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 "compat.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "ioctl.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
  60 /* Mask out flags that are inappropriate for the given type of inode. */
  61 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  62 {
  63         if (S_ISDIR(mode))
  64                 return flags;
  65         else if (S_ISREG(mode))
  66                 return flags & ~FS_DIRSYNC_FL;
  67         else
  68                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  69 }
  70
  71 /*
  72  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  73  */
  74 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  75 {
  76         unsigned int iflags = 0;
  77
  78         if (flags & BTRFS_INODE_SYNC)
  79                 iflags |= FS_SYNC_FL;
  80         if (flags & BTRFS_INODE_IMMUTABLE)
  81                 iflags |= FS_IMMUTABLE_FL;
  82         if (flags & BTRFS_INODE_APPEND)
  83                 iflags |= FS_APPEND_FL;
  84         if (flags & BTRFS_INODE_NODUMP)
  85                 iflags |= FS_NODUMP_FL;
  86         if (flags & BTRFS_INODE_NOATIME)
  87                 iflags |= FS_NOATIME_FL;
  88         if (flags & BTRFS_INODE_DIRSYNC)
  89                 iflags |= FS_DIRSYNC_FL;
  90         if (flags & BTRFS_INODE_NODATACOW)
  91                 iflags |= FS_NOCOW_FL;
  92
  93         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
  94                 iflags |= FS_COMPR_FL;
  95         else if (flags & BTRFS_INODE_NOCOMPRESS)
  96                 iflags |= FS_NOCOMP_FL;
  97
  98         return iflags;
  99 }
 100
 101 /*
 102  * Update inode->i_flags based on the btrfs internal flags.
 103  */
 104 void btrfs_update_iflags(struct inode *inode)
 105 {
 106         struct btrfs_inode *ip = BTRFS_I(inode);
 107
 108         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 109
 110         if (ip->flags & BTRFS_INODE_SYNC)
 111                 inode->i_flags |= S_SYNC;
 112         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 113                 inode->i_flags |= S_IMMUTABLE;
 114         if (ip->flags & BTRFS_INODE_APPEND)
 115                 inode->i_flags |= S_APPEND;
 116         if (ip->flags & BTRFS_INODE_NOATIME)
 117                 inode->i_flags |= S_NOATIME;
 118         if (ip->flags & BTRFS_INODE_DIRSYNC)
 119                 inode->i_flags |= S_DIRSYNC;
 120 }
 121
 122 /*
 123  * Inherit flags from the parent inode.
 124  *
 125  * Currently only the compression flags and the cow flags are inherited.
 126  */
 127 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 128 {
 129         unsigned int flags;
 130
 131         if (!dir)
 132                 return;
 133
 134         flags = BTRFS_I(dir)->flags;
 135
 136         if (flags & BTRFS_INODE_NOCOMPRESS) {
 137                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 138                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 139         } else if (flags & BTRFS_INODE_COMPRESS) {
 140                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 141                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 142         }
 143
 144         if (flags & BTRFS_INODE_NODATACOW)
 145                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 146
 147         btrfs_update_iflags(inode);
 148 }
 149
 150 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 151 {
 152         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 153         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 154
 155         if (copy_to_user(arg, &flags, sizeof(flags)))
 156                 return -EFAULT;
 157         return 0;
 158 }
 159
 160 static int check_flags(unsigned int flags)
 161 {
 162         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 163                       FS_NOATIME_FL | FS_NODUMP_FL | \
 164                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 165                       FS_NOCOMP_FL | FS_COMPR_FL |
 166                       FS_NOCOW_FL))
 167                 return -EOPNOTSUPP;
 168
 169         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 170                 return -EINVAL;
 171
 172         return 0;
 173 }
 174
 175 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 176 {
 177         struct inode *inode = file->f_path.dentry->d_inode;
 178         struct btrfs_inode *ip = BTRFS_I(inode);
 179         struct btrfs_root *root = ip->root;
 180         struct btrfs_trans_handle *trans;
 181         unsigned int flags, oldflags;
 182         int ret;
 183         u64 ip_oldflags;
 184         unsigned int i_oldflags;
 185         umode_t mode;
 186
 187         if (btrfs_root_readonly(root))
 188                 return -EROFS;
 189
 190         if (copy_from_user(&flags, arg, sizeof(flags)))
 191                 return -EFAULT;
 192
 193         ret = check_flags(flags);
 194         if (ret)
 195                 return ret;
 196
 197         if (!inode_owner_or_capable(inode))
 198                 return -EACCES;
 199
 200         ret = mnt_want_write_file(file);
 201         if (ret)
 202                 return ret;
 203
 204         mutex_lock(&inode->i_mutex);
 205
 206         ip_oldflags = ip->flags;
 207         i_oldflags = inode->i_flags;
 208         mode = inode->i_mode;
 209
 210         flags = btrfs_mask_flags(inode->i_mode, flags);
 211         oldflags = btrfs_flags_to_ioctl(ip->flags);
 212         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 213                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 214                         ret = -EPERM;
 215                         goto out_unlock;
 216                 }
 217         }
 218
 219         if (flags & FS_SYNC_FL)
 220                 ip->flags |= BTRFS_INODE_SYNC;
 221         else
 222                 ip->flags &= ~BTRFS_INODE_SYNC;
 223         if (flags & FS_IMMUTABLE_FL)
 224                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 225         else
 226                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 227         if (flags & FS_APPEND_FL)
 228                 ip->flags |= BTRFS_INODE_APPEND;
 229         else
 230                 ip->flags &= ~BTRFS_INODE_APPEND;
 231         if (flags & FS_NODUMP_FL)
 232                 ip->flags |= BTRFS_INODE_NODUMP;
 233         else
 234                 ip->flags &= ~BTRFS_INODE_NODUMP;
 235         if (flags & FS_NOATIME_FL)
 236                 ip->flags |= BTRFS_INODE_NOATIME;
 237         else
 238                 ip->flags &= ~BTRFS_INODE_NOATIME;
 239         if (flags & FS_DIRSYNC_FL)
 240                 ip->flags |= BTRFS_INODE_DIRSYNC;
 241         else
 242                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 243         if (flags & FS_NOCOW_FL) {
 244                 if (S_ISREG(mode)) {
 245                         /*
 246                          * It's safe to turn csums off here, no extents exist.
 247                          * Otherwise we want the flag to reflect the real COW
 248                          * status of the file and will not set it.
 249                          */
 250                         if (inode->i_size == 0)
 251                                 ip->flags |= BTRFS_INODE_NODATACOW
 252                                            | BTRFS_INODE_NODATASUM;
 253                 } else {
 254                         ip->flags |= BTRFS_INODE_NODATACOW;
 255                 }
 256         } else {
 257                 /*
 258                  * Revert back under same assuptions as above
 259                  */
 260                 if (S_ISREG(mode)) {
 261                         if (inode->i_size == 0)
 262                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 263                                              | BTRFS_INODE_NODATASUM);
 264                 } else {
 265                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 266                 }
 267         }
 268
 269         /*
 270          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 271          * flag may be changed automatically if compression code won't make
 272          * things smaller.
 273          */
 274         if (flags & FS_NOCOMP_FL) {
 275                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 276                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 277         } else if (flags & FS_COMPR_FL) {
 278                 ip->flags |= BTRFS_INODE_COMPRESS;
 279                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 280         } else {
 281                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 282         }
 283
 284         trans = btrfs_start_transaction(root, 1);
 285         if (IS_ERR(trans)) {
 286                 ret = PTR_ERR(trans);
 287                 goto out_drop;
 288         }
 289
 290         btrfs_update_iflags(inode);
 291         inode_inc_iversion(inode);
 292         inode->i_ctime = CURRENT_TIME;
 293         ret = btrfs_update_inode(trans, root, inode);
 294
 295         btrfs_end_transaction(trans, root);
 296  out_drop:
 297         if (ret) {
 298                 ip->flags = ip_oldflags;
 299                 inode->i_flags = i_oldflags;
 300         }
 301
 302  out_unlock:
 303         mutex_unlock(&inode->i_mutex);
 304         mnt_drop_write_file(file);
 305         return ret;
 306 }
 307
 308 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 309 {
 310         struct inode *inode = file->f_path.dentry->d_inode;
 311
 312         return put_user(inode->i_generation, arg);
 313 }
 314
 315 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 316 {
 317         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
 318         struct btrfs_device *device;
 319         struct request_queue *q;
 320         struct fstrim_range range;
 321         u64 minlen = ULLONG_MAX;
 322         u64 num_devices = 0;
 323         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 324         int ret;
 325
 326         if (!capable(CAP_SYS_ADMIN))
 327                 return -EPERM;
 328
 329         rcu_read_lock();
 330         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 331                                 dev_list) {
 332                 if (!device->bdev)
 333                         continue;
 334                 q = bdev_get_queue(device->bdev);
 335                 if (blk_queue_discard(q)) {
 336                         num_devices++;
 337                         minlen = min((u64)q->limits.discard_granularity,
 338                                      minlen);
 339                 }
 340         }
 341         rcu_read_unlock();
 342
 343         if (!num_devices)
 344                 return -EOPNOTSUPP;
 345         if (copy_from_user(&range, arg, sizeof(range)))
 346                 return -EFAULT;
 347         if (range.start > total_bytes ||
 348             range.len < fs_info->sb->s_blocksize)
 349                 return -EINVAL;
 350
 351         range.len = min(range.len, total_bytes - range.start);
 352         range.minlen = max(range.minlen, minlen);
 353         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 354         if (ret < 0)
 355                 return ret;
 356
 357         if (copy_to_user(arg, &range, sizeof(range)))
 358                 return -EFAULT;
 359
 360         return 0;
 361 }
 362
 363 static noinline int create_subvol(struct btrfs_root *root,
 364                                   struct dentry *dentry,
 365                                   char *name, int namelen,
 366                                   u64 *async_transid,
 367                                   struct btrfs_qgroup_inherit **inherit)
 368 {
 369         struct btrfs_trans_handle *trans;
 370         struct btrfs_key key;
 371         struct btrfs_root_item root_item;
 372         struct btrfs_inode_item *inode_item;
 373         struct extent_buffer *leaf;
 374         struct btrfs_root *new_root;
 375         struct dentry *parent = dentry->d_parent;
 376         struct inode *dir;
 377         struct timespec cur_time = CURRENT_TIME;
 378         int ret;
 379         int err;
 380         u64 objectid;
 381         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 382         u64 index = 0;
 383         uuid_le new_uuid;
 384
 385         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 386         if (ret)
 387                 return ret;
 388
 389         dir = parent->d_inode;
 390
 391         /*
 392          * 1 - inode item
 393          * 2 - refs
 394          * 1 - root item
 395          * 2 - dir items
 396          */
 397         trans = btrfs_start_transaction(root, 6);
 398         if (IS_ERR(trans))
 399                 return PTR_ERR(trans);
 400
 401         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid,
 402                                    inherit ? *inherit : NULL);
 403         if (ret)
 404                 goto fail;
 405
 406         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 407                                       0, objectid, NULL, 0, 0, 0);
 408         if (IS_ERR(leaf)) {
 409                 ret = PTR_ERR(leaf);
 410                 goto fail;
 411         }
 412
 413         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 414         btrfs_set_header_bytenr(leaf, leaf->start);
 415         btrfs_set_header_generation(leaf, trans->transid);
 416         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 417         btrfs_set_header_owner(leaf, objectid);
 418
 419         write_extent_buffer(leaf, root->fs_info->fsid,
 420                             (unsigned long)btrfs_header_fsid(leaf),
 421                             BTRFS_FSID_SIZE);
 422         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 423                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 424                             BTRFS_UUID_SIZE);
 425         btrfs_mark_buffer_dirty(leaf);
 426
 427         memset(&root_item, 0, sizeof(root_item));
 428
 429         inode_item = &root_item.inode;
 430         inode_item->generation = cpu_to_le64(1);
 431         inode_item->size = cpu_to_le64(3);
 432         inode_item->nlink = cpu_to_le32(1);
 433         inode_item->nbytes = cpu_to_le64(root->leafsize);
 434         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 435
 436         root_item.flags = 0;
 437         root_item.byte_limit = 0;
 438         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
 439
 440         btrfs_set_root_bytenr(&root_item, leaf->start);
 441         btrfs_set_root_generation(&root_item, trans->transid);
 442         btrfs_set_root_level(&root_item, 0);
 443         btrfs_set_root_refs(&root_item, 1);
 444         btrfs_set_root_used(&root_item, leaf->len);
 445         btrfs_set_root_last_snapshot(&root_item, 0);
 446
 447         btrfs_set_root_generation_v2(&root_item,
 448                         btrfs_root_generation(&root_item));
 449         uuid_le_gen(&new_uuid);
 450         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 451         root_item.otime.sec = cpu_to_le64(cur_time.tv_sec);
 452         root_item.otime.nsec = cpu_to_le32(cur_time.tv_nsec);
 453         root_item.ctime = root_item.otime;
 454         btrfs_set_root_ctransid(&root_item, trans->transid);
 455         btrfs_set_root_otransid(&root_item, trans->transid);
 456
 457         btrfs_tree_unlock(leaf);
 458         free_extent_buffer(leaf);
 459         leaf = NULL;
 460
 461         btrfs_set_root_dirid(&root_item, new_dirid);
 462
 463         key.objectid = objectid;
 464         key.offset = 0;
 465         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 466         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 467                                 &root_item);
 468         if (ret)
 469                 goto fail;
 470
 471         key.offset = (u64)-1;
 472         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 473         if (IS_ERR(new_root)) {
 474                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 475                 ret = PTR_ERR(new_root);
 476                 goto fail;
 477         }
 478
 479         btrfs_record_root_in_trans(trans, new_root);
 480
 481         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
 482         if (ret) {
 483                 /* We potentially lose an unused inode item here */
 484                 btrfs_abort_transaction(trans, root, ret);
 485                 goto fail;
 486         }
 487
 488         /*
 489          * insert the directory item
 490          */
 491         ret = btrfs_set_inode_index(dir, &index);
 492         if (ret) {
 493                 btrfs_abort_transaction(trans, root, ret);
 494                 goto fail;
 495         }
 496
 497         ret = btrfs_insert_dir_item(trans, root,
 498                                     name, namelen, dir, &key,
 499                                     BTRFS_FT_DIR, index);
 500         if (ret) {
 501                 btrfs_abort_transaction(trans, root, ret);
 502                 goto fail;
 503         }
 504
 505         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 506         ret = btrfs_update_inode(trans, root, dir);
 507         BUG_ON(ret);
 508
 509         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 510                                  objectid, root->root_key.objectid,
 511                                  btrfs_ino(dir), index, name, namelen);
 512
 513         BUG_ON(ret);
 514
 515         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 516 fail:
 517         if (async_transid) {
 518                 *async_transid = trans->transid;
 519                 err = btrfs_commit_transaction_async(trans, root, 1);
 520         } else {
 521                 err = btrfs_commit_transaction(trans, root);
 522         }
 523         if (err && !ret)
 524                 ret = err;
 525         return ret;
 526 }
 527
 528 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 529                            char *name, int namelen, u64 *async_transid,
 530                            bool readonly, struct btrfs_qgroup_inherit **inherit)
 531 {
 532         struct inode *inode;
 533         struct btrfs_pending_snapshot *pending_snapshot;
 534         struct btrfs_trans_handle *trans;
 535         int ret;
 536
 537         if (!root->ref_cows)
 538                 return -EINVAL;
 539
 540         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 541         if (!pending_snapshot)
 542                 return -ENOMEM;
 543
 544         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 545                              BTRFS_BLOCK_RSV_TEMP);
 546         pending_snapshot->dentry = dentry;
 547         pending_snapshot->root = root;
 548         pending_snapshot->readonly = readonly;
 549         if (inherit) {
 550                 pending_snapshot->inherit = *inherit;
 551                 *inherit = NULL;        /* take responsibility to free it */
 552         }
 553
 554         trans = btrfs_start_transaction(root->fs_info->extent_root, 6);
 555         if (IS_ERR(trans)) {
 556                 ret = PTR_ERR(trans);
 557                 goto fail;
 558         }
 559
 560         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 561         BUG_ON(ret);
 562
 563         spin_lock(&root->fs_info->trans_lock);
 564         list_add(&pending_snapshot->list,
 565                  &trans->transaction->pending_snapshots);
 566         spin_unlock(&root->fs_info->trans_lock);
 567         if (async_transid) {
 568                 *async_transid = trans->transid;
 569                 ret = btrfs_commit_transaction_async(trans,
 570                                      root->fs_info->extent_root, 1);
 571         } else {
 572                 ret = btrfs_commit_transaction(trans,
 573                                                root->fs_info->extent_root);
 574         }
 575         if (ret) {
 576                 /* cleanup_transaction has freed this for us */
 577                 if (trans->aborted)
 578                         pending_snapshot = NULL;
 579                 goto fail;
 580         }
 581
 582         ret = pending_snapshot->error;
 583         if (ret)
 584                 goto fail;
 585
 586         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 587         if (ret)
 588                 goto fail;
 589
 590         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 591         if (IS_ERR(inode)) {
 592                 ret = PTR_ERR(inode);
 593                 goto fail;
 594         }
 595         BUG_ON(!inode);
 596         d_instantiate(dentry, inode);
 597         ret = 0;
 598 fail:
 599         kfree(pending_snapshot);
 600         return ret;
 601 }
 602
 603 /*  copy of check_sticky in fs/namei.c()
 604 * It's inline, so penalty for filesystems that don't use sticky bit is
 605 * minimal.
