENOENT;
 794         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 795                 return -EBUSY;
 796         return 0;
 797 }
 798
 799 /* copy of may_create in fs/namei.c() */
 800 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 801 {
 802         if (d_really_is_positive(child))
 803                 return -EEXIST;
 804         if (IS_DEADDIR(dir))
 805                 return -ENOENT;
 806         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 807 }
 808
 809 /*
 810  * Create a new subvolume below @parent.  This is largely modeled after
 811  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 812  * inside this filesystem so it's quite a bit simpler.
 813  */
 814 static noinline int btrfs_mksubvol(struct path *parent,
 815                                    char *name, int namelen,
 816                                    struct btrfs_root *snap_src,
 817                                    u64 *async_transid, bool readonly,
 818                                    struct btrfs_qgroup_inherit *inherit)
 819 {
 820         struct inode *dir  = d_inode(parent->dentry);
 821         struct dentry *dentry;
 822         int error;
 823
 824         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 825         if (error == -EINTR)
 826                 return error;
 827
 828         dentry = lookup_one_len(name, parent->dentry, namelen);
 829         error = PTR_ERR(dentry);
 830         if (IS_ERR(dentry))
 831                 goto out_unlock;
 832
 833         error = -EEXIST;
 834         if (d_really_is_positive(dentry))
 835                 goto out_dput;
 836
 837         error = btrfs_may_create(dir, dentry);
 838         if (error)
 839                 goto out_dput;
 840
 841         /*
 842          * even if this name doesn't exist, we may get hash collisions.
 843          * check for them now when we can safely fail
 844          */
 845         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 846                                                dir->i_ino, name,
 847                                                namelen);
 848         if (error)
 849                 goto out_dput;
 850
 851         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 852
 853         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 854                 goto out_up_read;
 855
 856         if (snap_src) {
 857                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 858                                         async_transid, readonly, inherit);
 859         } else {
 860                 error = create_subvol(dir, dentry, name, namelen,
 861                                       async_transid, inherit);
 862         }
 863         if (!error)
 864                 fsnotify_mkdir(dir, dentry);
 865 out_up_read:
 866         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 867 out_dput:
 868         dput(dentry);
 869 out_unlock:
 870         mutex_unlock(&dir->i_mutex);
 871         return error;
 872 }
 873
 874 /*
 875  * When we're defragging a range, we don't want to kick it off again
 876  * if it is really just waiting for delalloc to send it down.
 877  * If we find a nice big extent or delalloc range for the bytes in the
 878  * file you want to defrag, we return 0 to let you know to skip this
 879  * part of the file
 880  */
 881 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 882 {
 883         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 884         struct extent_map *em = NULL;
 885         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 886         u64 end;
 887
 888         read_lock(&em_tree->lock);
 889         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 890         read_unlock(&em_tree->lock);
 891
 892         if (em) {
 893                 end = extent_map_end(em);
 894                 free_extent_map(em);
 895                 if (end - offset > thresh)
 896                         return 0;
 897         }
 898         /* if we already have a nice delalloc here, just stop */
 899         thresh /= 2;
 900         end = count_range_bits(io_tree, &offset, offset + thresh,
 901                                thresh, EXTENT_DELALLOC, 1);
 902         if (end >= thresh)
 903                 return 0;
 904         return 1;
 905 }
 906
 907 /*
 908  * helper function to walk through a file and find extents
 909  * newer than a specific transid, and smaller than thresh.
 910  *
 911  * This is used by the defragging code to find new and small
 912  * extents
 913  */
 914 static int find_new_extents(struct btrfs_root *root,
 915                             struct inode *inode, u64 newer_than,
 916                             u64 *off, u32 thresh)
 917 {
 918         struct btrfs_path *path;
 919         struct btrfs_key min_key;
 920         struct extent_buffer *leaf;
 921         struct btrfs_file_extent_item *extent;
 922         int type;
 923         int ret;
 924         u64 ino = btrfs_ino(inode);
 925
 926         path = btrfs_alloc_path();
 927         if (!path)
 928                 return -ENOMEM;
 929
 930         min_key.objectid = ino;
 931         min_key.type = BTRFS_EXTENT_DATA_KEY;
 932         min_key.offset = *off;
 933
 934         while (1) {
 935                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 936                 if (ret != 0)
 937                         goto none;
 938 process_slot:
 939                 if (min_key.objectid != ino)
 940                         goto none;
 941                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 942                         goto none;
 943
 944                 leaf = path->nodes[0];
 945                 extent = btrfs_item_ptr(leaf, path->slots[0],
 946                                         struct btrfs_file_extent_item);
 947
 948                 type = btrfs_file_extent_type(leaf, extent);
 949                 if (type == BTRFS_FILE_EXTENT_REG &&
 950                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 951                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 952                         *off = min_key.offset;
 953                         btrfs_free_path(path);
 954                         return 0;
 955                 }
 956
 957                 path->slots[0]++;
 958                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 959                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 960                         goto process_slot;
 961                 }
 962
 963                 if (min_key.offset == (u64)-1)
 964                         goto none;
 965
 966                 min_key.offset++;
 967                 btrfs_release_path(path);
 968         }
 969 none:
 970         btrfs_free_path(path);
 971         return -ENOENT;
 972 }
 973
 974 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 975 {
 976         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 977         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 978         struct extent_map *em;
 979         u64 len = PAGE_CACHE_SIZE;
 980
 981         /*
 982          * hopefully we have this extent in the tree already, try without
 983          * the full extent lock
 984          */
 985         read_lock(&em_tree->lock);
 986         em = lookup_extent_mapping(em_tree, start, len);
 987         read_unlock(&em_tree->lock);
 988
 989         if (!em) {
 990                 struct extent_state *cached = NULL;
 991                 u64 end = start + len - 1;
 992
 993                 /* get the big lock and read metadata off disk */
 994                 lock_extent_bits(io_tree, start, end, 0, &cached);
 995                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 996                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 997
 998                 if (IS_ERR(em))
 999                         return NULL;
1000         }
1001
1002         return em;
1003 }
1004
1005 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1006 {
1007         struct extent_map *next;
1008         bool ret = true;
1009
1010         /* this is the last extent */
1011         if (em->start + em->len >= i_size_read(inode))
1012                 return false;
1013
1014         next = defrag_lookup_extent(inode, em->start + em->len);
1015         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1016                 ret = false;
1017         else if ((em->block_start + em->block_len == next->block_start) &&
1018                  (em->block_len > 128 * 1024 && next->block_len > 128 * 1024))
1019                 ret = false;
1020
1021         free_extent_map(next);
1022         return ret;
1023 }
1024
1025 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1026                                u64 *last_len, u64 *skip, u64 *defrag_end,
1027                                int compress)
1028 {
1029         struct extent_map *em;
1030         int ret = 1;
1031         bool next_mergeable = true;
1032
1033         /*
1034          * make sure that once we start defragging an extent, we keep on
1035          * defragging it
1036          */
1037         if (start < *defrag_end)
1038                 return 1;
1039
1040         *skip = 0;
1041
1042         em = defrag_lookup_extent(inode, start);
1043         if (!em)
1044                 return 0;
1045
1046         /* this will cover holes, and inline extents */
1047         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1048                 ret = 0;
1049                 goto out;
1050         }
1051
1052         next_mergeable = defrag_check_next_extent(inode, em);
1053         /*
1054          * we hit a real extent, if it is big or the next extent is not a
1055          * real extent, don't bother defragging it
1056          */
1057         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1058             (em->len >= thresh || !next_mergeable))
1059                 ret = 0;
1060 out:
1061         /*
1062          * last_len ends up being a counter of how many bytes we've defragged.
1063          * every time we choose not to defrag an extent, we reset *last_len
1064          * so that the next tiny extent will force a defrag.
1065          *
1066          * The end result of this is that tiny extents before a single big
1067          * extent will force at least part of that big extent to be defragged.
1068          */
1069         if (ret) {
1070                 *defrag_end = extent_map_end(em);
1071         } else {
1072                 *last_len = 0;
1073                 *skip = extent_map_end(em);
1074                 *defrag_end = 0;
1075         }
1076
1077         free_extent_map(em);
1078         return ret;
1079 }
1080
1081 /*
1082  * it doesn't do much good to defrag one or two pages
1083  * at a time.  This pulls in a nice chunk of pages
1084  * to COW and defrag.
1085  *
1086  * It also makes sure the delalloc code has enough
1087  * dirty data to avoid making new small extents as part
1088  * of the defrag
1089  *
1090  * It's a good idea to start RA on this range
1091  * before calling this.
1092  */
1093 static int cluster_pages_for_defrag(struct inode *inode,
1094                                     struct page **pages,
1095                                     unsigned long start_index,
1096                                     unsigned long num_pages)
1097 {
1098         unsigned long file_end;
1099         u64 isize = i_size_read(inode);
1100         u64 page_start;
1101         u64 page_end;
1102         u64 page_cnt;
1103         int ret;
1104         int i;
1105         int i_done;
1106         struct btrfs_ordered_extent *ordered;
1107         struct extent_state *cached_state = NULL;
1108         struct extent_io_tree *tree;
1109         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1110
1111         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1112         if (!isize || start_index > file_end)
1113                 return 0;
1114
1115         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1116
1117         ret = btrfs_delalloc_reserve_space(inode,
1118                                            page_cnt << PAGE_CACHE_SHIFT);
1119         if (ret)
1120                 return ret;
1121         i_done = 0;
1122         tree = &BTRFS_I(inode)->io_tree;
1123
1124         /* step one, lock all the pages */
1125         for (i = 0; i < page_cnt; i++) {
1126                 struct page *page;
1127 again:
1128                 page = find_or_create_page(inode->i_mapping,
1129                                            start_index + i, mask);
1130                 if (!page)
1131                         break;
1132
1133                 page_start = page_offset(page);
1134                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1135                 while (1) {
1136                         lock_extent_bits(tree, page_start, page_end,
1137                                          0, &cached_state);
1138                         ordered = btrfs_lookup_ordered_extent(inode,
1139                                                               page_start);
1140                         unlock_extent_cached(tree, page_start, page_end,
1141                                              &cached_state, GFP_NOFS);
1142                         if (!ordered)
1143                                 break;
1144
1145                         unlock_page(page);
1146                         btrfs_start_ordered_extent(inode, ordered, 1);
1147                         btrfs_put_ordered_extent(ordered);
1148                         lock_page(page);
1149                         /*
1150                          * we unlocked the page above, so we need check if
1151                          * it was released or not.
1152                          */
1153                         if (page->mapping != inode->i_mapping) {
1154                                 unlock_page(page);
1155                                 page_cache_release(page);
1156                                 goto again;
1157                         }
1158                 }
1159
1160                 if (!PageUptodate(page)) {
1161                         btrfs_readpage(NULL, page);
1162                         lock_page(page);
1163                         if (!PageUptodate(page)) {
1164                                 unlock_page(page);
1165                                 page_cache_release(page);
1166                                 ret = -EIO;
1167                                 break;
1168                         }
1169                 }
1170
1171                 if (page->mapping != inode->i_mapping) {
1172                         unlock_page(page);
1173                         page_cache_release(page);
1174                         goto again;
1175                 }
1176
1177                 pages[i] = page;
1178                 i_done++;
1179         }
1180         if (!i_done || ret)
1181                 goto out;
1182
1183         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1184                 goto out;
1185
1186         /*
1187          * so now we have a nice long stream of locked
1188          * and up to date pages, lets wait on them
1189          */
1190         for (i = 0; i < i_done; i++)
1191                 wait_on_page_writeback(pages[i]);
1192
1193         page_start = page_offset(pages[0]);
1194         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1195
1196         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1197                          page_start, page_end - 1, 0, &cached_state);
1198         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1199                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1200                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1201                           &cached_state, GFP_NOFS);
1202
1203         if (i_done != page_cnt) {
1204                 spin_lock(&BTRFS_I(inode)->lock);
1205                 BTRFS_I(inode)->outstanding_extents++;
1206                 spin_unlock(&BTRFS_I(inode)->lock);
1207                 btrfs_delalloc_release_space(inode,
1208                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1209         }
1210
1211
1212         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1213                           &cached_state, GFP_NOFS);
1214
1215         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1216                              page_start, page_end - 1, &cached_state,
1217                              GFP_NOFS);
1218
1219         for (i = 0; i < i_done; i++) {
1220                 clear_page_dirty_for_io(pages[i]);
1221                 ClearPageChecked(pages[i]);
1222                 set_page_extent_mapped(pages[i]);
1223                 set_page_dirty(pages[i]);
1224                 unlock_page(pages[i]);
1225                 page_cache_release(pages[i]);
1226         }
1227         return i_done;
1228 out:
1229         for (i = 0; i < i_done; i++) {
1230                 unlock_page(pages[i]);
1231                 page_cache_release(pages[i]);
1232         }
1233         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1234         return ret;
1235
1236 }
1237
1238 int btrfs_defrag_file(struct inode *inode, struct file *file,
1239                       struct btrfs_ioctl_defrag_range_args *range,
1240                       u64 newer_than, unsigned long max_to_defrag)
1241 {
1242         struct btrfs_root *root = BTRFS_I(inode)->root;
1243         struct file_ra_state *ra = NULL;
1244         unsigned long last_index;
1245         u64 isize = i_size_read(inode);
1246         u64 last_len = 0;
1247         u64 skip = 0;
1248         u64 defrag_end = 0;
1249         u64 newer_off = range->start;
1250         unsigned long i;
1251         unsigned long ra_index = 0;
1252         int ret;
1253         int defrag_count = 0;
1254         int compress_type = BTRFS_COMPRESS_ZLIB;
1255         u32 extent_thresh = range->extent_thresh;
1256         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1257         unsigned long cluster = max_cluster;
1258         u64 new_align = ~((u64)128 * 1024 - 1);
1259         struct page **pages = NULL;
1260
1261         if (isize == 0)
1262                 return 0;
1263
1264         if (range->start >= isize)
1265                 return -EINVAL;
1266
1267         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1268                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1269                         return -EINVAL;
1270                 if (range->compress_type)
1271                         compress_type = range->compress_type;
1272         }
1273
1274         if (extent_thresh == 0)
1275                 extent_thresh = 256 * 1024;
1276
1277         /*
1278          * if we were not given a file, allocate a readahead
1279          * context
1280          */
1281         if (!file) {
1282                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1283                 if (!ra)
1284                         return -ENOMEM;
1285                 file_ra_state_init(ra, inode->i_mapping);
1286         } else {
1287                 ra = &file->f_ra;
1288         }
1289
1290         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1291                         GFP_NOFS);
1292         if (!pages) {
1293                 ret = -ENOMEM;
1294                 goto out_ra;
1295         }
1296
1297         /* find the last page to defrag */
1298         if (range->start + range->len > range->start) {
1299                 last_index = min_t(u64, isize - 1,
1300                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1301         } else {
1302                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1303         }
1304
1305         if (newer_than) {
1306                 ret = find_new_extents(root, inode, newer_than,
1307                                        &newer_off, 64 * 1024);
1308                 if (!ret) {
1309                         range->start = newer_off;
1310                         /*
1311                          * we always align our defrag to help keep
1312                          * the extents in the file evenly spaced
1313                          */
1314                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1315                 } else
1316                         goto out_ra;
1317         } else {
1318                 i = range->start >> PAGE_CACHE_SHIFT;
1319         }
1320         if (!max_to_defrag)
1321                 max_to_defrag = last_index + 1;
1322
1323         /*
1324          * make writeback starts from i, so the defrag range can be
1325          * written sequentially.
