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                 if (S_ISREG(inode->i_mode))
 147                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 148         }
 149
 150         btrfs_update_iflags(inode);
 151 }
 152
 153 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 154 {
 155         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 156         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 157
 158         if (copy_to_user(arg, &flags, sizeof(flags)))
 159                 return -EFAULT;
 160         return 0;
 161 }
 162
 163 static int check_flags(unsigned int flags)
 164 {
 165         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 166                       FS_NOATIME_FL | FS_NODUMP_FL | \
 167                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 168                       FS_NOCOMP_FL | FS_COMPR_FL |
 169                       FS_NOCOW_FL))
 170                 return -EOPNOTSUPP;
 171
 172         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 173                 return -EINVAL;
 174
 175         return 0;
 176 }
 177
 178 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 179 {
 180         struct inode *inode = file->f_path.dentry->d_inode;
 181         struct btrfs_inode *ip = BTRFS_I(inode);
 182         struct btrfs_root *root = ip->root;
 183         struct btrfs_trans_handle *trans;
 184         unsigned int flags, oldflags;
 185         int ret;
 186         u64 ip_oldflags;
 187         unsigned int i_oldflags;
 188         umode_t mode;
 189
 190         if (btrfs_root_readonly(root))
 191                 return -EROFS;
 192
 193         if (copy_from_user(&flags, arg, sizeof(flags)))
 194                 return -EFAULT;
 195
 196         ret = check_flags(flags);
 197         if (ret)
 198                 return ret;
 199
 200         if (!inode_owner_or_capable(inode))
 201                 return -EACCES;
 202
 203         ret = mnt_want_write_file(file);
 204         if (ret)
 205                 return ret;
 206
 207         mutex_lock(&inode->i_mutex);
 208
 209         ip_oldflags = ip->flags;
 210         i_oldflags = inode->i_flags;
 211         mode = inode->i_mode;
 212
 213         flags = btrfs_mask_flags(inode->i_mode, flags);
 214         oldflags = btrfs_flags_to_ioctl(ip->flags);
 215         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 216                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 217                         ret = -EPERM;
 218                         goto out_unlock;
 219                 }
 220         }
 221
 222         if (flags & FS_SYNC_FL)
 223                 ip->flags |= BTRFS_INODE_SYNC;
 224         else
 225                 ip->flags &= ~BTRFS_INODE_SYNC;
 226         if (flags & FS_IMMUTABLE_FL)
 227                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 228         else
 229                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 230         if (flags & FS_APPEND_FL)
 231                 ip->flags |= BTRFS_INODE_APPEND;
 232         else
 233                 ip->flags &= ~BTRFS_INODE_APPEND;
 234         if (flags & FS_NODUMP_FL)
 235                 ip->flags |= BTRFS_INODE_NODUMP;
 236         else
 237                 ip->flags &= ~BTRFS_INODE_NODUMP;
 238         if (flags & FS_NOATIME_FL)
 239                 ip->flags |= BTRFS_INODE_NOATIME;
 240         else
 241                 ip->flags &= ~BTRFS_INODE_NOATIME;
 242         if (flags & FS_DIRSYNC_FL)
 243                 ip->flags |= BTRFS_INODE_DIRSYNC;
 244         else
 245                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 246         if (flags & FS_NOCOW_FL) {
 247                 if (S_ISREG(mode)) {
 248                         /*
 249                          * It's safe to turn csums off here, no extents exist.
 250                          * Otherwise we want the flag to reflect the real COW
 251                          * status of the file and will not set it.
 252                          */
 253                         if (inode->i_size == 0)
 254                                 ip->flags |= BTRFS_INODE_NODATACOW
 255                                            | BTRFS_INODE_NODATASUM;
 256                 } else {
 257                         ip->flags |= BTRFS_INODE_NODATACOW;
 258                 }
 259         } else {
 260                 /*
 261                  * Revert back under same assuptions as above
 262                  */
 263                 if (S_ISREG(mode)) {
 264                         if (inode->i_size == 0)
 265                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 266                                              | BTRFS_INODE_NODATASUM);
 267                 } else {
 268                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 269                 }
 270         }
 271
 272         /*
 273          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 274          * flag may be changed automatically if compression code won't make
 275          * things smaller.
 276          */
 277         if (flags & FS_NOCOMP_FL) {
 278                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 279                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 280         } else if (flags & FS_COMPR_FL) {
 281                 ip->flags |= BTRFS_INODE_COMPRESS;
 282                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 283         } else {
 284                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 285         }
 286
 287         trans = btrfs_start_transaction(root, 1);
 288         if (IS_ERR(trans)) {
 289                 ret = PTR_ERR(trans);
 290                 goto out_drop;
 291         }
 292
 293         btrfs_update_iflags(inode);
 294         inode_inc_iversion(inode);
 295         inode->i_ctime = CURRENT_TIME;
 296         ret = btrfs_update_inode(trans, root, inode);
 297
 298         btrfs_end_transaction(trans, root);
 299  out_drop:
 300         if (ret) {
 301                 ip->flags = ip_oldflags;
 302                 inode->i_flags = i_oldflags;
 303         }
 304
 305  out_unlock:
 306         mutex_unlock(&inode->i_mutex);
 307         mnt_drop_write_file(file);
 308         return ret;
 309 }
 310
 311 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 312 {
 313         struct inode *inode = file->f_path.dentry->d_inode;
 314
 315         return put_user(inode->i_generation, arg);
 316 }
 317
 318 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 319 {
 320         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
 321         struct btrfs_device *device;
 322         struct request_queue *q;
 323         struct fstrim_range range;
 324         u64 minlen = ULLONG_MAX;
 325         u64 num_devices = 0;
 326         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 327         int ret;
 328
 329         if (!capable(CAP_SYS_ADMIN))
 330                 return -EPERM;
 331
 332         rcu_read_lock();
 333         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 334                                 dev_list) {
 335                 if (!device->bdev)
 336                         continue;
 337                 q = bdev_get_queue(device->bdev);
 338                 if (blk_queue_discard(q)) {
 339                         num_devices++;
 340                         minlen = min((u64)q->limits.discard_granularity,
 341                                      minlen);
 342                 }
 343         }
 344         rcu_read_unlock();
 345
 346         if (!num_devices)
 347                 return -EOPNOTSUPP;
 348         if (copy_from_user(&range, arg, sizeof(range)))
 349                 return -EFAULT;
 350         if (range.start > total_bytes ||
 351             range.len < fs_info->sb->s_blocksize)
 352                 return -EINVAL;
 353
 354         range.len = min(range.len, total_bytes - range.start);
 355         range.minlen = max(range.minlen, minlen);
 356         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 357         if (ret < 0)
 358                 return ret;
 359
 360         if (copy_to_user(arg, &range, sizeof(range)))
 361                 return -EFAULT;
 362
 363         return 0;
 364 }
 365
 366 static noinline int create_subvol(struct btrfs_root *root,
 367                                   struct dentry *dentry,
 368                                   char *name, int namelen,
 369                                   u64 *async_transid,
 370                                   struct btrfs_qgroup_inherit **inherit)
 371 {
 372         struct btrfs_trans_handle *trans;
 373         struct btrfs_key key;
 374         struct btrfs_root_item root_item;
 375         struct btrfs_inode_item *inode_item;
 376         struct extent_buffer *leaf;
 377         struct btrfs_root *new_root;
 378         struct dentry *parent = dentry->d_parent;
 379         struct inode *dir;
 380         struct timespec cur_time = CURRENT_TIME;
 381         int ret;
 382         int err;
 383         u64 objectid;
 384         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 385         u64 index = 0;
 386         uuid_le new_uuid;
 387
 388         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 389         if (ret)
 390                 return ret;
 391
 392         dir = parent->d_inode;
 393
 394         /*
 395          * 1 - inode item
 396          * 2 - refs
 397          * 1 - root item
 398          * 2 - dir items
 399          */
 400         trans = btrfs_start_transaction(root, 6);
 401         if (IS_ERR(trans))
 402                 return PTR_ERR(trans);
 403
 404         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid,
 405                                    inherit ? *inherit : NULL);
 406         if (ret)
 407                 goto fail;
 408
 409         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 410                                       0, objectid, NULL, 0, 0, 0);
 411         if (IS_ERR(leaf)) {
 412                 ret = PTR_ERR(leaf);
 413                 goto fail;
 414         }
 415
 416         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 417         btrfs_set_header_bytenr(leaf, leaf->start);
 418         btrfs_set_header_generation(leaf, trans->transid);
 419         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 420         btrfs_set_header_owner(leaf, objectid);
 421
 422         write_extent_buffer(leaf, root->fs_info->fsid,
 423                             (unsigned long)btrfs_header_fsid(leaf),
 424                             BTRFS_FSID_SIZE);
 425         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 426                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 427                             BTRFS_UUID_SIZE);
 428         btrfs_mark_buffer_dirty(leaf);
 429
 430         memset(&root_item, 0, sizeof(root_item));
 431
 432         inode_item = &root_item.inode;
 433         inode_item->generation = cpu_to_le64(1);
 434         inode_item->size = cpu_to_le64(3);
 435         inode_item->nlink = cpu_to_le32(1);
 436         inode_item->nbytes = cpu_to_le64(root->leafsize);
 437         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 438
 439         root_item.flags = 0;
 440         root_item.byte_limit = 0;
 441         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
 442
 443         btrfs_set_root_bytenr(&root_item, leaf->start);
 444         btrfs_set_root_generation(&root_item, trans->transid);
 445         btrfs_set_root_level(&root_item, 0);
 446         btrfs_set_root_refs(&root_item, 1);
 447         btrfs_set_root_used(&root_item, leaf->len);
 448         btrfs_set_root_last_snapshot(&root_item, 0);
 449
 450         btrfs_set_root_generation_v2(&root_item,
 451                         btrfs_root_generation(&root_item));
 452         uuid_le_gen(&new_uuid);
 453         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 454         root_item.otime.sec = cpu_to_le64(cur_time.tv_sec);
 455         root_item.otime.nsec = cpu_to_le32(cur_time.tv_nsec);
 456         root_item.ctime = root_item.otime;
 457         btrfs_set_root_ctransid(&root_item, trans->transid);
 458         btrfs_set_root_otransid(&root_item, trans->transid);
 459
 460         btrfs_tree_unlock(leaf);
 461         free_extent_buffer(leaf);
 462         leaf = NULL;
 463
 464         btrfs_set_root_dirid(&root_item, new_dirid);
 465
 466         key.objectid = objectid;
 467         key.offset = 0;
 468         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 469         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 470                                 &root_item);
 471         if (ret)
 472                 goto fail;
 473
 474         key.offset = (u64)-1;
 475         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 476         if (IS_ERR(new_root)) {
 477                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 478                 ret = PTR_ERR(new_root);
 479                 goto fail;
 480         }
 481
 482         btrfs_record_root_in_trans(trans, new_root);
 483
 484         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
 485         if (ret) {
 486                 /* We potentially lose an unused inode item here */
 487                 btrfs_abort_transaction(trans, root, ret);
 488                 goto fail;
 489         }
 490
 491         /*
 492          * insert the directory item
 493          */
 494         ret = btrfs_set_inode_index(dir, &index);
 495         if (ret) {
 496                 btrfs_abort_transaction(trans, root, ret);
 497                 goto fail;
 498         }
 499
 500         ret = btrfs_insert_dir_item(trans, root,
 501                                     name, namelen, dir, &key,
 502                                     BTRFS_FT_DIR, index);
 503         if (ret) {
 504                 btrfs_abort_transaction(trans, root, ret);
 505                 goto fail;
 506         }
 507
 508         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 509         ret = btrfs_update_inode(trans, root, dir);
 510         BUG_ON(ret);
 511
 512         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 513                                  objectid, root->root_key.objectid,
 514                                  btrfs_ino(dir), index, name, namelen);
 515
 516         BUG_ON(ret);
 517
 518         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 519 fail:
 520         if (async_transid) {
 521                 *async_transid = trans->transid;
 522                 err = btrfs_commit_transaction_async(trans, root, 1);
 523         } else {
 524                 err = btrfs_commit_transaction(trans, root);
 525         }
 526         if (err && !ret)
 527                 ret = err;
 528         return ret;
 529 }
 530
 531 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 532                            char *name, int namelen, u64 *async_transid,
 533                            bool readonly, struct btrfs_qgroup_inherit **inherit)
 534 {
 535         struct inode *inode;
 536         struct btrfs_pending_snapshot *pending_snapshot;
 537         struct btrfs_trans_handle *trans;
 538         int ret;
 539
 540         if (!root->ref_cows)
 541                 return -EINVAL;
 542
 543         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 544         if (!pending_snapshot)
 545                 return -ENOMEM;
 546
 547         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 548                              BTRFS_BLOCK_RSV_TEMP);
 549         pending_snapshot->dentry = dentry;
 550         pending_snapshot->root = root;
 551         pending_snapshot->readonly = readonly;
 552         if (inherit) {
 553                 pending_snapshot->inherit = *inherit;
 554                 *inherit = NULL;        /* take responsibility to free it */
 555         }
 556
 557         trans = btrfs_start_transaction(root->fs_info->extent_root, 6);
 558         if (IS_ERR(trans)) {
 559                 ret = PTR_ERR(trans);
 560                 goto fail;
 561         }
 562
 563         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 564         BUG_ON(ret);
 565
 566         spin_lock(&root->fs_info->trans_lock);
 567         list_add(&pending_snapshot->list,
 568                  &trans->transaction->pending_snapshots);
 569         spin_unlock(&root->fs_info->trans_lock);
 570         if (async_transid) {
 571                 *async_transid = trans->transid;
 572                 ret = btrfs_commit_transaction_async(trans,
 573                                      root->fs_info->extent_root, 1);
 574         } else {
 575                 ret = btrfs_commit_transaction(trans,
 576                                                root->fs_info->extent_root);
 577         }
 578         if (ret) {
 579                 /* cleanup_transaction has freed this for us */
 580                 if (trans->aborted)
 581                         pending_snapshot = NULL;
 582                 goto fail;
 583         }
 584
 585         ret = pending_snapshot->error;
 586         if (ret)
 587                 goto fail;
 588
 589         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 590         if (ret)
 591                 goto fail;
 592
 593         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 594         if (IS_ERR(inode)) {
 595                 ret = PTR_ERR(inode);
 596                 goto fail;
 597         }
 598         BUG_ON(!inode);
 599         d_instantiate(dentry, inode);
 600         ret = 0;
 601 fail:
 602         kfree(pending_snapshot);
 603         return ret;
 604 }
 605
 606 /*  copy of check_sticky in fs/namei.c()
 607 * It's inline, so penalty for filesystems that don't use sticky bit is
 608 * minimal.
