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