1326          */
1327         if (i < inode->i_mapping->writeback_index)
1328                 inode->i_mapping->writeback_index = i;
1329
1330         while (i <= last_index && defrag_count < max_to_defrag &&
1331                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1332                 /*
1333                  * make sure we stop running if someone unmounts
1334                  * the FS
1335                  */
1336                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1337                         break;
1338
1339                 if (btrfs_defrag_cancelled(root->fs_info)) {
1340                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1341                         ret = -EAGAIN;
1342                         break;
1343                 }
1344
1345                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1346                                          extent_thresh, &last_len, &skip,
1347                                          &defrag_end, range->flags &
1348                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1349                         unsigned long next;
1350                         /*
1351                          * the should_defrag function tells us how much to skip
1352                          * bump our counter by the suggested amount
1353                          */
1354                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1355                         i = max(i + 1, next);
1356                         continue;
1357                 }
1358
1359                 if (!newer_than) {
1360                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1361                                    PAGE_CACHE_SHIFT) - i;
1362                         cluster = min(cluster, max_cluster);
1363                 } else {
1364                         cluster = max_cluster;
1365                 }
1366
1367                 if (i + cluster > ra_index) {
1368                         ra_index = max(i, ra_index);
1369                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1370                                        cluster);
1371                         ra_index += max_cluster;
1372                 }
1373
1374                 mutex_lock(&inode->i_mutex);
1375                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1376                         BTRFS_I(inode)->force_compress = compress_type;
1377                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1378                 if (ret < 0) {
1379                         mutex_unlock(&inode->i_mutex);
1380                         goto out_ra;
1381                 }
1382
1383                 defrag_count += ret;
1384                 balance_dirty_pages_ratelimited(inode->i_mapping);
1385                 mutex_unlock(&inode->i_mutex);
1386
1387                 if (newer_than) {
1388                         if (newer_off == (u64)-1)
1389                                 break;
1390
1391                         if (ret > 0)
1392                                 i += ret;
1393
1394                         newer_off = max(newer_off + 1,
1395                                         (u64)i << PAGE_CACHE_SHIFT);
1396
1397                         ret = find_new_extents(root, inode,
1398                                                newer_than, &newer_off,
1399                                                64 * 1024);
1400                         if (!ret) {
1401                                 range->start = newer_off;
1402                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1403                         } else {
1404                                 break;
1405                         }
1406                 } else {
1407                         if (ret > 0) {
1408                                 i += ret;
1409                                 last_len += ret << PAGE_CACHE_SHIFT;
1410                         } else {
1411                                 i++;
1412                                 last_len = 0;
1413                         }
1414                 }
1415         }
1416
1417         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1418                 filemap_flush(inode->i_mapping);
1419                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1420                              &BTRFS_I(inode)->runtime_flags))
1421                         filemap_flush(inode->i_mapping);
1422         }
1423
1424         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1425                 /* the filemap_flush will queue IO into the worker threads, but
1426                  * we have to make sure the IO is actually started and that
1427                  * ordered extents get created before we return
1428                  */
1429                 atomic_inc(&root->fs_info->async_submit_draining);
1430                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1431                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1432                         wait_event(root->fs_info->async_submit_wait,
1433                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1434                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1435                 }
1436                 atomic_dec(&root->fs_info->async_submit_draining);
1437         }
1438
1439         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1440                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1441         }
1442
1443         ret = defrag_count;
1444
1445 out_ra:
1446         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1447                 mutex_lock(&inode->i_mutex);
1448                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1449                 mutex_unlock(&inode->i_mutex);
1450         }
1451         if (!file)
1452                 kfree(ra);
1453         kfree(pages);
1454         return ret;
1455 }
1456
1457 static noinline int btrfs_ioctl_resize(struct file *file,
1458                                         void __user *arg)
1459 {
1460         u64 new_size;
1461         u64 old_size;
1462         u64 devid = 1;
1463         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1464         struct btrfs_ioctl_vol_args *vol_args;
1465         struct btrfs_trans_handle *trans;
1466         struct btrfs_device *device = NULL;
1467         char *sizestr;
1468         char *retptr;
1469         char *devstr = NULL;
1470         int ret = 0;
1471         int mod = 0;
1472
1473         if (!capable(CAP_SYS_ADMIN))
1474                 return -EPERM;
1475
1476         ret = mnt_want_write_file(file);
1477         if (ret)
1478                 return ret;
1479
1480         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1481                         1)) {
1482                 mnt_drop_write_file(file);
1483                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1484         }
1485
1486         mutex_lock(&root->fs_info->volume_mutex);
1487         vol_args = memdup_user(arg, sizeof(*vol_args));
1488         if (IS_ERR(vol_args)) {
1489                 ret = PTR_ERR(vol_args);
1490                 goto out;
1491         }
1492
1493         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1494
1495         sizestr = vol_args->name;
1496         devstr = strchr(sizestr, ':');
1497         if (devstr) {
1498                 sizestr = devstr + 1;
1499                 *devstr = '\0';
1500                 devstr = vol_args->name;
1501                 ret = kstrtoull(devstr, 10, &devid);
1502                 if (ret)
1503                         goto out_free;
1504                 if (!devid) {
1505                         ret = -EINVAL;
1506                         goto out_free;
1507                 }
1508                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1509         }
1510
1511         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1512         if (!device) {
1513                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1514                        devid);
1515                 ret = -ENODEV;
1516                 goto out_free;
1517         }
1518
1519         if (!device->writeable) {
1520                 btrfs_info(root->fs_info,
1521                            "resizer unable to apply on readonly device %llu",
1522                        devid);
1523                 ret = -EPERM;
1524                 goto out_free;
1525         }
1526
1527         if (!strcmp(sizestr, "max"))
1528                 new_size = device->bdev->bd_inode->i_size;
1529         else {
1530                 if (sizestr[0] == '-') {
1531                         mod = -1;
1532                         sizestr++;
1533                 } else if (sizestr[0] == '+') {
1534                         mod = 1;
1535                         sizestr++;
1536                 }
1537                 new_size = memparse(sizestr, &retptr);
1538                 if (*retptr != '\0' || new_size == 0) {
1539                         ret = -EINVAL;
1540                         goto out_free;
1541                 }
1542         }
1543
1544         if (device->is_tgtdev_for_dev_replace) {
1545                 ret = -EPERM;
1546                 goto out_free;
1547         }
1548
1549         old_size = btrfs_device_get_total_bytes(device);
1550
1551         if (mod < 0) {
1552                 if (new_size > old_size) {
1553                         ret = -EINVAL;
1554                         goto out_free;
1555                 }
1556                 new_size = old_size - new_size;
1557         } else if (mod > 0) {
1558                 if (new_size > ULLONG_MAX - old_size) {
1559                         ret = -ERANGE;
1560                         goto out_free;
1561                 }
1562                 new_size = old_size + new_size;
1563         }
1564
1565         if (new_size < 256 * 1024 * 1024) {
1566                 ret = -EINVAL;
1567                 goto out_free;
1568         }
1569         if (new_size > device->bdev->bd_inode->i_size) {
1570                 ret = -EFBIG;
1571                 goto out_free;
1572         }
1573
1574         new_size = div_u64(new_size, root->sectorsize);
1575         new_size *= root->sectorsize;
1576
1577         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1578                       rcu_str_deref(device->name), new_size);
1579
1580         if (new_size > old_size) {
1581                 trans = btrfs_start_transaction(root, 0);
1582                 if (IS_ERR(trans)) {
1583                         ret = PTR_ERR(trans);
1584                         goto out_free;
1585                 }
1586                 ret = btrfs_grow_device(trans, device, new_size);
1587                 btrfs_commit_transaction(trans, root);
1588         } else if (new_size < old_size) {
1589                 ret = btrfs_shrink_device(device, new_size);
1590         } /* equal, nothing need to do */
1591
1592 out_free:
1593         kfree(vol_args);
1594 out:
1595         mutex_unlock(&root->fs_info->volume_mutex);
1596         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1597         mnt_drop_write_file(file);
1598         return ret;
1599 }
1600
1601 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1602                                 char *name, unsigned long fd, int subvol,
1603                                 u64 *transid, bool readonly,
1604                                 struct btrfs_qgroup_inherit *inherit)
1605 {
1606         int namelen;
1607         int ret = 0;
1608
1609         ret = mnt_want_write_file(file);
1610         if (ret)
1611                 goto out;
1612
1613         namelen = strlen(name);
1614         if (strchr(name, '/')) {
1615                 ret = -EINVAL;
1616                 goto out_drop_write;
1617         }
1618
1619         if (name[0] == '.' &&
1620            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1621                 ret = -EEXIST;
1622                 goto out_drop_write;
1623         }
1624
1625         if (subvol) {
1626                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1627                                      NULL, transid, readonly, inherit);
1628         } else {
1629                 struct fd src = fdget(fd);
1630                 struct inode *src_inode;
1631                 if (!src.file) {
1632                         ret = -EINVAL;
1633                         goto out_drop_write;
1634                 }
1635
1636                 src_inode = file_inode(src.file);
1637                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1638                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1639                                    "Snapshot src from another FS");
1640                         ret = -EXDEV;
1641                 } else if (!inode_owner_or_capable(src_inode)) {
1642                         /*
1643                          * Subvolume creation is not restricted, but snapshots
1644                          * are limited to own subvolumes only
1645                          */
1646                         ret = -EPERM;
1647                 } else {
1648                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1649                                              BTRFS_I(src_inode)->root,
1650                                              transid, readonly, inherit);
1651                 }
1652                 fdput(src);
1653         }
1654 out_drop_write:
1655         mnt_drop_write_file(file);
1656 out:
1657         return ret;
1658 }
1659
1660 static noinline int btrfs_ioctl_snap_create(struct file *file,
1661                                             void __user *arg, int subvol)
1662 {
1663         struct btrfs_ioctl_vol_args *vol_args;
1664         int ret;
1665
1666         vol_args = memdup_user(arg, sizeof(*vol_args));
1667         if (IS_ERR(vol_args))
1668                 return PTR_ERR(vol_args);
1669         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1670
1671         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1672                                               vol_args->fd, subvol,
1673                                               NULL, false, NULL);
1674
1675         kfree(vol_args);
1676         return ret;
1677 }
1678
1679 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1680                                                void __user *arg, int subvol)
1681 {
1682         struct btrfs_ioctl_vol_args_v2 *vol_args;
1683         int ret;
1684         u64 transid = 0;
1685         u64 *ptr = NULL;
1686         bool readonly = false;
1687         struct btrfs_qgroup_inherit *inherit = NULL;
1688
1689         vol_args = memdup_user(arg, sizeof(*vol_args));
1690         if (IS_ERR(vol_args))
1691                 return PTR_ERR(vol_args);
1692         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1693
1694         if (vol_args->flags &
1695             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1696               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1697                 ret = -EOPNOTSUPP;
1698                 goto free_args;
1699         }
1700
1701         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1702                 ptr = &transid;
1703         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1704                 readonly = true;
1705         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1706                 if (vol_args->size > PAGE_CACHE_SIZE) {
1707                         ret = -EINVAL;
1708                         goto free_args;
1709                 }
1710                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1711                 if (IS_ERR(inherit)) {
1712                         ret = PTR_ERR(inherit);
1713                         goto free_args;
1714                 }
1715         }
1716
1717         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1718                                               vol_args->fd, subvol, ptr,
1719                                               readonly, inherit);
1720         if (ret)
1721                 goto free_inherit;
1722
1723         if (ptr && copy_to_user(arg +
1724                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1725                                         transid),
1726                                 ptr, sizeof(*ptr)))
1727                 ret = -EFAULT;
1728
1729 free_inherit:
1730         kfree(inherit);
1731 free_args:
1732         kfree(vol_args);
1733         return ret;
1734 }
1735
1736 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1737                                                 void __user *arg)
1738 {
1739         struct inode *inode = file_inode(file);
1740         struct btrfs_root *root = BTRFS_I(inode)->root;
1741         int ret = 0;
1742         u64 flags = 0;
1743
1744         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1745                 return -EINVAL;
1746
1747         down_read(&root->fs_info->subvol_sem);
1748         if (btrfs_root_readonly(root))
1749                 flags |= BTRFS_SUBVOL_RDONLY;
1750         up_read(&root->fs_info->subvol_sem);
1751
1752         if (copy_to_user(arg, &flags, sizeof(flags)))
1753                 ret = -EFAULT;
1754
1755         return ret;
1756 }
1757
1758 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1759                                               void __user *arg)
1760 {
1761         struct inode *inode = file_inode(file);
1762         struct btrfs_root *root = BTRFS_I(inode)->root;
1763         struct btrfs_trans_handle *trans;
1764         u64 root_flags;
1765         u64 flags;
1766         int ret = 0;
1767
1768         if (!inode_owner_or_capable(inode))
1769                 return -EPERM;
1770
1771         ret = mnt_want_write_file(file);
1772         if (ret)
1773                 goto out;
1774
1775         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1776                 ret = -EINVAL;
1777                 goto out_drop_write;
1778         }
1779
1780         if (copy_from_user(&flags, arg, sizeof(flags))) {
1781                 ret = -EFAULT;
1782                 goto out_drop_write;
1783         }
1784
1785         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1786                 ret = -EINVAL;
1787                 goto out_drop_write;
1788         }
1789
1790         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1791                 ret = -EOPNOTSUPP;
1792                 goto out_drop_write;
1793         }
1794
1795         down_write(&root->fs_info->subvol_sem);
1796
1797         /* nothing to do */
1798         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1799                 goto out_drop_sem;
1800
1801         root_flags = btrfs_root_flags(&root->root_item);
1802         if (flags & BTRFS_SUBVOL_RDONLY) {
1803                 btrfs_set_root_flags(&root->root_item,
1804                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1805         } else {
1806                 /*
1807                  * Block RO -> RW transition if this subvolume is involved in
1808                  * send
1809                  */
1810                 spin_lock(&root->root_item_lock);
1811                 if (root->send_in_progress == 0) {
1812                         btrfs_set_root_flags(&root->root_item,
1813                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1814                         spin_unlock(&root->root_item_lock);
1815                 } else {
1816                         spin_unlock(&root->root_item_lock);
1817                         btrfs_warn(root->fs_info,
1818                         "Attempt to set subvolume %llu read-write during send",
1819                                         root->root_key.objectid);
1820                         ret = -EPERM;
1821                         goto out_drop_sem;
1822                 }
1823         }
1824
1825         trans = btrfs_start_transaction(root, 1);
1826         if (IS_ERR(trans)) {
1827                 ret = PTR_ERR(trans);
1828                 goto out_reset;
1829         }
1830
1831         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1832                                 &root->root_key, &root->root_item);
1833
1834         btrfs_commit_transaction(trans, root);
1835 out_reset:
1836         if (ret)
1837                 btrfs_set_root_flags(&root->root_item, root_flags);
1838 out_drop_sem:
1839         up_write(&root->fs_info->subvol_sem);
1840 out_drop_write:
1841         mnt_drop_write_file(file);
1842 out:
1843         return ret;
1844 }
1845
1846 /*
1847  * helper to check if the subvolume references other subvolumes
1848  */
1849 static noinline int may_destroy_subvol(struct btrfs_root *root)
1850 {
1851         struct btrfs_path *path;
1852         struct btrfs_dir_item *di;
1853         struct btrfs_key key;
1854         u64 dir_id;
1855         int ret;
1856
1857         path = btrfs_alloc_path();
1858         if (!path)
1859                 return -ENOMEM;
1860
1861         /* Make sure this root isn't set as the default subvol */
1862         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1863         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1864                                    dir_id, "default", 7, 0);
1865         if (di && !IS_ERR(di)) {
1866                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1867                 if (key.objectid == root->root_key.objectid) {
1868                         ret = -EPERM;
1869                         btrfs_err(root->fs_info, "deleting default subvolume "
1870                                   "%llu is not allowed", key.objectid);
1871                         goto out;
1872                 }
1873                 btrfs_release_path(path);
1874         }
1875
1876         key.objectid = root->root_key.objectid;
1877         key.type = BTRFS_ROOT_REF_KEY;
1878         key.offset = (u64)-1;
1879
1880         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1881                                 &key, path, 0, 0);
1882         if (ret < 0)
1883                 goto out;
1884         BUG_ON(ret == 0);
1885
1886         ret = 0;
1887         if (path->slots[0] > 0) {
1888                 path->slots[0]--;
1889                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1890                 if (key.objectid == root->root_key.objectid &&
1891                     key.type == BTRFS_ROOT_REF_KEY)
1892                         ret = -ENOTEMPTY;
1893         }
1894 out:
1895         btrfs_free_path(path);
1896         return ret;
1897 }
1898
1899 static noinline int key_in_sk(struct btrfs_key *key,
1900                               struct btrfs_ioctl_search_key *sk)
1901 {
1902         struct btrfs_key test;
1903         int ret;
1904
1905         test.objectid = sk->min_objectid;
1906         test.type = sk->min_type;
1907         test.offset = sk->min_offset;
1908
1909         ret = btrfs_comp_cpu_keys(key, &test);
1910         if (ret < 0)
1911                 return 0;
1912
1913         test.objectid = sk->max_objectid;
1914         test.type = sk->max_type;
1915         test.offset = sk->max_offset;
1916
1917         ret = btrfs_comp_cpu_keys(key, &test);
1918         if (ret > 0)
1919                 return 0;
1920         return 1;
1921 }
1922
1923 static noinline int copy_to_sk(struct btrfs_root *root,
1924                                struct btrfs_path *path,
1925                                struct btrfs_key *key,
1926                                struct btrfs_ioctl_search_key *sk,
1927                                size_t *buf_size,
1928                                char __user *ubuf,
1929                                unsigned long *sk_offset,
1930                                int *num_found)
1931 {
1932         u64 found_transid;
1933         struct extent_buffer *leaf;
1934         struct btrfs_ioctl_search_header sh;
1935         unsigned long item_off;
1936         unsigned long item_len;
1937         int nritems;
1938         int i;
1939         int slot;
1940         int ret = 0;
1941
1942         leaf = path->nodes[0];
1943         slot = path->slots[0];
1944         nritems = btrfs_header_nritems(leaf);
1945
1946         if (btrfs_header_generation(leaf) > sk->max_transid) {
1947                 i = nritems;
1948                 goto advance_key;
1949         }
1950         found_transid = btrfs_header_generation(leaf);
1951
1952         for (i = slot; i < nritems; i++) {
1953                 item_off = btrfs_item_ptr_offset(leaf, i);
1954                 item_len = btrfs_item_size_nr(leaf, i);
1955
1956                 btrfs_item_key_to_cpu(leaf, key, i);
1957                 if (!key_in_sk(key, sk))
1958                         continue;
1959
1960                 if (sizeof(sh) + item_len > *buf_size) {
1961                         if (*num_found) {
1962                                 ret = 1;
1963                                 goto out;
1964                         }
1965
1966                         /*
1967                          * return one empty item back for v1, which does not
1968                          * handle -EOVERFLOW
1969                          */
1970
1971                         *buf_size = sizeof(sh) + item_len;
1972                         item_len = 0;
1973                         ret = -EOVERFLOW;
1974                 }
1975
1976                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1977                         ret = 1;
1978                         goto out;
1979                 }
1980
1981                 sh.objectid = key->objectid;
1982                 sh.offset = key->offset;
1983                 sh.type = key->type;
1984                 sh.len = item_len;
1985                 sh.transid = found_transid;
1986
1987                 /* copy search result header */
1988                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1989                         ret = -EFAULT;
1990                         goto out;
1991                 }
1992
1993                 *sk_offset += sizeof(sh);
1994
1995                 if (item_len) {
1996                         char __user *up = ubuf + *sk_offset;
1997                         /* copy the item */
1998                         if (read_extent_buffer_to_user(leaf, up,
1999                                                        item_off, item_len)) {
2000                                 ret = -EFAULT;
2001                                 goto out;
2002                         }
2003
2004                         *sk_offset += item_len;
2005                 }
2006                 (*num_found)++;
2007
2008                 if (ret) /* -EOVERFLOW from above */
2009                         goto out;
2010
2011                 if (*num_found >= sk->nr_items) {
2012                         ret = 1;
2013                         goto out;
2014                 }
2015         }
2016 advance_key:
2017         ret = 0;
2018         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2019                 key->offset++;
2020         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2021                 key->offset = 0;
2022                 key->type++;
2023         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2024                 key->offset = 0;
2025                 key->type = 0;
2026                 key->objectid++;
2027         } else
2028                 ret = 1;
2029 out:
2030         /*
2031          *  0: all items from this leaf copied, continue with next
2032          *  1: * more items can be copied, but unused buffer is too small
2033          *     * all items were found
2034          *     Either way, it will stops the loop which iterates to the next
2035          *     leaf
2036          *  -EOVERFLOW: item was to large for buffer
2037          *  -EFAULT: could not copy extent buffer back to userspace
2038          */
2039         return ret;
2040 }
2041
2042 static noinline int search_ioctl(struct inode *inode,
2043                                  struct btrfs_ioctl_search_key *sk,
2044                                  size_t *buf_size,
2045                                  char __user *ubuf)
2046 {
2047         struct btrfs_root *root;
2048         struct btrfs_key key;
2049         struct btrfs_path *path;
2050         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2051         int ret;
2052         int num_found = 0;
2053         unsigned long sk_offset = 0;
2054
2055         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2056                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2057                 return -EOVERFLOW;
2058         }
2059
2060         path = btrfs_alloc_path();
2061         if (!path)
2062                 return -ENOMEM;
2063
2064         if (sk->tree_id == 0) {
2065                 /* search the root of the inode that was passed */
2066                 root = BTRFS_I(inode)->root;
2067         } else {
2068                 key.objectid = sk->tree_id;
2069                 key.type = BTRFS_ROOT_ITEM_KEY;
2070                 key.offset = (u64)-1;
2071                 root = btrfs_read_fs_root_no_name(info, &key);
2072                 if (IS_ERR(root)) {
2073                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2074                                sk->tree_id);
2075                         btrfs_free_path(path);
2076                         return -ENOENT;
2077                 }
2078         }
2079
2080         key.objectid = sk->min_objectid;
2081         key.type = sk->min_type;
2082         key.offset = sk->min_offset;
2083
2084         while (1) {
2085                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2086                 if (ret != 0) {
2087                         if (ret > 0)
2088                                 ret = 0;
2089                         goto err;
2090                 }
2091                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2092                                  &sk_offset, &num_found);
2093                 btrfs_release_path(path);
2094                 if (ret)
2095                         break;
2096
2097         }
2098         if (ret > 0)
2099                 ret = 0;
2100 err:
2101         sk->nr_items = num_found;
2102         btrfs_free_path(path);
2103         return ret;
2104 }
2105
2106 static noinline int btrfs_ioctl_tree_search(struct file *file,
2107                                            void __user *argp)
2108 {
2109         struct btrfs_ioctl_search_args __user *uargs;
2110         struct btrfs_ioctl_search_key sk;
2111         struct inode *inode;
2112         int ret;
2113         size_t buf_size;
2114
2115         if (!capable(CAP_SYS_ADMIN))
2116                 return -EPERM;
2117
2118         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2119
2120         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2121                 return -EFAULT;
2122
2123         buf_size = sizeof(uargs->buf);
2124
2125         inode = file_inode(file);
2126         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2127
2128         /*
2129          * In the origin implementation an overflow is handled by returning a
2130          * search header with a len of zero, so reset ret.