 606 */
 607 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 608 {
 609         kuid_t fsuid = current_fsuid();
 610
 611         if (!(dir->i_mode & S_ISVTX))
 612                 return 0;
 613         if (uid_eq(inode->i_uid, fsuid))
 614                 return 0;
 615         if (uid_eq(dir->i_uid, fsuid))
 616                 return 0;
 617         return !capable(CAP_FOWNER);
 618 }
 619
 620 /*  copy of may_delete in fs/namei.c()
 621  *      Check whether we can remove a link victim from directory dir, check
 622  *  whether the type of victim is right.
 623  *  1. We can't do it if dir is read-only (done in permission())
 624  *  2. We should have write and exec permissions on dir
 625  *  3. We can't remove anything from append-only dir
 626  *  4. We can't do anything with immutable dir (done in permission())
 627  *  5. If the sticky bit on dir is set we should either
 628  *      a. be owner of dir, or
 629  *      b. be owner of victim, or
 630  *      c. have CAP_FOWNER capability
 631  *  6. If the victim is append-only or immutable we can't do antyhing with
 632  *     links pointing to it.
 633  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 634  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 635  *  9. We can't remove a root or mountpoint.
 636  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 637  *     nfs_async_unlink().
 638  */
 639
 640 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 641 {
 642         int error;
 643
 644         if (!victim->d_inode)
 645                 return -ENOENT;
 646
 647         BUG_ON(victim->d_parent->d_inode != dir);
 648         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 649
 650         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 651         if (error)
 652                 return error;
 653         if (IS_APPEND(dir))
 654                 return -EPERM;
 655         if (btrfs_check_sticky(dir, victim->d_inode)||
 656                 IS_APPEND(victim->d_inode)||
 657             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 658                 return -EPERM;
 659         if (isdir) {
 660                 if (!S_ISDIR(victim->d_inode->i_mode))
 661                         return -ENOTDIR;
 662                 if (IS_ROOT(victim))
 663                         return -EBUSY;
 664         } else if (S_ISDIR(victim->d_inode->i_mode))
 665                 return -EISDIR;
 666         if (IS_DEADDIR(dir))
 667                 return -ENOENT;
 668         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 669                 return -EBUSY;
 670         return 0;
 671 }
 672
 673 /* copy of may_create in fs/namei.c() */
 674 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 675 {
 676         if (child->d_inode)
 677                 return -EEXIST;
 678         if (IS_DEADDIR(dir))
 679                 return -ENOENT;
 680         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 681 }
 682
 683 /*
 684  * Create a new subvolume below @parent.  This is largely modeled after
 685  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 686  * inside this filesystem so it's quite a bit simpler.
 687  */
 688 static noinline int btrfs_mksubvol(struct path *parent,
 689                                    char *name, int namelen,
 690                                    struct btrfs_root *snap_src,
 691                                    u64 *async_transid, bool readonly,
 692                                    struct btrfs_qgroup_inherit **inherit)
 693 {
 694         struct inode *dir  = parent->dentry->d_inode;
 695         struct dentry *dentry;
 696         int error;
 697
 698         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 699
 700         dentry = lookup_one_len(name, parent->dentry, namelen);
 701         error = PTR_ERR(dentry);
 702         if (IS_ERR(dentry))
 703                 goto out_unlock;
 704
 705         error = -EEXIST;
 706         if (dentry->d_inode)
 707                 goto out_dput;
 708
 709         error = btrfs_may_create(dir, dentry);
 710         if (error)
 711                 goto out_dput;
 712
 713         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 714
 715         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 716                 goto out_up_read;
 717
 718         if (snap_src) {
 719                 error = create_snapshot(snap_src, dentry, name, namelen,
 720                                         async_transid, readonly, inherit);
 721         } else {
 722                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 723                                       name, namelen, async_transid, inherit);
 724         }
 725         if (!error)
 726                 fsnotify_mkdir(dir, dentry);
 727 out_up_read:
 728         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 729 out_dput:
 730         dput(dentry);
 731 out_unlock:
 732         mutex_unlock(&dir->i_mutex);
 733         return error;
 734 }
 735
 736 /*
 737  * When we're defragging a range, we don't want to kick it off again
 738  * if it is really just waiting for delalloc to send it down.
 739  * If we find a nice big extent or delalloc range for the bytes in the
 740  * file you want to defrag, we return 0 to let you know to skip this
 741  * part of the file
 742  */
 743 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 744 {
 745         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 746         struct extent_map *em = NULL;
 747         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 748         u64 end;
 749
 750         read_lock(&em_tree->lock);
 751         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 752         read_unlock(&em_tree->lock);
 753
 754         if (em) {
 755                 end = extent_map_end(em);
 756                 free_extent_map(em);
 757                 if (end - offset > thresh)
 758                         return 0;
 759         }
 760         /* if we already have a nice delalloc here, just stop */
 761         thresh /= 2;
 762         end = count_range_bits(io_tree, &offset, offset + thresh,
 763                                thresh, EXTENT_DELALLOC, 1);
 764         if (end >= thresh)
 765                 return 0;
 766         return 1;
 767 }
 768
 769 /*
 770  * helper function to walk through a file and find extents
 771  * newer than a specific transid, and smaller than thresh.
 772  *
 773  * This is used by the defragging code to find new and small
 774  * extents
 775  */
 776 static int find_new_extents(struct btrfs_root *root,
 777                             struct inode *inode, u64 newer_than,
 778                             u64 *off, int thresh)
 779 {
 780         struct btrfs_path *path;
 781         struct btrfs_key min_key;
 782         struct btrfs_key max_key;
 783         struct extent_buffer *leaf;
 784         struct btrfs_file_extent_item *extent;
 785         int type;
 786         int ret;
 787         u64 ino = btrfs_ino(inode);
 788
 789         path = btrfs_alloc_path();
 790         if (!path)
 791                 return -ENOMEM;
 792
 793         min_key.objectid = ino;
 794         min_key.type = BTRFS_EXTENT_DATA_KEY;
 795         min_key.offset = *off;
 796
 797         max_key.objectid = ino;
 798         max_key.type = (u8)-1;
 799         max_key.offset = (u64)-1;
 800
 801         path->keep_locks = 1;
 802
 803         while(1) {
 804                 ret = btrfs_search_forward(root, &min_key, &max_key,
 805                                            path, 0, newer_than);
 806                 if (ret != 0)
 807                         goto none;
 808                 if (min_key.objectid != ino)
 809                         goto none;
 810                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 811                         goto none;
 812
 813                 leaf = path->nodes[0];
 814                 extent = btrfs_item_ptr(leaf, path->slots[0],
 815                                         struct btrfs_file_extent_item);
 816
 817                 type = btrfs_file_extent_type(leaf, extent);
 818                 if (type == BTRFS_FILE_EXTENT_REG &&
 819                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 820                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 821                         *off = min_key.offset;
 822                         btrfs_free_path(path);
 823                         return 0;
 824                 }
 825
 826                 if (min_key.offset == (u64)-1)
 827                         goto none;
 828
 829                 min_key.offset++;
 830                 btrfs_release_path(path);
 831         }
 832 none:
 833         btrfs_free_path(path);
 834         return -ENOENT;
 835 }
 836
 837 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 838 {
 839         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 840         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 841         struct extent_map *em;
 842         u64 len = PAGE_CACHE_SIZE;
 843
 844         /*
 845          * hopefully we have this extent in the tree already, try without
 846          * the full extent lock
 847          */
 848         read_lock(&em_tree->lock);
 849         em = lookup_extent_mapping(em_tree, start, len);
 850         read_unlock(&em_tree->lock);
 851
 852         if (!em) {
 853                 /* get the big lock and read metadata off disk */
 854                 lock_extent(io_tree, start, start + len - 1);
 855                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 856                 unlock_extent(io_tree, start, start + len - 1);
 857
 858                 if (IS_ERR(em))
 859                         return NULL;
 860         }
 861
 862         return em;
 863 }
 864
 865 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 866 {
 867         struct extent_map *next;
 868         bool ret = true;
 869
 870         /* this is the last extent */
 871         if (em->start + em->len >= i_size_read(inode))
 872                 return false;
 873
 874         next = defrag_lookup_extent(inode, em->start + em->len);
 875         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 876                 ret = false;
 877
 878         free_extent_map(next);
 879         return ret;
 880 }
 881
 882 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
 883                                u64 *last_len, u64 *skip, u64 *defrag_end,
 884                                int compress)
 885 {
 886         struct extent_map *em;
 887         int ret = 1;
 888         bool next_mergeable = true;
 889
 890         /*
 891          * make sure that once we start defragging an extent, we keep on
 892          * defragging it
 893          */
 894         if (start < *defrag_end)
 895                 return 1;
 896
 897         *skip = 0;
 898
 899         em = defrag_lookup_extent(inode, start);
 900         if (!em)
 901                 return 0;
 902
 903         /* this will cover holes, and inline extents */
 904         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 905                 ret = 0;
 906                 goto out;
 907         }
 908
 909         next_mergeable = defrag_check_next_extent(inode, em);
 910
 911         /*
 912          * we hit a real extent, if it is big or the next extent is not a
 913          * real extent, don't bother defragging it
 914          */
 915         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
 916             (em->len >= thresh || !next_mergeable))
 917                 ret = 0;
 918 out:
 919         /*
 920          * last_len ends up being a counter of how many bytes we've defragged.
 921          * every time we choose not to defrag an extent, we reset *last_len
 922          * so that the next tiny extent will force a defrag.
 923          *
 924          * The end result of this is that tiny extents before a single big
 925          * extent will force at least part of that big extent to be defragged.
 926          */
 927         if (ret) {
 928                 *defrag_end = extent_map_end(em);
 929         } else {
 930                 *last_len = 0;
 931                 *skip = extent_map_end(em);
 932                 *defrag_end = 0;
 933         }
 934
 935         free_extent_map(em);
 936         return ret;
 937 }
 938
 939 /*
 940  * it doesn't do much good to defrag one or two pages
 941  * at a time.  This pulls in a nice chunk of pages
 942  * to COW and defrag.
 943  *
 944  * It also makes sure the delalloc code has enough
 945  * dirty data to avoid making new small extents as part
 946  * of the defrag
 947  *
 948  * It's a good idea to start RA on this range
 949  * before calling this.
 950  */
 951 static int cluster_pages_for_defrag(struct inode *inode,
 952                                     struct page **pages,
 953                                     unsigned long start_index,
 954                                     int num_pages)
 955 {
 956         unsigned long file_end;
 957         u64 isize = i_size_read(inode);
 958         u64 page_start;
 959         u64 page_end;
 960         u64 page_cnt;
 961         int ret;
 962         int i;
 963         int i_done;
 964         struct btrfs_ordered_extent *ordered;
 965         struct extent_state *cached_state = NULL;
 966         struct extent_io_tree *tree;
 967         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 968
 969         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 970         if (!isize || start_index > file_end)
 971                 return 0;
 972
 973         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
 974
 975         ret = btrfs_delalloc_reserve_space(inode,
 976                                            page_cnt << PAGE_CACHE_SHIFT);
 977         if (ret)
 978                 return ret;
 979         i_done = 0;
 980         tree = &BTRFS_I(inode)->io_tree;
 981
 982         /* step one, lock all the pages */
 983         for (i = 0; i < page_cnt; i++) {
 984                 struct page *page;
 985 again:
 986                 page = find_or_create_page(inode->i_mapping,
 987                                            start_index + i, mask);
 988                 if (!page)
 989                         break;
 990
 991                 page_start = page_offset(page);
 992                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 993                 while (1) {
 994                         lock_extent(tree, page_start, page_end);
 995                         ordered = btrfs_lookup_ordered_extent(inode,
 996                                                               page_start);
 997                         unlock_extent(tree, page_start, page_end);
 998                         if (!ordered)
 999                                 break;
1000
1001                         unlock_page(page);
1002                         btrfs_start_ordered_extent(inode, ordered, 1);
1003                         btrfs_put_ordered_extent(ordered);
1004                         lock_page(page);
1005                         /*
1006                          * we unlocked the page above, so we need check if
1007                          * it was released or not.