 609 */
 610 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 611 {
 612         kuid_t fsuid = current_fsuid();
 613
 614         if (!(dir->i_mode & S_ISVTX))
 615                 return 0;
 616         if (uid_eq(inode->i_uid, fsuid))
 617                 return 0;
 618         if (uid_eq(dir->i_uid, fsuid))
 619                 return 0;
 620         return !capable(CAP_FOWNER);
 621 }
 622
 623 /*  copy of may_delete in fs/namei.c()
 624  *      Check whether we can remove a link victim from directory dir, check
 625  *  whether the type of victim is right.
 626  *  1. We can't do it if dir is read-only (done in permission())
 627  *  2. We should have write and exec permissions on dir
 628  *  3. We can't remove anything from append-only dir
 629  *  4. We can't do anything with immutable dir (done in permission())
 630  *  5. If the sticky bit on dir is set we should either
 631  *      a. be owner of dir, or
 632  *      b. be owner of victim, or
 633  *      c. have CAP_FOWNER capability
 634  *  6. If the victim is append-only or immutable we can't do antyhing with
 635  *     links pointing to it.
 636  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 637  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 638  *  9. We can't remove a root or mountpoint.
 639  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 640  *     nfs_async_unlink().
 641  */
 642
 643 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 644 {
 645         int error;
 646
 647         if (!victim->d_inode)
 648                 return -ENOENT;
 649
 650         BUG_ON(victim->d_parent->d_inode != dir);
 651         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 652
 653         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 654         if (error)
 655                 return error;
 656         if (IS_APPEND(dir))
 657                 return -EPERM;
 658         if (btrfs_check_sticky(dir, victim->d_inode)||
 659                 IS_APPEND(victim->d_inode)||
 660             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 661                 return -EPERM;
 662         if (isdir) {
 663                 if (!S_ISDIR(victim->d_inode->i_mode))
 664                         return -ENOTDIR;
 665                 if (IS_ROOT(victim))
 666                         return -EBUSY;
 667         } else if (S_ISDIR(victim->d_inode->i_mode))
 668                 return -EISDIR;
 669         if (IS_DEADDIR(dir))
 670                 return -ENOENT;
 671         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 672                 return -EBUSY;
 673         return 0;
 674 }
 675
 676 /* copy of may_create in fs/namei.c() */
 677 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 678 {
 679         if (child->d_inode)
 680                 return -EEXIST;
 681         if (IS_DEADDIR(dir))
 682                 return -ENOENT;
 683         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 684 }
 685
 686 /*
 687  * Create a new subvolume below @parent.  This is largely modeled after
 688  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 689  * inside this filesystem so it's quite a bit simpler.
 690  */
 691 static noinline int btrfs_mksubvol(struct path *parent,
 692                                    char *name, int namelen,
 693                                    struct btrfs_root *snap_src,
 694                                    u64 *async_transid, bool readonly,
 695                                    struct btrfs_qgroup_inherit **inherit)
 696 {
 697         struct inode *dir  = parent->dentry->d_inode;
 698         struct dentry *dentry;
 699         int error;
 700
 701         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 702
 703         dentry = lookup_one_len(name, parent->dentry, namelen);
 704         error = PTR_ERR(dentry);
 705         if (IS_ERR(dentry))
 706                 goto out_unlock;
 707
 708         error = -EEXIST;
 709         if (dentry->d_inode)
 710                 goto out_dput;
 711
 712         error = btrfs_may_create(dir, dentry);
 713         if (error)
 714                 goto out_dput;
 715
 716         /*
 717          * even if this name doesn't exist, we may get hash collisions.
 718          * check for them now when we can safely fail
 719          */
 720         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 721                                                dir->i_ino, name,
 722                                                namelen);
 723         if (error)
 724                 goto out_dput;
 725
 726         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 727
 728         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 729                 goto out_up_read;
 730
 731         if (snap_src) {
 732                 error = create_snapshot(snap_src, dentry, name, namelen,
 733                                         async_transid, readonly, inherit);
 734         } else {
 735                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 736                                       name, namelen, async_transid, inherit);
 737         }
 738         if (!error)
 739                 fsnotify_mkdir(dir, dentry);
 740 out_up_read:
 741         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 742 out_dput:
 743         dput(dentry);
 744 out_unlock:
 745         mutex_unlock(&dir->i_mutex);
 746         return error;
 747 }
 748
 749 /*
 750  * When we're defragging a range, we don't want to kick it off again
 751  * if it is really just waiting for delalloc to send it down.
 752  * If we find a nice big extent or delalloc range for the bytes in the
 753  * file you want to defrag, we return 0 to let you know to skip this
 754  * part of the file
 755  */
 756 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 757 {
 758         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 759         struct extent_map *em = NULL;
 760         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 761         u64 end;
 762
 763         read_lock(&em_tree->lock);
 764         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 765         read_unlock(&em_tree->lock);
 766
 767         if (em) {
 768                 end = extent_map_end(em);
 769                 free_extent_map(em);
 770                 if (end - offset > thresh)
 771                         return 0;
 772         }
 773         /* if we already have a nice delalloc here, just stop */
 774         thresh /= 2;
 775         end = count_range_bits(io_tree, &offset, offset + thresh,
 776                                thresh, EXTENT_DELALLOC, 1);
 777         if (end >= thresh)
 778                 return 0;
 779         return 1;
 780 }
 781
 782 /*
 783  * helper function to walk through a file and find extents
 784  * newer than a specific transid, and smaller than thresh.
 785  *
 786  * This is used by the defragging code to find new and small
 787  * extents
 788  */
 789 static int find_new_extents(struct btrfs_root *root,
 790                             struct inode *inode, u64 newer_than,
 791                             u64 *off, int thresh)
 792 {
 793         struct btrfs_path *path;
 794         struct btrfs_key min_key;
 795         struct btrfs_key max_key;
 796         struct extent_buffer *leaf;
 797         struct btrfs_file_extent_item *extent;
 798         int type;
 799         int ret;
 800         u64 ino = btrfs_ino(inode);
 801
 802         path = btrfs_alloc_path();
 803         if (!path)
 804                 return -ENOMEM;
 805
 806         min_key.objectid = ino;
 807         min_key.type = BTRFS_EXTENT_DATA_KEY;
 808         min_key.offset = *off;
 809
 810         max_key.objectid = ino;
 811         max_key.type = (u8)-1;
 812         max_key.offset = (u64)-1;
 813
 814         path->keep_locks = 1;
 815
 816         while(1) {
 817                 ret = btrfs_search_forward(root, &min_key, &max_key,
 818                                            path, 0, newer_than);
 819                 if (ret != 0)
 820                         goto none;
 821                 if (min_key.objectid != ino)
 822                         goto none;
 823                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 824                         goto none;
 825
 826                 leaf = path->nodes[0];
 827                 extent = btrfs_item_ptr(leaf, path->slots[0],
 828                                         struct btrfs_file_extent_item);
 829
 830                 type = btrfs_file_extent_type(leaf, extent);
 831                 if (type == BTRFS_FILE_EXTENT_REG &&
 832                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 833                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 834                         *off = min_key.offset;
 835                         btrfs_free_path(path);
 836                         return 0;
 837                 }
 838
 839                 if (min_key.offset == (u64)-1)
 840                         goto none;
 841
 842                 min_key.offset++;
 843                 btrfs_release_path(path);
 844         }
 845 none:
 846         btrfs_free_path(path);
 847         return -ENOENT;
 848 }
 849
 850 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 851 {
 852         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 853         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 854         struct extent_map *em;
 855         u64 len = PAGE_CACHE_SIZE;
 856
 857         /*
 858          * hopefully we have this extent in the tree already, try without
 859          * the full extent lock
 860          */
 861         read_lock(&em_tree->lock);
 862         em = lookup_extent_mapping(em_tree, start, len);
 863         read_unlock(&em_tree->lock);
 864
 865         if (!em) {
 866                 /* get the big lock and read metadata off disk */
 867                 lock_extent(io_tree, start, start + len - 1);
 868                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 869                 unlock_extent(io_tree, start, start + len - 1);
 870
 871                 if (IS_ERR(em))
 872                         return NULL;
 873         }
 874
 875         return em;
 876 }
 877
 878 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 879 {
 880         struct extent_map *next;
 881         bool ret = true;
 882
 883         /* this is the last extent */
 884         if (em->start + em->len >= i_size_read(inode))
 885                 return false;
 886
 887         next = defrag_lookup_extent(inode, em->start + em->len);
 888         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 889                 ret = false;
 890
 891         free_extent_map(next);
 892         return ret;
 893 }
 894
 895 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
 896                                u64 *last_len, u64 *skip, u64 *defrag_end,
 897                                int compress)
 898 {
 899         struct extent_map *em;
 900         int ret = 1;
 901         bool next_mergeable = true;
 902
 903         /*
 904          * make sure that once we start defragging an extent, we keep on
 905          * defragging it
 906          */
 907         if (start < *defrag_end)
 908                 return 1;
 909
 910         *skip = 0;
 911
 912         em = defrag_lookup_extent(inode, start);
 913         if (!em)
 914                 return 0;
 915
 916         /* this will cover holes, and inline extents */
 917         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 918                 ret = 0;
 919                 goto out;
 920         }
 921
 922         next_mergeable = defrag_check_next_extent(inode, em);
 923
 924         /*
 925          * we hit a real extent, if it is big or the next extent is not a
 926          * real extent, don't bother defragging it
 927          */
 928         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
 929             (em->len >= thresh || !next_mergeable))
 930                 ret = 0;
 931 out:
 932         /*
 933          * last_len ends up being a counter of how many bytes we've defragged.
 934          * every time we choose not to defrag an extent, we reset *last_len
 935          * so that the next tiny extent will force a defrag.
 936          *
 937          * The end result of this is that tiny extents before a single big
 938          * extent will force at least part of that big extent to be defragged.
 939          */
 940         if (ret) {
 941                 *defrag_end = extent_map_end(em);
 942         } else {
 943                 *last_len = 0;
 944                 *skip = extent_map_end(em);
 945                 *defrag_end = 0;
 946         }
 947
 948         free_extent_map(em);
 949         return ret;
 950 }
 951
 952 /*
 953  * it doesn't do much good to defrag one or two pages
 954  * at a time.  This pulls in a nice chunk of pages
 955  * to COW and defrag.