2131          */
2132         if (ret == -EOVERFLOW)
2133                 ret = 0;
2134
2135         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2136                 ret = -EFAULT;
2137         return ret;
2138 }
2139
2140 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2141                                                void __user *argp)
2142 {
2143         struct btrfs_ioctl_search_args_v2 __user *uarg;
2144         struct btrfs_ioctl_search_args_v2 args;
2145         struct inode *inode;
2146         int ret;
2147         size_t buf_size;
2148         const size_t buf_limit = 16 * 1024 * 1024;
2149
2150         if (!capable(CAP_SYS_ADMIN))
2151                 return -EPERM;
2152
2153         /* copy search header and buffer size */
2154         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2155         if (copy_from_user(&args, uarg, sizeof(args)))
2156                 return -EFAULT;
2157
2158         buf_size = args.buf_size;
2159
2160         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2161                 return -EOVERFLOW;
2162
2163         /* limit result size to 16MB */
2164         if (buf_size > buf_limit)
2165                 buf_size = buf_limit;
2166
2167         inode = file_inode(file);
2168         ret = search_ioctl(inode, &args.key, &buf_size,
2169                            (char *)(&uarg->buf[0]));
2170         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2171                 ret = -EFAULT;
2172         else if (ret == -EOVERFLOW &&
2173                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2174                 ret = -EFAULT;
2175
2176         return ret;
2177 }
2178
2179 /*
2180  * Search INODE_REFs to identify path name of 'dirid' directory
2181  * in a 'tree_id' tree. and sets path name to 'name'.
2182  */
2183 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2184                                 u64 tree_id, u64 dirid, char *name)
2185 {
2186         struct btrfs_root *root;
2187         struct btrfs_key key;
2188         char *ptr;
2189         int ret = -1;
2190         int slot;
2191         int len;
2192         int total_len = 0;
2193         struct btrfs_inode_ref *iref;
2194         struct extent_buffer *l;
2195         struct btrfs_path *path;
2196
2197         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2198                 name[0]='\0';
2199                 return 0;
2200         }
2201
2202         path = btrfs_alloc_path();
2203         if (!path)
2204                 return -ENOMEM;
2205
2206         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2207
2208         key.objectid = tree_id;
2209         key.type = BTRFS_ROOT_ITEM_KEY;
2210         key.offset = (u64)-1;
2211         root = btrfs_read_fs_root_no_name(info, &key);
2212         if (IS_ERR(root)) {
2213                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2214                 ret = -ENOENT;
2215                 goto out;
2216         }
2217
2218         key.objectid = dirid;
2219         key.type = BTRFS_INODE_REF_KEY;
2220         key.offset = (u64)-1;
2221
2222         while (1) {
2223                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2224                 if (ret < 0)
2225                         goto out;
2226                 else if (ret > 0) {
2227                         ret = btrfs_previous_item(root, path, dirid,
2228                                                   BTRFS_INODE_REF_KEY);
2229                         if (ret < 0)
2230                                 goto out;
2231                         else if (ret > 0) {
2232                                 ret = -ENOENT;
2233                                 goto out;
2234                         }
2235                 }
2236
2237                 l = path->nodes[0];
2238                 slot = path->slots[0];
2239                 btrfs_item_key_to_cpu(l, &key, slot);
2240
2241                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2242                 len = btrfs_inode_ref_name_len(l, iref);
2243                 ptr -= len + 1;
2244                 total_len += len + 1;
2245                 if (ptr < name) {
2246                         ret = -ENAMETOOLONG;
2247                         goto out;
2248                 }
2249
2250                 *(ptr + len) = '/';
2251                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2252
2253                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2254                         break;
2255
2256                 btrfs_release_path(path);
2257                 key.objectid = key.offset;
2258                 key.offset = (u64)-1;
2259                 dirid = key.objectid;
2260         }
2261         memmove(name, ptr, total_len);
2262         name[total_len] = '\0';
2263         ret = 0;
2264 out:
2265         btrfs_free_path(path);
2266         return ret;
2267 }
2268
2269 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2270                                            void __user *argp)
2271 {
2272          struct btrfs_ioctl_ino_lookup_args *args;
2273          struct inode *inode;
2274          int ret;
2275
2276         if (!capable(CAP_SYS_ADMIN))
2277                 return -EPERM;
2278
2279         args = memdup_user(argp, sizeof(*args));
2280         if (IS_ERR(args))
2281                 return PTR_ERR(args);
2282
2283         inode = file_inode(file);
2284
2285         if (args->treeid == 0)
2286                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2287
2288         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2289                                         args->treeid, args->objectid,
2290                                         args->name);
2291
2292         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2293                 ret = -EFAULT;
2294
2295         kfree(args);
2296         return ret;
2297 }
2298
2299 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2300                                              void __user *arg)
2301 {
2302         struct dentry *parent = file->f_path.dentry;
2303         struct dentry *dentry;
2304         struct inode *dir = d_inode(parent);
2305         struct inode *inode;
2306         struct btrfs_root *root = BTRFS_I(dir)->root;
2307         struct btrfs_root *dest = NULL;
2308         struct btrfs_ioctl_vol_args *vol_args;
2309         struct btrfs_trans_handle *trans;
2310         struct btrfs_block_rsv block_rsv;
2311         u64 root_flags;
2312         u64 qgroup_reserved;
2313         int namelen;
2314         int ret;
2315         int err = 0;
2316
2317         vol_args = memdup_user(arg, sizeof(*vol_args));
2318         if (IS_ERR(vol_args))
2319                 return PTR_ERR(vol_args);
2320
2321         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2322         namelen = strlen(vol_args->name);
2323         if (strchr(vol_args->name, '/') ||
2324             strncmp(vol_args->name, "..", namelen) == 0) {
2325                 err = -EINVAL;
2326                 goto out;
2327         }
2328
2329         err = mnt_want_write_file(file);
2330         if (err)
2331                 goto out;
2332
2333
2334         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2335         if (err == -EINTR)
2336                 goto out_drop_write;
2337         dentry = lookup_one_len(vol_args->name, parent, namelen);
2338         if (IS_ERR(dentry)) {
2339                 err = PTR_ERR(dentry);
2340                 goto out_unlock_dir;
2341         }
2342
2343         if (d_really_is_negative(dentry)) {
2344                 err = -ENOENT;
2345                 goto out_dput;
2346         }
2347
2348         inode = d_inode(dentry);
2349         dest = BTRFS_I(inode)->root;
2350         if (!capable(CAP_SYS_ADMIN)) {
2351                 /*
2352                  * Regular user.  Only allow this with a special mount
2353                  * option, when the user has write+exec access to the
2354                  * subvol root, and when rmdir(2) would have been
2355                  * allowed.
2356                  *
2357                  * Note that this is _not_ check that the subvol is
2358                  * empty or doesn't contain data that we wouldn't
2359                  * otherwise be able to delete.
2360                  *
2361                  * Users who want to delete empty subvols should try
2362                  * rmdir(2).
2363                  */
2364                 err = -EPERM;
2365                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2366                         goto out_dput;
2367
2368                 /*
2369                  * Do not allow deletion if the parent dir is the same
2370                  * as the dir to be deleted.  That means the ioctl
2371                  * must be called on the dentry referencing the root
2372                  * of the subvol, not a random directory contained
2373                  * within it.
2374                  */
2375                 err = -EINVAL;
2376                 if (root == dest)
2377                         goto out_dput;
2378
2379                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2380                 if (err)
2381                         goto out_dput;
2382         }
2383
2384         /* check if subvolume may be deleted by a user */
2385         err = btrfs_may_delete(dir, dentry, 1);
2386         if (err)
2387                 goto out_dput;
2388
2389         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2390                 err = -EINVAL;
2391                 goto out_dput;
2392         }
2393
2394         mutex_lock(&inode->i_mutex);
2395
2396         /*
2397          * Don't allow to delete a subvolume with send in progress. This is
2398          * inside the i_mutex so the error handling that has to drop the bit
2399          * again is not run concurrently.
2400          */
2401         spin_lock(&dest->root_item_lock);
2402         root_flags = btrfs_root_flags(&dest->root_item);
2403         if (dest->send_in_progress == 0) {
2404                 btrfs_set_root_flags(&dest->root_item,
2405                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2406                 spin_unlock(&dest->root_item_lock);
2407         } else {
2408                 spin_unlock(&dest->root_item_lock);
2409                 btrfs_warn(root->fs_info,
2410                         "Attempt to delete subvolume %llu during send",
2411                         dest->root_key.objectid);
2412                 err = -EPERM;
2413                 goto out_dput;
2414         }
2415
2416         d_invalidate(dentry);
2417
2418         down_write(&root->fs_info->subvol_sem);
2419
2420         err = may_destroy_subvol(dest);
2421         if (err)
2422                 goto out_up_write;
2423
2424         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2425         /*
2426          * One for dir inode, two for dir entries, two for root
2427          * ref/backref.