1008                          */
1009                         if (page->mapping != inode->i_mapping) {
1010                                 unlock_page(page);
1011                                 page_cache_release(page);
1012                                 goto again;
1013                         }
1014                 }
1015
1016                 if (!PageUptodate(page)) {
1017                         btrfs_readpage(NULL, page);
1018                         lock_page(page);
1019                         if (!PageUptodate(page)) {
1020                                 unlock_page(page);
1021                                 page_cache_release(page);
1022                                 ret = -EIO;
1023                                 break;
1024                         }
1025                 }
1026
1027                 if (page->mapping != inode->i_mapping) {
1028                         unlock_page(page);
1029                         page_cache_release(page);
1030                         goto again;
1031                 }
1032
1033                 pages[i] = page;
1034                 i_done++;
1035         }
1036         if (!i_done || ret)
1037                 goto out;
1038
1039         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1040                 goto out;
1041
1042         /*
1043          * so now we have a nice long stream of locked
1044          * and up to date pages, lets wait on them
1045          */
1046         for (i = 0; i < i_done; i++)
1047                 wait_on_page_writeback(pages[i]);
1048
1049         page_start = page_offset(pages[0]);
1050         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1051
1052         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1053                          page_start, page_end - 1, 0, &cached_state);
1054         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1055                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1056                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1057                           &cached_state, GFP_NOFS);
1058
1059         if (i_done != page_cnt) {
1060                 spin_lock(&BTRFS_I(inode)->lock);
1061                 BTRFS_I(inode)->outstanding_extents++;
1062                 spin_unlock(&BTRFS_I(inode)->lock);
1063                 btrfs_delalloc_release_space(inode,
1064                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1065         }
1066
1067
1068         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1069                           &cached_state, GFP_NOFS);
1070
1071         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1072                              page_start, page_end - 1, &cached_state,
1073                              GFP_NOFS);
1074
1075         for (i = 0; i < i_done; i++) {
1076                 clear_page_dirty_for_io(pages[i]);
1077                 ClearPageChecked(pages[i]);
1078                 set_page_extent_mapped(pages[i]);
1079                 set_page_dirty(pages[i]);
1080                 unlock_page(pages[i]);
1081                 page_cache_release(pages[i]);
1082         }
1083         return i_done;
1084 out:
1085         for (i = 0; i < i_done; i++) {
1086                 unlock_page(pages[i]);
1087                 page_cache_release(pages[i]);
1088         }
1089         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1090         return ret;
1091
1092 }
1093
1094 int btrfs_defrag_file(struct inode *inode, struct file *file,
1095                       struct btrfs_ioctl_defrag_range_args *range,
1096                       u64 newer_than, unsigned long max_to_defrag)
1097 {
1098         struct btrfs_root *root = BTRFS_I(inode)->root;
1099         struct file_ra_state *ra = NULL;
1100         unsigned long last_index;
1101         u64 isize = i_size_read(inode);
1102         u64 last_len = 0;
1103         u64 skip = 0;
1104         u64 defrag_end = 0;
1105         u64 newer_off = range->start;
1106         unsigned long i;
1107         unsigned long ra_index = 0;
1108         int ret;
1109         int defrag_count = 0;
1110         int compress_type = BTRFS_COMPRESS_ZLIB;
1111         int extent_thresh = range->extent_thresh;
1112         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1113         int cluster = max_cluster;
1114         u64 new_align = ~((u64)128 * 1024 - 1);
1115         struct page **pages = NULL;
1116
1117         if (extent_thresh == 0)
1118                 extent_thresh = 256 * 1024;
1119
1120         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1121                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1122                         return -EINVAL;
1123                 if (range->compress_type)
1124                         compress_type = range->compress_type;
1125         }
1126
1127         if (isize == 0)
1128                 return 0;
1129
1130         /*
1131          * if we were not given a file, allocate a readahead
1132          * context
1133          */
1134         if (!file) {
1135                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1136                 if (!ra)
1137                         return -ENOMEM;
1138                 file_ra_state_init(ra, inode->i_mapping);
1139         } else {
1140                 ra = &file->f_ra;
1141         }
1142
1143         pages = kmalloc(sizeof(struct page *) * max_cluster,
1144                         GFP_NOFS);
1145         if (!pages) {
1146                 ret = -ENOMEM;
1147                 goto out_ra;
1148         }
1149
1150         /* find the last page to defrag */
1151         if (range->start + range->len > range->start) {
1152                 last_index = min_t(u64, isize - 1,
1153                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1154         } else {
1155                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1156         }
1157
1158         if (newer_than) {
1159                 ret = find_new_extents(root, inode, newer_than,
1160                                        &newer_off, 64 * 1024);
1161                 if (!ret) {
1162                         range->start = newer_off;
1163                         /*
1164                          * we always align our defrag to help keep
1165                          * the extents in the file evenly spaced
1166                          */
1167                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1168                 } else
1169                         goto out_ra;
1170         } else {
1171                 i = range->start >> PAGE_CACHE_SHIFT;
1172         }
1173         if (!max_to_defrag)
1174                 max_to_defrag = last_index + 1;
1175
1176         /*
1177          * make writeback starts from i, so the defrag range can be
1178          * written sequentially.
1179          */
1180         if (i < inode->i_mapping->writeback_index)
1181                 inode->i_mapping->writeback_index = i;
1182
1183         while (i <= last_index && defrag_count < max_to_defrag &&
1184                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1185                 PAGE_CACHE_SHIFT)) {
1186                 /*
1187                  * make sure we stop running if someone unmounts
1188                  * the FS
1189                  */
1190                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1191                         break;
1192
1193                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1194                                          extent_thresh, &last_len, &skip,
1195                                          &defrag_end, range->flags &
1196                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1197                         unsigned long next;
1198                         /*
1199                          * the should_defrag function tells us how much to skip
1200                          * bump our counter by the suggested amount
1201                          */
1202                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1203                         i = max(i + 1, next);
1204                         continue;
1205                 }
1206
1207                 if (!newer_than) {
1208                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1209                                    PAGE_CACHE_SHIFT) - i;
1210                         cluster = min(cluster, max_cluster);
1211                 } else {
1212                         cluster = max_cluster;
1213                 }
1214
1215                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1216                         BTRFS_I(inode)->force_compress = compress_type;
1217
1218                 if (i + cluster > ra_index) {
1219                         ra_index = max(i, ra_index);
1220                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1221                                        cluster);
1222                         ra_index += max_cluster;
1223                 }
1224
1225                 mutex_lock(&inode->i_mutex);
1226                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1227                 if (ret < 0) {
1228                         mutex_unlock(&inode->i_mutex);
1229                         goto out_ra;
1230                 }
1231
1232                 defrag_count += ret;
1233                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1234                 mutex_unlock(&inode->i_mutex);
1235
1236                 if (newer_than) {
1237                         if (newer_off == (u64)-1)
1238                                 break;
1239
1240                         if (ret > 0)
1241                                 i += ret;
1242
1243                         newer_off = max(newer_off + 1,
1244                                         (u64)i << PAGE_CACHE_SHIFT);
1245
1246                         ret = find_new_extents(root, inode,
1247                                                newer_than, &newer_off,
1248                                                64 * 1024);
1249                         if (!ret) {
1250                                 range->start = newer_off;
1251                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1252                         } else {
1253                                 break;
1254                         }
1255                 } else {
1256                         if (ret > 0) {
1257                                 i += ret;
1258                                 last_len += ret << PAGE_CACHE_SHIFT;
1259                         } else {
1260                                 i++;
1261                                 last_len = 0;
1262                         }
1263                 }
1264         }
1265
1266         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1267                 filemap_flush(inode->i_mapping);
1268
1269         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1270                 /* the filemap_flush will queue IO into the worker threads, but
1271                  * we have to make sure the IO is actually started and that
1272                  * ordered extents get created before we return
1273                  */
1274                 atomic_inc(&root->fs_info->async_submit_draining);
1275                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1276                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1277                         wait_event(root->fs_info->async_submit_wait,
1278                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1279                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1280                 }
1281                 atomic_dec(&root->fs_info->async_submit_draining);
1282
1283                 mutex_lock(&inode->i_mutex);
1284                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1285                 mutex_unlock(&inode->i_mutex);
1286         }
1287
1288         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1289                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1290         }
1291
1292         ret = defrag_count;
1293
1294 out_ra:
1295         if (!file)
1296                 kfree(ra);
1297         kfree(pages);
1298         return ret;
1299 }
1300
1301 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1302                                         void __user *arg)
1303 {
1304         u64 new_size;
1305         u64 old_size;
1306         u64 devid = 1;
1307         struct btrfs_ioctl_vol_args *vol_args;
1308         struct btrfs_trans_handle *trans;
1309         struct btrfs_device *device = NULL;
1310         char *sizestr;
1311         char *devstr = NULL;
1312         int ret = 0;
1313         int mod = 0;
1314
1315         if (root->fs_info->sb->s_flags & MS_RDONLY)
1316                 return -EROFS;
1317
1318         if (!capable(CAP_SYS_ADMIN))
1319                 return -EPERM;
1320
1321         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1322                         1)) {
1323                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
1324                 return -EINPROGRESS;
1325         }
1326
1327         mutex_lock(&root->fs_info->volume_mutex);
1328         vol_args = memdup_user(arg, sizeof(*vol_args));
1329         if (IS_ERR(vol_args)) {
1330                 ret = PTR_ERR(vol_args);
1331                 goto out;
1332         }
1333
1334         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1335
1336         sizestr = vol_args->name;
1337         devstr = strchr(sizestr, ':');
1338         if (devstr) {
1339                 char *end;
1340                 sizestr = devstr + 1;
1341                 *devstr = '\0';
1342                 devstr = vol_args->name;
1343                 devid = simple_strtoull(devstr, &end, 10);
1344                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1345                        (unsigned long long)devid);
1346         }
1347         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1348         if (!device) {
1349                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1350                        (unsigned long long)devid);
1351                 ret = -EINVAL;
1352                 goto out_free;
1353         }
1354         if (device->fs_devices && device->fs_devices->seeding) {
1355                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1356                        "seeding device %llu\n",
1357                        (unsigned long long)devid);
1358                 ret = -EINVAL;
1359                 goto out_free;
1360         }
1361
1362         if (!strcmp(sizestr, "max"))
1363                 new_size = device->bdev->bd_inode->i_size;
1364         else {
1365                 if (sizestr[0] == '-') {
1366                         mod = -1;
1367                         sizestr++;
1368                 } else if (sizestr[0] == '+') {
1369                         mod = 1;
1370                         sizestr++;
1371                 }
1372                 new_size = memparse(sizestr, NULL);
1373                 if (new_size == 0) {
1374                         ret = -EINVAL;
1375                         goto out_free;
1376                 }
1377         }
1378
1379         if (device->is_tgtdev_for_dev_replace) {
1380                 ret = -EINVAL;
1381                 goto out_free;
1382         }
1383
1384         old_size = device->total_bytes;
1385
1386         if (mod < 0) {
1387                 if (new_size > old_size) {
1388                         ret = -EINVAL;
1389                         goto out_free;
1390                 }
1391                 new_size = old_size - new_size;
1392         } else if (mod > 0) {
1393                 new_size = old_size + new_size;
1394         }
1395
1396         if (new_size < 256 * 1024 * 1024) {
1397                 ret = -EINVAL;
1398                 goto out_free;
1399         }
1400         if (new_size > device->bdev->bd_inode->i_size) {
1401                 ret = -EFBIG;
1402                 goto out_free;
1403         }
1404
1405         do_div(new_size, root->sectorsize);
1406         new_size *= root->sectorsize;
1407
1408         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1409                       rcu_str_deref(device->name),
1410                       (unsigned long long)new_size);
1411
1412         if (new_size > old_size) {
1413                 trans = btrfs_start_transaction(root, 0);
1414                 if (IS_ERR(trans)) {
1415                         ret = PTR_ERR(trans);
1416                         goto out_free;
1417                 }
1418                 ret = btrfs_grow_device(trans, device, new_size);
1419                 btrfs_commit_transaction(trans, root);
1420         } else if (new_size < old_size) {
1421                 ret = btrfs_shrink_device(device, new_size);
1422         } /* equal, nothing need to do */
1423
1424 out_free:
1425         kfree(vol_args);
1426 out:
1427         mutex_unlock(&root->fs_info->volume_mutex);
1428         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1429         return ret;
1430 }
1431
1432 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1433                                 char *name, unsigned long fd, int subvol,
1434                                 u64 *transid, bool readonly,
1435                                 struct btrfs_qgroup_inherit **inherit)
1436 {
1437         int namelen;
1438         int ret = 0;
1439
1440         ret = mnt_want_write_file(file);
1441         if (ret)
1442                 goto out;
1443
1444         namelen = strlen(name);
1445         if (strchr(name, '/')) {
1446                 ret = -EINVAL;
1447                 goto out_drop_write;
1448         }
1449
1450         if (name[0] == '.' &&
1451            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1452                 ret = -EEXIST;
1453                 goto out_drop_write;
1454         }
1455
1456         if (subvol) {
1457                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1458                                      NULL, transid, readonly, inherit);
1459         } else {
1460                 struct fd src = fdget(fd);
1461                 struct inode *src_inode;
1462                 if (!src.file) {
1463                         ret = -EINVAL;
1464                         goto out_drop_write;
1465                 }
1466
1467                 src_inode = src.file->f_path.dentry->d_inode;
1468                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1469                         printk(KERN_INFO "btrfs: Snapshot src from "
1470                                "another FS\n");
1471                         ret = -EINVAL;
1472                 } else {
1473                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1474                                              BTRFS_I(src_inode)->root,
1475                                              transid, readonly, inherit);
1476                 }
1477                 fdput(src);
1478         }
1479 out_drop_write:
1480         mnt_drop_write_file(file);
1481 out:
1482         return ret;
1483 }
1484
1485 static noinline int btrfs_ioctl_snap_create(struct file *file,
1486                                             void __user *arg, int subvol)
1487 {
1488         struct btrfs_ioctl_vol_args *vol_args;
1489         int ret;
1490
1491         vol_args = memdup_user(arg, sizeof(*vol_args));
1492         if (IS_ERR(vol_args))
1493                 return PTR_ERR(vol_args);
1494         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1495
1496         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1497                                               vol_args->fd, subvol,
1498                                               NULL, false, NULL);
1499