 956  *
 957  * It also makes sure the delalloc code has enough
 958  * dirty data to avoid making new small extents as part
 959  * of the defrag
 960  *
 961  * It's a good idea to start RA on this range
 962  * before calling this.
 963  */
 964 static int cluster_pages_for_defrag(struct inode *inode,
 965                                     struct page **pages,
 966                                     unsigned long start_index,
 967                                     int num_pages)
 968 {
 969         unsigned long file_end;
 970         u64 isize = i_size_read(inode);
 971         u64 page_start;
 972         u64 page_end;
 973         u64 page_cnt;
 974         int ret;
 975         int i;
 976         int i_done;
 977         struct btrfs_ordered_extent *ordered;
 978         struct extent_state *cached_state = NULL;
 979         struct extent_io_tree *tree;
 980         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 981
 982         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 983         if (!isize || start_index > file_end)
 984                 return 0;
 985
 986         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
 987
 988         ret = btrfs_delalloc_reserve_space(inode,
 989                                            page_cnt << PAGE_CACHE_SHIFT);
 990         if (ret)
 991                 return ret;
 992         i_done = 0;
 993         tree = &BTRFS_I(inode)->io_tree;
 994
 995         /* step one, lock all the pages */
 996         for (i = 0; i < page_cnt; i++) {
 997                 struct page *page;
 998 again:
 999                 page = find_or_create_page(inode->i_mapping,
1000                                            start_index + i, mask);
1001                 if (!page)
1002                         break;
1003
1004                 page_start = page_offset(page);
1005                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1006                 while (1) {
1007                         lock_extent(tree, page_start, page_end);
1008                         ordered = btrfs_lookup_ordered_extent(inode,
1009                                                               page_start);
1010                         unlock_extent(tree, page_start, page_end);
1011                         if (!ordered)
1012                                 break;
1013
1014                         unlock_page(page);
1015                         btrfs_start_ordered_extent(inode, ordered, 1);
1016                         btrfs_put_ordered_extent(ordered);
1017                         lock_page(page);
1018                         /*
1019                          * we unlocked the page above, so we need check if
1020                          * it was released or not.
1021                          */
1022                         if (page->mapping != inode->i_mapping) {
1023                                 unlock_page(page);
1024                                 page_cache_release(page);
1025                                 goto again;
1026                         }
1027                 }
1028
1029                 if (!PageUptodate(page)) {
1030                         btrfs_readpage(NULL, page);
1031                         lock_page(page);
1032                         if (!PageUptodate(page)) {
1033                                 unlock_page(page);
1034                                 page_cache_release(page);
1035                                 ret = -EIO;
1036                                 break;
1037                         }
1038                 }
1039
1040                 if (page->mapping != inode->i_mapping) {
1041                         unlock_page(page);
1042                         page_cache_release(page);
1043                         goto again;
1044                 }
1045
1046                 pages[i] = page;
1047                 i_done++;
1048         }
1049         if (!i_done || ret)
1050                 goto out;
1051
1052         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1053                 goto out;
1054
1055         /*
1056          * so now we have a nice long stream of locked
1057          * and up to date pages, lets wait on them
1058          */
1059         for (i = 0; i < i_done; i++)
1060                 wait_on_page_writeback(pages[i]);
1061
1062         page_start = page_offset(pages[0]);
1063         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1064
1065         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1066                          page_start, page_end - 1, 0, &cached_state);
1067         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1068                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1069                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1070                           &cached_state, GFP_NOFS);
1071
1072         if (i_done != page_cnt) {
1073                 spin_lock(&BTRFS_I(inode)->lock);
1074                 BTRFS_I(inode)->outstanding_extents++;
1075                 spin_unlock(&BTRFS_I(inode)->lock);
1076                 btrfs_delalloc_release_space(inode,
1077                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1078         }
1079
1080
1081         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1082                           &cached_state, GFP_NOFS);
1083
1084         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1085                              page_start, page_end - 1, &cached_state,
1086                              GFP_NOFS);
1087
1088         for (i = 0; i < i_done; i++) {
1089                 clear_page_dirty_for_io(pages[i]);
1090                 ClearPageChecked(pages[i]);
1091                 set_page_extent_mapped(pages[i]);
1092                 set_page_dirty(pages[i]);
1093                 unlock_page(pages[i]);
1094                 page_cache_release(pages[i]);
1095         }
1096         return i_done;
1097 out:
1098         for (i = 0; i < i_done; i++) {
1099                 unlock_page(pages[i]);
1100                 page_cache_release(pages[i]);
1101         }
1102         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1103         return ret;
1104
1105 }
1106
1107 int btrfs_defrag_file(struct inode *inode, struct file *file,
1108                       struct btrfs_ioctl_defrag_range_args *range,
1109                       u64 newer_than, unsigned long max_to_defrag)
1110 {
1111         struct btrfs_root *root = BTRFS_I(inode)->root;
1112         struct file_ra_state *ra = NULL;
1113         unsigned long last_index;
1114         u64 isize = i_size_read(inode);
1115         u64 last_len = 0;
1116         u64 skip = 0;
1117         u64 defrag_end = 0;
1118         u64 newer_off = range->start;
1119         unsigned long i;
1120         unsigned long ra_index = 0;
1121         int ret;
1122         int defrag_count = 0;
1123         int compress_type = BTRFS_COMPRESS_ZLIB;
1124         int extent_thresh = range->extent_thresh;
1125         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1126         int cluster = max_cluster;
1127         u64 new_align = ~((u64)128 * 1024 - 1);
1128         struct page **pages = NULL;
1129
1130         if (extent_thresh == 0)
1131                 extent_thresh = 256 * 1024;
1132
1133         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1134                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1135                         return -EINVAL;
1136                 if (range->compress_type)
1137                         compress_type = range->compress_type;
1138         }
1139
1140         if (isize == 0)
1141                 return 0;
1142
1143         /*
1144          * if we were not given a file, allocate a readahead
1145          * context
1146          */
1147         if (!file) {
1148                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1149                 if (!ra)
1150                         return -ENOMEM;
1151                 file_ra_state_init(ra, inode->i_mapping);
1152         } else {
1153                 ra = &file->f_ra;
1154         }
1155
1156         pages = kmalloc(sizeof(struct page *) * max_cluster,
1157                         GFP_NOFS);
1158         if (!pages) {
1159                 ret = -ENOMEM;
1160                 goto out_ra;
1161         }
1162
1163         /* find the last page to defrag */
1164         if (range->start + range->len > range->start) {
1165                 last_index = min_t(u64, isize - 1,
1166                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1167         } else {
1168                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1169         }
1170
1171         if (newer_than) {
1172                 ret = find_new_extents(root, inode, newer_than,
1173                                        &newer_off, 64 * 1024);
1174                 if (!ret) {
1175                         range->start = newer_off;
1176                         /*
1177                          * we always align our defrag to help keep
1178                          * the extents in the file evenly spaced
1179                          */
1180                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1181                 } else
1182                         goto out_ra;
1183         } else {
1184                 i = range->start >> PAGE_CACHE_SHIFT;
1185         }
1186         if (!max_to_defrag)
1187                 max_to_defrag = last_index + 1;
1188
1189         /*
1190          * make writeback starts from i, so the defrag range can be
1191          * written sequentially.
1192          */
1193         if (i < inode->i_mapping->writeback_index)
1194                 inode->i_mapping->writeback_index = i;
1195
1196         while (i <= last_index && defrag_count < max_to_defrag &&
1197                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1198                 PAGE_CACHE_SHIFT)) {
1199                 /*
1200                  * make sure we stop running if someone unmounts
1201                  * the FS
1202                  */
1203                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1204                         break;
1205
1206                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1207                                          extent_thresh, &last_len, &skip,
1208                                          &defrag_end, range->flags &
1209                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1210                         unsigned long next;
1211                         /*
1212                          * the should_defrag function tells us how much to skip
1213                          * bump our counter by the suggested amount
1214                          */
1215                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1216                         i = max(i + 1, next);
1217                         continue;
1218                 }
1219
1220                 if (!newer_than) {
1221                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1222                                    PAGE_CACHE_SHIFT) - i;
1223                         cluster = min(cluster, max_cluster);
1224                 } else {
1225                         cluster = max_cluster;
1226                 }
1227
1228                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1229                         BTRFS_I(inode)->force_compress = compress_type;
1230
1231                 if (i + cluster > ra_index) {
1232                         ra_index = max(i, ra_index);
1233                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1234                                        cluster);
1235                         ra_index += max_cluster;
1236                 }
1237
1238                 mutex_lock(&inode->i_mutex);
1239                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1240                 if (ret < 0) {
1241                         mutex_unlock(&inode->i_mutex);
1242                         goto out_ra;
1243                 }
1244
1245                 defrag_count += ret;
1246                 balance_dirty_pages_ratelimited(inode->i_mapping);
1247                 mutex_unlock(&inode->i_mutex);
1248
1249                 if (newer_than) {
1250                         if (newer_off == (u64)-1)
1251                                 break;
1252
1253                         if (ret > 0)
1254                                 i += ret;
1255
1256                         newer_off = max(newer_off + 1,
1257                                         (u64)i << PAGE_CACHE_SHIFT);
1258
1259                         ret = find_new_extents(root, inode,
1260                                                newer_than, &newer_off,
1261                                                64 * 1024);
1262                         if (!ret) {
1263                                 range->start = newer_off;
1264                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1265                         } else {
1266                                 break;
1267                         }
1268                 } else {
1269                         if (ret > 0) {
1270                                 i += ret;
1271                                 last_len += ret << PAGE_CACHE_SHIFT;
1272                         } else {
1273                                 i++;
1274                                 last_len = 0;
1275                         }
1276                 }
1277         }
1278
1279         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1280                 filemap_flush(inode->i_mapping);
1281
1282         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1283                 /* the filemap_flush will queue IO into the worker threads, but
1284                  * we have to make sure the IO is actually started and that
1285                  * ordered extents get created before we return
1286                  */
1287                 atomic_inc(&root->fs_info->async_submit_draining);
1288                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1289                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1290                         wait_event(root->fs_info->async_submit_wait,
1291                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1292                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1293                 }
1294                 atomic_dec(&root->fs_info->async_submit_draining);
1295
1296                 mutex_lock(&inode->i_mutex);
1297                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1298                 mutex_unlock(&inode->i_mutex);
1299         }
1300
1301         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1302                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1303         }
1304
1305         ret = defrag_count;
1306
1307 out_ra:
1308         if (!file)
1309                 kfree(ra);
1310         kfree(pages);
1311         return ret;
1312 }
1313
1314 static noinline int btrfs_ioctl_resize(struct file *file,
1315                                         void __user *arg)
1316 {
1317         u64 new_size;
1318         u64 old_size;
1319         u64 devid = 1;
1320         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1321         struct btrfs_ioctl_vol_args *vol_args;
1322         struct btrfs_trans_handle *trans;
1323         struct btrfs_device *device = NULL;
1324         char *sizestr;
1325         char *devstr = NULL;
1326         int ret = 0;
1327         int mod = 0;
1328
1329         if (root->fs_info->sb->s_flags & MS_RDONLY)
1330                 return -EROFS;
1331
1332         if (!capable(CAP_SYS_ADMIN))
1333                 return -EPERM;
1334
1335         ret = mnt_want_write_file(file);
1336         if (ret)
1337                 return ret;
1338
1339         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1340                         1)) {
1341                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
1342                 return -EINPROGRESS;
1343         }
1344
1345         mutex_lock(&root->fs_info->volume_mutex);
1346         vol_args = memdup_user(arg, sizeof(*vol_args));
1347         if (IS_ERR(vol_args)) {
1348                 ret = PTR_ERR(vol_args);
1349                 goto out;
1350         }
1351
1352         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1353
1354         sizestr = vol_args->name;
1355         devstr = strchr(sizestr, ':');
1356         if (devstr) {
1357                 char *end;
1358                 sizestr = devstr + 1;
1359                 *devstr = '\0';
1360                 devstr = vol_args->name;
1361                 devid = simple_strtoull(devstr, &end, 10);
1362                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1363                        (unsigned long long)devid);
1364         }
1365         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1366         if (!device) {
1367                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1368                        (unsigned long long)devid);
1369                 ret = -EINVAL;
1370                 goto out_free;
1371         }
1372         if (device->fs_devices && device->fs_devices->seeding) {
1373                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1374                        "seeding device %llu\n",
1375                        (unsigned long long)devid);
1376                 ret = -EINVAL;
1377                 goto out_free;
1378         }
1379
1380         if (!strcmp(sizestr, "max"))
1381                 new_size = device->bdev->bd_inode->i_size;
1382         else {
1383                 if (sizestr[0] == '-') {
1384                         mod = -1;
1385                         sizestr++;
1386                 } else if (sizestr[0] == '+') {
1387                         mod = 1;
1388                         sizestr++;
1389                 }
1390                 new_size = memparse(sizestr, NULL);
1391                 if (new_size == 0) {
1392                         ret = -EINVAL;
1393                         goto out_free;
1394                 }
1395         }
1396
1397         if (device->is_tgtdev_for_dev_replace) {
1398                 ret = -EINVAL;
1399                 goto out_free;
1400         }
1401
1402         old_size = device->total_bytes;
1403