2428          */
2429         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2430                                                5, &qgroup_reserved, true);
2431         if (err)
2432                 goto out_up_write;
2433
2434         trans = btrfs_start_transaction(root, 0);
2435         if (IS_ERR(trans)) {
2436                 err = PTR_ERR(trans);
2437                 goto out_release;
2438         }
2439         trans->block_rsv = &block_rsv;
2440         trans->bytes_reserved = block_rsv.size;
2441
2442         ret = btrfs_unlink_subvol(trans, root, dir,
2443                                 dest->root_key.objectid,
2444                                 dentry->d_name.name,
2445                                 dentry->d_name.len);
2446         if (ret) {
2447                 err = ret;
2448                 btrfs_abort_transaction(trans, root, ret);
2449                 goto out_end_trans;
2450         }
2451
2452         btrfs_record_root_in_trans(trans, dest);
2453
2454         memset(&dest->root_item.drop_progress, 0,
2455                 sizeof(dest->root_item.drop_progress));
2456         dest->root_item.drop_level = 0;
2457         btrfs_set_root_refs(&dest->root_item, 0);
2458
2459         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2460                 ret = btrfs_insert_orphan_item(trans,
2461                                         root->fs_info->tree_root,
2462                                         dest->root_key.objectid);
2463                 if (ret) {
2464                         btrfs_abort_transaction(trans, root, ret);
2465                         err = ret;
2466                         goto out_end_trans;
2467                 }
2468         }
2469
2470         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2471                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2472                                   dest->root_key.objectid);
2473         if (ret && ret != -ENOENT) {
2474                 btrfs_abort_transaction(trans, root, ret);
2475                 err = ret;
2476                 goto out_end_trans;
2477         }
2478         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2479                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2480                                           dest->root_item.received_uuid,
2481                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2482                                           dest->root_key.objectid);
2483                 if (ret && ret != -ENOENT) {
2484                         btrfs_abort_transaction(trans, root, ret);
2485                         err = ret;
2486                         goto out_end_trans;
2487                 }
2488         }
2489
2490 out_end_trans:
2491         trans->block_rsv = NULL;
2492         trans->bytes_reserved = 0;
2493         ret = btrfs_end_transaction(trans, root);
2494         if (ret && !err)
2495                 err = ret;
2496         inode->i_flags |= S_DEAD;
2497 out_release:
2498         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2499 out_up_write:
2500         up_write(&root->fs_info->subvol_sem);
2501         if (err) {
2502                 spin_lock(&dest->root_item_lock);
2503                 root_flags = btrfs_root_flags(&dest->root_item);
2504                 btrfs_set_root_flags(&dest->root_item,
2505                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2506                 spin_unlock(&dest->root_item_lock);
2507         }
2508         mutex_unlock(&inode->i_mutex);
2509         if (!err) {
2510                 shrink_dcache_sb(root->fs_info->sb);
2511                 btrfs_invalidate_inodes(dest);
2512                 d_delete(dentry);
2513                 ASSERT(dest->send_in_progress == 0);
2514
2515                 /* the last ref */
2516                 if (dest->ino_cache_inode) {
2517                         iput(dest->ino_cache_inode);
2518                         dest->ino_cache_inode = NULL;
2519                 }
2520         }
2521 out_dput:
2522         dput(dentry);
2523 out_unlock_dir:
2524         mutex_unlock(&dir->i_mutex);
2525 out_drop_write:
2526         mnt_drop_write_file(file);
2527 out:
2528         kfree(vol_args);
2529         return err;
2530 }
2531
2532 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2533 {
2534         struct inode *inode = file_inode(file);
2535         struct btrfs_root *root = BTRFS_I(inode)->root;
2536         struct btrfs_ioctl_defrag_range_args *range;
2537         int ret;
2538
2539         ret = mnt_want_write_file(file);
2540         if (ret)
2541                 return ret;
2542
2543         if (btrfs_root_readonly(root)) {
2544                 ret = -EROFS;
2545                 goto out;
2546         }
2547
2548         switch (inode->i_mode & S_IFMT) {
2549         case S_IFDIR:
2550                 if (!capable(CAP_SYS_ADMIN)) {
2551                         ret = -EPERM;
2552                         goto out;
2553                 }
2554                 ret = btrfs_defrag_root(root);
2555                 if (ret)
2556                         goto out;
2557                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2558                 break;
2559         case S_IFREG:
2560                 if (!(file->f_mode & FMODE_WRITE)) {
2561                         ret = -EINVAL;
2562                         goto out;
2563                 }
2564
2565                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2566                 if (!range) {
2567                         ret = -ENOMEM;
2568                         goto out;
2569                 }
2570
2571                 if (argp) {
2572                         if (copy_from_user(range, argp,
2573                                            sizeof(*range))) {
2574                                 ret = -EFAULT;
2575                                 kfree(range);
2576                                 goto out;
2577                         }
2578                         /* compression requires us to start the IO */
2579                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2580                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2581                                 range->extent_thresh = (u32)-1;
2582                         }
2583                 } else {
2584                         /* the rest are all set to zero by kzalloc */
2585                         range->len = (u64)-1;
2586                 }
2587                 ret = btrfs_defrag_file(file_inode(file), file,
2588                                         range, 0, 0);
2589                 if (ret > 0)
2590                         ret = 0;
2591                 kfree(range);
2592                 break;
2593         default:
2594                 ret = -EINVAL;
2595         }
2596 out:
2597         mnt_drop_write_file(file);
2598         return ret;
2599 }
2600
2601 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2602 {
2603         struct btrfs_ioctl_vol_args *vol_args;
2604         int ret;
2605
2606         if (!capable(CAP_SYS_ADMIN))
2607                 return -EPERM;
2608
2609         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2610                         1)) {
2611                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2612         }
2613
2614         mutex_lock(&root->fs_info->volume_mutex);
2615         vol_args = memdup_user(arg, sizeof(*vol_args));
2616         if (IS_ERR(vol_args)) {
2617                 ret = PTR_ERR(vol_args);
2618                 goto out;
2619         }
2620
2621         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2622         ret = btrfs_init_new_device(root, vol_args->name);
2623
2624         if (!ret)
2625                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2626
2627         kfree(vol_args);
2628 out:
2629         mutex_unlock(&root->fs_info->volume_mutex);
2630         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2631         return ret;
2632 }
2633
2634 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2635 {
2636         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2637         struct btrfs_ioctl_vol_args *vol_args;
2638         int ret;
2639
2640         if (!capable(CAP_SYS_ADMIN))
2641                 return -EPERM;
2642
2643         ret = mnt_want_write_file(file);
2644         if (ret)
2645                 return ret;
2646
2647         vol_args = memdup_user(arg, sizeof(*vol_args));
2648         if (IS_ERR(vol_args)) {
2649                 ret = PTR_ERR(vol_args);
2650                 goto err_drop;
2651         }
2652
2653         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2654
2655         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2656                         1)) {
2657                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2658                 goto out;
2659         }
2660
2661         mutex_lock(&root->fs_info->volume_mutex);
2662         ret = btrfs_rm_device(root, vol_args->name);
2663         mutex_unlock(&root->fs_info->volume_mutex);
2664         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2665
2666         if (!ret)
2667                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2668
2669 out:
2670         kfree(vol_args);
2671 err_drop:
2672         mnt_drop_write_file(file);
2673         return ret;
2674 }
2675
2676 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2677 {
2678         struct btrfs_ioctl_fs_info_args *fi_args;
2679         struct btrfs_device *device;
2680         struct btrfs_device *next;
2681         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2682         int ret = 0;
2683
2684         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2685         if (!fi_args)
2686                 return -ENOMEM;
2687
2688         mutex_lock(&fs_devices->device_list_mutex);
2689         fi_args->num_devices = fs_devices->num_devices;
2690         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2691
2692         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2693                 if (device->devid > fi_args->max_id)
2694                         fi_args->max_id = device->devid;
2695         }
2696         mutex_unlock(&fs_devices->device_list_mutex);
2697
2698         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2699         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2700         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2701
2702         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2703                 ret = -EFAULT;
2704
2705         kfree(fi_args);
2706         return ret;
2707 }
2708
2709 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2710 {
2711         struct btrfs_ioctl_dev_info_args *di_args;
2712         struct btrfs_device *dev;
2713         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2714         int ret = 0;
2715         char *s_uuid = NULL;
2716
2717         di_args = memdup_user(arg, sizeof(*di_args));
2718         if (IS_ERR(di_args))
2719                 return PTR_ERR(di_args);
2720
2721         if (!btrfs_is_empty_uuid(di_args->uuid))
2722                 s_uuid = di_args->uuid;
2723
2724         mutex_lock(&fs_devices->device_list_mutex);
2725         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2726
2727         if (!dev) {
2728                 ret = -ENODEV;
2729                 goto out;
2730         }
2731
2732         di_args->devid = dev->devid;
2733         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2734         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2735         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2736         if (dev->name) {
2737                 struct rcu_string *name;
2738
2739                 rcu_read_lock();
2740                 name = rcu_dereference(dev->name);
2741                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2742                 rcu_read_unlock();
2743                 di_args->path[sizeof(di_args->path) - 1] = 0;
2744         } else {
2745                 di_args->path[0] = '\0';
2746         }
2747
2748 out:
2749         mutex_unlock(&fs_devices->device_list_mutex);
2750         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2751                 ret = -EFAULT;
2752
2753         kfree(di_args);
2754         return ret;
2755 }
2756
2757 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2758 {
2759         struct page *page;
2760         pgoff_t index;
2761         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2762
2763         index = off >> PAGE_CACHE_SHIFT;
2764
2765         page = grab_cache_page(inode->i_mapping, index);
2766         if (!page)
2767                 return NULL;
2768
2769         if (!PageUptodate(page)) {
2770                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2771                                                  0))
2772                         return NULL;
2773                 lock_page(page);
2774                 if (!PageUptodate(page)) {
2775                         unlock_page(page);
2776                         page_cache_release(page);
2777                         return NULL;
2778                 }
2779         }
2780         unlock_page(page);
2781
2782         return page;
2783 }
2784
2785 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2786 {
2787         /* do any pending delalloc/csum calc on src, one way or
2788            another, and lock file content */
2789         while (1) {
2790                 struct btrfs_ordered_extent *ordered;
2791                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2792                 ordered = btrfs_lookup_first_ordered_extent(inode,
2793                                                             off + len - 1);
2794                 if ((!ordered ||
2795                      ordered->file_offset + ordered->len <= off ||
2796                      ordered->file_offset >= off + len) &&
2797                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2798                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2799                         if (ordered)
2800                                 btrfs_put_ordered_extent(ordered);
2801                         break;
2802                 }
2803                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2804                 if (ordered)
2805                         btrfs_put_ordered_extent(ordered);
2806                 btrfs_wait_ordered_range(inode, off, len);
2807         }
2808 }
2809
2810 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2811                                 struct inode *inode2, u64 loff2, u64 len)
2812 {
2813         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2814         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2815
2816         mutex_unlock(&inode1->i_mutex);
2817         mutex_unlock(&inode2->i_mutex);
2818 }
2819
2820 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2821                               struct inode *inode2, u64 loff2, u64 len)
2822 {
2823         if (inode1 < inode2) {
2824                 swap(inode1, inode2);
2825                 swap(loff1, loff2);
2826         }
2827
2828         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2829         lock_extent_range(inode1, loff1, len);
2830         if (inode1 != inode2) {
2831                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2832                 lock_extent_range(inode2, loff2, len);
2833         }
2834 }
2835
2836 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2837                           u64 dst_loff, u64 len)
2838 {
2839         int ret = 0;
2840         struct page *src_page, *dst_page;
2841         unsigned int cmp_len = PAGE_CACHE_SIZE;
2842         void *addr, *dst_addr;
2843
2844         while (len) {
2845                 if (len < PAGE_CACHE_SIZE)
2846                         cmp_len = len;
2847
2848                 src_page = extent_same_get_page(src, loff);
2849                 if (!src_page)
2850                         return -EINVAL;
2851                 dst_page = extent_same_get_page(dst, dst_loff);
2852                 if (!dst_page) {
2853                         page_cache_release(src_page);
2854                         return -EINVAL;
2855                 }
2856                 addr = kmap_atomic(src_page);
2857                 dst_addr = kmap_atomic(dst_page);
2858
2859                 flush_dcache_page(src_page);
2860                 flush_dcache_page(dst_page);
2861
2862                 if (memcmp(addr, dst_addr, cmp_len))
2863                         ret = BTRFS_SAME_DATA_DIFFERS;
2864
2865                 kunmap_atomic(addr);
2866                 kunmap_atomic(dst_addr);
2867                 page_cache_release(src_page);
2868                 page_cache_release(dst_page);
2869
2870                 if (ret)
2871                         break;
2872
2873                 loff += cmp_len;
2874                 dst_loff += cmp_len;
2875                 len -= cmp_len;
2876         }
2877
2878         return ret;
2879 }
2880
2881 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2882 {
2883         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2884
2885         if (off + len > inode->i_size || off + len < off)
2886                 return -EINVAL;
2887         /* Check that we are block aligned - btrfs_clone() requires this */
2888         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2889                 return -EINVAL;
2890
2891         return 0;
2892 }
2893
2894 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2895                              struct inode *dst, u64 dst_loff)
2896 {
2897         int ret;
2898
2899         /*
2900          * btrfs_clone() can't handle extents in the same file
2901          * yet. Once that works, we can drop this check and replace it
2902          * with a check for the same inode, but overlapping extents.
2903          */
2904         if (src == dst)
2905                 return -EINVAL;
2906
2907         if (len == 0)
2908                 return 0;
2909
2910         btrfs_double_lock(src, loff, dst, dst_loff, len);
2911
2912         ret = extent_same_check_offsets(src, loff, len);
2913         if (ret)
2914                 goto out_unlock;
2915
2916         ret = extent_same_check_offsets(dst, dst_loff, len);
2917         if (ret)
2918                 goto out_unlock;
2919
2920         /* don't make the dst file partly checksummed */
2921         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2922             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2923                 ret = -EINVAL;
2924                 goto out_unlock;
2925         }
2926
2927         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2928         if (ret == 0)
2929                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2930
2931 out_unlock:
2932         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2933
2934         return ret;
2935 }
2936
2937 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2938
2939 static long btrfs_ioctl_file_extent_same(struct file *file,
2940                         struct btrfs_ioctl_same_args __user *argp)
2941 {
2942         struct btrfs_ioctl_same_args *same;
2943         struct btrfs_ioctl_same_extent_info *info;
2944         struct inode *src = file_inode(file);
2945         u64 off;
2946         u64 len;
2947         int i;
2948         int ret;
2949         unsigned long size;
2950         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2951         bool is_admin = capable(CAP_SYS_ADMIN);
2952         u16 count;
2953
2954         if (!(file->f_mode & FMODE_READ))
2955                 return -EINVAL;
2956
2957         ret = mnt_want_write_file(file);
2958         if (ret)
2959                 return ret;
2960
2961         if (get_user(count, &argp->dest_count)) {
2962                 ret = -EFAULT;
2963                 goto out;
2964         }
2965
2966         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2967
2968         same = memdup_user(argp, size);
2969
2970         if (IS_ERR(same)) {
2971                 ret = PTR_ERR(same);
2972                 goto out;
2973         }
2974
2975         off = same->logical_offset;
2976         len = same->length;
2977
2978         /*
2979          * Limit the total length we will dedupe for each operation.
2980          * This is intended to bound the total time spent in this
2981          * ioctl to something sane.
2982          */
2983         if (len > BTRFS_MAX_DEDUPE_LEN)
2984                 len = BTRFS_MAX_DEDUPE_LEN;
2985
2986         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2987                 /*
2988                  * Btrfs does not support blocksize < page_size. As a
2989                  * result, btrfs_cmp_data() won't correctly handle
2990                  * this situation without an update.
2991                  */
2992                 ret = -EINVAL;
2993                 goto out;
2994         }
2995
2996         ret = -EISDIR;
2997         if (S_ISDIR(src->i_mode))
2998                 goto out;
2999
3000         ret = -EACCES;
3001         if (!S_ISREG(src->i_mode))
3002                 goto out;
3003
3004         /* pre-format output fields to sane values */
3005         for (i = 0; i < count; i++) {
3006                 same->info[i].bytes_deduped = 0ULL;
3007                 same->info[i].status = 0;
3008         }
3009
3010         for (i = 0, info = same->info; i < count; i++, info++) {
3011                 struct inode *dst;
3012                 struct fd dst_file = fdget(info->fd);
3013                 if (!dst_file.file) {
3014                         info->status = -EBADF;
3015                         continue;
3016                 }
3017                 dst = file_inode(dst_file.file);
3018
3019                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3020                         info->status = -EINVAL;
3021                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3022                         info->status = -EXDEV;
3023                 } else if (S_ISDIR(dst->i_mode)) {
3024                         info->status = -EISDIR;
3025                 } else if (!S_ISREG(dst->i_mode)) {
3026                         info->status = -EACCES;
3027                 } else {
3028                         info->status = btrfs_extent_same(src, off, len, dst,
3029                                                         info->logical_offset);
3030                         if (info->status == 0)
3031                                 info->bytes_deduped += len;
3032                 }
3033                 fdput(dst_file);
3034         }
3035
3036         ret = copy_to_user(argp, same, size);
3037         if (ret)
3038                 ret = -EFAULT;
3039
3040 out:
3041         mnt_drop_write_file(file);
3042         return ret;
3043 }
3044
3045 /* Helper to check and see if this root currently has a ref on the given disk
3046  * bytenr.  If it does then we need to update the quota for this root.  This
3047  * doesn't do anything if quotas aren't enabled.
3048  */
3049 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3050                      u64 disko)
3051 {
3052         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
3053         struct ulist *roots;
3054         struct ulist_iterator uiter;
3055         struct ulist_node *root_node = NULL;
3056         int ret;
3057
3058         if (!root->fs_info->quota_enabled)
3059                 return 1;
3060
3061         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3062         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3063                                    tree_mod_seq_elem.seq, &roots);
3064         if (ret < 0)
3065                 goto out;
3066         ret = 0;
3067         ULIST_ITER_INIT(&uiter);
3068         while ((root_node = ulist_next(roots, &uiter))) {
3069                 if (root_node->val == root->objectid) {
3070                         ret = 1;
3071                         break;
3072                 }
3073         }
3074         ulist_free(roots);
3075 out:
3076         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3077         return ret;
3078 }
3079
3080 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3081                                      struct inode *inode,
3082                                      u64 endoff,
3083                                      const u64 destoff,
3084                                      const u64 olen)
3085 {
3086         struct btrfs_root *root = BTRFS_I(inode)->root;
3087         int ret;
3088
3089         inode_inc_iversion(inode);
3090         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3091         /*
3092          * We round up to the block size at eof when determining which
3093          * extents to clone above, but shouldn't round up the file size.
3094          */
3095         if (endoff > destoff + olen)
3096                 endoff = destoff + olen;
3097         if (endoff > inode->i_size)
3098                 btrfs_i_size_write(inode, endoff);
3099
3100         ret = btrfs_update_inode(trans, root, inode);
3101         if (ret) {
3102                 btrfs_abort_transaction(trans, root, ret);
3103                 btrfs_end_transaction(trans, root);
3104                 goto out;
3105         }
3106         ret = btrfs_end_transaction(trans, root);
3107 out:
3108         return ret;
3109 }
3110
3111 static void clone_update_extent_map(struct inode *inode,
3112                                     const struct btrfs_trans_handle *trans,
3113                                     const struct btrfs_path *path,
3114                                     const u64 hole_offset,
3115                                     const u64 hole_len)
3116 {
3117         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3118         struct extent_map *em;
3119         int ret;
3120
3121         em = alloc_extent_map();
3122         if (!em) {
3123                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3124                         &BTRFS_I(inode)->runtime_flags);
3125                 return;
3126         }
3127
3128         if (path) {
3129                 struct btrfs_file_extent_item *fi;
3130
3131                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3132                                     struct btrfs_file_extent_item);
3133                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3134                 em->generation = -1;
3135                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3136                     BTRFS_FILE_EXTENT_INLINE)
3137                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3138                                 &BTRFS_I(inode)->runtime_flags);
3139         } else {
3140                 em->start = hole_offset;
3141                 em->len = hole_len;
3142                 em->ram_bytes = em->len;
3143                 em->orig_start = hole_offset;
3144                 em->block_start = EXTENT_MAP_HOLE;
3145                 em->block_len = 0;
3146                 em->orig_block_len = 0;
3147                 em->compress_type = BTRFS_COMPRESS_NONE;
3148                 em->generation = trans->transid;
3149         }
3150
3151         while (1) {
3152                 write_lock(&em_tree->lock);
3153                 ret = add_extent_mapping(em_tree, em, 1);
3154                 write_unlock(&em_tree->lock);
3155                 if (ret != -EEXIST) {
3156                         free_extent_map(em);
3157                         break;
3158                 }
3159                 btrfs_drop_extent_cache(inode, em->start,
3160                                         em->start + em->len - 1, 0);
3161         }
3162
3163         if (ret)
3164                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3165                         &BTRFS_I(inode)->runtime_flags);
3166 }
3167
3168 /**
3169  * btrfs_clone() - clone a range from inode file to another
3170  *
3171  * @src: Inode to clone from
3172  * @inode: Inode to clone to
3173  * @off: Offset within source to start clone from
3174  * @olen: Original length, passed by user, of range to clone
3175  * @olen_aligned: Block-aligned value of olen, extent_same uses
3176  *               identical values here
3177  * @destoff: Offset within @inode to start clone
3178  */
3179 static int btrfs_clone(struct inode *src, struct inode *inode,
3180                        const u64 off, const u64 olen, const u64 olen_aligned,
3181                        const u64 destoff)
3182 {
3183         struct btrfs_root *root = BTRFS_I(inode)->root;
3184         struct btrfs_path *path = NULL;
3185         struct extent_buffer *leaf;
3186         struct btrfs_trans_handle *trans;
3187         char *buf = NULL;
3188         struct btrfs_key key;
3189         u32 nritems;
3190         int slot;
3191         int ret;
3192         int no_quota;
3193         const u64 len = olen_aligned;
3194         u64 last_disko = 0;
3195         u64 last_dest_end = destoff;
3196
3197         ret = -ENOMEM;
3198         buf = vmalloc(root->nodesize);
3199         if (!buf)
3200                 return ret;
3201
3202         path = btrfs_alloc_path();
3203         if (!path) {
3204                 vfree(buf);
3205                 return ret;
3206         }
3207
3208         path->reada = 2;
3209         /* clone data */
3210         key.objectid = btrfs_ino(src);
3211         key.type = BTRFS_EXTENT_DATA_KEY;
3212         key.offset = off;
3213
3214         while (1) {
3215                 u64 next_key_min_offset = key.offset + 1;
3216
3217                 /*
3218                  * note the key will change type as we walk through the
3219                  * tree.