1500         kfree(vol_args);
1501         return ret;
1502 }
1503
1504 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1505                                                void __user *arg, int subvol)
1506 {
1507         struct btrfs_ioctl_vol_args_v2 *vol_args;
1508         int ret;
1509         u64 transid = 0;
1510         u64 *ptr = NULL;
1511         bool readonly = false;
1512         struct btrfs_qgroup_inherit *inherit = NULL;
1513
1514         vol_args = memdup_user(arg, sizeof(*vol_args));
1515         if (IS_ERR(vol_args))
1516                 return PTR_ERR(vol_args);
1517         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1518
1519         if (vol_args->flags &
1520             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1521               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1522                 ret = -EOPNOTSUPP;
1523                 goto out;
1524         }
1525
1526         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1527                 ptr = &transid;
1528         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1529                 readonly = true;
1530         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1531                 if (vol_args->size > PAGE_CACHE_SIZE) {
1532                         ret = -EINVAL;
1533                         goto out;
1534                 }
1535                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1536                 if (IS_ERR(inherit)) {
1537                         ret = PTR_ERR(inherit);
1538                         goto out;
1539                 }
1540         }
1541
1542         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1543                                               vol_args->fd, subvol, ptr,
1544                                               readonly, &inherit);
1545
1546         if (ret == 0 && ptr &&
1547             copy_to_user(arg +
1548                          offsetof(struct btrfs_ioctl_vol_args_v2,
1549                                   transid), ptr, sizeof(*ptr)))
1550                 ret = -EFAULT;
1551 out:
1552         kfree(vol_args);
1553         kfree(inherit);
1554         return ret;
1555 }
1556
1557 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1558                                                 void __user *arg)
1559 {
1560         struct inode *inode = fdentry(file)->d_inode;
1561         struct btrfs_root *root = BTRFS_I(inode)->root;
1562         int ret = 0;
1563         u64 flags = 0;
1564
1565         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1566                 return -EINVAL;
1567
1568         down_read(&root->fs_info->subvol_sem);
1569         if (btrfs_root_readonly(root))
1570                 flags |= BTRFS_SUBVOL_RDONLY;
1571         up_read(&root->fs_info->subvol_sem);
1572
1573         if (copy_to_user(arg, &flags, sizeof(flags)))
1574                 ret = -EFAULT;
1575
1576         return ret;
1577 }
1578
1579 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1580                                               void __user *arg)
1581 {
1582         struct inode *inode = fdentry(file)->d_inode;
1583         struct btrfs_root *root = BTRFS_I(inode)->root;
1584         struct btrfs_trans_handle *trans;
1585         u64 root_flags;
1586         u64 flags;
1587         int ret = 0;
1588
1589         ret = mnt_want_write_file(file);
1590         if (ret)
1591                 goto out;
1592
1593         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1594                 ret = -EINVAL;
1595                 goto out_drop_write;
1596         }
1597
1598         if (copy_from_user(&flags, arg, sizeof(flags))) {
1599                 ret = -EFAULT;
1600                 goto out_drop_write;
1601         }
1602
1603         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1604                 ret = -EINVAL;
1605                 goto out_drop_write;
1606         }
1607
1608         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1609                 ret = -EOPNOTSUPP;
1610                 goto out_drop_write;
1611         }
1612
1613         if (!inode_owner_or_capable(inode)) {
1614                 ret = -EACCES;
1615                 goto out_drop_write;
1616         }
1617
1618         down_write(&root->fs_info->subvol_sem);
1619
1620         /* nothing to do */
1621         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1622                 goto out_drop_sem;
1623
1624         root_flags = btrfs_root_flags(&root->root_item);
1625         if (flags & BTRFS_SUBVOL_RDONLY)
1626                 btrfs_set_root_flags(&root->root_item,
1627                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1628         else
1629                 btrfs_set_root_flags(&root->root_item,
1630                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1631
1632         trans = btrfs_start_transaction(root, 1);
1633         if (IS_ERR(trans)) {
1634                 ret = PTR_ERR(trans);
1635                 goto out_reset;
1636         }
1637
1638         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1639                                 &root->root_key, &root->root_item);
1640
1641         btrfs_commit_transaction(trans, root);
1642 out_reset:
1643         if (ret)
1644                 btrfs_set_root_flags(&root->root_item, root_flags);
1645 out_drop_sem:
1646         up_write(&root->fs_info->subvol_sem);
1647 out_drop_write:
1648         mnt_drop_write_file(file);
1649 out:
1650         return ret;
1651 }
1652
1653 /*
1654  * helper to check if the subvolume references other subvolumes
1655  */
1656 static noinline int may_destroy_subvol(struct btrfs_root *root)
1657 {
1658         struct btrfs_path *path;
1659         struct btrfs_key key;
1660         int ret;
1661
1662         path = btrfs_alloc_path();
1663         if (!path)
1664                 return -ENOMEM;
1665
1666         key.objectid = root->root_key.objectid;
1667         key.type = BTRFS_ROOT_REF_KEY;
1668         key.offset = (u64)-1;
1669
1670         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1671                                 &key, path, 0, 0);
1672         if (ret < 0)
1673                 goto out;
1674         BUG_ON(ret == 0);
1675
1676         ret = 0;
1677         if (path->slots[0] > 0) {
1678                 path->slots[0]--;
1679                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1680                 if (key.objectid == root->root_key.objectid &&
1681                     key.type == BTRFS_ROOT_REF_KEY)
1682                         ret = -ENOTEMPTY;
1683         }
1684 out:
1685         btrfs_free_path(path);
1686         return ret;
1687 }
1688
1689 static noinline int key_in_sk(struct btrfs_key *key,
1690                               struct btrfs_ioctl_search_key *sk)
1691 {
1692         struct btrfs_key test;
1693         int ret;
1694
1695         test.objectid = sk->min_objectid;
1696         test.type = sk->min_type;
1697         test.offset = sk->min_offset;
1698
1699         ret = btrfs_comp_cpu_keys(key, &test);
1700         if (ret < 0)
1701                 return 0;
1702
1703         test.objectid = sk->max_objectid;
1704         test.type = sk->max_type;
1705         test.offset = sk->max_offset;
1706
1707         ret = btrfs_comp_cpu_keys(key, &test);
1708         if (ret > 0)
1709                 return 0;
1710         return 1;
1711 }
1712
1713 static noinline int copy_to_sk(struct btrfs_root *root,
1714                                struct btrfs_path *path,
1715                                struct btrfs_key *key,
1716                                struct btrfs_ioctl_search_key *sk,
1717                                char *buf,
1718                                unsigned long *sk_offset,
1719                                int *num_found)
1720 {
1721         u64 found_transid;
1722         struct extent_buffer *leaf;
1723         struct btrfs_ioctl_search_header sh;
1724         unsigned long item_off;
1725         unsigned long item_len;
1726         int nritems;
1727         int i;
1728         int slot;
1729         int ret = 0;
1730
1731         leaf = path->nodes[0];
1732         slot = path->slots[0];
1733         nritems = btrfs_header_nritems(leaf);
1734
1735         if (btrfs_header_generation(leaf) > sk->max_transid) {
1736                 i = nritems;
1737                 goto advance_key;
1738         }
1739         found_transid = btrfs_header_generation(leaf);
1740
1741         for (i = slot; i < nritems; i++) {
1742                 item_off = btrfs_item_ptr_offset(leaf, i);
1743                 item_len = btrfs_item_size_nr(leaf, i);
1744
1745                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1746                         item_len = 0;
1747
1748                 if (sizeof(sh) + item_len + *sk_offset >
1749                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1750                         ret = 1;
1751                         goto overflow;
1752                 }
1753
1754                 btrfs_item_key_to_cpu(leaf, key, i);
1755                 if (!key_in_sk(key, sk))
1756                         continue;
1757
1758                 sh.objectid = key->objectid;
1759                 sh.offset = key->offset;
1760                 sh.type = key->type;
1761                 sh.len = item_len;
1762                 sh.transid = found_transid;
1763
1764                 /* copy search result header */
1765                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1766                 *sk_offset += sizeof(sh);
1767
1768                 if (item_len) {
1769                         char *p = buf + *sk_offset;
1770                         /* copy the item */
1771                         read_extent_buffer(leaf, p,
1772                                            item_off, item_len);
1773                         *sk_offset += item_len;
1774                 }
1775                 (*num_found)++;
1776
1777                 if (*num_found >= sk->nr_items)
1778                         break;
1779         }
1780 advance_key:
1781         ret = 0;
1782         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1783                 key->offset++;
1784         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1785                 key->offset = 0;
1786                 key->type++;
1787         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1788                 key->offset = 0;
1789                 key->type = 0;
1790                 key->objectid++;
1791         } else
1792                 ret = 1;
1793 overflow:
1794         return ret;
1795 }
1796
1797 static noinline int search_ioctl(struct inode *inode,
1798                                  struct btrfs_ioctl_search_args *args)
1799 {
1800         struct btrfs_root *root;
1801         struct btrfs_key key;
1802         struct btrfs_key max_key;
1803         struct btrfs_path *path;
1804         struct btrfs_ioctl_search_key *sk = &args->key;
1805         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1806         int ret;
1807         int num_found = 0;
1808         unsigned long sk_offset = 0;
1809
1810         path = btrfs_alloc_path();
1811         if (!path)
1812                 return -ENOMEM;
1813
1814         if (sk->tree_id == 0) {
1815                 /* search the root of the inode that was passed */
1816                 root = BTRFS_I(inode)->root;
1817         } else {
1818                 key.objectid = sk->tree_id;
1819                 key.type = BTRFS_ROOT_ITEM_KEY;
1820                 key.offset = (u64)-1;
1821                 root = btrfs_read_fs_root_no_name(info, &key);
1822                 if (IS_ERR(root)) {
1823                         printk(KERN_ERR "could not find root %llu\n",
1824                                sk->tree_id);
1825                         btrfs_free_path(path);
1826                         return -ENOENT;
1827                 }
1828         }
1829
1830         key.objectid = sk->min_objectid;
1831         key.type = sk->min_type;
1832         key.offset = sk->min_offset;
1833
1834         max_key.objectid = sk->max_objectid;
1835         max_key.type = sk->max_type;
1836         max_key.offset = sk->max_offset;
1837
1838         path->keep_locks = 1;
1839
1840         while(1) {
1841                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1842                                            sk->min_transid);
1843                 if (ret != 0) {
1844                         if (ret > 0)
1845                                 ret = 0;
1846                         goto err;
1847                 }
1848                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1849                                  &sk_offset, &num_found);
1850                 btrfs_release_path(path);
1851                 if (ret || num_found >= sk->nr_items)
1852                         break;
1853
1854         }
1855         ret = 0;
1856 err:
1857         sk->nr_items = num_found;
1858         btrfs_free_path(path);
1859         return ret;
1860 }
1861
1862 static noinline int btrfs_ioctl_tree_search(struct file *file,
1863                                            void __user *argp)
1864 {
1865          struct btrfs_ioctl_search_args *args;
1866          struct inode *inode;
1867          int ret;
1868
1869         if (!capable(CAP_SYS_ADMIN))
1870                 return -EPERM;
1871
1872         args = memdup_user(argp, sizeof(*args));
1873         if (IS_ERR(args))
1874                 return PTR_ERR(args);
1875
1876         inode = fdentry(file)->d_inode;
1877         ret = search_ioctl(inode, args);
1878         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1879                 ret = -EFAULT;
1880         kfree(args);
1881         return ret;
1882 }
1883
1884 /*
1885  * Search INODE_REFs to identify path name of 'dirid' directory
1886  * in a 'tree_id' tree. and sets path name to 'name'.
1887  */
1888 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1889                                 u64 tree_id, u64 dirid, char *name)
1890 {
1891         struct btrfs_root *root;
1892         struct btrfs_key key;
1893         char *ptr;
1894         int ret = -1;
1895         int slot;
1896         int len;
1897         int total_len = 0;
1898         struct btrfs_inode_ref *iref;
1899         struct extent_buffer *l;
1900         struct btrfs_path *path;
1901
1902         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1903                 name[0]='\0';
1904                 return 0;
1905         }
1906
1907         path = btrfs_alloc_path();
1908         if (!path)
1909                 return -ENOMEM;
1910
1911         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1912
1913         key.objectid = tree_id;
1914         key.type = BTRFS_ROOT_ITEM_KEY;
1915         key.offset = (u64)-1;
1916         root = btrfs_read_fs_root_no_name(info, &key);
1917         if (IS_ERR(root)) {
1918                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1919                 ret = -ENOENT;
1920                 goto out;
1921         }
1922
1923         key.objectid = dirid;
1924         key.type = BTRFS_INODE_REF_KEY;
1925         key.offset = (u64)-1;
1926
1927         while(1) {
1928                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1929                 if (ret < 0)
1930                         goto out;
1931
1932                 l = path->nodes[0];
1933                 slot = path->slots[0];
1934                 if (ret > 0 && slot > 0)
1935                         slot--;
1936                 btrfs_item_key_to_cpu(l, &key, slot);
1937
1938                 if (ret > 0 && (key.objectid != dirid ||
1939                                 key.type != BTRFS_INODE_REF_KEY)) {
1940                         ret = -ENOENT;
1941                         goto out;
1942                 }
1943
1944                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1945                 len = btrfs_inode_ref_name_len(l, iref);
1946                 ptr -= len + 1;
1947                 total_len += len + 1;
1948                 if (ptr < name)
1949                         goto out;
1950
1951                 *(ptr + len) = '/';
1952                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1953
1954                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1955                         break;
1956
1957                 btrfs_release_path(path);
1958                 key.objectid = key.offset;
1959                 key.offset = (u64)-1;
1960                 dirid = key.objectid;
1961         }
1962         if (ptr < name)
1963                 goto out;
1964         memmove(name, ptr, total_len);
1965         name[total_len]='\0';
1966         ret = 0;
1967 out:
1968         btrfs_free_path(path);
1969         return ret;
1970 }
1971
1972 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1973                                            void __user *argp)
1974 {
1975          struct btrfs_ioctl_ino_lookup_args *args;
1976          struct inode *inode;
1977          int ret;
1978
1979         if (!capable(CAP_SYS_ADMIN))
1980                 return -EPERM;
1981
1982         args = memdup_user(argp, sizeof(*args));
1983         if (IS_ERR(args))
1984                 return PTR_ERR(args);
1985
1986         inode = fdentry(file)->d_inode;
1987
1988         if (args->treeid == 0)
1989                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1990
1991         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1992                                         args->treeid, args->objectid,
1993                                         args->name);
1994
1995         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1996                 ret = -EFAULT;
1997
1998         kfree(args);
1999         return ret;
2000 }
2001
2002 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2003                                              void __user *arg)
2004 {
2005         struct dentry *parent = fdentry(file);
2006         struct dentry *dentry;
2007         struct inode *dir = parent->d_inode;
2008         struct inode *inode;
2009         struct btrfs_root *root = BTRFS_I(dir)->root;
2010         struct btrfs_root *dest = NULL;
2011         struct btrfs_ioctl_vol_args *vol_args;
2012         struct btrfs_trans_handle *trans;
2013         int namelen;
2014         int ret;
2015         int err = 0;
2016
2017         vol_args = memdup_user(arg, sizeof(*vol_args));
2018         if (IS_ERR(vol_args))
2019                 return PTR_ERR(vol_args);
2020
2021         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2022         namelen = strlen(vol_args->name);
2023         if (strchr(vol_args->name, '/') ||
2024             strncmp(vol_args->name, "..", namelen) == 0) {
2025                 err = -EINVAL;
2026                 goto out;
2027         }
2028
2029         err = mnt_want_write_file(file);
2030         if (err)
2031                 goto out;
2032
2033         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
2034         dentry = lookup_one_len(vol_args->name, parent, namelen);
2035         if (IS_ERR(dentry)) {
2036                 err = PTR_ERR(dentry);
2037                 goto out_unlock_dir;
2038         }
2039
2040         if (!dentry->d_inode) {
2041                 err = -ENOENT;
2042                 goto out_dput;
2043         }
2044
2045         inode = dentry->d_inode;
2046         dest = BTRFS_I(inode)->root;
2047         if (!capable(CAP_SYS_ADMIN)){
2048                 /*
2049                  * Regular user.  Only allow this with a special mount
2050                  * option, when the user has write+exec access to the
2051                  * subvol root, and when rmdir(2) would have been
2052                  * allowed.