1404         if (mod < 0) {
1405                 if (new_size > old_size) {
1406                         ret = -EINVAL;
1407                         goto out_free;
1408                 }
1409                 new_size = old_size - new_size;
1410         } else if (mod > 0) {
1411                 new_size = old_size + new_size;
1412         }
1413
1414         if (new_size < 256 * 1024 * 1024) {
1415                 ret = -EINVAL;
1416                 goto out_free;
1417         }
1418         if (new_size > device->bdev->bd_inode->i_size) {
1419                 ret = -EFBIG;
1420                 goto out_free;
1421         }
1422
1423         do_div(new_size, root->sectorsize);
1424         new_size *= root->sectorsize;
1425
1426         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1427                       rcu_str_deref(device->name),
1428                       (unsigned long long)new_size);
1429
1430         if (new_size > old_size) {
1431                 trans = btrfs_start_transaction(root, 0);
1432                 if (IS_ERR(trans)) {
1433                         ret = PTR_ERR(trans);
1434                         goto out_free;
1435                 }
1436                 ret = btrfs_grow_device(trans, device, new_size);
1437                 btrfs_commit_transaction(trans, root);
1438         } else if (new_size < old_size) {
1439                 ret = btrfs_shrink_device(device, new_size);
1440         } /* equal, nothing need to do */
1441
1442 out_free:
1443         kfree(vol_args);
1444 out:
1445         mutex_unlock(&root->fs_info->volume_mutex);
1446         mnt_drop_write_file(file);
1447         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1448         return ret;
1449 }
1450
1451 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1452                                 char *name, unsigned long fd, int subvol,
1453                                 u64 *transid, bool readonly,
1454                                 struct btrfs_qgroup_inherit **inherit)
1455 {
1456         int namelen;
1457         int ret = 0;
1458
1459         ret = mnt_want_write_file(file);
1460         if (ret)
1461                 goto out;
1462
1463         namelen = strlen(name);
1464         if (strchr(name, '/')) {
1465                 ret = -EINVAL;
1466                 goto out_drop_write;
1467         }
1468
1469         if (name[0] == '.' &&
1470            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1471                 ret = -EEXIST;
1472                 goto out_drop_write;
1473         }
1474
1475         if (subvol) {
1476                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1477                                      NULL, transid, readonly, inherit);
1478         } else {
1479                 struct fd src = fdget(fd);
1480                 struct inode *src_inode;
1481                 if (!src.file) {
1482                         ret = -EINVAL;
1483                         goto out_drop_write;
1484                 }
1485
1486                 src_inode = src.file->f_path.dentry->d_inode;
1487                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1488                         printk(KERN_INFO "btrfs: Snapshot src from "
1489                                "another FS\n");
1490                         ret = -EINVAL;
1491                 } else {
1492                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1493                                              BTRFS_I(src_inode)->root,
1494                                              transid, readonly, inherit);
1495                 }
1496                 fdput(src);
1497         }
1498 out_drop_write:
1499         mnt_drop_write_file(file);
1500 out:
1501         return ret;
1502 }
1503
1504 static noinline int btrfs_ioctl_snap_create(struct file *file,
1505                                             void __user *arg, int subvol)
1506 {
1507         struct btrfs_ioctl_vol_args *vol_args;
1508         int ret;
1509
1510         vol_args = memdup_user(arg, sizeof(*vol_args));
1511         if (IS_ERR(vol_args))
1512                 return PTR_ERR(vol_args);
1513         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1514
1515         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1516                                               vol_args->fd, subvol,
1517                                               NULL, false, NULL);
1518
1519         kfree(vol_args);
1520         return ret;
1521 }
1522
1523 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1524                                                void __user *arg, int subvol)
1525 {
1526         struct btrfs_ioctl_vol_args_v2 *vol_args;
1527         int ret;
1528         u64 transid = 0;
1529         u64 *ptr = NULL;
1530         bool readonly = false;
1531         struct btrfs_qgroup_inherit *inherit = NULL;
1532
1533         vol_args = memdup_user(arg, sizeof(*vol_args));
1534         if (IS_ERR(vol_args))
1535                 return PTR_ERR(vol_args);
1536         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1537
1538         if (vol_args->flags &
1539             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1540               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1541                 ret = -EOPNOTSUPP;
1542                 goto out;
1543         }
1544
1545         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1546                 ptr = &transid;
1547         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1548                 readonly = true;
1549         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1550                 if (vol_args->size > PAGE_CACHE_SIZE) {
1551                         ret = -EINVAL;
1552                         goto out;
1553                 }
1554                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1555                 if (IS_ERR(inherit)) {
1556                         ret = PTR_ERR(inherit);
1557                         goto out;
1558                 }
1559         }
1560
1561         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1562                                               vol_args->fd, subvol, ptr,
1563                                               readonly, &inherit);
1564
1565         if (ret == 0 && ptr &&
1566             copy_to_user(arg +
1567                          offsetof(struct btrfs_ioctl_vol_args_v2,
1568                                   transid), ptr, sizeof(*ptr)))
1569                 ret = -EFAULT;
1570 out:
1571         kfree(vol_args);
1572         kfree(inherit);
1573         return ret;
1574 }
1575
1576 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1577                                                 void __user *arg)
1578 {
1579         struct inode *inode = fdentry(file)->d_inode;
1580         struct btrfs_root *root = BTRFS_I(inode)->root;
1581         int ret = 0;
1582         u64 flags = 0;
1583
1584         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1585                 return -EINVAL;
1586
1587         down_read(&root->fs_info->subvol_sem);
1588         if (btrfs_root_readonly(root))
1589                 flags |= BTRFS_SUBVOL_RDONLY;
1590         up_read(&root->fs_info->subvol_sem);
1591
1592         if (copy_to_user(arg, &flags, sizeof(flags)))
1593                 ret = -EFAULT;
1594
1595         return ret;
1596 }
1597
1598 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1599                                               void __user *arg)
1600 {
1601         struct inode *inode = fdentry(file)->d_inode;
1602         struct btrfs_root *root = BTRFS_I(inode)->root;
1603         struct btrfs_trans_handle *trans;
1604         u64 root_flags;
1605         u64 flags;
1606         int ret = 0;
1607
1608         ret = mnt_want_write_file(file);
1609         if (ret)
1610                 goto out;
1611
1612         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1613                 ret = -EINVAL;
1614                 goto out_drop_write;
1615         }
1616
1617         if (copy_from_user(&flags, arg, sizeof(flags))) {
1618                 ret = -EFAULT;
1619                 goto out_drop_write;
1620         }
1621
1622         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1623                 ret = -EINVAL;
1624                 goto out_drop_write;
1625         }
1626
1627         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1628                 ret = -EOPNOTSUPP;
1629                 goto out_drop_write;
1630         }
1631
1632         if (!inode_owner_or_capable(inode)) {
1633                 ret = -EACCES;
1634                 goto out_drop_write;
1635         }
1636
1637         down_write(&root->fs_info->subvol_sem);
1638
1639         /* nothing to do */
1640         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1641                 goto out_drop_sem;
1642
1643         root_flags = btrfs_root_flags(&root->root_item);
1644         if (flags & BTRFS_SUBVOL_RDONLY)
1645                 btrfs_set_root_flags(&root->root_item,
1646                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1647         else
1648                 btrfs_set_root_flags(&root->root_item,
1649                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1650
1651         trans = btrfs_start_transaction(root, 1);
1652         if (IS_ERR(trans)) {
1653                 ret = PTR_ERR(trans);
1654                 goto out_reset;
1655         }
1656
1657         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1658                                 &root->root_key, &root->root_item);
1659
1660         btrfs_commit_transaction(trans, root);
1661 out_reset:
1662         if (ret)
1663                 btrfs_set_root_flags(&root->root_item, root_flags);
1664 out_drop_sem:
1665         up_write(&root->fs_info->subvol_sem);
1666 out_drop_write:
1667         mnt_drop_write_file(file);
1668 out:
1669         return ret;
1670 }
1671
1672 /*
1673  * helper to check if the subvolume references other subvolumes
1674  */
1675 static noinline int may_destroy_subvol(struct btrfs_root *root)
1676 {
1677         struct btrfs_path *path;
1678         struct btrfs_key key;
1679         int ret;
1680
1681         path = btrfs_alloc_path();
1682         if (!path)
1683                 return -ENOMEM;
1684
1685         key.objectid = root->root_key.objectid;
1686         key.type = BTRFS_ROOT_REF_KEY;
1687         key.offset = (u64)-1;
1688
1689         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1690                                 &key, path, 0, 0);
1691         if (ret < 0)
1692                 goto out;
1693         BUG_ON(ret == 0);
1694
1695         ret = 0;
1696         if (path->slots[0] > 0) {
1697                 path->slots[0]--;
1698                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1699                 if (key.objectid == root->root_key.objectid &&
1700                     key.type == BTRFS_ROOT_REF_KEY)
1701                         ret = -ENOTEMPTY;
1702         }
1703 out:
1704         btrfs_free_path(path);
1705         return ret;
1706 }
1707
1708 static noinline int key_in_sk(struct btrfs_key *key,
1709                               struct btrfs_ioctl_search_key *sk)
1710 {
1711         struct btrfs_key test;
1712         int ret;
1713
1714         test.objectid = sk->min_objectid;
1715         test.type = sk->min_type;
1716         test.offset = sk->min_offset;
1717
1718         ret = btrfs_comp_cpu_keys(key, &test);
1719         if (ret < 0)
1720                 return 0;
1721
1722         test.objectid = sk->max_objectid;
1723         test.type = sk->max_type;
1724         test.offset = sk->max_offset;
1725
1726         ret = btrfs_comp_cpu_keys(key, &test);
1727         if (ret > 0)
1728                 return 0;
1729         return 1;
1730 }
1731
1732 static noinline int copy_to_sk(struct btrfs_root *root,
1733                                struct btrfs_path *path,
1734                                struct btrfs_key *key,
1735                                struct btrfs_ioctl_search_key *sk,
1736                                char *buf,
1737                                unsigned long *sk_offset,
1738                                int *num_found)
1739 {
1740         u64 found_transid;
1741         struct extent_buffer *leaf;
1742         struct btrfs_ioctl_search_header sh;
1743         unsigned long item_off;
1744         unsigned long item_len;
1745         int nritems;
1746         int i;
1747         int slot;
1748         int ret = 0;
1749
1750         leaf = path->nodes[0];
1751         slot = path->slots[0];
1752         nritems = btrfs_header_nritems(leaf);
1753
1754         if (btrfs_header_generation(leaf) > sk->max_transid) {
1755                 i = nritems;
1756                 goto advance_key;
1757         }
1758         found_transid = btrfs_header_generation(leaf);
1759
1760         for (i = slot; i < nritems; i++) {
1761                 item_off = btrfs_item_ptr_offset(leaf, i);
1762                 item_len = btrfs_item_size_nr(leaf, i);
1763
1764                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1765                         item_len = 0;
1766
1767                 if (sizeof(sh) + item_len + *sk_offset >
1768                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1769                         ret = 1;
1770                         goto overflow;
1771                 }
1772
1773                 btrfs_item_key_to_cpu(leaf, key, i);
1774                 if (!key_in_sk(key, sk))
1775                         continue;
1776
1777                 sh.objectid = key->objectid;
1778                 sh.offset = key->offset;
1779                 sh.type = key->type;
1780                 sh.len = item_len;
1781                 sh.transid = found_transid;
1782
1783                 /* copy search result header */
1784                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1785                 *sk_offset += sizeof(sh);
1786
1787                 if (item_len) {
1788                         char *p = buf + *sk_offset;
1789                         /* copy the item */
1790                         read_extent_buffer(leaf, p,
1791                                            item_off, item_len);
1792                         *sk_offset += item_len;
1793                 }
1794                 (*num_found)++;
1795
1796                 if (*num_found >= sk->nr_items)
1797                         break;
1798         }
1799 advance_key:
1800         ret = 0;
1801         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1802                 key->offset++;
1803         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1804                 key->offset = 0;
1805                 key->type++;
1806         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1807                 key->offset = 0;
1808                 key->type = 0;
1809                 key->objectid++;
1810         } else
1811                 ret = 1;
1812 overflow:
1813         return ret;
1814 }
1815
1816 static noinline int search_ioctl(struct inode *inode,
1817                                  struct btrfs_ioctl_search_args *args)
1818 {
1819         struct btrfs_root *root;
1820         struct btrfs_key key;
1821         struct btrfs_key max_key;
1822         struct btrfs_path *path;
1823         struct btrfs_ioctl_search_key *sk = &args->key;
1824         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1825         int ret;
1826         int num_found = 0;
1827         unsigned long sk_offset = 0;
1828
1829         path = btrfs_alloc_path();
1830         if (!path)
1831                 return -ENOMEM;
1832
1833         if (sk->tree_id == 0) {
1834                 /* search the root of the inode that was passed */
1835                 root = BTRFS_I(inode)->root;
1836         } else {
1837                 key.objectid = sk->tree_id;
1838                 key.type = BTRFS_ROOT_ITEM_KEY;
1839                 key.offset = (u64)-1;
1840                 root = btrfs_read_fs_root_no_name(info, &key);
1841                 if (IS_ERR(root)) {
1842                         printk(KERN_ERR "could not find root %llu\n",
1843                                sk->tree_id);
1844                         btrfs_free_path(path);
1845                         return -ENOENT;
1846                 }
1847         }
1848
1849         key.objectid = sk->min_objectid;
1850         key.type = sk->min_type;
1851         key.offset = sk->min_offset;
1852
1853         max_key.objectid = sk->max_objectid;
1854         max_key.type = sk->max_type;
1855         max_key.offset = sk->max_offset;
1856
1857         path->keep_locks = 1;
1858
1859         while(1) {
1860                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1861                                            sk->min_transid);
1862                 if (ret != 0) {
1863                         if (ret > 0)
1864                                 ret = 0;
1865                         goto err;
1866                 }
1867                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1868                                  &sk_offset, &num_found);
1869                 btrfs_release_path(path);
1870                 if (ret || num_found >= sk->nr_items)
1871                         break;
1872
1873         }
1874         ret = 0;
1875 err:
1876         sk->nr_items = num_found;
1877         btrfs_free_path(path);
1878         return ret;
1879 }
1880
1881 static noinline int btrfs_ioctl_tree_search(struct file *file,
1882                                            void __user *argp)
1883 {
1884          struct btrfs_ioctl_search_args *args;
1885          struct inode *inode;
1886          int ret;
1887
1888         if (!capable(CAP_SYS_ADMIN))
1889                 return -EPERM;
1890
1891         args = memdup_user(argp, sizeof(*args));
1892         if (IS_ERR(args))
1893                 return PTR_ERR(args);
1894
1895         inode = fdentry(file)->d_inode;
1896         ret = search_ioctl(inode, args);
1897         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1898                 ret = -EFAULT;
1899         kfree(args);
1900         return ret;
1901 }
1902
1903 /*
1904  * Search INODE_REFs to identify path name of 'dirid' directory
1905  * in a 'tree_id' tree. and sets path name to 'name'.