3220                  */
3221                 path->leave_spinning = 1;
3222                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3223                                 0, 0);
3224                 if (ret < 0)
3225                         goto out;
3226                 /*
3227                  * First search, if no extent item that starts at offset off was
3228                  * found but the previous item is an extent item, it's possible
3229                  * it might overlap our target range, therefore process it.
3230                  */
3231                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3232                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3233                                               path->slots[0] - 1);
3234                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3235                                 path->slots[0]--;
3236                 }
3237
3238                 nritems = btrfs_header_nritems(path->nodes[0]);
3239 process_slot:
3240                 no_quota = 1;
3241                 if (path->slots[0] >= nritems) {
3242                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3243                         if (ret < 0)
3244                                 goto out;
3245                         if (ret > 0)
3246                                 break;
3247                         nritems = btrfs_header_nritems(path->nodes[0]);
3248                 }
3249                 leaf = path->nodes[0];
3250                 slot = path->slots[0];
3251
3252                 btrfs_item_key_to_cpu(leaf, &key, slot);
3253                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3254                     key.objectid != btrfs_ino(src))
3255                         break;
3256
3257                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3258                         struct btrfs_file_extent_item *extent;
3259                         int type;
3260                         u32 size;
3261                         struct btrfs_key new_key;
3262                         u64 disko = 0, diskl = 0;
3263                         u64 datao = 0, datal = 0;
3264                         u8 comp;
3265                         u64 drop_start;
3266
3267                         extent = btrfs_item_ptr(leaf, slot,
3268                                                 struct btrfs_file_extent_item);
3269                         comp = btrfs_file_extent_compression(leaf, extent);
3270                         type = btrfs_file_extent_type(leaf, extent);
3271                         if (type == BTRFS_FILE_EXTENT_REG ||
3272                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3273                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3274                                                                       extent);
3275                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3276                                                                  extent);
3277                                 datao = btrfs_file_extent_offset(leaf, extent);
3278                                 datal = btrfs_file_extent_num_bytes(leaf,
3279                                                                     extent);
3280                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3281                                 /* take upper bound, may be compressed */
3282                                 datal = btrfs_file_extent_ram_bytes(leaf,
3283                                                                     extent);
3284                         }
3285
3286                         /*
3287                          * The first search might have left us at an extent
3288                          * item that ends before our target range's start, can
3289                          * happen if we have holes and NO_HOLES feature enabled.
3290                          */
3291                         if (key.offset + datal <= off) {
3292                                 path->slots[0]++;
3293                                 goto process_slot;
3294                         } else if (key.offset >= off + len) {
3295                                 break;
3296                         }
3297                         next_key_min_offset = key.offset + datal;
3298                         size = btrfs_item_size_nr(leaf, slot);
3299                         read_extent_buffer(leaf, buf,
3300                                            btrfs_item_ptr_offset(leaf, slot),
3301                                            size);
3302
3303                         btrfs_release_path(path);
3304                         path->leave_spinning = 0;
3305
3306                         memcpy(&new_key, &key, sizeof(new_key));
3307                         new_key.objectid = btrfs_ino(inode);
3308                         if (off <= key.offset)
3309                                 new_key.offset = key.offset + destoff - off;
3310                         else
3311                                 new_key.offset = destoff;
3312
3313                         /*
3314                          * Deal with a hole that doesn't have an extent item
3315                          * that represents it (NO_HOLES feature enabled).
3316                          * This hole is either in the middle of the cloning
3317                          * range or at the beginning (fully overlaps it or
3318                          * partially overlaps it).
3319                          */
3320                         if (new_key.offset != last_dest_end)
3321                                 drop_start = last_dest_end;
3322                         else
3323                                 drop_start = new_key.offset;
3324
3325                         /*
3326                          * 1 - adjusting old extent (we may have to split it)
3327                          * 1 - add new extent
3328                          * 1 - inode update
3329                          */
3330                         trans = btrfs_start_transaction(root, 3);
3331                         if (IS_ERR(trans)) {
3332                                 ret = PTR_ERR(trans);
3333                                 goto out;
3334                         }
3335
3336                         if (type == BTRFS_FILE_EXTENT_REG ||
3337                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3338                                 /*
3339                                  *    a  | --- range to clone ---|  b
3340                                  * | ------------- extent ------------- |
3341                                  */
3342
3343                                 /* subtract range b */
3344                                 if (key.offset + datal > off + len)
3345                                         datal = off + len - key.offset;
3346
3347                                 /* subtract range a */
3348                                 if (off > key.offset) {
3349                                         datao += off - key.offset;
3350                                         datal -= off - key.offset;
3351                                 }
3352
3353                                 ret = btrfs_drop_extents(trans, root, inode,
3354                                                          drop_start,
3355                                                          new_key.offset + datal,
3356                                                          1);
3357                                 if (ret) {
3358                                         if (ret != -EOPNOTSUPP)
3359                                                 btrfs_abort_transaction(trans,
3360                                                                 root, ret);
3361                                         btrfs_end_transaction(trans, root);
3362                                         goto out;
3363                                 }
3364
3365                                 ret = btrfs_insert_empty_item(trans, root, path,
3366                                                               &new_key, size);
3367                                 if (ret) {
3368                                         btrfs_abort_transaction(trans, root,
3369                                                                 ret);
3370                                         btrfs_end_transaction(trans, root);
3371                                         goto out;
3372                                 }
3373
3374                                 leaf = path->nodes[0];
3375                                 slot = path->slots[0];
3376                                 write_extent_buffer(leaf, buf,
3377                                             btrfs_item_ptr_offset(leaf, slot),
3378                                             size);
3379
3380                                 extent = btrfs_item_ptr(leaf, slot,
3381                                                 struct btrfs_file_extent_item);
3382
3383                                 /* disko == 0 means it's a hole */
3384                                 if (!disko)
3385                                         datao = 0;
3386
3387                                 btrfs_set_file_extent_offset(leaf, extent,
3388                                                              datao);
3389                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3390                                                                 datal);
3391
3392                                 /*
3393                                  * We need to look up the roots that point at
3394                                  * this bytenr and see if the new root does.  If
3395                                  * it does not we need to make sure we update
3396                                  * quotas appropriately.
3397                                  */
3398                                 if (disko && root != BTRFS_I(src)->root &&
3399                                     disko != last_disko) {
3400                                         no_quota = check_ref(trans, root,
3401                                                              disko);
3402                                         if (no_quota < 0) {
3403                                                 btrfs_abort_transaction(trans,
3404                                                                         root,
3405                                                                         ret);
3406                                                 btrfs_end_transaction(trans,
3407                                                                       root);
3408                                                 ret = no_quota;
3409                                                 goto out;
3410                                         }
3411                                 }
3412
3413                                 if (disko) {
3414                                         inode_add_bytes(inode, datal);
3415                                         ret = btrfs_inc_extent_ref(trans, root,
3416                                                         disko, diskl, 0,
3417                                                         root->root_key.objectid,
3418                                                         btrfs_ino(inode),
3419                                                         new_key.offset - datao,
3420                                                         no_quota);
3421                                         if (ret) {
3422                                                 btrfs_abort_transaction(trans,
3423                                                                         root,
3424                                                                         ret);
3425                                                 btrfs_end_transaction(trans,
3426                                                                       root);
3427                                                 goto out;
3428
3429                                         }
3430                                 }
3431                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3432                                 u64 skip = 0;
3433                                 u64 trim = 0;
3434                                 u64 aligned_end = 0;
3435
3436                                 if (off > key.offset) {
3437                                         skip = off - key.offset;
3438                                         new_key.offset += skip;
3439                                 }
3440
3441                                 if (key.offset + datal > off + len)
3442                                         trim = key.offset + datal - (off + len);
3443
3444                                 if (comp && (skip || trim)) {
3445                                         ret = -EINVAL;
3446                                         btrfs_end_transaction(trans, root);
3447                                         goto out;
3448                                 }
3449                                 size -= skip + trim;
3450                                 datal -= skip + trim;
3451
3452                                 aligned_end = ALIGN(new_key.offset + datal,
3453                                                     root->sectorsize);
3454                                 ret = btrfs_drop_extents(trans, root, inode,
3455                                                          drop_start,
3456                                                          aligned_end,
3457                                                          1);
3458                                 if (ret) {
3459                                         if (ret != -EOPNOTSUPP)
3460                                                 btrfs_abort_transaction(trans,
3461                                                         root, ret);
3462                                         btrfs_end_transaction(trans, root);
3463                                         goto out;
3464                                 }
3465
3466                                 ret = btrfs_insert_empty_item(trans, root, path,
3467                                                               &new_key, size);
3468                                 if (ret) {
3469                                         btrfs_abort_transaction(trans, root,
3470                                                                 ret);
3471                                         btrfs_end_transaction(trans, root);
3472                                         goto out;
3473                                 }
3474
3475                                 if (skip) {
3476                                         u32 start =
3477                                           btrfs_file_extent_calc_inline_size(0);
3478                                         memmove(buf+start, buf+start+skip,
3479                                                 datal);
3480                                 }
3481
3482                                 leaf = path->nodes[0];
3483                                 slot = path->slots[0];
3484                                 write_extent_buffer(leaf, buf,
3485                                             btrfs_item_ptr_offset(leaf, slot),
3486                                             size);
3487                                 inode_add_bytes(inode, datal);
3488                         }
3489
3490                         /* If we have an implicit hole (NO_HOLES feature). */
3491                         if (drop_start < new_key.offset)
3492                                 clone_update_extent_map(inode, trans,
3493                                                 NULL, drop_start,
3494                                                 new_key.offset - drop_start);
3495
3496                         clone_update_extent_map(inode, trans, path, 0, 0);
3497
3498                         btrfs_mark_buffer_dirty(leaf);
3499                         btrfs_release_path(path);
3500
3501                         last_dest_end = ALIGN(new_key.offset + datal,
3502                                               root->sectorsize);
3503                         ret = clone_finish_inode_update(trans, inode,
3504                                                         last_dest_end,
3505                                                         destoff, olen);
3506                         if (ret)
3507                                 goto out;
3508                         if (new_key.offset + datal >= destoff + len)
3509                                 break;
3510                 }
3511                 btrfs_release_path(path);
3512                 key.offset = next_key_min_offset;
3513         }
3514         ret = 0;
3515
3516         if (last_dest_end < destoff + len) {
3517                 /*
3518                  * We have an implicit hole (NO_HOLES feature is enabled) that
3519                  * fully or partially overlaps our cloning range at its end.
3520                  */
3521                 btrfs_release_path(path);
3522
3523                 /*
3524                  * 1 - remove extent(s)
3525                  * 1 - inode update
3526                  */
3527                 trans = btrfs_start_transaction(root, 2);
3528                 if (IS_ERR(trans)) {
3529                         ret = PTR_ERR(trans);
3530                         goto out;
3531                 }
3532                 ret = btrfs_drop_extents(trans, root, inode,
3533                                          last_dest_end, destoff + len, 1);
3534                 if (ret) {
3535                         if (ret != -EOPNOTSUPP)
3536                                 btrfs_abort_transaction(trans, root, ret);
3537                         btrfs_end_transaction(trans, root);
3538                         goto out;
3539                 }
3540                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3541                                         destoff + len - last_dest_end);
3542                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3543                                                 destoff, olen);
3544         }
3545
3546 out:
3547         btrfs_free_path(path);
3548         vfree(buf);
3549         return ret;
3550 }
3551
3552 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3553                                        u64 off, u64 olen, u64 destoff)
3554 {
3555         struct inode *inode = file_inode(file);
3556         struct btrfs_root *root = BTRFS_I(inode)->root;
3557         struct fd src_file;
3558         struct inode *src;
3559         int ret;
3560         u64 len = olen;
3561         u64 bs = root->fs_info->sb->s_blocksize;
3562         int same_inode = 0;
3563
3564         /*
3565          * TODO:
3566          * - split compressed inline extents.  annoying: we need to
3567          *   decompress into destination's address_space (the file offset
3568          *   may change, so source mapping won't do), then recompress (or
3569          *   otherwise reinsert) a subrange.
3570          *
3571          * - split destination inode's inline extents.  The inline extents can
3572          *   be either compressed or non-compressed.
3573          */
3574
3575         /* the destination must be opened for writing */
3576         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3577                 return -EINVAL;
3578
3579         if (btrfs_root_readonly(root))
3580                 return -EROFS;
3581
3582         ret = mnt_want_write_file(file);
3583         if (ret)
3584                 return ret;
3585
3586         src_file = fdget(srcfd);
3587         if (!src_file.file) {
3588                 ret = -EBADF;
3589                 goto out_drop_write;
3590         }
3591
3592         ret = -EXDEV;
3593         if (src_file.file->f_path.mnt != file->f_path.mnt)
3594                 goto out_fput;
3595
3596         src = file_inode(src_file.file);
3597
3598         ret = -EINVAL;
3599         if (src == inode)
3600                 same_inode = 1;
3601
3602         /* the src must be open for reading */
3603         if (!(src_file.file->f_mode & FMODE_READ))
3604                 goto out_fput;
3605
3606         /* don't make the dst file partly checksummed */
3607         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3608             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3609                 goto out_fput;
3610
3611         ret = -EISDIR;
3612         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3613                 goto out_fput;
3614
3615         ret = -EXDEV;
3616         if (src->i_sb != inode->i_sb)
3617                 goto out_fput;
3618
3619         if (!same_inode) {
3620                 if (inode < src) {
3621                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3622                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3623                 } else {
3624                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3625                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3626                 }
3627         } else {
3628                 mutex_lock(&src->i_mutex);
3629         }
3630
3631         /* determine range to clone */
3632         ret = -EINVAL;
3633         if (off + len > src->i_size || off + len < off)
3634                 goto out_unlock;
3635         if (len == 0)
3636                 olen = len = src->i_size - off;
3637         /* if we extend to eof, continue to block boundary */
3638         if (off + len == src->i_size)
3639                 len = ALIGN(src->i_size, bs) - off;
3640
3641         if (len == 0) {
3642                 ret = 0;
3643                 goto out_unlock;
3644         }
3645
3646         /* verify the end result is block aligned */
3647         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3648             !IS_ALIGNED(destoff, bs))
3649                 goto out_unlock;
3650
3651         /* verify if ranges are overlapped within the same file */
3652         if (same_inode) {
3653                 if (destoff + len > off && destoff < off + len)
3654                         goto out_unlock;
3655         }
3656
3657         if (destoff > inode->i_size) {
3658                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3659                 if (ret)
3660                         goto out_unlock;
3661         }
3662
3663         /*
3664          * Lock the target range too. Right after we replace the file extent
3665          * items in the fs tree (which now point to the cloned data), we might
3666          * have a worker replace them with extent items relative to a write
3667          * operation that was issued before this clone operation (i.e. confront
3668          * with inode.c:btrfs_finish_ordered_io).
3669          */
3670         if (same_inode) {
3671                 u64 lock_start = min_t(u64, off, destoff);
3672                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3673
3674                 lock_extent_range(src, lock_start, lock_len);
3675         } else {
3676                 lock_extent_range(src, off, len);
3677                 lock_extent_range(inode, destoff, len);
3678         }
3679
3680         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3681
3682         if (same_inode) {
3683                 u64 lock_start = min_t(u64, off, destoff);
3684                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3685
3686                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3687         } else {
3688                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3689                 unlock_extent(&BTRFS_I(inode)->io_tree, destoff,
3690                               destoff + len - 1);
3691         }
3692         /*
3693          * Truncate page cache pages so that future reads will see the cloned
3694          * data immediately and not the previous data.