2053                  *
2054                  * Note that this is _not_ check that the subvol is
2055                  * empty or doesn't contain data that we wouldn't
2056                  * otherwise be able to delete.
2057                  *
2058                  * Users who want to delete empty subvols should try
2059                  * rmdir(2).
2060                  */
2061                 err = -EPERM;
2062                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2063                         goto out_dput;
2064
2065                 /*
2066                  * Do not allow deletion if the parent dir is the same
2067                  * as the dir to be deleted.  That means the ioctl
2068                  * must be called on the dentry referencing the root
2069                  * of the subvol, not a random directory contained
2070                  * within it.
2071                  */
2072                 err = -EINVAL;
2073                 if (root == dest)
2074                         goto out_dput;
2075
2076                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2077                 if (err)
2078                         goto out_dput;
2079
2080                 /* check if subvolume may be deleted by a non-root user */
2081                 err = btrfs_may_delete(dir, dentry, 1);
2082                 if (err)
2083                         goto out_dput;
2084         }
2085
2086         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2087                 err = -EINVAL;
2088                 goto out_dput;
2089         }
2090
2091         mutex_lock(&inode->i_mutex);
2092         err = d_invalidate(dentry);
2093         if (err)
2094                 goto out_unlock;
2095
2096         down_write(&root->fs_info->subvol_sem);
2097
2098         err = may_destroy_subvol(dest);
2099         if (err)
2100                 goto out_up_write;
2101
2102         trans = btrfs_start_transaction(root, 0);
2103         if (IS_ERR(trans)) {
2104                 err = PTR_ERR(trans);
2105                 goto out_up_write;
2106         }
2107         trans->block_rsv = &root->fs_info->global_block_rsv;
2108
2109         ret = btrfs_unlink_subvol(trans, root, dir,
2110                                 dest->root_key.objectid,
2111                                 dentry->d_name.name,
2112                                 dentry->d_name.len);
2113         if (ret) {
2114                 err = ret;
2115                 btrfs_abort_transaction(trans, root, ret);
2116                 goto out_end_trans;
2117         }
2118
2119         btrfs_record_root_in_trans(trans, dest);
2120
2121         memset(&dest->root_item.drop_progress, 0,
2122                 sizeof(dest->root_item.drop_progress));
2123         dest->root_item.drop_level = 0;
2124         btrfs_set_root_refs(&dest->root_item, 0);
2125
2126         if (!xchg(&dest->orphan_item_inserted, 1)) {
2127                 ret = btrfs_insert_orphan_item(trans,
2128                                         root->fs_info->tree_root,
2129                                         dest->root_key.objectid);
2130                 if (ret) {
2131                         btrfs_abort_transaction(trans, root, ret);
2132                         err = ret;
2133                         goto out_end_trans;
2134                 }
2135         }
2136 out_end_trans:
2137         ret = btrfs_end_transaction(trans, root);
2138         if (ret && !err)
2139                 err = ret;
2140         inode->i_flags |= S_DEAD;
2141 out_up_write:
2142         up_write(&root->fs_info->subvol_sem);
2143 out_unlock:
2144         mutex_unlock(&inode->i_mutex);
2145         if (!err) {
2146                 shrink_dcache_sb(root->fs_info->sb);
2147                 btrfs_invalidate_inodes(dest);
2148                 d_delete(dentry);
2149         }
2150 out_dput:
2151         dput(dentry);
2152 out_unlock_dir:
2153         mutex_unlock(&dir->i_mutex);
2154         mnt_drop_write_file(file);
2155 out:
2156         kfree(vol_args);
2157         return err;
2158 }
2159
2160 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2161 {
2162         struct inode *inode = fdentry(file)->d_inode;
2163         struct btrfs_root *root = BTRFS_I(inode)->root;
2164         struct btrfs_ioctl_defrag_range_args *range;
2165         int ret;
2166
2167         if (btrfs_root_readonly(root))
2168                 return -EROFS;
2169
2170         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2171                         1)) {
2172                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2173                 return -EINPROGRESS;
2174         }
2175         ret = mnt_want_write_file(file);
2176         if (ret) {
2177                 atomic_set(&root->fs_info->mutually_exclusive_operation_running,
2178                            0);
2179                 return ret;
2180         }
2181
2182         switch (inode->i_mode & S_IFMT) {
2183         case S_IFDIR:
2184                 if (!capable(CAP_SYS_ADMIN)) {
2185                         ret = -EPERM;
2186                         goto out;
2187                 }
2188                 ret = btrfs_defrag_root(root, 0);
2189                 if (ret)
2190                         goto out;
2191                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2192                 break;
2193         case S_IFREG:
2194                 if (!(file->f_mode & FMODE_WRITE)) {
2195                         ret = -EINVAL;
2196                         goto out;
2197                 }
2198
2199                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2200                 if (!range) {
2201                         ret = -ENOMEM;
2202                         goto out;
2203                 }
2204
2205                 if (argp) {
2206                         if (copy_from_user(range, argp,
2207                                            sizeof(*range))) {
2208                                 ret = -EFAULT;
2209                                 kfree(range);
2210                                 goto out;
2211                         }
2212                         /* compression requires us to start the IO */
2213                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2214                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2215                                 range->extent_thresh = (u32)-1;
2216                         }
2217                 } else {
2218                         /* the rest are all set to zero by kzalloc */
2219                         range->len = (u64)-1;
2220                 }
2221                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2222                                         range, 0, 0);
2223                 if (ret > 0)
2224                         ret = 0;
2225                 kfree(range);
2226                 break;
2227         default:
2228                 ret = -EINVAL;
2229         }
2230 out:
2231         mnt_drop_write_file(file);
2232         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2233         return ret;
2234 }
2235
2236 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2237 {
2238         struct btrfs_ioctl_vol_args *vol_args;
2239         int ret;
2240
2241         if (!capable(CAP_SYS_ADMIN))
2242                 return -EPERM;
2243
2244         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2245                         1)) {
2246                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2247                 return -EINPROGRESS;
2248         }
2249
2250         mutex_lock(&root->fs_info->volume_mutex);
2251         vol_args = memdup_user(arg, sizeof(*vol_args));
2252         if (IS_ERR(vol_args)) {
2253                 ret = PTR_ERR(vol_args);
2254                 goto out;
2255         }
2256
2257         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2258         ret = btrfs_init_new_device(root, vol_args->name);
2259
2260         kfree(vol_args);
2261 out:
2262         mutex_unlock(&root->fs_info->volume_mutex);
2263         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2264         return ret;
2265 }
2266
2267 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2268 {
2269         struct btrfs_ioctl_vol_args *vol_args;
2270         int ret;
2271
2272         if (!capable(CAP_SYS_ADMIN))
2273                 return -EPERM;
2274
2275         if (root->fs_info->sb->s_flags & MS_RDONLY)
2276                 return -EROFS;
2277
2278         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2279                         1)) {
2280                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2281                 return -EINPROGRESS;
2282         }
2283
2284         mutex_lock(&root->fs_info->volume_mutex);
2285         vol_args = memdup_user(arg, sizeof(*vol_args));
2286         if (IS_ERR(vol_args)) {
2287                 ret = PTR_ERR(vol_args);
2288                 goto out;
2289         }
2290
2291         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2292         ret = btrfs_rm_device(root, vol_args->name);
2293
2294         kfree(vol_args);
2295 out:
2296         mutex_unlock(&root->fs_info->volume_mutex);
2297         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2298         return ret;
2299 }
2300
2301 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2302 {
2303         struct btrfs_ioctl_fs_info_args *fi_args;
2304         struct btrfs_device *device;
2305         struct btrfs_device *next;
2306         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2307         int ret = 0;
2308
2309         if (!capable(CAP_SYS_ADMIN))
2310                 return -EPERM;
2311
2312         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2313         if (!fi_args)
2314                 return -ENOMEM;
2315
2316         fi_args->num_devices = fs_devices->num_devices;
2317         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2318
2319         mutex_lock(&fs_devices->device_list_mutex);
2320         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2321                 if (device->devid > fi_args->max_id)
2322                         fi_args->max_id = device->devid;
2323         }
2324         mutex_unlock(&fs_devices->device_list_mutex);
2325
2326         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2327                 ret = -EFAULT;
2328
2329         kfree(fi_args);
2330         return ret;
2331 }
2332
2333 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2334 {
2335         struct btrfs_ioctl_dev_info_args *di_args;
2336         struct btrfs_device *dev;
2337         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2338         int ret = 0;
2339         char *s_uuid = NULL;
2340         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2341
2342         if (!capable(CAP_SYS_ADMIN))
2343                 return -EPERM;
2344
2345         di_args = memdup_user(arg, sizeof(*di_args));
2346         if (IS_ERR(di_args))
2347                 return PTR_ERR(di_args);
2348
2349         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2350                 s_uuid = di_args->uuid;
2351
2352         mutex_lock(&fs_devices->device_list_mutex);
2353         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2354         mutex_unlock(&fs_devices->device_list_mutex);
2355
2356         if (!dev) {
2357                 ret = -ENODEV;
2358                 goto out;
2359         }
2360
2361         di_args->devid = dev->devid;
2362         di_args->bytes_used = dev->bytes_used;
2363         di_args->total_bytes = dev->total_bytes;
2364         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2365         if (dev->name) {
2366                 struct rcu_string *name;
2367
2368                 rcu_read_lock();
2369                 name = rcu_dereference(dev->name);
2370                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2371                 rcu_read_unlock();
2372                 di_args->path[sizeof(di_args->path) - 1] = 0;
2373         } else {
2374                 di_args->path[0] = '\0';
2375         }
2376
2377 out:
2378         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2379                 ret = -EFAULT;
2380
2381         kfree(di_args);
2382         return ret;
2383 }
2384
2385 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2386                                        u64 off, u64 olen, u64 destoff)
2387 {
2388         struct inode *inode = fdentry(file)->d_inode;
2389         struct btrfs_root *root = BTRFS_I(inode)->root;
2390         struct fd src_file;
2391         struct inode *src;
2392         struct btrfs_trans_handle *trans;
2393         struct btrfs_path *path;
2394         struct extent_buffer *leaf;
2395         char *buf;
2396         struct btrfs_key key;
2397         u32 nritems;
2398         int slot;
2399         int ret;
2400         u64 len = olen;
2401         u64 bs = root->fs_info->sb->s_blocksize;
2402
2403         /*
2404          * TODO:
2405          * - split compressed inline extents.  annoying: we need to
2406          *   decompress into destination's address_space (the file offset
2407          *   may change, so source mapping won't do), then recompress (or
2408          *   otherwise reinsert) a subrange.
2409          * - allow ranges within the same file to be cloned (provided
2410          *   they don't overlap)?
2411          */
2412
2413         /* the destination must be opened for writing */
2414         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2415                 return -EINVAL;
2416
2417         if (btrfs_root_readonly(root))
2418                 return -EROFS;
2419
2420         ret = mnt_want_write_file(file);
2421         if (ret)
2422                 return ret;
2423
2424         src_file = fdget(srcfd);
2425         if (!src_file.file) {
2426                 ret = -EBADF;
2427                 goto out_drop_write;
2428         }
2429
2430         ret = -EXDEV;
2431         if (src_file.file->f_path.mnt != file->f_path.mnt)
2432                 goto out_fput;
2433
2434         src = src_file.file->f_dentry->d_inode;
2435
2436         ret = -EINVAL;
2437         if (src == inode)
2438                 goto out_fput;
2439
2440         /* the src must be open for reading */
2441         if (!(src_file.file->f_mode & FMODE_READ))
2442                 goto out_fput;
2443
2444         /* don't make the dst file partly checksummed */
2445         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2446             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2447                 goto out_fput;
2448
2449         ret = -EISDIR;
2450         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2451                 goto out_fput;
2452
2453         ret = -EXDEV;
2454         if (src->i_sb != inode->i_sb)
2455                 goto out_fput;
2456
2457         ret = -ENOMEM;
2458         buf = vmalloc(btrfs_level_size(root, 0));
2459         if (!buf)
2460                 goto out_fput;
2461
2462         path = btrfs_alloc_path();
2463         if (!path) {
2464                 vfree(buf);
2465                 goto out_fput;
2466         }
2467         path->reada = 2;
2468
2469         if (inode < src) {
2470                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2471                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2472         } else {
2473                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2474                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2475         }
2476
2477         /* determine range to clone */
2478         ret = -EINVAL;
2479         if (off + len > src->i_size || off + len < off)
2480                 goto out_unlock;
2481         if (len == 0)
2482                 olen = len = src->i_size - off;
2483         /* if we extend to eof, continue to block boundary */
2484         if (off + len == src->i_size)
2485                 len = ALIGN(src->i_size, bs) - off;
2486
2487         /* verify the end result is block aligned */
2488         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2489             !IS_ALIGNED(destoff, bs))
2490                 goto out_unlock;
2491
2492         if (destoff > inode->i_size) {
2493                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2494                 if (ret)
2495                         goto out_unlock;
2496         }
2497
2498         /* truncate page cache pages from target inode range */
2499         truncate_inode_pages_range(&inode->i_data, destoff,
2500                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2501
2502         /* do any pending delalloc/csum calc on src, one way or
2503            another, and lock file content */
2504         while (1) {
2505                 struct btrfs_ordered_extent *ordered;
2506                 lock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2507                 ordered = btrfs_lookup_first_ordered_extent(src, off + len - 1);
2508                 if (!ordered &&
2509                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off + len - 1,
2510                                     EXTENT_DELALLOC, 0, NULL))
2511                         break;
2512                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2513                 if (ordered)
2514                         btrfs_put_ordered_extent(ordered);
2515                 btrfs_wait_ordered_range(src, off, len);
2516         }
2517
2518         /* clone data */
2519         key.objectid = btrfs_ino(src);
2520         key.type = BTRFS_EXTENT_DATA_KEY;
2521         key.offset = 0;
2522
2523         while (1) {
2524                 /*
2525                  * note the key will change type as we walk through the
2526                  * tree.