1906  */
1907 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1908                                 u64 tree_id, u64 dirid, char *name)
1909 {
1910         struct btrfs_root *root;
1911         struct btrfs_key key;
1912         char *ptr;
1913         int ret = -1;
1914         int slot;
1915         int len;
1916         int total_len = 0;
1917         struct btrfs_inode_ref *iref;
1918         struct extent_buffer *l;
1919         struct btrfs_path *path;
1920
1921         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1922                 name[0]='\0';
1923                 return 0;
1924         }
1925
1926         path = btrfs_alloc_path();
1927         if (!path)
1928                 return -ENOMEM;
1929
1930         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1931
1932         key.objectid = tree_id;
1933         key.type = BTRFS_ROOT_ITEM_KEY;
1934         key.offset = (u64)-1;
1935         root = btrfs_read_fs_root_no_name(info, &key);
1936         if (IS_ERR(root)) {
1937                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1938                 ret = -ENOENT;
1939                 goto out;
1940         }
1941
1942         key.objectid = dirid;
1943         key.type = BTRFS_INODE_REF_KEY;
1944         key.offset = (u64)-1;
1945
1946         while(1) {
1947                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1948                 if (ret < 0)
1949                         goto out;
1950
1951                 l = path->nodes[0];
1952                 slot = path->slots[0];
1953                 if (ret > 0 && slot > 0)
1954                         slot--;
1955                 btrfs_item_key_to_cpu(l, &key, slot);
1956
1957                 if (ret > 0 && (key.objectid != dirid ||
1958                                 key.type != BTRFS_INODE_REF_KEY)) {
1959                         ret = -ENOENT;
1960                         goto out;
1961                 }
1962
1963                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1964                 len = btrfs_inode_ref_name_len(l, iref);
1965                 ptr -= len + 1;
1966                 total_len += len + 1;
1967                 if (ptr < name)
1968                         goto out;
1969
1970                 *(ptr + len) = '/';
1971                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1972
1973                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1974                         break;
1975
1976                 btrfs_release_path(path);
1977                 key.objectid = key.offset;
1978                 key.offset = (u64)-1;
1979                 dirid = key.objectid;
1980         }
1981         if (ptr < name)
1982                 goto out;
1983         memmove(name, ptr, total_len);
1984         name[total_len]='\0';
1985         ret = 0;
1986 out:
1987         btrfs_free_path(path);
1988         return ret;
1989 }
1990
1991 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1992                                            void __user *argp)
1993 {
1994          struct btrfs_ioctl_ino_lookup_args *args;
1995          struct inode *inode;
1996          int ret;
1997
1998         if (!capable(CAP_SYS_ADMIN))
1999                 return -EPERM;
2000
2001         args = memdup_user(argp, sizeof(*args));
2002         if (IS_ERR(args))
2003                 return PTR_ERR(args);
2004
2005         inode = fdentry(file)->d_inode;
2006
2007         if (args->treeid == 0)
2008                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2009
2010         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2011                                         args->treeid, args->objectid,
2012                                         args->name);
2013
2014         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2015                 ret = -EFAULT;
2016
2017         kfree(args);
2018         return ret;
2019 }
2020
2021 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2022                                              void __user *arg)
2023 {
2024         struct dentry *parent = fdentry(file);
2025         struct dentry *dentry;
2026         struct inode *dir = parent->d_inode;
2027         struct inode *inode;
2028         struct btrfs_root *root = BTRFS_I(dir)->root;
2029         struct btrfs_root *dest = NULL;
2030         struct btrfs_ioctl_vol_args *vol_args;
2031         struct btrfs_trans_handle *trans;
2032         int namelen;
2033         int ret;
2034         int err = 0;
2035
2036         vol_args = memdup_user(arg, sizeof(*vol_args));
2037         if (IS_ERR(vol_args))
2038                 return PTR_ERR(vol_args);
2039
2040         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2041         namelen = strlen(vol_args->name);
2042         if (strchr(vol_args->name, '/') ||
2043             strncmp(vol_args->name, "..", namelen) == 0) {
2044                 err = -EINVAL;
2045                 goto out;
2046         }
2047
2048         err = mnt_want_write_file(file);
2049         if (err)
2050                 goto out;
2051
2052         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
2053         dentry = lookup_one_len(vol_args->name, parent, namelen);
2054         if (IS_ERR(dentry)) {
2055                 err = PTR_ERR(dentry);
2056                 goto out_unlock_dir;
2057         }
2058
2059         if (!dentry->d_inode) {
2060                 err = -ENOENT;
2061                 goto out_dput;
2062         }
2063
2064         inode = dentry->d_inode;
2065         dest = BTRFS_I(inode)->root;
2066         if (!capable(CAP_SYS_ADMIN)){
2067                 /*
2068                  * Regular user.  Only allow this with a special mount
2069                  * option, when the user has write+exec access to the
2070                  * subvol root, and when rmdir(2) would have been
2071                  * allowed.
2072                  *
2073                  * Note that this is _not_ check that the subvol is
2074                  * empty or doesn't contain data that we wouldn't
2075                  * otherwise be able to delete.
2076                  *
2077                  * Users who want to delete empty subvols should try
2078                  * rmdir(2).
2079                  */
2080                 err = -EPERM;
2081                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2082                         goto out_dput;
2083
2084                 /*
2085                  * Do not allow deletion if the parent dir is the same
2086                  * as the dir to be deleted.  That means the ioctl
2087                  * must be called on the dentry referencing the root
2088                  * of the subvol, not a random directory contained
2089                  * within it.