3695          */
3696         truncate_inode_pages_range(&inode->i_data, destoff,
3697                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3698 out_unlock:
3699         if (!same_inode) {
3700                 if (inode < src) {
3701                         mutex_unlock(&src->i_mutex);
3702                         mutex_unlock(&inode->i_mutex);
3703                 } else {
3704                         mutex_unlock(&inode->i_mutex);
3705                         mutex_unlock(&src->i_mutex);
3706                 }
3707         } else {
3708                 mutex_unlock(&src->i_mutex);
3709         }
3710 out_fput:
3711         fdput(src_file);
3712 out_drop_write:
3713         mnt_drop_write_file(file);
3714         return ret;
3715 }
3716
3717 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3718 {
3719         struct btrfs_ioctl_clone_range_args args;
3720
3721         if (copy_from_user(&args, argp, sizeof(args)))
3722                 return -EFAULT;
3723         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3724                                  args.src_length, args.dest_offset);
3725 }
3726
3727 /*
3728  * there are many ways the trans_start and trans_end ioctls can lead
3729  * to deadlocks.  They should only be used by applications that
3730  * basically own the machine, and have a very in depth understanding
3731  * of all the possible deadlocks and enospc problems.
3732  */
3733 static long btrfs_ioctl_trans_start(struct file *file)
3734 {
3735         struct inode *inode = file_inode(file);
3736         struct btrfs_root *root = BTRFS_I(inode)->root;
3737         struct btrfs_trans_handle *trans;
3738         int ret;
3739
3740         ret = -EPERM;
3741         if (!capable(CAP_SYS_ADMIN))
3742                 goto out;
3743
3744         ret = -EINPROGRESS;
3745         if (file->private_data)
3746                 goto out;
3747
3748         ret = -EROFS;
3749         if (btrfs_root_readonly(root))
3750                 goto out;
3751
3752         ret = mnt_want_write_file(file);
3753         if (ret)
3754                 goto out;
3755
3756         atomic_inc(&root->fs_info->open_ioctl_trans);
3757
3758         ret = -ENOMEM;
3759         trans = btrfs_start_ioctl_transaction(root);
3760         if (IS_ERR(trans))
3761                 goto out_drop;
3762
3763         file->private_data = trans;
3764         return 0;
3765
3766 out_drop:
3767         atomic_dec(&root->fs_info->open_ioctl_trans);
3768         mnt_drop_write_file(file);
3769 out:
3770         return ret;
3771 }
3772
3773 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3774 {
3775         struct inode *inode = file_inode(file);
3776         struct btrfs_root *root = BTRFS_I(inode)->root;
3777         struct btrfs_root *new_root;
3778         struct btrfs_dir_item *di;
3779         struct btrfs_trans_handle *trans;
3780         struct btrfs_path *path;
3781         struct btrfs_key location;
3782         struct btrfs_disk_key disk_key;
3783         u64 objectid = 0;
3784         u64 dir_id;
3785         int ret;
3786
3787         if (!capable(CAP_SYS_ADMIN))
3788                 return -EPERM;
3789
3790         ret = mnt_want_write_file(file);
3791         if (ret)
3792                 return ret;
3793
3794         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3795                 ret = -EFAULT;
3796                 goto out;
3797         }
3798
3799         if (!objectid)
3800                 objectid = BTRFS_FS_TREE_OBJECTID;
3801
3802         location.objectid = objectid;
3803         location.type = BTRFS_ROOT_ITEM_KEY;
3804         location.offset = (u64)-1;
3805
3806         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3807         if (IS_ERR(new_root)) {
3808                 ret = PTR_ERR(new_root);
3809                 goto out;
3810         }
3811
3812         path = btrfs_alloc_path();
3813         if (!path) {
3814                 ret = -ENOMEM;
3815                 goto out;
3816         }
3817         path->leave_spinning = 1;
3818
3819         trans = btrfs_start_transaction(root, 1);
3820         if (IS_ERR(trans)) {
3821                 btrfs_free_path(path);
3822                 ret = PTR_ERR(trans);
3823                 goto out;
3824         }
3825
3826         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3827         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3828                                    dir_id, "default", 7, 1);
3829         if (IS_ERR_OR_NULL(di)) {
3830                 btrfs_free_path(path);
3831                 btrfs_end_transaction(trans, root);
3832                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3833                            "item, this isn't going to work");
3834                 ret = -ENOENT;
3835                 goto out;
3836         }
3837
3838         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3839         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3840         btrfs_mark_buffer_dirty(path->nodes[0]);
3841         btrfs_free_path(path);
3842
3843         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3844         btrfs_end_transaction(trans, root);
3845 out:
3846         mnt_drop_write_file(file);
3847         return ret;
3848 }
3849
3850 void btrfs_get_block_group_info(struct list_head *groups_list,
3851                                 struct btrfs_ioctl_space_info *space)
3852 {
3853         struct btrfs_block_group_cache *block_group;
3854
3855         space->total_bytes = 0;
3856         space->used_bytes = 0;
3857         space->flags = 0;
3858         list_for_each_entry(block_group, groups_list, list) {
3859                 space->flags = block_group->flags;
3860                 space->total_bytes += block_group->key.offset;
3861                 space->used_bytes +=
3862                         btrfs_block_group_used(&block_group->item);
3863         }
3864 }
3865
3866 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3867 {
3868         struct btrfs_ioctl_space_args space_args;
3869         struct btrfs_ioctl_space_info space;
3870         struct btrfs_ioctl_space_info *dest;
3871         struct btrfs_ioctl_space_info *dest_orig;
3872         struct btrfs_ioctl_space_info __user *user_dest;
3873         struct btrfs_space_info *info;
3874         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3875                        BTRFS_BLOCK_GROUP_SYSTEM,
3876                        BTRFS_BLOCK_GROUP_METADATA,
3877                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3878         int num_types = 4;
3879         int alloc_size;
3880         int ret = 0;
3881         u64 slot_count = 0;
3882         int i, c;
3883
3884         if (copy_from_user(&space_args,
3885                            (struct btrfs_ioctl_space_args __user *)arg,
3886                            sizeof(space_args)))
3887                 return -EFAULT;
3888
3889         for (i = 0; i < num_types; i++) {
3890                 struct btrfs_space_info *tmp;
3891
3892                 info = NULL;
3893                 rcu_read_lock();
3894                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3895                                         list) {
3896                         if (tmp->flags == types[i]) {
3897                                 info = tmp;
3898                                 break;
3899                         }
3900                 }
3901                 rcu_read_unlock();
3902
3903                 if (!info)
3904                         continue;
3905
3906                 down_read(&info->groups_sem);
3907                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3908                         if (!list_empty(&info->block_groups[c]))
3909                                 slot_count++;
3910                 }
3911                 up_read(&info->groups_sem);
3912         }
3913
3914         /*
3915          * Global block reserve, exported as a space_info
3916          */
3917         slot_count++;
3918
3919         /* space_slots == 0 means they are asking for a count */
3920         if (space_args.space_slots == 0) {
3921                 space_args.total_spaces = slot_count;
3922                 goto out;
3923         }
3924
3925         slot_count = min_t(u64, space_args.space_slots, slot_count);
3926
3927         alloc_size = sizeof(*dest) * slot_count;
3928
3929         /* we generally have at most 6 or so space infos, one for each raid
3930          * level.  So, a whole page should be more than enough for everyone
3931          */
3932         if (alloc_size > PAGE_CACHE_SIZE)
3933                 return -ENOMEM;
3934
3935         space_args.total_spaces = 0;
3936         dest = kmalloc(alloc_size, GFP_NOFS);
3937         if (!dest)
3938                 return -ENOMEM;
3939         dest_orig = dest;
3940
3941         /* now we have a buffer to copy into */
3942         for (i = 0; i < num_types; i++) {
3943                 struct btrfs_space_info *tmp;
3944
3945                 if (!slot_count)
3946                         break;
3947
3948                 info = NULL;
3949                 rcu_read_lock();
3950                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3951                                         list) {
3952                         if (tmp->flags == types[i]) {
3953                                 info = tmp;
3954                                 break;
3955                         }
3956                 }
3957                 rcu_read_unlock();
3958
3959                 if (!info)
3960                         continue;
3961                 down_read(&info->groups_sem);
3962                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3963                         if (!list_empty(&info->block_groups[c])) {
3964                                 btrfs_get_block_group_info(
3965                                         &info->block_groups[c], &space);
3966                                 memcpy(dest, &space, sizeof(space));
3967                                 dest++;
3968                                 space_args.total_spaces++;
3969                                 slot_count--;
3970                         }
3971                         if (!slot_count)
3972                                 break;
3973                 }
3974                 up_read(&info->groups_sem);
3975         }
3976
3977         /*
3978          * Add global block reserve
3979          */
3980         if (slot_count) {
3981                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3982
3983                 spin_lock(&block_rsv->lock);
3984                 space.total_bytes = block_rsv->size;
3985                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3986                 spin_unlock(&block_rsv->lock);
3987                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3988                 memcpy(dest, &space, sizeof(space));
3989                 space_args.total_spaces++;
3990         }
3991
3992         user_dest = (struct btrfs_ioctl_space_info __user *)
3993                 (arg + sizeof(struct btrfs_ioctl_space_args));
3994
3995         if (copy_to_user(user_dest, dest_orig, alloc_size))
3996                 ret = -EFAULT;
3997
3998         kfree(dest_orig);
3999 out:
4000         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4001                 ret = -EFAULT;
4002
4003         return ret;
4004 }
4005
4006 /*
4007  * there are many ways the trans_start and trans_end ioctls can lead
4008  * to deadlocks.  They should only be used by applications that
4009  * basically own the machine, and have a very in depth understanding
4010  * of all the possible deadlocks and enospc problems.
4011  */
4012 long btrfs_ioctl_trans_end(struct file *file)
4013 {
4014         struct inode *inode = file_inode(file);
4015         struct btrfs_root *root = BTRFS_I(inode)->root;
4016         struct btrfs_trans_handle *trans;
4017
4018         trans = file->private_data;
4019         if (!trans)
4020                 return -EINVAL;
4021         file->private_data = NULL;
4022
4023         btrfs_end_transaction(trans, root);
4024
4025         atomic_dec(&root->fs_info->open_ioctl_trans);
4026
4027         mnt_drop_write_file(file);
4028         return 0;
4029 }
4030
4031 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4032                                             void __user *argp)
4033 {
4034         struct btrfs_trans_handle *trans;
4035         u64 transid;
4036         int ret;
4037
4038         trans = btrfs_attach_transaction_barrier(root);
4039         if (IS_ERR(trans)) {
4040                 if (PTR_ERR(trans) != -ENOENT)
4041                         return PTR_ERR(trans);
4042
4043                 /* No running transaction, don't bother */
4044                 transid = root->fs_info->last_trans_committed;
4045                 goto out;
4046         }
4047         transid = trans->transid;
4048         ret = btrfs_commit_transaction_async(trans, root, 0);
4049         if (ret) {
4050                 btrfs_end_transaction(trans, root);
4051                 return ret;
4052         }
4053 out:
4054         if (argp)
4055                 if (copy_to_user(argp, &transid, sizeof(transid)))
4056                         return -EFAULT;
4057         return 0;
4058 }
4059
4060 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4061                                            void __user *argp)
4062 {
4063         u64 transid;
4064
4065         if (argp) {
4066                 if (copy_from_user(&transid, argp, sizeof(transid)))
4067                         return -EFAULT;
4068         } else {
4069                 transid = 0;  /* current trans */
4070         }
4071         return btrfs_wait_for_commit(root, transid);
4072 }
4073
4074 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4075 {
4076         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4077         struct btrfs_ioctl_scrub_args *sa;
4078         int ret;
4079
4080         if (!capable(CAP_SYS_ADMIN))
4081                 return -EPERM;
4082
4083         sa = memdup_user(arg, sizeof(*sa));
4084         if (IS_ERR(sa))
4085                 return PTR_ERR(sa);
4086
4087         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4088                 ret = mnt_want_write_file(file);
4089                 if (ret)
4090                         goto out;
4091         }
4092
4093         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4094                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4095                               0);
4096
4097         if (copy_to_user(arg, sa, sizeof(*sa)))
4098                 ret = -EFAULT;
4099
4100         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4101                 mnt_drop_write_file(file);
4102 out:
4103         kfree(sa);
4104         return ret;
4105 }
4106
4107 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4108 {
4109         if (!capable(CAP_SYS_ADMIN))
4110                 return -EPERM;
4111
4112         return btrfs_scrub_cancel(root->fs_info);
4113 }
4114
4115 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4116                                        void __user *arg)
4117 {
4118         struct btrfs_ioctl_scrub_args *sa;
4119         int ret;
4120
4121         if (!capable(CAP_SYS_ADMIN))
4122                 return -EPERM;
4123
4124         sa = memdup_user(arg, sizeof(*sa));
4125         if (IS_ERR(sa))
4126                 return PTR_ERR(sa);
4127
4128         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4129
4130         if (copy_to_user(arg, sa, sizeof(*sa)))
4131                 ret = -EFAULT;
4132
4133         kfree(sa);
4134         return ret;
4135 }
4136
4137 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4138                                       void __user *arg)
4139 {
4140         struct btrfs_ioctl_get_dev_stats *sa;
4141         int ret;
4142
4143         sa = memdup_user(arg, sizeof(*sa));
4144         if (IS_ERR(sa))
4145                 return PTR_ERR(sa);
4146
4147         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4148                 kfree(sa);
4149                 return -EPERM;
4150         }
4151
4152         ret = btrfs_get_dev_stats(root, sa);
4153
4154         if (copy_to_user(arg, sa, sizeof(*sa)))
4155                 ret = -EFAULT;
4156
4157         kfree(sa);
4158         return ret;
4159 }
4160
4161 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4162 {
4163         struct btrfs_ioctl_dev_replace_args *p;
4164         int ret;
4165
4166         if (!capable(CAP_SYS_ADMIN))
4167                 return -EPERM;
4168
4169         p = memdup_user(arg, sizeof(*p));
4170         if (IS_ERR(p))
4171                 return PTR_ERR(p);
4172
4173         switch (p->cmd) {
4174         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4175                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4176                         ret = -EROFS;
4177                         goto out;
4178                 }
4179                 if (atomic_xchg(
4180                         &root->fs_info->mutually_exclusive_operation_running,
4181                         1)) {
4182                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4183                 } else {
4184                         ret = btrfs_dev_replace_start(root, p);
4185                         atomic_set(
4186                          &root->fs_info->mutually_exclusive_operation_running,
4187                          0);
4188                 }
4189                 break;
4190         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4191                 btrfs_dev_replace_status(root->fs_info, p);
4192                 ret = 0;
4193                 break;
4194         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4195                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4196                 break;
4197         default:
4198                 ret = -EINVAL;
4199                 break;
4200         }
4201
4202         if (copy_to_user(arg, p, sizeof(*p)))
4203                 ret = -EFAULT;
4204 out:
4205         kfree(p);
4206         return ret;
4207 }
4208
4209 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4210 {
4211         int ret = 0;
4212         int i;
4213         u64 rel_ptr;
4214         int size;
4215         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4216         struct inode_fs_paths *ipath = NULL;
4217         struct btrfs_path *path;
4218
4219         if (!capable(CAP_DAC_READ_SEARCH))
4220                 return -EPERM;
4221
4222         path = btrfs_alloc_path();
4223         if (!path) {
4224                 ret = -ENOMEM;
4225                 goto out;
4226         }
4227
4228         ipa = memdup_user(arg, sizeof(*ipa));
4229         if (IS_ERR(ipa)) {
4230                 ret = PTR_ERR(ipa);
4231                 ipa = NULL;
4232                 goto out;
4233         }
4234
4235         size = min_t(u32, ipa->size, 4096);
4236         ipath = init_ipath(size, root, path);
4237         if (IS_ERR(ipath)) {
4238                 ret = PTR_ERR(ipath);
4239                 ipath = NULL;
4240                 goto out;
4241         }
4242
4243         ret = paths_from_inode(ipa->inum, ipath);
4244         if (ret < 0)
4245                 goto out;
4246
4247         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4248                 rel_ptr = ipath->fspath->val[i] -
4249                           (u64)(unsigned long)ipath->fspath->val;
4250                 ipath->fspath->val[i] = rel_ptr;
4251         }
4252
4253         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4254                            (void *)(unsigned long)ipath->fspath, size);
4255         if (ret) {
4256                 ret = -EFAULT;
4257                 goto out;
4258         }
4259
4260 out:
4261         btrfs_free_path(path);
4262         free_ipath(ipath);
4263         kfree(ipa);
4264
4265         return ret;
4266 }
4267
4268 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4269 {
4270         struct btrfs_data_container *inodes = ctx;
4271         const size_t c = 3 * sizeof(u64);
4272
4273         if (inodes->bytes_left >= c) {
4274                 inodes->bytes_left -= c;
4275                 inodes->val[inodes->elem_cnt] = inum;
4276                 inodes->val[inodes->elem_cnt + 1] = offset;