2527                  */
2528                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2529                                 0, 0);
2530                 if (ret < 0)
2531                         goto out;
2532
2533                 nritems = btrfs_header_nritems(path->nodes[0]);
2534                 if (path->slots[0] >= nritems) {
2535                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2536                         if (ret < 0)
2537                                 goto out;
2538                         if (ret > 0)
2539                                 break;
2540                         nritems = btrfs_header_nritems(path->nodes[0]);
2541                 }
2542                 leaf = path->nodes[0];
2543                 slot = path->slots[0];
2544
2545                 btrfs_item_key_to_cpu(leaf, &key, slot);
2546                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2547                     key.objectid != btrfs_ino(src))
2548                         break;
2549
2550                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2551                         struct btrfs_file_extent_item *extent;
2552                         int type;
2553                         u32 size;
2554                         struct btrfs_key new_key;
2555                         u64 disko = 0, diskl = 0;
2556                         u64 datao = 0, datal = 0;
2557                         u8 comp;
2558                         u64 endoff;
2559
2560                         size = btrfs_item_size_nr(leaf, slot);
2561                         read_extent_buffer(leaf, buf,
2562                                            btrfs_item_ptr_offset(leaf, slot),
2563                                            size);
2564
2565                         extent = btrfs_item_ptr(leaf, slot,
2566                                                 struct btrfs_file_extent_item);
2567                         comp = btrfs_file_extent_compression(leaf, extent);
2568                         type = btrfs_file_extent_type(leaf, extent);
2569                         if (type == BTRFS_FILE_EXTENT_REG ||
2570                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2571                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2572                                                                       extent);
2573                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2574                                                                  extent);
2575                                 datao = btrfs_file_extent_offset(leaf, extent);
2576                                 datal = btrfs_file_extent_num_bytes(leaf,
2577                                                                     extent);
2578                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2579                                 /* take upper bound, may be compressed */
2580                                 datal = btrfs_file_extent_ram_bytes(leaf,
2581                                                                     extent);
2582                         }
2583                         btrfs_release_path(path);
2584
2585                         if (key.offset + datal <= off ||
2586                             key.offset >= off + len - 1)
2587                                 goto next;
2588
2589                         memcpy(&new_key, &key, sizeof(new_key));
2590                         new_key.objectid = btrfs_ino(inode);
2591                         if (off <= key.offset)
2592                                 new_key.offset = key.offset + destoff - off;
2593                         else
2594                                 new_key.offset = destoff;
2595
2596                         /*
2597                          * 1 - adjusting old extent (we may have to split it)
2598                          * 1 - add new extent
2599                          * 1 - inode update
2600                          */
2601                         trans = btrfs_start_transaction(root, 3);
2602                         if (IS_ERR(trans)) {
2603                                 ret = PTR_ERR(trans);
2604                                 goto out;
2605                         }
2606
2607                         if (type == BTRFS_FILE_EXTENT_REG ||
2608                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2609                                 /*
2610                                  *    a  | --- range to clone ---|  b
2611                                  * | ------------- extent ------------- |
2612                                  */
2613
2614                                 /* substract range b */
2615                                 if (key.offset + datal > off + len)
2616                                         datal = off + len - key.offset;
2617
2618                                 /* substract range a */
2619                                 if (off > key.offset) {
2620                                         datao += off - key.offset;
2621                                         datal -= off - key.offset;
2622                                 }
2623
2624                                 ret = btrfs_drop_extents(trans, root, inode,
2625                                                          new_key.offset,
2626                                                          new_key.offset + datal,
2627                                                          1);
2628                                 if (ret) {
2629                                         btrfs_abort_transaction(trans, root,
2630                                                                 ret);
2631                                         btrfs_end_transaction(trans, root);
2632                                         goto out;
2633                                 }
2634
2635                                 ret = btrfs_insert_empty_item(trans, root, path,
2636                                                               &new_key, size);
2637                                 if (ret) {
2638                                         btrfs_abort_transaction(trans, root,
2639                                                                 ret);
2640                                         btrfs_end_transaction(trans, root);
2641                                         goto out;
2642                                 }
2643
2644                                 leaf = path->nodes[0];
2645                                 slot = path->slots[0];
2646                                 write_extent_buffer(leaf, buf,
2647                                             btrfs_item_ptr_offset(leaf, slot),
2648                                             size);
2649
2650                                 extent = btrfs_item_ptr(leaf, slot,
2651                                                 struct btrfs_file_extent_item);
2652
2653                                 /* disko == 0 means it's a hole */
2654                                 if (!disko)
2655                                         datao = 0;
2656
2657                                 btrfs_set_file_extent_offset(leaf, extent,
2658                                                              datao);
2659                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2660                                                                 datal);
2661                                 if (disko) {
2662                                         inode_add_bytes(inode, datal);
2663                                         ret = btrfs_inc_extent_ref(trans, root,
2664                                                         disko, diskl, 0,
2665                                                         root->root_key.objectid,
2666                                                         btrfs_ino(inode),
2667                                                         new_key.offset - datao,
2668                                                         0);
2669                                         if (ret) {
2670                                                 btrfs_abort_transaction(trans,
2671                                                                         root,
2672                                                                         ret);
2673                                                 btrfs_end_transaction(trans,
2674                                                                       root);
2675                                                 goto out;
2676
2677                                         }
2678                                 }
2679                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2680                                 u64 skip = 0;
2681                                 u64 trim = 0;
2682                                 if (off > key.offset) {
2683                                         skip = off - key.offset;
2684                                         new_key.offset += skip;
2685                                 }
2686
2687                                 if (key.offset + datal > off + len)
2688                                         trim = key.offset + datal - (off + len);
2689
2690                                 if (comp && (skip || trim)) {
2691                                         ret = -EINVAL;
2692                                         btrfs_end_transaction(trans, root);
2693                                         goto out;
2694                                 }
2695                                 size -= skip + trim;
2696                                 datal -= skip + trim;
2697
2698                                 ret = btrfs_drop_extents(trans, root, inode,
2699                                                          new_key.offset,
2700                                                          new_key.offset + datal,
2701                                                          1);
2702                                 if (ret) {
2703                                         btrfs_abort_transaction(trans, root,
2704                                                                 ret);
2705                                         btrfs_end_transaction(trans, root);
2706                                         goto out;
2707                                 }
2708
2709                                 ret = btrfs_insert_empty_item(trans, root, path,
2710                                                               &new_key, size);
2711                                 if (ret) {
2712                                         btrfs_abort_transaction(trans, root,
2713                                                                 ret);
2714                                         btrfs_end_transaction(trans, root);
2715                                         goto out;
2716                                 }
2717
2718                                 if (skip) {
2719                                         u32 start =
2720                                           btrfs_file_extent_calc_inline_size(0);
2721                                         memmove(buf+start, buf+start+skip,
2722                                                 datal);
2723                                 }
2724
2725                                 leaf = path->nodes[0];
2726                                 slot = path->slots[0];
2727                                 write_extent_buffer(leaf, buf,
2728                                             btrfs_item_ptr_offset(leaf, slot),
2729                                             size);
2730                                 inode_add_bytes(inode, datal);
2731                         }
2732
2733                         btrfs_mark_buffer_dirty(leaf);
2734                         btrfs_release_path(path);
2735
2736                         inode_inc_iversion(inode);
2737                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2738
2739                         /*
2740                          * we round up to the block size at eof when
2741                          * determining which extents to clone above,
2742                          * but shouldn't round up the file size
2743                          */
2744                         endoff = new_key.offset + datal;
2745                         if (endoff > destoff+olen)
2746                                 endoff = destoff+olen;
2747                         if (endoff > inode->i_size)
2748                                 btrfs_i_size_write(inode, endoff);
2749
2750                         ret = btrfs_update_inode(trans, root, inode);
2751                         if (ret) {
2752                                 btrfs_abort_transaction(trans, root, ret);
2753                                 btrfs_end_transaction(trans, root);
2754                                 goto out;
2755                         }
2756                         ret = btrfs_end_transaction(trans, root);
2757                 }
2758 next:
2759                 btrfs_release_path(path);
2760                 key.offset++;
2761         }
2762         ret = 0;
2763 out:
2764         btrfs_release_path(path);
2765         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2766 out_unlock:
2767         mutex_unlock(&src->i_mutex);
2768         mutex_unlock(&inode->i_mutex);
2769         vfree(buf);
2770         btrfs_free_path(path);
2771 out_fput:
2772         fdput(src_file);
2773 out_drop_write:
2774         mnt_drop_write_file(file);
2775         return ret;
2776 }
2777
2778 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2779 {
2780         struct btrfs_ioctl_clone_range_args args;
2781
2782         if (copy_from_user(&args, argp, sizeof(args)))
2783                 return -EFAULT;
2784         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2785                                  args.src_length, args.dest_offset);
2786 }
2787
2788 /*
2789  * there are many ways the trans_start and trans_end ioctls can lead
2790  * to deadlocks.  They should only be used by applications that
2791  * basically own the machine, and have a very in depth understanding
2792  * of all the possible deadlocks and enospc problems.
2793  */
2794 static long btrfs_ioctl_trans_start(struct file *file)
2795 {
2796         struct inode *inode = fdentry(file)->d_inode;
2797         struct btrfs_root *root = BTRFS_I(inode)->root;
2798         struct btrfs_trans_handle *trans;
2799         int ret;
2800
2801         ret = -EPERM;
2802         if (!capable(CAP_SYS_ADMIN))
2803                 goto out;
2804
2805         ret = -EINPROGRESS;
2806         if (file->private_data)
2807                 goto out;
2808
2809         ret = -EROFS;
2810         if (btrfs_root_readonly(root))
2811                 goto out;
2812
2813         ret = mnt_want_write_file(file);
2814         if (ret)
2815                 goto out;
2816
2817         atomic_inc(&root->fs_info->open_ioctl_trans);
2818
2819         ret = -ENOMEM;
2820         trans = btrfs_start_ioctl_transaction(root);
2821         if (IS_ERR(trans))
2822                 goto out_drop;
2823
2824         file->private_data = trans;
2825         return 0;
2826
2827 out_drop:
2828         atomic_dec(&root->fs_info->open_ioctl_trans);
2829         mnt_drop_write_file(file);
2830 out:
2831         return ret;
2832 }
2833
2834 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2835 {
2836         struct inode *inode = fdentry(file)->d_inode;
2837         struct btrfs_root *root = BTRFS_I(inode)->root;
2838         struct btrfs_root *new_root;
2839         struct btrfs_dir_item *di;
2840         struct btrfs_trans_handle *trans;
2841         struct btrfs_path *path;
2842         struct btrfs_key location;
2843         struct btrfs_disk_key disk_key;
2844         u64 objectid = 0;
2845         u64 dir_id;
2846         int ret;
2847
2848         if (!capable(CAP_SYS_ADMIN))
2849                 return -EPERM;
2850
2851         ret = mnt_want_write_file(file);
2852         if (ret)
2853                 return ret;
2854
2855         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
2856                 ret = -EFAULT;
2857                 goto out;
2858         }
2859
2860         if (!objectid)
2861                 objectid = root->root_key.objectid;
2862
2863         location.objectid = objectid;
2864         location.type = BTRFS_ROOT_ITEM_KEY;
2865         location.offset = (u64)-1;
2866
2867         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2868         if (IS_ERR(new_root)) {
2869                 ret = PTR_ERR(new_root);
2870                 goto out;
2871         }
2872
2873         if (btrfs_root_refs(&new_root->root_item) == 0) {
2874                 ret = -ENOENT;
2875                 goto out;
2876         }
2877
2878         path = btrfs_alloc_path();
2879         if (!path) {
2880                 ret = -ENOMEM;
2881                 goto out;
2882         }
2883         path->leave_spinning = 1;
2884
2885         trans = btrfs_start_transaction(root, 1);
2886         if (IS_ERR(trans)) {
2887                 btrfs_free_path(path);
2888                 ret = PTR_ERR(trans);
2889                 goto out;
2890         }
2891
2892         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2893         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2894                                    dir_id, "default", 7, 1);
2895         if (IS_ERR_OR_NULL(di)) {
2896                 btrfs_free_path(path);
2897                 btrfs_end_transaction(trans, root);
2898                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2899                        "this isn't going to work\n");
2900                 ret = -ENOENT;
2901                 goto out;
2902         }
2903
2904         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2905         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2906         btrfs_mark_buffer_dirty(path->nodes[0]);
2907         btrfs_free_path(path);
2908
2909         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
2910         btrfs_end_transaction(trans, root);
2911 out:
2912         mnt_drop_write_file(file);
2913         return ret;
2914 }
2915
2916 void btrfs_get_block_group_info(struct list_head *groups_list,
2917                                 struct btrfs_ioctl_space_info *space)
2918 {
2919         struct btrfs_block_group_cache *block_group;
2920
2921         space->total_bytes = 0;
2922         space->used_bytes = 0;
2923         space->flags = 0;
2924         list_for_each_entry(block_group, groups_list, list) {
2925                 space->flags = block_group->flags;
2926                 space->total_bytes += block_group->key.offset;
2927                 space->used_bytes +=
2928                         btrfs_block_group_used(&block_group->item);
2929         }
2930 }
2931
2932 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2933 {
2934         struct btrfs_ioctl_space_args space_args;
2935         struct btrfs_ioctl_space_info space;
2936         struct btrfs_ioctl_space_info *dest;
2937         struct btrfs_ioctl_space_info *dest_orig;
2938         struct btrfs_ioctl_space_info __user *user_dest;
2939         struct btrfs_space_info *info;
2940         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2941                        BTRFS_BLOCK_GROUP_SYSTEM,
2942                        BTRFS_BLOCK_GROUP_METADATA,
2943                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2944         int num_types = 4;
2945         int alloc_size;
2946         int ret = 0;
2947         u64 slot_count = 0;
2948         int i, c;
2949
2950         if (copy_from_user(&space_args,
2951                            (struct btrfs_ioctl_space_args __user *)arg,
2952                            sizeof(space_args)))
2953                 return -EFAULT;
2954
2955         for (i = 0; i < num_types; i++) {
2956                 struct btrfs_space_info *tmp;
2957
2958                 info = NULL;
2959                 rcu_read_lock();
2960                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2961                                         list) {
2962                         if (tmp->flags == types[i]) {
2963                                 info = tmp;
2964                                 break;
2965                         }
2966                 }
2967                 rcu_read_unlock();
2968
2969                 if (!info)
2970                         continue;
2971
2972                 down_read(&info->groups_sem);
2973                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2974                         if (!list_empty(&info->block_groups[c]))
2975                                 slot_count++;
2976                 }
2977                 up_read(&info->groups_sem);
2978         }
2979
2980         /* space_slots == 0 means they are asking for a count */
2981         if (space_args.space_slots == 0) {
2982                 space_args.total_spaces = slot_count;
2983                 goto out;
2984         }
2985
2986         slot_count = min_t(u64, space_args.space_slots, slot_count);
2987
2988         alloc_size = sizeof(*dest) * slot_count;
2989
2990         /* we generally have at most 6 or so space infos, one for each raid
2991          * level.  So, a whole page should be more than enough for everyone
2992          */
2993         if (alloc_size > PAGE_CACHE_SIZE)
2994                 return -ENOMEM;
2995
2996         space_args.total_spaces = 0;
2997         dest = kmalloc(alloc_size, GFP_NOFS);
2998         if (!dest)
2999                 return -ENOMEM;
3000         dest_orig = dest;
3001
3002         /* now we have a buffer to copy into */
3003         for (i = 0; i < num_types; i++) {
3004                 struct btrfs_space_info *tmp;
3005
3006                 if (!slot_count)
3007                         break;
3008
3009                 info = NULL;
3010                 rcu_read_lock();
3011                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3012                                         list) {
3013                         if (tmp->flags == types[i]) {
3014                                 info = tmp;
3015                                 break;
3016                         }
3017                 }
3018                 rcu_read_unlock();
3019
3020                 if (!info)
3021                         continue;
3022                 down_read(&info->groups_sem);
3023                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3024                         if (!list_empty(&info->block_groups[c])) {
3025                                 btrfs_get_block_group_info(
3026                                         &info->block_groups[c], &space);
3027                                 memcpy(dest, &space, sizeof(space));
3028                                 dest++;
3029                                 space_args.total_spaces++;
3030                                 slot_count--;
3031                         }
3032                         if (!slot_count)
3033                                 break;
3034                 }
3035                 up_read(&info->groups_sem);
3036         }
3037
3038         user_dest = (struct btrfs_ioctl_space_info __user *)
3039                 (arg + sizeof(struct btrfs_ioctl_space_args));
3040
3041         if (copy_to_user(user_dest, dest_orig, alloc_size))
3042                 ret = -EFAULT;
3043
3044         kfree(dest_orig);
3045 out:
3046         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3047                 ret = -EFAULT;
3048
3049         return ret;
3050 }
3051
3052 /*
3053  * there are many ways the trans_start and trans_end ioctls can lead
3054  * to deadlocks.  They should only be used by applications that
3055  * basically own the machine, and have a very in depth understanding
3056  * of all the possible deadlocks and enospc problems.