2090                  */
2091                 err = -EINVAL;
2092                 if (root == dest)
2093                         goto out_dput;
2094
2095                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2096                 if (err)
2097                         goto out_dput;
2098
2099                 /* check if subvolume may be deleted by a non-root user */
2100                 err = btrfs_may_delete(dir, dentry, 1);
2101                 if (err)
2102                         goto out_dput;
2103         }
2104
2105         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2106                 err = -EINVAL;
2107                 goto out_dput;
2108         }
2109
2110         mutex_lock(&inode->i_mutex);
2111         err = d_invalidate(dentry);
2112         if (err)
2113                 goto out_unlock;
2114
2115         down_write(&root->fs_info->subvol_sem);
2116
2117         err = may_destroy_subvol(dest);
2118         if (err)
2119                 goto out_up_write;
2120
2121         trans = btrfs_start_transaction(root, 0);
2122         if (IS_ERR(trans)) {
2123                 err = PTR_ERR(trans);
2124                 goto out_up_write;
2125         }
2126         trans->block_rsv = &root->fs_info->global_block_rsv;
2127
2128         ret = btrfs_unlink_subvol(trans, root, dir,
2129                                 dest->root_key.objectid,
2130                                 dentry->d_name.name,
2131                                 dentry->d_name.len);
2132         if (ret) {
2133                 err = ret;
2134                 btrfs_abort_transaction(trans, root, ret);
2135                 goto out_end_trans;
2136         }
2137
2138         btrfs_record_root_in_trans(trans, dest);
2139
2140         memset(&dest->root_item.drop_progress, 0,
2141                 sizeof(dest->root_item.drop_progress));
2142         dest->root_item.drop_level = 0;
2143         btrfs_set_root_refs(&dest->root_item, 0);
2144
2145         if (!xchg(&dest->orphan_item_inserted, 1)) {
2146                 ret = btrfs_insert_orphan_item(trans,
2147                                         root->fs_info->tree_root,
2148                                         dest->root_key.objectid);
2149                 if (ret) {
2150                         btrfs_abort_transaction(trans, root, ret);
2151                         err = ret;
2152                         goto out_end_trans;
2153                 }
2154         }
2155 out_end_trans:
2156         ret = btrfs_end_transaction(trans, root);
2157         if (ret && !err)
2158                 err = ret;
2159         inode->i_flags |= S_DEAD;
2160 out_up_write:
2161         up_write(&root->fs_info->subvol_sem);
2162 out_unlock:
2163         mutex_unlock(&inode->i_mutex);
2164         if (!err) {
2165                 shrink_dcache_sb(root->fs_info->sb);
2166                 btrfs_invalidate_inodes(dest);
2167                 d_delete(dentry);
2168         }
2169 out_dput:
2170         dput(dentry);
2171 out_unlock_dir:
2172         mutex_unlock(&dir->i_mutex);
2173         mnt_drop_write_file(file);
2174 out:
2175         kfree(vol_args);
2176         return err;
2177 }
2178
2179 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2180 {
2181         struct inode *inode = fdentry(file)->d_inode;
2182         struct btrfs_root *root = BTRFS_I(inode)->root;
2183         struct btrfs_ioctl_defrag_range_args *range;
2184         int ret;
2185
2186         if (btrfs_root_readonly(root))
2187                 return -EROFS;
2188
2189         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2190                         1)) {
2191                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2192                 return -EINPROGRESS;
2193         }
2194         ret = mnt_want_write_file(file);
2195         if (ret) {
2196                 atomic_set(&root->fs_info->mutually_exclusive_operation_running,
2197                            0);
2198                 return ret;
2199         }
2200
2201         switch (inode->i_mode & S_IFMT) {
2202         case S_IFDIR:
2203                 if (!capable(CAP_SYS_ADMIN)) {
2204                         ret = -EPERM;
2205                         goto out;
2206                 }
2207                 ret = btrfs_defrag_root(root, 0);
2208                 if (ret)
2209                         goto out;
2210                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2211                 break;
2212         case S_IFREG:
2213                 if (!(file->f_mode & FMODE_WRITE)) {
2214                         ret = -EINVAL;
2215                         goto out;
2216                 }
2217
2218                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2219                 if (!range) {
2220                         ret = -ENOMEM;
2221                         goto out;
2222                 }
2223
2224                 if (argp) {
2225                         if (copy_from_user(range, argp,
2226                                            sizeof(*range))) {
2227                                 ret = -EFAULT;
2228                                 kfree(range);
2229                                 goto out;
2230                         }
2231                         /* compression requires us to start the IO */
2232                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2233                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2234                                 range->extent_thresh = (u32)-1;
2235                         }
2236                 } else {
2237                         /* the rest are all set to zero by kzalloc */
2238                         range->len = (u64)-1;
2239                 }
2240                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2241                                         range, 0, 0);
2242                 if (ret > 0)
2243                         ret = 0;
2244                 kfree(range);
2245                 break;
2246         default:
2247                 ret = -EINVAL;
2248         }
2249 out:
2250         mnt_drop_write_file(file);
2251         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2252         return ret;
2253 }
2254
2255 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2256 {
2257         struct btrfs_ioctl_vol_args *vol_args;
2258         int ret;
2259
2260         if (!capable(CAP_SYS_ADMIN))
2261                 return -EPERM;
2262
2263         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2264                         1)) {
2265                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2266                 return -EINPROGRESS;
2267         }
2268
2269         mutex_lock(&root->fs_info->volume_mutex);
2270         vol_args = memdup_user(arg, sizeof(*vol_args));
2271         if (IS_ERR(vol_args)) {
2272                 ret = PTR_ERR(vol_args);
2273                 goto out;
2274         }
2275
2276         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2277         ret = btrfs_init_new_device(root, vol_args->name);
2278
2279         kfree(vol_args);
2280 out:
2281         mutex_unlock(&root->fs_info->volume_mutex);
2282         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2283         return ret;
2284 }
2285
2286 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2287 {
2288         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2289         struct btrfs_ioctl_vol_args *vol_args;
2290         int ret;
2291
2292         if (!capable(CAP_SYS_ADMIN))
2293                 return -EPERM;
2294
2295         ret = mnt_want_write_file(file);
2296         if (ret)
2297                 return ret;
2298
2299         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2300                         1)) {
2301                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2302                 mnt_drop_write_file(file);
2303                 return -EINPROGRESS;
2304         }
2305
2306         mutex_lock(&root->fs_info->volume_mutex);
2307         vol_args = memdup_user(arg, sizeof(*vol_args));
2308         if (IS_ERR(vol_args)) {
2309                 ret = PTR_ERR(vol_args);
2310                 goto out;
2311         }
2312
2313         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2314         ret = btrfs_rm_device(root, vol_args->name);
2315
2316         kfree(vol_args);
2317 out:
2318         mutex_unlock(&root->fs_info->volume_mutex);
2319         mnt_drop_write_file(file);
2320         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2321         return ret;
2322 }
2323
2324 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2325 {
2326         struct btrfs_ioctl_fs_info_args *fi_args;
2327         struct btrfs_device *device;
2328         struct btrfs_device *next;
2329         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2330         int ret = 0;
2331
2332         if (!capable(CAP_SYS_ADMIN))
2333                 return -EPERM;
2334
2335         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2336         if (!fi_args)
2337                 return -ENOMEM;
2338
2339         fi_args->num_devices = fs_devices->num_devices;
2340         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2341
2342         mutex_lock(&fs_devices->device_list_mutex);
2343         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2344                 if (device->devid > fi_args->max_id)
2345                         fi_args->max_id = device->devid;
2346         }
2347         mutex_unlock(&fs_devices->device_list_mutex);
2348
2349         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2350                 ret = -EFAULT;
2351
2352         kfree(fi_args);
2353         return ret;
2354 }
2355
2356 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2357 {
2358         struct btrfs_ioctl_dev_info_args *di_args;
2359         struct btrfs_device *dev;
2360         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2361         int ret = 0;
2362         char *s_uuid = NULL;
2363         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2364
2365         if (!capable(CAP_SYS_ADMIN))
2366                 return -EPERM;
2367
2368         di_args = memdup_user(arg, sizeof(*di_args));
2369         if (IS_ERR(di_args))
2370                 return PTR_ERR(di_args);
2371
2372         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2373                 s_uuid = di_args->uuid;
2374
2375         mutex_lock(&fs_devices->device_list_mutex);
2376         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2377         mutex_unlock(&fs_devices->device_list_mutex);
2378
2379         if (!dev) {
2380                 ret = -ENODEV;
2381                 goto out;
2382         }
2383
2384         di_args->devid = dev->devid;
2385         di_args->bytes_used = dev->bytes_used;
2386         di_args->total_bytes = dev->total_bytes;
2387         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2388         if (dev->name) {
2389                 struct rcu_string *name;
2390
2391                 rcu_read_lock();
2392                 name = rcu_dereference(dev->name);
2393                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2394                 rcu_read_unlock();
2395                 di_args->path[sizeof(di_args->path) - 1] = 0;
2396         } else {
2397                 di_args->path[0] = '\0';
2398         }
2399
2400 out:
2401         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2402                 ret = -EFAULT;
2403
2404         kfree(di_args);
2405         return ret;
2406 }
2407
2408 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2409                                        u64 off, u64 olen, u64 destoff)
2410 {
2411         struct inode *inode = fdentry(file)->d_inode;
2412         struct btrfs_root *root = BTRFS_I(inode)->root;
2413         struct fd src_file;
2414         struct inode *src;
2415         struct btrfs_trans_handle *trans;
2416         struct btrfs_path *path;
2417         struct extent_buffer *leaf;
2418         char *buf;
2419         struct btrfs_key key;
2420         u32 nritems;
2421         int slot;
2422         int ret;
2423         u64 len = olen;
2424         u64 bs = root->fs_info->sb->s_blocksize;
2425
2426         /*
2427          * TODO:
2428          * - split compressed inline extents.  annoying: we need to
2429          *   decompress into destination's address_space (the file offset
2430          *   may change, so source mapping won't do), then recompress (or
2431          *   otherwise reinsert) a subrange.
2432          * - allow ranges within the same file to be cloned (provided
2433          *   they don't overlap)?
2434          */
2435
2436         /* the destination must be opened for writing */
2437         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2438                 return -EINVAL;
2439
2440         if (btrfs_root_readonly(root))
2441                 return -EROFS;
2442
2443         ret = mnt_want_write_file(file);
2444         if (ret)
2445                 return ret;
2446
2447         src_file = fdget(srcfd);
2448         if (!src_file.file) {
2449                 ret = -EBADF;
2450                 goto out_drop_write;
2451         }
2452
2453         ret = -EXDEV;
2454         if (src_file.file->f_path.mnt != file->f_path.mnt)
2455                 goto out_fput;
2456
2457         src = src_file.file->f_dentry->d_inode;
2458
2459         ret = -EINVAL;
2460         if (src == inode)
2461                 goto out_fput;
2462
2463         /* the src must be open for reading */
2464         if (!(src_file.file->f_mode & FMODE_READ))
2465                 goto out_fput;
2466
2467         /* don't make the dst file partly checksummed */
2468         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2469             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2470                 goto out_fput;
2471
2472         ret = -EISDIR;
2473         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2474                 goto out_fput;
2475
2476         ret = -EXDEV;
2477         if (src->i_sb != inode->i_sb)
2478                 goto out_fput;
2479
2480         ret = -ENOMEM;
2481         buf = vmalloc(btrfs_level_size(root, 0));
2482         if (!buf)
2483                 goto out_fput;
2484
2485         path = btrfs_alloc_path();
2486         if (!path) {
2487                 vfree(buf);
2488                 goto out_fput;
2489         }
2490         path->reada = 2;
2491
2492         if (inode < src) {
2493                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2494                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2495         } else {
2496                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2497                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2498         }
2499
2500         /* determine range to clone */
2501         ret = -EINVAL;
2502         if (off + len > src->i_size || off + len < off)
2503                 goto out_unlock;
2504         if (len == 0)
2505                 olen = len = src->i_size - off;
2506         /* if we extend to eof, continue to block boundary */
2507         if (off + len == src->i_size)
2508                 len = ALIGN(src->i_size, bs) - off;
2509
2510         /* verify the end result is block aligned */
2511         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2512             !IS_ALIGNED(destoff, bs))
2513                 goto out_unlock;
2514
2515         if (destoff > inode->i_size) {
2516                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2517                 if (ret)
2518                         goto out_unlock;
2519         }
2520
2521         /* truncate page cache pages from target inode range */
2522         truncate_inode_pages_range(&inode->i_data, destoff,
2523                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2524
2525         /* do any pending delalloc/csum calc on src, one way or
2526            another, and lock file content */
2527         while (1) {
2528                 struct btrfs_ordered_extent *ordered;
2529                 lock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2530                 ordered = btrfs_lookup_first_ordered_extent(src, off + len - 1);
2531                 if (!ordered &&
2532                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off + len - 1,
2533                                     EXTENT_DELALLOC, 0, NULL))
2534                         break;
2535                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2536                 if (ordered)
2537                         btrfs_put_ordered_extent(ordered);
2538                 btrfs_wait_ordered_range(src, off, len);
2539         }
2540
2541         /* clone data */
2542         key.objectid = btrfs_ino(src);
2543         key.type = BTRFS_EXTENT_DATA_KEY;
2544         key.offset = 0;
2545
2546         while (1) {
2547                 /*
2548                  * note the key will change type as we walk through the
2549                  * tree.