4277                 inodes->val[inodes->elem_cnt + 2] = root;
4278                 inodes->elem_cnt += 3;
4279         } else {
4280                 inodes->bytes_missing += c - inodes->bytes_left;
4281                 inodes->bytes_left = 0;
4282                 inodes->elem_missed += 3;
4283         }
4284
4285         return 0;
4286 }
4287
4288 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4289                                         void __user *arg)
4290 {
4291         int ret = 0;
4292         int size;
4293         struct btrfs_ioctl_logical_ino_args *loi;
4294         struct btrfs_data_container *inodes = NULL;
4295         struct btrfs_path *path = NULL;
4296
4297         if (!capable(CAP_SYS_ADMIN))
4298                 return -EPERM;
4299
4300         loi = memdup_user(arg, sizeof(*loi));
4301         if (IS_ERR(loi)) {
4302                 ret = PTR_ERR(loi);
4303                 loi = NULL;
4304                 goto out;
4305         }
4306
4307         path = btrfs_alloc_path();
4308         if (!path) {
4309                 ret = -ENOMEM;
4310                 goto out;
4311         }
4312
4313         size = min_t(u32, loi->size, 64 * 1024);
4314         inodes = init_data_container(size);
4315         if (IS_ERR(inodes)) {
4316                 ret = PTR_ERR(inodes);
4317                 inodes = NULL;
4318                 goto out;
4319         }
4320
4321         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4322                                           build_ino_list, inodes);
4323         if (ret == -EINVAL)
4324                 ret = -ENOENT;
4325         if (ret < 0)
4326                 goto out;
4327
4328         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4329                            (void *)(unsigned long)inodes, size);
4330         if (ret)
4331                 ret = -EFAULT;
4332
4333 out:
4334         btrfs_free_path(path);
4335         vfree(inodes);
4336         kfree(loi);
4337
4338         return ret;
4339 }
4340
4341 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4342                                struct btrfs_ioctl_balance_args *bargs)
4343 {
4344         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4345
4346         bargs->flags = bctl->flags;
4347
4348         if (atomic_read(&fs_info->balance_running))
4349                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4350         if (atomic_read(&fs_info->balance_pause_req))
4351                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4352         if (atomic_read(&fs_info->balance_cancel_req))
4353                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4354
4355         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4356         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4357         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4358
4359         if (lock) {
4360                 spin_lock(&fs_info->balance_lock);
4361                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4362                 spin_unlock(&fs_info->balance_lock);
4363         } else {
4364                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4365         }
4366 }
4367
4368 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4369 {
4370         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4371         struct btrfs_fs_info *fs_info = root->fs_info;
4372         struct btrfs_ioctl_balance_args *bargs;
4373         struct btrfs_balance_control *bctl;
4374         bool need_unlock; /* for mut. excl. ops lock */
4375         int ret;
4376
4377         if (!capable(CAP_SYS_ADMIN))
4378                 return -EPERM;
4379
4380         ret = mnt_want_write_file(file);
4381         if (ret)
4382                 return ret;
4383
4384 again:
4385         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4386                 mutex_lock(&fs_info->volume_mutex);
4387                 mutex_lock(&fs_info->balance_mutex);
4388                 need_unlock = true;
4389                 goto locked;
4390         }
4391
4392         /*
4393          * mut. excl. ops lock is locked.  Three possibilites:
4394          *   (1) some other op is running
4395          *   (2) balance is running
4396          *   (3) balance is paused -- special case (think resume)
4397          */
4398         mutex_lock(&fs_info->balance_mutex);
4399         if (fs_info->balance_ctl) {
4400                 /* this is either (2) or (3) */
4401                 if (!atomic_read(&fs_info->balance_running)) {
4402                         mutex_unlock(&fs_info->balance_mutex);
4403                         if (!mutex_trylock(&fs_info->volume_mutex))
4404                                 goto again;
4405                         mutex_lock(&fs_info->balance_mutex);
4406
4407                         if (fs_info->balance_ctl &&
4408                             !atomic_read(&fs_info->balance_running)) {
4409                                 /* this is (3) */
4410                                 need_unlock = false;
4411                                 goto locked;
4412                         }
4413
4414                         mutex_unlock(&fs_info->balance_mutex);
4415                         mutex_unlock(&fs_info->volume_mutex);
4416                         goto again;
4417                 } else {
4418                         /* this is (2) */
4419                         mutex_unlock(&fs_info->balance_mutex);
4420                         ret = -EINPROGRESS;
4421                         goto out;
4422                 }
4423         } else {
4424                 /* this is (1) */
4425                 mutex_unlock(&fs_info->balance_mutex);
4426                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4427                 goto out;
4428         }
4429
4430 locked:
4431         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4432
4433         if (arg) {
4434                 bargs = memdup_user(arg, sizeof(*bargs));
4435                 if (IS_ERR(bargs)) {
4436                         ret = PTR_ERR(bargs);
4437                         goto out_unlock;
4438                 }
4439
4440                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4441                         if (!fs_info->balance_ctl) {
4442                                 ret = -ENOTCONN;
4443                                 goto out_bargs;
4444                         }
4445
4446                         bctl = fs_info->balance_ctl;
4447                         spin_lock(&fs_info->balance_lock);
4448                         bctl->flags |= BTRFS_BALANCE_RESUME;
4449                         spin_unlock(&fs_info->balance_lock);
4450
4451                         goto do_balance;
4452                 }
4453         } else {
4454                 bargs = NULL;
4455         }
4456
4457         if (fs_info->balance_ctl) {
4458                 ret = -EINPROGRESS;
4459                 goto out_bargs;
4460         }
4461
4462         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4463         if (!bctl) {
4464                 ret = -ENOMEM;
4465                 goto out_bargs;
4466         }
4467
4468         bctl->fs_info = fs_info;
4469         if (arg) {
4470                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4471                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4472                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4473
4474                 bctl->flags = bargs->flags;
4475         } else {
4476                 /* balance everything - no filters */
4477                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4478         }
4479
4480 do_balance:
4481         /*
4482          * Ownership of bctl and mutually_exclusive_operation_running
4483          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4484          * or, if restriper was paused all the way until unmount, in
4485          * free_fs_info.  mutually_exclusive_operation_running is
4486          * cleared in __cancel_balance.
4487          */
4488         need_unlock = false;
4489
4490         ret = btrfs_balance(bctl, bargs);
4491
4492         if (arg) {
4493                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4494                         ret = -EFAULT;
4495         }
4496
4497 out_bargs:
4498         kfree(bargs);
4499 out_unlock:
4500         mutex_unlock(&fs_info->balance_mutex);
4501         mutex_unlock(&fs_info->volume_mutex);
4502         if (need_unlock)
4503                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4504 out:
4505         mnt_drop_write_file(file);
4506         return ret;
4507 }
4508
4509 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4510 {
4511         if (!capable(CAP_SYS_ADMIN))
4512                 return -EPERM;
4513
4514         switch (cmd) {
4515         case BTRFS_BALANCE_CTL_PAUSE:
4516                 return btrfs_pause_balance(root->fs_info);
4517         case BTRFS_BALANCE_CTL_CANCEL:
4518                 return btrfs_cancel_balance(root->fs_info);
4519         }
4520
4521         return -EINVAL;
4522 }
4523
4524 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4525                                          void __user *arg)
4526 {
4527         struct btrfs_fs_info *fs_info = root->fs_info;
4528         struct btrfs_ioctl_balance_args *bargs;
4529         int ret = 0;
4530
4531         if (!capable(CAP_SYS_ADMIN))
4532                 return -EPERM;
4533
4534         mutex_lock(&fs_info->balance_mutex);
4535         if (!fs_info->balance_ctl) {
4536                 ret = -ENOTCONN;
4537                 goto out;
4538         }
4539
4540         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4541         if (!bargs) {
4542                 ret = -ENOMEM;
4543                 goto out;
4544         }
4545
4546         update_ioctl_balance_args(fs_info, 1, bargs);
4547
4548         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4549                 ret = -EFAULT;
4550
4551         kfree(bargs);
4552 out:
4553         mutex_unlock(&fs_info->balance_mutex);
4554         return ret;
4555 }
4556
4557 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4558 {
4559         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4560         struct btrfs_ioctl_quota_ctl_args *sa;
4561         struct btrfs_trans_handle *trans = NULL;
4562         int ret;
4563         int err;
4564
4565         if (!capable(CAP_SYS_ADMIN))
4566                 return -EPERM;
4567
4568         ret = mnt_want_write_file(file);
4569         if (ret)
4570                 return ret;
4571
4572         sa = memdup_user(arg, sizeof(*sa));
4573         if (IS_ERR(sa)) {
4574                 ret = PTR_ERR(sa);
4575                 goto drop_write;
4576         }
4577
4578         down_write(&root->fs_info->subvol_sem);
4579         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4580         if (IS_ERR(trans)) {
4581                 ret = PTR_ERR(trans);
4582                 goto out;
4583         }
4584
4585         switch (sa->cmd) {
4586         case BTRFS_QUOTA_CTL_ENABLE:
4587                 ret = btrfs_quota_enable(trans, root->fs_info);
4588                 break;
4589         case BTRFS_QUOTA_CTL_DISABLE:
4590                 ret = btrfs_quota_disable(trans, root->fs_info);
4591                 break;
4592         default:
4593                 ret = -EINVAL;
4594                 break;
4595         }
4596
4597         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4598         if (err && !ret)
4599                 ret = err;
4600 out:
4601         kfree(sa);
4602         up_write(&root->fs_info->subvol_sem);
4603 drop_write:
4604         mnt_drop_write_file(file);
4605         return ret;
4606 }
4607
4608 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4609 {
4610         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4611         struct btrfs_ioctl_qgroup_assign_args *sa;
4612         struct btrfs_trans_handle *trans;
4613         int ret;
4614         int err;
4615
4616         if (!capable(CAP_SYS_ADMIN))
4617                 return -EPERM;
4618
4619         ret = mnt_want_write_file(file);
4620         if (ret)
4621                 return ret;
4622
4623         sa = memdup_user(arg, sizeof(*sa));
4624         if (IS_ERR(sa)) {
4625                 ret = PTR_ERR(sa);
4626                 goto drop_write;
4627         }
4628
4629         trans = btrfs_join_transaction(root);
4630         if (IS_ERR(trans)) {
4631                 ret = PTR_ERR(trans);
4632                 goto out;
4633         }
4634
4635         /* FIXME: check if the IDs really exist */
4636         if (sa->assign) {
4637                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4638                                                 sa->src, sa->dst);
4639         } else {
4640                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4641                                                 sa->src, sa->dst);
4642         }
4643
4644         /* update qgroup status and info */
4645         err = btrfs_run_qgroups(trans, root->fs_info);
4646         if (err < 0)
4647                 btrfs_error(root->fs_info, ret,
4648                             "failed to update qgroup status and info\n");
4649         err = btrfs_end_transaction(trans, root);
4650         if (err && !ret)
4651                 ret = err;
4652
4653 out:
4654         kfree(sa);
4655 drop_write:
4656         mnt_drop_write_file(file);
4657         return ret;
4658 }
4659
4660 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4661 {
4662         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4663         struct btrfs_ioctl_qgroup_create_args *sa;
4664         struct btrfs_trans_handle *trans;
4665         int ret;
4666         int err;
4667
4668         if (!capable(CAP_SYS_ADMIN))
4669                 return -EPERM;
4670
4671         ret = mnt_want_write_file(file);
4672         if (ret)
4673                 return ret;
4674
4675         sa = memdup_user(arg, sizeof(*sa));
4676         if (IS_ERR(sa)) {
4677                 ret = PTR_ERR(sa);
4678                 goto drop_write;
4679         }
4680
4681         if (!sa->qgroupid) {
4682                 ret = -EINVAL;
4683                 goto out;
4684         }
4685
4686         trans = btrfs_join_transaction(root);
4687         if (IS_ERR(trans)) {
4688                 ret = PTR_ERR(trans);
4689                 goto out;
4690         }
4691
4692         /* FIXME: check if the IDs really exist */
4693         if (sa->create) {
4694                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4695         } else {
4696                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4697         }
4698
4699         err = btrfs_end_transaction(trans, root);
4700         if (err && !ret)
4701                 ret = err;
4702
4703 out:
4704         kfree(sa);
4705 drop_write:
4706         mnt_drop_write_file(file);
4707         return ret;
4708 }
4709
4710 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4711 {
4712         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4713         struct btrfs_ioctl_qgroup_limit_args *sa;
4714         struct btrfs_trans_handle *trans;
4715         int ret;
4716         int err;
4717         u64 qgroupid;
4718
4719         if (!capable(CAP_SYS_ADMIN))
4720                 return -EPERM;
4721
4722         ret = mnt_want_write_file(file);
4723         if (ret)
4724                 return ret;
4725
4726         sa = memdup_user(arg, sizeof(*sa));
4727         if (IS_ERR(sa)) {
4728                 ret = PTR_ERR(sa);
4729                 goto drop_write;
4730         }
4731
4732         trans = btrfs_join_transaction(root);
4733         if (IS_ERR(trans)) {
4734                 ret = PTR_ERR(trans);
4735                 goto out;
4736         }
4737
4738         qgroupid = sa->qgroupid;
4739         if (!qgroupid) {
4740                 /* take the current subvol as qgroup */
4741                 qgroupid = root->root_key.objectid;
4742         }
4743
4744         /* FIXME: check if the IDs really exist */
4745         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4746
4747         err = btrfs_end_transaction(trans, root);
4748         if (err && !ret)
4749                 ret = err;
4750
4751 out:
4752         kfree(sa);
4753 drop_write:
4754         mnt_drop_write_file(file);
4755         return ret;
4756 }
4757
4758 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4759 {
4760         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4761         struct btrfs_ioctl_quota_rescan_args *qsa;
4762         int ret;
4763
4764         if (!capable(CAP_SYS_ADMIN))
4765                 return -EPERM;
4766
4767         ret = mnt_want_write_file(file);
4768         if (ret)
4769                 return ret;
4770
4771         qsa = memdup_user(arg, sizeof(*qsa));
4772         if (IS_ERR(qsa)) {
4773                 ret = PTR_ERR(qsa);
4774                 goto drop_write;
4775         }
4776
4777         if (qsa->flags) {
4778                 ret = -EINVAL;
4779                 goto out;
4780         }
4781
4782         ret = btrfs_qgroup_rescan(root->fs_info);
4783
4784 out:
4785         kfree(qsa);
4786 drop_write:
4787         mnt_drop_write_file(file);
4788         return ret;
4789 }
4790
4791 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4792 {
4793         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4794         struct btrfs_ioctl_quota_rescan_args *qsa;
4795         int ret = 0;
4796
4797         if (!capable(CAP_SYS_ADMIN))
4798                 return -EPERM;
4799
4800         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4801         if (!qsa)
4802                 return -ENOMEM;
4803
4804         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4805                 qsa->flags = 1;
4806                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4807         }
4808
4809         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4810                 ret = -EFAULT;
4811
4812         kfree(qsa);
4813         return ret;
4814 }
4815
4816 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4817 {
4818         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4819
4820         if (!capable(CAP_SYS_ADMIN))
4821                 return -EPERM;
4822
4823         return btrfs_qgroup_wait_for_completion(root->fs_info);
4824 }
4825
4826 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4827                                             struct btrfs_ioctl_received_subvol_args *sa)
4828 {
4829         struct inode *inode = file_inode(file);
4830         struct btrfs_root *root = BTRFS_I(inode)->root;
4831         struct btrfs_root_item *root_item = &root->root_item;
4832         struct btrfs_trans_handle *trans;
4833         struct timespec ct = CURRENT_TIME;
4834         int ret = 0;
4835         int received_uuid_changed;
4836
4837         if (!inode_owner_or_capable(inode))
4838                 return -EPERM;
4839
4840         ret = mnt_want_write_file(file);
4841         if (ret < 0)
4842                 return ret;
4843
4844         down_write(&root->fs_info->subvol_sem);
4845
4846         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4847                 ret = -EINVAL;
4848                 goto out;
4849         }
4850
4851         if (btrfs_root_readonly(root)) {
4852                 ret = -EROFS;
4853                 goto out;
4854         }
4855
4856         /*
4857          * 1 - root item
4858          * 2 - uuid items (received uuid + subvol uuid)
4859          */
4860         trans = btrfs_start_transaction(root, 3);
4861         if (IS_ERR(trans)) {
4862                 ret = PTR_ERR(trans);
4863                 trans = NULL;
4864                 goto out;
4865         }
4866
4867         sa->rtransid = trans->transid;
4868         sa->rtime.sec = ct.tv_sec;
4869         sa->rtime.nsec = ct.tv_nsec;
4870
4871         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4872                                        BTRFS_UUID_SIZE);
4873         if (received_uuid_changed &&
4874             !btrfs_is_empty_uuid(root_item->received_uuid))
4875                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4876                                     root_item->received_uuid,
4877                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4878                                     root->root_key.objectid);