3057  */
3058 long btrfs_ioctl_trans_end(struct file *file)
3059 {
3060         struct inode *inode = fdentry(file)->d_inode;
3061         struct btrfs_root *root = BTRFS_I(inode)->root;
3062         struct btrfs_trans_handle *trans;
3063
3064         trans = file->private_data;
3065         if (!trans)
3066                 return -EINVAL;
3067         file->private_data = NULL;
3068
3069         btrfs_end_transaction(trans, root);
3070
3071         atomic_dec(&root->fs_info->open_ioctl_trans);
3072
3073         mnt_drop_write_file(file);
3074         return 0;
3075 }
3076
3077 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3078                                             void __user *argp)
3079 {
3080         struct btrfs_trans_handle *trans;
3081         u64 transid;
3082         int ret;
3083
3084         trans = btrfs_attach_transaction(root);
3085         if (IS_ERR(trans)) {
3086                 if (PTR_ERR(trans) != -ENOENT)
3087                         return PTR_ERR(trans);
3088
3089                 /* No running transaction, don't bother */
3090                 transid = root->fs_info->last_trans_committed;
3091                 goto out;
3092         }
3093         transid = trans->transid;
3094         ret = btrfs_commit_transaction_async(trans, root, 0);
3095         if (ret) {
3096                 btrfs_end_transaction(trans, root);
3097                 return ret;
3098         }
3099 out:
3100         if (argp)
3101                 if (copy_to_user(argp, &transid, sizeof(transid)))
3102                         return -EFAULT;
3103         return 0;
3104 }
3105
3106 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3107                                            void __user *argp)
3108 {
3109         u64 transid;
3110
3111         if (argp) {
3112                 if (copy_from_user(&transid, argp, sizeof(transid)))
3113                         return -EFAULT;
3114         } else {
3115                 transid = 0;  /* current trans */
3116         }
3117         return btrfs_wait_for_commit(root, transid);
3118 }
3119
3120 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
3121 {
3122         int ret;
3123         struct btrfs_ioctl_scrub_args *sa;
3124
3125         if (!capable(CAP_SYS_ADMIN))
3126                 return -EPERM;
3127
3128         sa = memdup_user(arg, sizeof(*sa));
3129         if (IS_ERR(sa))
3130                 return PTR_ERR(sa);
3131
3132         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3133                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3134                               0);
3135
3136         if (copy_to_user(arg, sa, sizeof(*sa)))
3137                 ret = -EFAULT;
3138
3139         kfree(sa);
3140         return ret;
3141 }
3142
3143 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3144 {
3145         if (!capable(CAP_SYS_ADMIN))
3146                 return -EPERM;
3147
3148         return btrfs_scrub_cancel(root->fs_info);
3149 }
3150
3151 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3152                                        void __user *arg)
3153 {
3154         struct btrfs_ioctl_scrub_args *sa;
3155         int ret;
3156
3157         if (!capable(CAP_SYS_ADMIN))
3158                 return -EPERM;
3159
3160         sa = memdup_user(arg, sizeof(*sa));
3161         if (IS_ERR(sa))
3162                 return PTR_ERR(sa);
3163
3164         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3165
3166         if (copy_to_user(arg, sa, sizeof(*sa)))
3167                 ret = -EFAULT;
3168
3169         kfree(sa);
3170         return ret;
3171 }
3172
3173 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3174                                       void __user *arg)
3175 {
3176         struct btrfs_ioctl_get_dev_stats *sa;
3177         int ret;
3178
3179         sa = memdup_user(arg, sizeof(*sa));
3180         if (IS_ERR(sa))
3181                 return PTR_ERR(sa);
3182
3183         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3184                 kfree(sa);
3185                 return -EPERM;
3186         }
3187
3188         ret = btrfs_get_dev_stats(root, sa);
3189
3190         if (copy_to_user(arg, sa, sizeof(*sa)))
3191                 ret = -EFAULT;
3192
3193         kfree(sa);
3194         return ret;
3195 }
3196
3197 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3198 {
3199         struct btrfs_ioctl_dev_replace_args *p;
3200         int ret;
3201
3202         if (!capable(CAP_SYS_ADMIN))
3203                 return -EPERM;
3204
3205         p = memdup_user(arg, sizeof(*p));
3206         if (IS_ERR(p))
3207                 return PTR_ERR(p);
3208
3209         switch (p->cmd) {
3210         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3211                 if (atomic_xchg(
3212                         &root->fs_info->mutually_exclusive_operation_running,
3213                         1)) {
3214                         pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3215                         ret = -EINPROGRESS;
3216                 } else {
3217                         ret = btrfs_dev_replace_start(root, p);
3218                         atomic_set(
3219                          &root->fs_info->mutually_exclusive_operation_running,
3220                          0);
3221                 }
3222                 break;
3223         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3224                 btrfs_dev_replace_status(root->fs_info, p);
3225                 ret = 0;
3226                 break;
3227         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3228                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3229                 break;
3230         default:
3231                 ret = -EINVAL;
3232                 break;
3233         }
3234
3235         if (copy_to_user(arg, p, sizeof(*p)))
3236                 ret = -EFAULT;
3237
3238         kfree(p);
3239         return ret;
3240 }
3241
3242 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3243 {
3244         int ret = 0;
3245         int i;
3246         u64 rel_ptr;
3247         int size;
3248         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3249         struct inode_fs_paths *ipath = NULL;
3250         struct btrfs_path *path;
3251
3252         if (!capable(CAP_SYS_ADMIN))
3253                 return -EPERM;
3254
3255         path = btrfs_alloc_path();
3256         if (!path) {
3257                 ret = -ENOMEM;
3258                 goto out;
3259         }
3260
3261         ipa = memdup_user(arg, sizeof(*ipa));
3262         if (IS_ERR(ipa)) {
3263                 ret = PTR_ERR(ipa);
3264                 ipa = NULL;
3265                 goto out;
3266         }
3267
3268         size = min_t(u32, ipa->size, 4096);
3269         ipath = init_ipath(size, root, path);
3270         if (IS_ERR(ipath)) {
3271                 ret = PTR_ERR(ipath);
3272                 ipath = NULL;
3273                 goto out;
3274         }
3275
3276         ret = paths_from_inode(ipa->inum, ipath);
3277         if (ret < 0)
3278                 goto out;
3279
3280         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3281                 rel_ptr = ipath->fspath->val[i] -
3282                           (u64)(unsigned long)ipath->fspath->val;
3283                 ipath->fspath->val[i] = rel_ptr;
3284         }
3285
3286         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3287                            (void *)(unsigned long)ipath->fspath, size);
3288         if (ret) {
3289                 ret = -EFAULT;
3290                 goto out;
3291         }
3292
3293 out:
3294         btrfs_free_path(path);
3295         free_ipath(ipath);
3296         kfree(ipa);
3297
3298         return ret;
3299 }
3300
3301 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3302 {
3303         struct btrfs_data_container *inodes = ctx;
3304         const size_t c = 3 * sizeof(u64);
3305
3306         if (inodes->bytes_left >= c) {
3307                 inodes->bytes_left -= c;
3308                 inodes->val[inodes->elem_cnt] = inum;
3309                 inodes->val[inodes->elem_cnt + 1] = offset;
3310                 inodes->val[inodes->elem_cnt + 2] = root;
3311                 inodes->elem_cnt += 3;
3312         } else {
3313                 inodes->bytes_missing += c - inodes->bytes_left;
3314                 inodes->bytes_left = 0;
3315                 inodes->elem_missed += 3;
3316         }
3317
3318         return 0;
3319 }
3320
3321 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3322                                         void __user *arg)
3323 {
3324         int ret = 0;
3325         int size;
3326         struct btrfs_ioctl_logical_ino_args *loi;
3327         struct btrfs_data_container *inodes = NULL;
3328         struct btrfs_path *path = NULL;
3329
3330         if (!capable(CAP_SYS_ADMIN))
3331                 return -EPERM;
3332
3333         loi = memdup_user(arg, sizeof(*loi));
3334         if (IS_ERR(loi)) {
3335                 ret = PTR_ERR(loi);
3336                 loi = NULL;
3337                 goto out;
3338         }
3339
3340         path = btrfs_alloc_path();
3341         if (!path) {
3342                 ret = -ENOMEM;
3343                 goto out;
3344         }
3345
3346         size = min_t(u32, loi->size, 64 * 1024);
3347         inodes = init_data_container(size);
3348         if (IS_ERR(inodes)) {
3349                 ret = PTR_ERR(inodes);
3350                 inodes = NULL;
3351                 goto out;
3352         }
3353
3354         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3355                                           build_ino_list, inodes);
3356         if (ret == -EINVAL)
3357                 ret = -ENOENT;
3358         if (ret < 0)
3359                 goto out;
3360
3361         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3362                            (void *)(unsigned long)inodes, size);
3363         if (ret)
3364                 ret = -EFAULT;
3365
3366 out:
3367         btrfs_free_path(path);
3368         vfree(inodes);
3369         kfree(loi);
3370
3371         return ret;
3372 }
3373
3374 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3375                                struct btrfs_ioctl_balance_args *bargs)
3376 {
3377         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3378
3379         bargs->flags = bctl->flags;
3380
3381         if (atomic_read(&fs_info->balance_running))
3382                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3383         if (atomic_read(&fs_info->balance_pause_req))
3384                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3385         if (atomic_read(&fs_info->balance_cancel_req))
3386                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3387
3388         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3389         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3390         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3391
3392         if (lock) {
3393                 spin_lock(&fs_info->balance_lock);
3394                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3395                 spin_unlock(&fs_info->balance_lock);
3396         } else {
3397                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3398         }
3399 }
3400
3401 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3402 {
3403         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3404         struct btrfs_fs_info *fs_info = root->fs_info;
3405         struct btrfs_ioctl_balance_args *bargs;
3406         struct btrfs_balance_control *bctl;
3407         int ret;
3408         int need_to_clear_lock = 0;
3409
3410         if (!capable(CAP_SYS_ADMIN))
3411                 return -EPERM;
3412
3413         ret = mnt_want_write_file(file);
3414         if (ret)
3415                 return ret;
3416
3417         mutex_lock(&fs_info->volume_mutex);
3418         mutex_lock(&fs_info->balance_mutex);
3419
3420         if (arg) {
3421                 bargs = memdup_user(arg, sizeof(*bargs));
3422                 if (IS_ERR(bargs)) {
3423                         ret = PTR_ERR(bargs);
3424                         goto out;
3425                 }
3426
3427                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3428                         if (!fs_info->balance_ctl) {
3429                                 ret = -ENOTCONN;
3430                                 goto out_bargs;
3431                         }
3432
3433                         bctl = fs_info->balance_ctl;
3434                         spin_lock(&fs_info->balance_lock);
3435                         bctl->flags |= BTRFS_BALANCE_RESUME;
3436                         spin_unlock(&fs_info->balance_lock);
3437
3438                         goto do_balance;
3439                 }
3440         } else {
3441                 bargs = NULL;
3442         }
3443
3444         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
3445                         1)) {
3446                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3447                 ret = -EINPROGRESS;
3448                 goto out_bargs;
3449         }
3450         need_to_clear_lock = 1;
3451
3452         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3453         if (!bctl) {
3454                 ret = -ENOMEM;
3455                 goto out_bargs;
3456         }
3457
3458         bctl->fs_info = fs_info;
3459         if (arg) {
3460                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3461                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3462                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3463
3464                 bctl->flags = bargs->flags;
3465         } else {
3466                 /* balance everything - no filters */
3467                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3468         }
3469
3470 do_balance:
3471         ret = btrfs_balance(bctl, bargs);
3472         /*
3473          * bctl is freed in __cancel_balance or in free_fs_info if
3474          * restriper was paused all the way until unmount
3475          */
3476         if (arg) {
3477                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3478                         ret = -EFAULT;
3479         }
3480
3481 out_bargs:
3482         kfree(bargs);
3483 out:
3484         if (need_to_clear_lock)
3485                 atomic_set(&root->fs_info->mutually_exclusive_operation_running,
3486                            0);
3487         mutex_unlock(&fs_info->balance_mutex);
3488         mutex_unlock(&fs_info->volume_mutex);
3489         mnt_drop_write_file(file);
3490         return ret;
3491 }
3492
3493 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3494 {
3495         if (!capable(CAP_SYS_ADMIN))
3496                 return -EPERM;
3497
3498         switch (cmd) {
3499         case BTRFS_BALANCE_CTL_PAUSE:
3500                 return btrfs_pause_balance(root->fs_info);
3501         case BTRFS_BALANCE_CTL_CANCEL:
3502                 return btrfs_cancel_balance(root->fs_info);
3503         }
3504
3505         return -EINVAL;
3506 }
3507
3508 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3509                                          void __user *arg)
3510 {
3511         struct btrfs_fs_info *fs_info = root->fs_info;
3512         struct btrfs_ioctl_balance_args *bargs;
3513         int ret = 0;
3514
3515         if (!capable(CAP_SYS_ADMIN))
3516                 return -EPERM;
3517
3518         mutex_lock(&fs_info->balance_mutex);
3519         if (!fs_info->balance_ctl) {
3520                 ret = -ENOTCONN;
3521                 goto out;
3522         }
3523
3524         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3525         if (!bargs) {
3526                 ret = -ENOMEM;
3527                 goto out;
3528         }
3529
3530         update_ioctl_balance_args(fs_info, 1, bargs);
3531
3532         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3533                 ret = -EFAULT;
3534
3535         kfree(bargs);
3536 out:
3537         mutex_unlock(&fs_info->balance_mutex);
3538         return ret;
3539 }
3540
3541 static long btrfs_ioctl_quota_ctl(struct btrfs_root *root, void __user *arg)
3542 {
3543         struct btrfs_ioctl_quota_ctl_args *sa;
3544         struct btrfs_trans_handle *trans = NULL;
3545         int ret;
3546         int err;
3547
3548         if (!capable(CAP_SYS_ADMIN))
3549                 return -EPERM;
3550
3551         if (root->fs_info->sb->s_flags & MS_RDONLY)
3552                 return -EROFS;
3553
3554         sa = memdup_user(arg, sizeof(*sa));
3555         if (IS_ERR(sa))
3556                 return PTR_ERR(sa);
3557
3558         if (sa->cmd != BTRFS_QUOTA_CTL_RESCAN) {
3559                 trans = btrfs_start_transaction(root, 2);
3560                 if (IS_ERR(trans)) {
3561                         ret = PTR_ERR(trans);
3562                         goto out;
3563                 }
3564         }
3565
3566         switch (sa->cmd) {
3567         case BTRFS_QUOTA_CTL_ENABLE:
3568                 ret = btrfs_quota_enable(trans, root->fs_info);
3569                 break;
3570         case BTRFS_QUOTA_CTL_DISABLE:
3571                 ret = btrfs_quota_disable(trans, root->fs_info);
3572                 break;
3573         case BTRFS_QUOTA_CTL_RESCAN:
3574                 ret = btrfs_quota_rescan(root->fs_info);
3575                 break;
3576         default:
3577                 ret = -EINVAL;
3578                 break;
3579         }
3580
3581         if (copy_to_user(arg, sa, sizeof(*sa)))
3582                 ret = -EFAULT;
3583
3584         if (trans) {
3585                 err = btrfs_commit_transaction(trans, root);
3586                 if (err && !ret)
3587                         ret = err;
3588         }
3589
3590 out:
3591         kfree(sa);
3592         return ret;
3593 }
3594
3595 static long btrfs_ioctl_qgroup_assign(struct btrfs_root *root, void __user *arg)
3596 {
3597         struct btrfs_ioctl_qgroup_assign_args *sa;
3598         struct btrfs_trans_handle *trans;
3599         int ret;
3600         int err;
3601
3602         if (!capable(CAP_SYS_ADMIN))
3603                 return -EPERM;
3604
3605         if (root->fs_info->sb->s_flags & MS_RDONLY)
3606                 return -EROFS;
3607
3608         sa = memdup_user(arg, sizeof(*sa));
3609         if (IS_ERR(sa))
3610                 return PTR_ERR(sa);
3611
3612         trans = btrfs_join_transaction(root);
3613         if (IS_ERR(trans)) {
3614                 ret = PTR_ERR(trans);
3615                 goto out;
3616         }
3617
3618         /* FIXME: check if the IDs really exist */
3619         if (sa->assign) {
3620                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
3621                                                 sa->src, sa->dst);
3622         } else {
3623                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
3624                                                 sa->src, sa->dst);
3625         }
3626
3627         err = btrfs_end_transaction(trans, root);
3628         if (err && !ret)
3629                 ret = err;
3630
3631 out:
3632         kfree(sa);
3633         return ret;
3634 }
3635
3636 static long btrfs_ioctl_qgroup_create(struct btrfs_root *root, void __user *arg)
3637 {
3638         struct btrfs_ioctl_qgroup_create_args *sa;
3639         struct btrfs_trans_handle *trans;
3640         int ret;
3641         int err;
3642
3643         if (!capable(CAP_SYS_ADMIN))
3644                 return -EPERM;
3645
3646         if (root->fs_info->sb->s_flags & MS_RDONLY)
3647                 return -EROFS;
3648
3649         sa = memdup_user(arg, sizeof(*sa));
3650         if (IS_ERR(sa))
3651                 return PTR_ERR(sa);
3652
3653         trans = btrfs_join_transaction(root);
3654         if (IS_ERR(trans)) {
3655                 ret = PTR_ERR(trans);
3656                 goto out;
3657         }
3658
3659         /* FIXME: check if the IDs really exist */
3660         if (sa->create) {
3661                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
3662                                           NULL);
3663         } else {
3664                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
3665         }
3666
3667         err = btrfs_end_transaction(trans, root);
3668         if (err && !ret)
3669                 ret = err;
3670
3671 out:
3672         kfree(sa);
3673         return ret;
3674 }
3675
3676 static long btrfs_ioctl_qgroup_limit(struct btrfs_root *root, void __user *arg)
3677 {
3678         struct btrfs_ioctl_qgroup_limit_args *sa;
3679         struct btrfs_trans_handle *trans;
3680         int ret;
3681         int err;
3682         u64 qgroupid;
3683
3684         if (!capable(CAP_SYS_ADMIN))
3685                 return -EPERM;
3686
3687         if (root->fs_info->sb->s_flags & MS_RDONLY)
3688                 return -EROFS;
3689
3690         sa = memdup_user(arg, sizeof(*sa));
3691         if (IS_ERR(sa))
3692                 return PTR_ERR(sa);
3693
3694         trans = btrfs_join_transaction(root);
3695         if (IS_ERR(trans)) {
3696                 ret = PTR_ERR(trans);
3697                 goto out;
3698         }
3699
3700         qgroupid = sa->qgroupid;
3701         if (!qgroupid) {
3702                 /* take the current subvol as qgroup */
3703                 qgroupid = root->root_key.objectid;
3704         }
3705
3706         /* FIXME: check if the IDs really exist */
3707         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
3708
3709         err = btrfs_end_transaction(trans, root);
3710         if (err && !ret)
3711                 ret = err;
3712
3713 out:
3714         kfree(sa);
3715         return ret;
3716 }
3717
3718 static long btrfs_ioctl_set_received_subvol(struct file *file,
3719                                             void __user *arg)
3720 {
3721         struct btrfs_ioctl_received_subvol_args *sa = NULL;
3722         struct inode *inode = fdentry(file)->d_inode;
3723         struct btrfs_root *root = BTRFS_I(inode)->root;
3724         struct btrfs_root_item *root_item = &root->root_item;
3725         struct btrfs_trans_handle *trans;
3726         struct timespec ct = CURRENT_TIME;
3727         int ret = 0;
3728
3729         ret = mnt_want_write_file(file);
3730         if (ret < 0)
3731                 return ret;
3732
3733         down_write(&root->fs_info->subvol_sem);
3734
3735         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
3736                 ret = -EINVAL;
3737                 goto out;
3738         }
3739
3740         if (btrfs_root_readonly(root)) {
3741                 ret = -EROFS;
3742                 goto out;
3743         }
3744
3745         if (!inode_owner_or_capable(inode)) {
3746                 ret = -EACCES;
3747                 goto out;
3748         }
3749
3750         sa = memdup_user(arg, sizeof(*sa));
3751         if (IS_ERR(sa)) {
3752                 ret = PTR_ERR(sa);
3753                 sa = NULL;
3754                 goto out;
3755         }
3756
3757         trans = btrfs_start_transaction(root, 1);
3758         if (IS_ERR(trans)) {
3759                 ret = PTR_ERR(trans);
3760                 trans = NULL;
3761                 goto out;
3762         }
3763
3764         sa->rtransid = trans->transid;
3765         sa->rtime.sec = ct.tv_sec;
3766         sa->rtime.nsec = ct.tv_nsec;
3767
3768         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
3769         btrfs_set_root_stransid(root_item, sa->stransid);
3770         btrfs_set_root_rtransid(root_item, sa->rtransid);
3771         root_item->stime.sec = cpu_to_le64(sa->stime.sec);
3772         root_item->stime.nsec = cpu_to_le32(sa->stime.nsec);
3773         root_item->rtime.sec = cpu_to_le64(sa->rtime.sec);
3774         root_item->rtime.nsec = cpu_to_le32(sa->rtime.nsec);
3775
3776         ret = btrfs_update_root(trans, root->fs_info->tree_root,
3777                                 &root->root_key, &root->root_item);
3778         if (ret < 0) {
3779                 btrfs_end_transaction(trans, root);
3780                 trans = NULL;
3781                 goto out;
3782         } else {
3783                 ret = btrfs_commit_transaction(trans, root);
3784                 if (ret < 0)
3785                         goto out;
3786         }
3787
3788         ret = copy_to_user(arg, sa, sizeof(*sa));
3789         if (ret)
3790                 ret = -EFAULT;
3791
3792 out:
3793         kfree(sa);
3794         up_write(&root->fs_info->subvol_sem);
3795         mnt_drop_write_file(file);
3796         return ret;
3797 }
3798
3799 long btrfs_ioctl(struct file *file, unsigned int
3800                 cmd, unsigned long arg)
3801 {
3802         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3803         void __user *argp = (void __user *)arg;
3804
3805         switch (cmd) {
3806         case FS_IOC_GETFLAGS:
3807                 return btrfs_ioctl_getflags(file, argp);
3808         case FS_IOC_SETFLAGS:
3809                 return btrfs_ioctl_setflags(file, argp);
3810         case FS_IOC_GETVERSION:
3811                 return btrfs_ioctl_getversion(file, argp);
3812         case FITRIM:
3813                 return btrfs_ioctl_fitrim(file, argp);
3814         case BTRFS_IOC_SNAP_CREATE:
3815                 return btrfs_ioctl_snap_create(file, argp, 0);
3816         case BTRFS_IOC_SNAP_CREATE_V2:
3817                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3818         case BTRFS_IOC_SUBVOL_CREATE:
3819                 return btrfs_ioctl_snap_create(file, argp, 1);
3820         case BTRFS_IOC_SUBVOL_CREATE_V2:
3821                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
3822         case BTRFS_IOC_SNAP_DESTROY:
3823                 return btrfs_ioctl_snap_destroy(file, argp);
3824         case BTRFS_IOC_SUBVOL_GETFLAGS:
3825                 return btrfs_ioctl_subvol_getflags(file, argp);
3826         case BTRFS_IOC_SUBVOL_SETFLAGS:
3827                 return btrfs_ioctl_subvol_setflags(file, argp);
3828         case BTRFS_IOC_DEFAULT_SUBVOL:
3829                 return btrfs_ioctl_default_subvol(file, argp);
3830         case BTRFS_IOC_DEFRAG:
3831                 return btrfs_ioctl_defrag(file, NULL);
3832         case BTRFS_IOC_DEFRAG_RANGE:
3833                 return btrfs_ioctl_defrag(file, argp);
3834         case BTRFS_IOC_RESIZE:
3835                 return btrfs_ioctl_resize(root, argp);
3836         case BTRFS_IOC_ADD_DEV:
3837                 return btrfs_ioctl_add_dev(root, argp);
3838         case BTRFS_IOC_RM_DEV:
3839                 return btrfs_ioctl_rm_dev(root, argp);
3840         case BTRFS_IOC_FS_INFO:
3841                 return btrfs_ioctl_fs_info(root, argp);
3842         case BTRFS_IOC_DEV_INFO:
3843                 return btrfs_ioctl_dev_info(root, argp);
3844         case BTRFS_IOC_BALANCE:
3845                 return btrfs_ioctl_balance(file, NULL);
3846         case BTRFS_IOC_CLONE:
3847                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3848         case BTRFS_IOC_CLONE_RANGE:
3849                 return btrfs_ioctl_clone_range(file, argp);
3850         case BTRFS_IOC_TRANS_START:
3851                 return btrfs_ioctl_trans_start(file);
3852         case BTRFS_IOC_TRANS_END:
3853                 return btrfs_ioctl_trans_end(file);
3854         case BTRFS_IOC_TREE_SEARCH:
3855                 return btrfs_ioctl_tree_search(file, argp);
3856         case BTRFS_IOC_INO_LOOKUP:
3857                 return btrfs_ioctl_ino_lookup(file, argp);
3858         case BTRFS_IOC_INO_PATHS:
3859                 return btrfs_ioctl_ino_to_path(root, argp);
3860         case BTRFS_IOC_LOGICAL_INO:
3861                 return btrfs_ioctl_logical_to_ino(root, argp);
3862         case BTRFS_IOC_SPACE_INFO:
3863                 return btrfs_ioctl_space_info(root, argp);
3864         case BTRFS_IOC_SYNC:
3865                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3866                 return 0;
3867         case BTRFS_IOC_START_SYNC:
3868                 return btrfs_ioctl_start_sync(root, argp);
3869         case BTRFS_IOC_WAIT_SYNC:
3870                 return btrfs_ioctl_wait_sync(root, argp);
3871         case BTRFS_IOC_SCRUB:
3872                 return btrfs_ioctl_scrub(root, argp);
3873         case BTRFS_IOC_SCRUB_CANCEL:
3874                 return btrfs_ioctl_scrub_cancel(root, argp);
3875         case BTRFS_IOC_SCRUB_PROGRESS:
3876                 return btrfs_ioctl_scrub_progress(root, argp);
3877         case BTRFS_IOC_BALANCE_V2:
3878                 return btrfs_ioctl_balance(file, argp);
3879         case BTRFS_IOC_BALANCE_CTL:
3880                 return btrfs_ioctl_balance_ctl(root, arg);
3881         case BTRFS_IOC_BALANCE_PROGRESS:
3882                 return btrfs_ioctl_balance_progress(root, argp);
3883         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
3884                 return btrfs_ioctl_set_received_subvol(file, argp);
3885         case BTRFS_IOC_SEND:
3886                 return btrfs_ioctl_send(file, argp);
3887         case BTRFS_IOC_GET_DEV_STATS:
3888                 return btrfs_ioctl_get_dev_stats(root, argp);
3889         case BTRFS_IOC_QUOTA_CTL:
3890                 return btrfs_ioctl_quota_ctl(root, argp);
3891         case BTRFS_IOC_QGROUP_ASSIGN:
3892                 return btrfs_ioctl_qgroup_assign(root, argp);
3893         case BTRFS_IOC_QGROUP_CREATE:
3894                 return btrfs_ioctl_qgroup_create(root, argp);
3895         case BTRFS_IOC_QGROUP_LIMIT:
3896                 return btrfs_ioctl_qgroup_limit(root, argp);
3897         case BTRFS_IOC_DEV_REPLACE:
3898                 return btrfs_ioctl_dev_replace(root, argp);
3899         }
3900
3901         return -ENOTTY;
3902 }