2550                  */
2551                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2552                                 0, 0);
2553                 if (ret < 0)
2554                         goto out;
2555
2556                 nritems = btrfs_header_nritems(path->nodes[0]);
2557                 if (path->slots[0] >= nritems) {
2558                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2559                         if (ret < 0)
2560                                 goto out;
2561                         if (ret > 0)
2562                                 break;
2563                         nritems = btrfs_header_nritems(path->nodes[0]);
2564                 }
2565                 leaf = path->nodes[0];
2566                 slot = path->slots[0];
2567
2568                 btrfs_item_key_to_cpu(leaf, &key, slot);
2569                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2570                     key.objectid != btrfs_ino(src))
2571                         break;
2572
2573                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2574                         struct btrfs_file_extent_item *extent;
2575                         int type;
2576                         u32 size;
2577                         struct btrfs_key new_key;
2578                         u64 disko = 0, diskl = 0;
2579                         u64 datao = 0, datal = 0;
2580                         u8 comp;
2581                         u64 endoff;
2582
2583                         size = btrfs_item_size_nr(leaf, slot);
2584                         read_extent_buffer(leaf, buf,
2585                                            btrfs_item_ptr_offset(leaf, slot),
2586                                            size);
2587
2588                         extent = btrfs_item_ptr(leaf, slot,
2589                                                 struct btrfs_file_extent_item);
2590                         comp = btrfs_file_extent_compression(leaf, extent);
2591                         type = btrfs_file_extent_type(leaf, extent);
2592                         if (type == BTRFS_FILE_EXTENT_REG ||
2593                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2594                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2595                                                                       extent);
2596                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2597                                                                  extent);
2598                                 datao = btrfs_file_extent_offset(leaf, extent);
2599                                 datal = btrfs_file_extent_num_bytes(leaf,
2600                                                                     extent);
2601                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2602                                 /* take upper bound, may be compressed */
2603                                 datal = btrfs_file_extent_ram_bytes(leaf,
2604                                                                     extent);
2605                         }
2606                         btrfs_release_path(path);
2607
2608                         if (key.offset + datal <= off ||
2609                             key.offset >= off + len - 1)
2610                                 goto next;
2611
2612                         memcpy(&new_key, &key, sizeof(new_key));
2613                         new_key.objectid = btrfs_ino(inode);
2614                         if (off <= key.offset)
2615                                 new_key.offset = key.offset + destoff - off;
2616                         else
2617                                 new_key.offset = destoff;
2618
2619                         /*
2620                          * 1 - adjusting old extent (we may have to split it)
2621                          * 1 - add new extent
2622                          * 1 - inode update
2623                          */
2624                         trans = btrfs_start_transaction(root, 3);
2625                         if (IS_ERR(trans)) {
2626                                 ret = PTR_ERR(trans);
2627                                 goto out;
2628                         }
2629
2630                         if (type == BTRFS_FILE_EXTENT_REG ||
2631                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2632                                 /*
2633                                  *    a  | --- range to clone ---|  b
2634                                  * | ------------- extent ------------- |
2635                                  */
2636
2637                                 /* substract range b */
2638                                 if (key.offset + datal > off + len)
2639                                         datal = off + len - key.offset;
2640
2641                                 /* substract range a */
2642                                 if (off > key.offset) {
2643                                         datao += off - key.offset;
2644                                         datal -= off - key.offset;
2645                                 }
2646
2647                                 ret = btrfs_drop_extents(trans, root, inode,
2648                                                          new_key.offset,
2649                                                          new_key.offset + datal,
2650                                                          1);
2651                                 if (ret) {
2652                                         btrfs_abort_transaction(trans, root,
2653                                                                 ret);
2654                                         btrfs_end_transaction(trans, root);
2655                                         goto out;
2656                                 }
2657
2658                                 ret = btrfs_insert_empty_item(trans, root, path,
2659                                                               &new_key, size);
2660                                 if (ret) {
2661                                         btrfs_abort_transaction(trans, root,
2662                                                                 ret);
2663                                         btrfs_end_transaction(trans, root);
2664                                         goto out;
2665                                 }
2666
2667                                 leaf = path->nodes[0];
2668                                 slot = path->slots[0];
2669                                 write_extent_buffer(leaf, buf,
2670                                             btrfs_item_ptr_offset(leaf, slot),
2671                                             size);
2672
2673                                 extent = btrfs_item_ptr(leaf, slot,
2674                                                 struct btrfs_file_extent_item);
2675
2676                                 /* disko == 0 means it's a hole */
2677                                 if (!disko)
2678                                         datao = 0;
2679
2680                                 btrfs_set_file_extent_offset(leaf, extent,
2681                                                              datao);
2682                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2683                                                                 datal);
2684                                 if (disko) {
2685                                         inode_add_bytes(inode, datal);
2686                                         ret = btrfs_inc_extent_ref(trans, root,
2687                                                         disko, diskl, 0,
2688                                                         root->root_key.objectid,
2689                                                         btrfs_ino(inode),
2690                                                         new_key.offset - datao,
2691                                                         0);
2692                                         if (ret) {
2693                                                 btrfs_abort_transaction(trans,
2694                                                                         root,
2695                                                                         ret);
2696                                                 btrfs_end_transaction(trans,
2697                                                                       root);
2698                                                 goto out;
2699
2700                                         }
2701                                 }
2702                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2703                                 u64 skip = 0;
2704                                 u64 trim = 0;
2705                                 if (off > key.offset) {
2706                                         skip = off - key.offset;
2707                                         new_key.offset += skip;
2708                                 }
2709
2710                                 if (key.offset + datal > off + len)
2711                                         trim = key.offset + datal - (off + len);
2712
2713                                 if (comp && (skip || trim)) {
2714                                         ret = -EINVAL;
2715                                         btrfs_end_transaction(trans, root);
2716                                         goto out;
2717                                 }
2718                                 size -= skip + trim;
2719                                 datal -= skip + trim;
2720
2721                                 ret = btrfs_drop_extents(trans, root, inode,
2722                                                          new_key.offset,
2723                                                          new_key.offset + datal,
2724                                                          1);
2725                                 if (ret) {
2726                                         btrfs_abort_transaction(trans, root,
2727                                                                 ret);
2728                                         btrfs_end_transaction(trans, root);
2729                                         goto out;
2730                                 }
2731
2732                                 ret = btrfs_insert_empty_item(trans, root, path,
2733                                                               &new_key, size);
2734                                 if (ret) {
2735                                         btrfs_abort_transaction(trans, root,
2736                                                                 ret);
2737                                         btrfs_end_transaction(trans, root);
2738                                         goto out;
2739                                 }
2740
2741                                 if (skip) {
2742                                         u32 start =
2743                                           btrfs_file_extent_calc_inline_size(0);
2744                                         memmove(buf+start, buf+start+skip,
2745                                                 datal);
2746                                 }
2747
2748                                 leaf = path->nodes[0];
2749                                 slot = path->slots[0];
2750                                 write_extent_buffer(leaf, buf,
2751                                             btrfs_item_ptr_offset(leaf, slot),
2752                                             size);
2753                                 inode_add_bytes(inode, datal);
2754                         }
2755
2756                         btrfs_mark_buffer_dirty(leaf);
2757                         btrfs_release_path(path);
2758
2759                         inode_inc_iversion(inode);
2760                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2761
2762                         /*
2763                          * we round up to the block size at eof when
2764                          * determining which extents to clone above,
2765                          * but shouldn't round up the file size
2766                          */
2767                         endoff = new_key.offset + datal;
2768                         if (endoff > destoff+olen)
2769                                 endoff = destoff+olen;
2770                         if (endoff > inode->i_size)
2771                                 btrfs_i_size_write(inode, endoff);
2772
2773                         ret = btrfs_update_inode(trans, root, inode);
2774                         if (ret) {
2775                                 btrfs_abort_transaction(trans, root, ret);
2776                                 btrfs_end_transaction(trans, root);
2777                                 goto out;
2778                         }
2779                         ret = btrfs_end_transaction(trans, root);
2780                 }
2781 next:
2782                 btrfs_release_path(path);
2783                 key.offset++;
2784         }
2785         ret = 0;
2786 out:
2787         btrfs_release_path(path);
2788         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2789 out_unlock:
2790         mutex_unlock(&src->i_mutex);
2791         mutex_unlock(&inode->i_mutex);
2792         vfree(buf);
2793         btrfs_free_path(path);
2794 out_fput:
2795         fdput(src_file);
2796 out_drop_write:
2797         mnt_drop_write_file(file);
2798         return ret;
2799 }
2800
2801 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2802 {
2803         struct btrfs_ioctl_clone_range_args args;
2804
2805         if (copy_from_user(&args, argp, sizeof(args)))
2806                 return -EFAULT;
2807         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2808                                  args.src_length, args.dest_offset);
2809 }
2810
2811 /*
2812  * there are many ways the trans_start and trans_end ioctls can lead
2813  * to deadlocks.  They should only be used by applications that
2814  * basically own the machine, and have a very in depth understanding
2815  * of all the possible deadlocks and enospc problems.
2816  */
2817 static long btrfs_ioctl_trans_start(struct file *file)
2818 {
2819         struct inode *inode = fdentry(file)->d_inode;
2820         struct btrfs_root *root = BTRFS_I(inode)->root;
2821         struct btrfs_trans_handle *trans;
2822         int ret;
2823
2824         ret = -EPERM;
2825         if (!capable(CAP_SYS_ADMIN))
2826                 goto out;
2827
2828         ret = -EINPROGRESS;
2829         if (file->private_data)
2830                 goto out;
2831
2832         ret = -EROFS;
2833         if (btrfs_root_readonly(root))
2834                 goto out;
2835
2836         ret = mnt_want_write_file(file);
2837         if (ret)
2838                 goto out;
2839
2840         atomic_inc(&root->fs_info->open_ioctl_trans);
2841
2842         ret = -ENOMEM;
2843         trans = btrfs_start_ioctl_transaction(root);
2844         if (IS_ERR(trans))
2845                 goto out_drop;
2846
2847         file->private_data = trans;
2848         return 0;
2849
2850 out_drop:
2851         atomic_dec(&root->fs_info->open_ioctl_trans);
2852         mnt_drop_write_file(file);
2853 out:
2854         return ret;
2855 }
2856
2857 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2858 {
2859         struct inode *inode = fdentry(file)->d_inode;
2860         struct btrfs_root *root = BTRFS_I(inode)->root;
2861         struct btrfs_root *new_root;
2862         struct btrfs_dir_item *di;
2863         struct btrfs_trans_handle *trans;
2864         struct btrfs_path *path;
2865         struct btrfs_key location;
2866         struct btrfs_disk_key disk_key;
2867         u64 objectid = 0;
2868         u64 dir_id;
2869         int ret;
2870
2871         if (!capable(CAP_SYS_ADMIN))
2872                 return -EPERM;
2873
2874         ret = mnt_want_write_file(file);
2875         if (ret)
2876                 return ret;
2877
2878         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
2879                 ret = -EFAULT;
2880                 goto out;
2881         }
2882
2883         if (!objectid)
2884                 objectid = root->root_key.objectid;
2885
2886         location.objectid = objectid;
2887         location.type = BTRFS_ROOT_ITEM_KEY;
2888         location.offset = (u64)-1;
2889
2890         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2891         if (IS_ERR(new_root)) {
2892                 ret = PTR_ERR(new_root);
2893                 goto out;
2894         }
2895
2896         if (btrfs_root_refs(&new_root->root_item) == 0) {
2897                 ret = -ENOENT;
2898                 goto out;
2899         }
2900
2901         path = btrfs_alloc_path();
2902         if (!path) {
2903                 ret = -ENOMEM;
2904                 goto out;
2905         }
2906         path->leave_spinning = 1;
2907
2908         trans = btrfs_start_transaction(root, 1);
2909         if (IS_ERR(trans)) {
2910                 btrfs_free_path(path);
2911                 ret = PTR_ERR(trans);
2912                 goto out;
2913         }
2914
2915         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2916         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2917                                    dir_id, "default", 7, 1);
2918         if (IS_ERR_OR_NULL(di)) {
2919                 btrfs_free_path(path);
2920                 btrfs_end_transaction(trans, root);
2921                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2922                        "this isn't going to work\n");
2923                 ret = -ENOENT;
2924                 goto out;
2925         }
2926
2927         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2928         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2929         btrfs_mark_buffer_dirty(path->nodes[0]);
2930         btrfs_free_path(path);
2931
2932         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
2933         btrfs_end_transaction(trans, root);
2934 out:
2935         mnt_drop_write_file(file);
2936         return ret;
2937 }
2938
2939 void btrfs_get_block_group_info(struct list_head *groups_list,
2940                                 struct btrfs_ioctl_space_info *space)
2941 {
2942         struct btrfs_block_group_cache *block_group;
2943
2944         space->total_bytes = 0;
2945         space->used_bytes = 0;
2946         space->flags = 0;
2947         list_for_each_entry(block_group, groups_list, list) {
2948                 space->flags = block_group->flags;
2949                 space->total_bytes += block_group->key.offset;
2950                 space->used_bytes +=
2951                         btrfs_block_group_used(&block_group->item);
2952         }
2953 }
2954
2955 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2956 {
2957         struct btrfs_ioctl_space_args space_args;
2958         struct btrfs_ioctl_space_info space;
2959         struct btrfs_ioctl_space_info *dest;
2960         struct btrfs_ioctl_space_info *dest_orig;
2961         struct btrfs_ioctl_space_info __user *user_dest;
2962         struct btrfs_space_info *info;
2963         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2964                        BTRFS_BLOCK_GROUP_SYSTEM,
2965                        BTRFS_BLOCK_GROUP_METADATA,
2966                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2967         int num_types = 4;
2968         int alloc_size;
2969         int ret = 0;
2970         u64 slot_count = 0;
2971         int i, c;
2972
2973         if (copy_from_user(&space_args,
2974                            (struct btrfs_ioctl_space_args __user *)arg,
2975                            sizeof(space_args)))
2976                 return -EFAULT;
2977
2978         for (i = 0; i < num_types; i++) {
2979                 struct btrfs_space_info *tmp;
2980
2981                 info = NULL;
2982                 rcu_read_lock();
2983                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2984                                         list) {
2985                         if (tmp->flags == types[i]) {
2986                                 info = tmp;
2987                                 break;
2988                         }
2989                 }
2990                 rcu_read_unlock();
2991
2992                 if (!info)
2993                         continue;
2994
2995                 down_read(&info->groups_sem);
2996                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2997                         if (!list_empty(&info->block_groups[c]))
2998                                 slot_count++;
2999                 }
3000                 up_read(&info->groups_sem);
3001         }
3002
3003         /* space_slots == 0 means they are asking for a count */
3004         if (space_args.space_slots == 0) {
3005                 space_args.total_spaces = slot_count;
3006                 goto out;
3007         }
3008
3009         slot_count = min_t(u64, space_args.space_slots, slot_count);
3010
3011         alloc_size = sizeof(*dest) * slot_count;
3012
3013         /* we generally have at most 6 or so space infos, one for each raid
3014          * level.  So, a whole page should be more than enough for everyone
3015          */
3016         if (alloc_size > PAGE_CACHE_SIZE)
3017                 return -ENOMEM;
3018
3019         space_args.total_spaces = 0;
3020         dest = kmalloc(alloc_size, GFP_NOFS);
3021         if (!dest)
3022                 return -ENOMEM;
3023         dest_orig = dest;
3024
3025         /* now we have a buffer to copy into */
3026         for (i = 0; i < num_types; i++) {
3027                 struct btrfs_space_info *tmp;
3028
3029                 if (!slot_count)
3030                         break;
3031
3032                 info = NULL;
3033                 rcu_read_lock();
3034                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3035                                         list) {
3036                         if (tmp->flags == types[i]) {
3037                                 info = tmp;
3038                                 break;
3039                         }
3040                 }
3041                 rcu_read_unlock();
3042
3043                 if (!info)
3044                         continue;
3045                 down_read(&info->groups_sem);
3046                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3047                         if (!list_empty(&info->block_groups[c])) {
3048                                 btrfs_get_block_group_info(
3049                                         &info->block_groups[c], &space);
3050                                 memcpy(dest, &space, sizeof(space));
3051                                 dest++;
3052                                 space_args.total_spaces++;
3053                                 slot_count--;
3054                         }
3055                         if (!slot_count)
3056                                 break;
3057                 }
3058                 up_read(&info->groups_sem);
3059         }
3060
3061         user_dest = (struct btrfs_ioctl_space_info __user *)
3062                 (arg + sizeof(struct btrfs_ioctl_space_args));
3063
3064         if (copy_to_user(user_dest, dest_orig, alloc_size))
3065                 ret = -EFAULT;
3066
3067         kfree(dest_orig);
3068 out:
3069         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3070                 ret = -EFAULT;
3071
3072         return ret;
3073 }
3074
3075 /*
3076  * there are many ways the trans_start and trans_end ioctls can lead
3077  * to deadlocks.  They should only be used by applications that
3078  * basically own the machine, and have a very in depth understanding
3079  * of all the possible deadlocks and enospc problems.