4879         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4880         btrfs_set_root_stransid(root_item, sa->stransid);
4881         btrfs_set_root_rtransid(root_item, sa->rtransid);
4882         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4883         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4884         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4885         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4886
4887         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4888                                 &root->root_key, &root->root_item);
4889         if (ret < 0) {
4890                 btrfs_end_transaction(trans, root);
4891                 goto out;
4892         }
4893         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4894                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4895                                           sa->uuid,
4896                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4897                                           root->root_key.objectid);
4898                 if (ret < 0 && ret != -EEXIST) {
4899                         btrfs_abort_transaction(trans, root, ret);
4900                         goto out;
4901                 }
4902         }
4903         ret = btrfs_commit_transaction(trans, root);
4904         if (ret < 0) {
4905                 btrfs_abort_transaction(trans, root, ret);
4906                 goto out;
4907         }
4908
4909 out:
4910         up_write(&root->fs_info->subvol_sem);
4911         mnt_drop_write_file(file);
4912         return ret;
4913 }
4914
4915 #ifdef CONFIG_64BIT
4916 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4917                                                 void __user *arg)
4918 {
4919         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4920         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4921         int ret = 0;
4922
4923         args32 = memdup_user(arg, sizeof(*args32));
4924         if (IS_ERR(args32)) {
4925                 ret = PTR_ERR(args32);
4926                 args32 = NULL;
4927                 goto out;
4928         }
4929
4930         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4931         if (!args64) {
4932                 ret = -ENOMEM;
4933                 goto out;
4934         }
4935
4936         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4937         args64->stransid = args32->stransid;
4938         args64->rtransid = args32->rtransid;
4939         args64->stime.sec = args32->stime.sec;
4940         args64->stime.nsec = args32->stime.nsec;
4941         args64->rtime.sec = args32->rtime.sec;
4942         args64->rtime.nsec = args32->rtime.nsec;
4943         args64->flags = args32->flags;
4944
4945         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4946         if (ret)
4947                 goto out;
4948
4949         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4950         args32->stransid = args64->stransid;
4951         args32->rtransid = args64->rtransid;
4952         args32->stime.sec = args64->stime.sec;
4953         args32->stime.nsec = args64->stime.nsec;
4954         args32->rtime.sec = args64->rtime.sec;
4955         args32->rtime.nsec = args64->rtime.nsec;
4956         args32->flags = args64->flags;
4957
4958         ret = copy_to_user(arg, args32, sizeof(*args32));
4959         if (ret)
4960                 ret = -EFAULT;
4961
4962 out:
4963         kfree(args32);
4964         kfree(args64);
4965         return ret;
4966 }
4967 #endif
4968
4969 static long btrfs_ioctl_set_received_subvol(struct file *file,
4970                                             void __user *arg)
4971 {
4972         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4973         int ret = 0;
4974
4975         sa = memdup_user(arg, sizeof(*sa));
4976         if (IS_ERR(sa)) {
4977                 ret = PTR_ERR(sa);
4978                 sa = NULL;
4979                 goto out;
4980         }
4981
4982         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4983
4984         if (ret)
4985                 goto out;
4986
4987         ret = copy_to_user(arg, sa, sizeof(*sa));
4988         if (ret)
4989                 ret = -EFAULT;
4990
4991 out:
4992         kfree(sa);
4993         return ret;
4994 }
4995
4996 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4997 {
4998         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4999         size_t len;
5000         int ret;
5001         char label[BTRFS_LABEL_SIZE];
5002
5003         spin_lock(&root->fs_info->super_lock);
5004         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5005         spin_unlock(&root->fs_info->super_lock);
5006
5007         len = strnlen(label, BTRFS_LABEL_SIZE);
5008
5009         if (len == BTRFS_LABEL_SIZE) {
5010                 btrfs_warn(root->fs_info,
5011                         "label is too long, return the first %zu bytes", --len);
5012         }
5013
5014         ret = copy_to_user(arg, label, len);
5015
5016         return ret ? -EFAULT : 0;
5017 }
5018
5019 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5020 {
5021         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5022         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5023         struct btrfs_trans_handle *trans;
5024         char label[BTRFS_LABEL_SIZE];
5025         int ret;
5026
5027         if (!capable(CAP_SYS_ADMIN))
5028                 return -EPERM;
5029
5030         if (copy_from_user(label, arg, sizeof(label)))
5031                 return -EFAULT;
5032
5033         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5034                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5035                        BTRFS_LABEL_SIZE - 1);
5036                 return -EINVAL;
5037         }
5038
5039         ret = mnt_want_write_file(file);
5040         if (ret)
5041                 return ret;
5042
5043         trans = btrfs_start_transaction(root, 0);
5044         if (IS_ERR(trans)) {
5045                 ret = PTR_ERR(trans);
5046                 goto out_unlock;
5047         }
5048
5049         spin_lock(&root->fs_info->super_lock);
5050         strcpy(super_block->label, label);
5051         spin_unlock(&root->fs_info->super_lock);
5052         ret = btrfs_commit_transaction(trans, root);
5053
5054 out_unlock:
5055         mnt_drop_write_file(file);
5056         return ret;
5057 }
5058
5059 #define INIT_FEATURE_FLAGS(suffix) \
5060         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5061           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5062           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5063
5064 static int btrfs_ioctl_get_supported_features(struct file *file,
5065                                               void __user *arg)
5066 {
5067         static struct btrfs_ioctl_feature_flags features[3] = {
5068                 INIT_FEATURE_FLAGS(SUPP),
5069                 INIT_FEATURE_FLAGS(SAFE_SET),
5070                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5071         };
5072
5073         if (copy_to_user(arg, &features, sizeof(features)))
5074                 return -EFAULT;
5075
5076         return 0;
5077 }
5078
5079 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5080 {
5081         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5082         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5083         struct btrfs_ioctl_feature_flags features;
5084
5085         features.compat_flags = btrfs_super_compat_flags(super_block);
5086         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5087         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5088
5089         if (copy_to_user(arg, &features, sizeof(features)))
5090                 return -EFAULT;
5091
5092         return 0;
5093 }
5094
5095 static int check_feature_bits(struct btrfs_root *root,
5096                               enum btrfs_feature_set set,
5097                               u64 change_mask, u64 flags, u64 supported_flags,
5098                               u64 safe_set, u64 safe_clear)
5099 {
5100         const char *type = btrfs_feature_set_names[set];
5101         char *names;
5102         u64 disallowed, unsupported;
5103         u64 set_mask = flags & change_mask;
5104         u64 clear_mask = ~flags & change_mask;
5105
5106         unsupported = set_mask & ~supported_flags;
5107         if (unsupported) {
5108                 names = btrfs_printable_features(set, unsupported);
5109                 if (names) {
5110                         btrfs_warn(root->fs_info,
5111                            "this kernel does not support the %s feature bit%s",
5112                            names, strchr(names, ',') ? "s" : "");
5113                         kfree(names);
5114                 } else
5115                         btrfs_warn(root->fs_info,
5116                            "this kernel does not support %s bits 0x%llx",
5117                            type, unsupported);
5118                 return -EOPNOTSUPP;
5119         }
5120
5121         disallowed = set_mask & ~safe_set;
5122         if (disallowed) {
5123                 names = btrfs_printable_features(set, disallowed);
5124                 if (names) {
5125                         btrfs_warn(root->fs_info,
5126                            "can't set the %s feature bit%s while mounted",
5127                            names, strchr(names, ',') ? "s" : "");
5128                         kfree(names);
5129                 } else
5130                         btrfs_warn(root->fs_info,
5131                            "can't set %s bits 0x%llx while mounted",
5132                            type, disallowed);
5133                 return -EPERM;
5134         }
5135
5136         disallowed = clear_mask & ~safe_clear;
5137         if (disallowed) {
5138                 names = btrfs_printable_features(set, disallowed);
5139                 if (names) {
5140                         btrfs_warn(root->fs_info,
5141                            "can't clear the %s feature bit%s while mounted",
5142                            names, strchr(names, ',') ? "s" : "");
5143                         kfree(names);
5144                 } else
5145                         btrfs_warn(root->fs_info,
5146                            "can't clear %s bits 0x%llx while mounted",
5147                            type, disallowed);
5148                 return -EPERM;
5149         }
5150
5151         return 0;
5152 }
5153
5154 #define check_feature(root, change_mask, flags, mask_base)      \
5155 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5156                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5157                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5158                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5159
5160 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5161 {
5162         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5163         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5164         struct btrfs_ioctl_feature_flags flags[2];
5165         struct btrfs_trans_handle *trans;
5166         u64 newflags;
5167         int ret;
5168
5169         if (!capable(CAP_SYS_ADMIN))
5170                 return -EPERM;
5171
5172         if (copy_from_user(flags, arg, sizeof(flags)))
5173                 return -EFAULT;
5174
5175         /* Nothing to do */
5176         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5177             !flags[0].incompat_flags)
5178                 return 0;
5179
5180         ret = check_feature(root, flags[0].compat_flags,
5181                             flags[1].compat_flags, COMPAT);
5182         if (ret)
5183                 return ret;
5184
5185         ret = check_feature(root, flags[0].compat_ro_flags,
5186                             flags[1].compat_ro_flags, COMPAT_RO);
5187         if (ret)
5188                 return ret;
5189
5190         ret = check_feature(root, flags[0].incompat_flags,
5191                             flags[1].incompat_flags, INCOMPAT);
5192         if (ret)
5193                 return ret;
5194
5195         trans = btrfs_start_transaction(root, 0);
5196         if (IS_ERR(trans))
5197                 return PTR_ERR(trans);
5198
5199         spin_lock(&root->fs_info->super_lock);
5200         newflags = btrfs_super_compat_flags(super_block);
5201         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5202         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5203         btrfs_set_super_compat_flags(super_block, newflags);
5204
5205         newflags = btrfs_super_compat_ro_flags(super_block);
5206         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5207         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5208         btrfs_set_super_compat_ro_flags(super_block, newflags);
5209
5210         newflags = btrfs_super_incompat_flags(super_block);
5211         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5212         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5213         btrfs_set_super_incompat_flags(super_block, newflags);
5214         spin_unlock(&root->fs_info->super_lock);
5215
5216         return btrfs_commit_transaction(trans, root);
5217 }
5218
5219 long btrfs_ioctl(struct file *file, unsigned int
5220                 cmd, unsigned long arg)
5221 {
5222         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5223         void __user *argp = (void __user *)arg;
5224
5225         switch (cmd) {
5226         case FS_IOC_GETFLAGS:
5227                 return btrfs_ioctl_getflags(file, argp);
5228         case FS_IOC_SETFLAGS:
5229                 return btrfs_ioctl_setflags(file, argp);
5230         case FS_IOC_GETVERSION:
5231                 return btrfs_ioctl_getversion(file, argp);
5232         case FITRIM:
5233                 return btrfs_ioctl_fitrim(file, argp);
5234         case BTRFS_IOC_SNAP_CREATE:
5235                 return btrfs_ioctl_snap_create(file, argp, 0);
5236         case BTRFS_IOC_SNAP_CREATE_V2:
5237                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5238         case BTRFS_IOC_SUBVOL_CREATE:
5239                 return btrfs_ioctl_snap_create(file, argp, 1);
5240         case BTRFS_IOC_SUBVOL_CREATE_V2:
5241                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5242         case BTRFS_IOC_SNAP_DESTROY:
5243                 return btrfs_ioctl_snap_destroy(file, argp);
5244         case BTRFS_IOC_SUBVOL_GETFLAGS:
5245                 return btrfs_ioctl_subvol_getflags(file, argp);
5246         case BTRFS_IOC_SUBVOL_SETFLAGS:
5247                 return btrfs_ioctl_subvol_setflags(file, argp);
5248         case BTRFS_IOC_DEFAULT_SUBVOL:
5249                 return btrfs_ioctl_default_subvol(file, argp);
5250         case BTRFS_IOC_DEFRAG:
5251                 return btrfs_ioctl_defrag(file, NULL);
5252         case BTRFS_IOC_DEFRAG_RANGE:
5253                 return btrfs_ioctl_defrag(file, argp);
5254         case BTRFS_IOC_RESIZE:
5255                 return btrfs_ioctl_resize(file, argp);
5256         case BTRFS_IOC_ADD_DEV:
5257                 return btrfs_ioctl_add_dev(root, argp);
5258         case BTRFS_IOC_RM_DEV:
5259                 return btrfs_ioctl_rm_dev(file, argp);
5260         case BTRFS_IOC_FS_INFO:
5261                 return btrfs_ioctl_fs_info(root, argp);
5262         case BTRFS_IOC_DEV_INFO:
5263                 return btrfs_ioctl_dev_info(root, argp);
5264         case BTRFS_IOC_BALANCE:
5265                 return btrfs_ioctl_balance(file, NULL);
5266         case BTRFS_IOC_CLONE:
5267                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5268         case BTRFS_IOC_CLONE_RANGE:
5269                 return btrfs_ioctl_clone_range(file, argp);
5270         case BTRFS_IOC_TRANS_START:
5271                 return btrfs_ioctl_trans_start(file);
5272         case BTRFS_IOC_TRANS_END:
5273                 return btrfs_ioctl_trans_end(file);
5274         case BTRFS_IOC_TREE_SEARCH:
5275                 return btrfs_ioctl_tree_search(file, argp);
5276         case BTRFS_IOC_TREE_SEARCH_V2:
5277                 return btrfs_ioctl_tree_search_v2(file, argp);
5278         case BTRFS_IOC_INO_LOOKUP:
5279                 return btrfs_ioctl_ino_lookup(file, argp);
5280         case BTRFS_IOC_INO_PATHS:
5281                 return btrfs_ioctl_ino_to_path(root, argp);
5282         case BTRFS_IOC_LOGICAL_INO:
5283                 return btrfs_ioctl_logical_to_ino(root, argp);
5284         case BTRFS_IOC_SPACE_INFO:
5285                 return btrfs_ioctl_space_info(root, argp);
5286         case BTRFS_IOC_SYNC: {
5287                 int ret;
5288
5289                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5290                 if (ret)
5291                         return ret;
5292                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5293                 /*
5294                  * The transaction thread may want to do more work,
5295                  * namely it pokes the cleaner ktread that will start
5296                  * processing uncleaned subvols.
5297                  */
5298                 wake_up_process(root->fs_info->transaction_kthread);
5299                 return ret;
5300         }
5301         case BTRFS_IOC_START_SYNC:
5302                 return btrfs_ioctl_start_sync(root, argp);
5303         case BTRFS_IOC_WAIT_SYNC:
5304                 return btrfs_ioctl_wait_sync(root, argp);
5305         case BTRFS_IOC_SCRUB:
5306                 return btrfs_ioctl_scrub(file, argp);
5307         case BTRFS_IOC_SCRUB_CANCEL:
5308                 return btrfs_ioctl_scrub_cancel(root, argp);
5309         case BTRFS_IOC_SCRUB_PROGRESS:
5310                 return btrfs_ioctl_scrub_progress(root, argp);
5311         case BTRFS_IOC_BALANCE_V2:
5312                 return btrfs_ioctl_balance(file, argp);
5313         case BTRFS_IOC_BALANCE_CTL:
5314                 return btrfs_ioctl_balance_ctl(root, arg);
5315         case BTRFS_IOC_BALANCE_PROGRESS:
5316                 return btrfs_ioctl_balance_progress(root, argp);
5317         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5318                 return btrfs_ioctl_set_received_subvol(file, argp);
5319 #ifdef CONFIG_64BIT
5320         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5321                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5322 #endif
5323         case BTRFS_IOC_SEND:
5324                 return btrfs_ioctl_send(file, argp);
5325         case BTRFS_IOC_GET_DEV_STATS:
5326                 return btrfs_ioctl_get_dev_stats(root, argp);
5327         case BTRFS_IOC_QUOTA_CTL:
5328                 return btrfs_ioctl_quota_ctl(file, argp);
5329         case BTRFS_IOC_QGROUP_ASSIGN:
5330                 return btrfs_ioctl_qgroup_assign(file, argp);
5331         case BTRFS_IOC_QGROUP_CREATE:
5332                 return btrfs_ioctl_qgroup_create(file, argp);
5333         case BTRFS_IOC_QGROUP_LIMIT:
5334                 return btrfs_ioctl_qgroup_limit(file, argp);
5335         case BTRFS_IOC_QUOTA_RESCAN:
5336                 return btrfs_ioctl_quota_rescan(file, argp);
5337         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5338                 return btrfs_ioctl_quota_rescan_status(file, argp);
5339         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5340                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5341         case BTRFS_IOC_DEV_REPLACE:
5342                 return btrfs_ioctl_dev_replace(root, argp);
5343         case BTRFS_IOC_GET_FSLABEL:
5344                 return btrfs_ioctl_get_fslabel(file, argp);
5345         case BTRFS_IOC_SET_FSLABEL:
5346                 return btrfs_ioctl_set_fslabel(file, argp);
5347         case BTRFS_IOC_FILE_EXTENT_SAME:
5348                 return btrfs_ioctl_file_extent_same(file, argp);
5349         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5350                 return btrfs_ioctl_get_supported_features(file, argp);
5351         case BTRFS_IOC_GET_FEATURES:
5352                 return btrfs_ioctl_get_features(file, argp);
5353         case BTRFS_IOC_SET_FEATURES:
5354                 return btrfs_ioctl_set_features(file, argp);
5355         }
5356
5357         return -ENOTTY;
5358 }