3080  */
3081 long btrfs_ioctl_trans_end(struct file *file)
3082 {
3083         struct inode *inode = fdentry(file)->d_inode;
3084         struct btrfs_root *root = BTRFS_I(inode)->root;
3085         struct btrfs_trans_handle *trans;
3086
3087         trans = file->private_data;
3088         if (!trans)
3089                 return -EINVAL;
3090         file->private_data = NULL;
3091
3092         btrfs_end_transaction(trans, root);
3093
3094         atomic_dec(&root->fs_info->open_ioctl_trans);
3095
3096         mnt_drop_write_file(file);
3097         return 0;
3098 }
3099
3100 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3101                                             void __user *argp)
3102 {
3103         struct btrfs_trans_handle *trans;
3104         u64 transid;
3105         int ret;
3106
3107         trans = btrfs_attach_transaction(root);
3108         if (IS_ERR(trans)) {
3109                 if (PTR_ERR(trans) != -ENOENT)
3110                         return PTR_ERR(trans);
3111
3112                 /* No running transaction, don't bother */
3113                 transid = root->fs_info->last_trans_committed;
3114                 goto out;
3115         }
3116         transid = trans->transid;
3117         ret = btrfs_commit_transaction_async(trans, root, 0);
3118         if (ret) {
3119                 btrfs_end_transaction(trans, root);
3120                 return ret;
3121         }
3122 out:
3123         if (argp)
3124                 if (copy_to_user(argp, &transid, sizeof(transid)))
3125                         return -EFAULT;
3126         return 0;
3127 }
3128
3129 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3130                                            void __user *argp)
3131 {
3132         u64 transid;
3133
3134         if (argp) {
3135                 if (copy_from_user(&transid, argp, sizeof(transid)))
3136                         return -EFAULT;
3137         } else {
3138                 transid = 0;  /* current trans */
3139         }
3140         return btrfs_wait_for_commit(root, transid);
3141 }
3142
3143 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3144 {
3145         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3146         struct btrfs_ioctl_scrub_args *sa;
3147         int ret;
3148
3149         if (!capable(CAP_SYS_ADMIN))
3150                 return -EPERM;
3151
3152         sa = memdup_user(arg, sizeof(*sa));
3153         if (IS_ERR(sa))
3154                 return PTR_ERR(sa);
3155
3156         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3157                 ret = mnt_want_write_file(file);
3158                 if (ret)
3159                         goto out;
3160         }
3161
3162         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3163                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3164                               0);
3165
3166         if (copy_to_user(arg, sa, sizeof(*sa)))
3167                 ret = -EFAULT;
3168
3169         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3170                 mnt_drop_write_file(file);
3171 out:
3172         kfree(sa);
3173         return ret;
3174 }
3175
3176 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3177 {
3178         if (!capable(CAP_SYS_ADMIN))
3179                 return -EPERM;
3180
3181         return btrfs_scrub_cancel(root->fs_info);
3182 }
3183
3184 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3185                                        void __user *arg)
3186 {
3187         struct btrfs_ioctl_scrub_args *sa;
3188         int ret;
3189
3190         if (!capable(CAP_SYS_ADMIN))
3191                 return -EPERM;
3192
3193         sa = memdup_user(arg, sizeof(*sa));
3194         if (IS_ERR(sa))
3195                 return PTR_ERR(sa);
3196
3197         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3198
3199         if (copy_to_user(arg, sa, sizeof(*sa)))
3200                 ret = -EFAULT;
3201
3202         kfree(sa);
3203         return ret;
3204 }
3205
3206 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3207                                       void __user *arg)
3208 {
3209         struct btrfs_ioctl_get_dev_stats *sa;
3210         int ret;
3211
3212         sa = memdup_user(arg, sizeof(*sa));
3213         if (IS_ERR(sa))
3214                 return PTR_ERR(sa);
3215
3216         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3217                 kfree(sa);
3218                 return -EPERM;
3219         }
3220
3221         ret = btrfs_get_dev_stats(root, sa);
3222
3223         if (copy_to_user(arg, sa, sizeof(*sa)))
3224                 ret = -EFAULT;
3225
3226         kfree(sa);
3227         return ret;
3228 }
3229
3230 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3231 {
3232         struct btrfs_ioctl_dev_replace_args *p;
3233         int ret;
3234
3235         if (!capable(CAP_SYS_ADMIN))
3236                 return -EPERM;
3237
3238         p = memdup_user(arg, sizeof(*p));
3239         if (IS_ERR(p))
3240                 return PTR_ERR(p);
3241
3242         switch (p->cmd) {
3243         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3244                 if (atomic_xchg(
3245                         &root->fs_info->mutually_exclusive_operation_running,
3246                         1)) {
3247                         pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3248                         ret = -EINPROGRESS;
3249                 } else {
3250                         ret = btrfs_dev_replace_start(root, p);
3251                         atomic_set(
3252                          &root->fs_info->mutually_exclusive_operation_running,
3253                          0);
3254                 }
3255                 break;
3256         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3257                 btrfs_dev_replace_status(root->fs_info, p);
3258                 ret = 0;
3259                 break;
3260         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3261                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3262                 break;
3263         default:
3264                 ret = -EINVAL;
3265                 break;
3266         }
3267
3268         if (copy_to_user(arg, p, sizeof(*p)))
3269                 ret = -EFAULT;
3270
3271         kfree(p);
3272         return ret;
3273 }
3274
3275 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3276 {
3277         int ret = 0;
3278         int i;
3279         u64 rel_ptr;
3280         int size;
3281         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3282         struct inode_fs_paths *ipath = NULL;
3283         struct btrfs_path *path;
3284
3285         if (!capable(CAP_SYS_ADMIN))
3286                 return -EPERM;
3287
3288         path = btrfs_alloc_path();
3289         if (!path) {
3290                 ret = -ENOMEM;
3291                 goto out;
3292         }
3293
3294         ipa = memdup_user(arg, sizeof(*ipa));
3295         if (IS_ERR(ipa)) {
3296                 ret = PTR_ERR(ipa);
3297                 ipa = NULL;
3298                 goto out;
3299         }
3300
3301         size = min_t(u32, ipa->size, 4096);
3302         ipath = init_ipath(size, root, path);
3303         if (IS_ERR(ipath)) {
3304                 ret = PTR_ERR(ipath);
3305                 ipath = NULL;
3306                 goto out;
3307         }
3308
3309         ret = paths_from_inode(ipa->inum, ipath);
3310         if (ret < 0)
3311                 goto out;
3312
3313         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3314                 rel_ptr = ipath->fspath->val[i] -
3315                           (u64)(unsigned long)ipath->fspath->val;
3316                 ipath->fspath->val[i] = rel_ptr;
3317         }
3318
3319         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3320                            (void *)(unsigned long)ipath->fspath, size);
3321         if (ret) {
3322                 ret = -EFAULT;
3323                 goto out;
3324         }
3325
3326 out:
3327         btrfs_free_path(path);
3328         free_ipath(ipath);
3329         kfree(ipa);
3330
3331         return ret;
3332 }
3333
3334 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3335 {
3336         struct btrfs_data_container *inodes = ctx;
3337         const size_t c = 3 * sizeof(u64);
3338
3339         if (inodes->bytes_left >= c) {
3340                 inodes->bytes_left -= c;
3341                 inodes->val[inodes->elem_cnt] = inum;
3342                 inodes->val[inodes->elem_cnt + 1] = offset;
3343                 inodes->val[inodes->elem_cnt + 2] = root;
3344                 inodes->elem_cnt += 3;
3345         } else {
3346                 inodes->bytes_missing += c - inodes->bytes_left;
3347                 inodes->bytes_left = 0;
3348                 inodes->elem_missed += 3;
3349         }
3350
3351         return 0;
3352 }
3353
3354 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3355                                         void __user *arg)
3356 {
3357         int ret = 0;
3358         int size;
3359         struct btrfs_ioctl_logical_ino_args *loi;
3360         struct btrfs_data_container *inodes = NULL;
3361         struct btrfs_path *path = NULL;
3362
3363         if (!capable(CAP_SYS_ADMIN))
3364                 return -EPERM;
3365
3366         loi = memdup_user(arg, sizeof(*loi));
3367         if (IS_ERR(loi)) {
3368                 ret = PTR_ERR(loi);
3369                 loi = NULL;
3370                 goto out;
3371         }
3372
3373         path = btrfs_alloc_path();
3374         if (!path) {
3375                 ret = -ENOMEM;
3376                 goto out;
3377         }
3378
3379         size = min_t(u32, loi->size, 64 * 1024);
3380         inodes = init_data_container(size);
3381         if (IS_ERR(inodes)) {
3382                 ret = PTR_ERR(inodes);
3383                 inodes = NULL;
3384                 goto out;
3385         }
3386
3387         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3388                                           build_ino_list, inodes);
3389         if (ret == -EINVAL)
3390                 ret = -ENOENT;
3391         if (ret < 0)
3392                 goto out;
3393
3394         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3395                            (void *)(unsigned long)inodes, size);
3396         if (ret)
3397                 ret = -EFAULT;
3398
3399 out:
3400         btrfs_free_path(path);
3401         vfree(inodes);
3402         kfree(loi);
3403
3404         return ret;
3405 }
3406
3407 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3408                                struct btrfs_ioctl_balance_args *bargs)
3409 {
3410         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3411
3412         bargs->flags = bctl->flags;
3413
3414         if (atomic_read(&fs_info->balance_running))
3415                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3416         if (atomic_read(&fs_info->balance_pause_req))
3417                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3418         if (atomic_read(&fs_info->balance_cancel_req))
3419                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3420
3421         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3422         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3423         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3424
3425         if (lock) {
3426                 spin_lock(&fs_info->balance_lock);
3427                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3428                 spin_unlock(&fs_info->balance_lock);
3429         } else {
3430                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3431         }
3432 }
3433
3434 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3435 {
3436         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3437         struct btrfs_fs_info *fs_info = root->fs_info;
3438         struct btrfs_ioctl_balance_args *bargs;
3439         struct btrfs_balance_control *bctl;
3440         int ret;
3441         int need_to_clear_lock = 0;
3442
3443         if (!capable(CAP_SYS_ADMIN))
3444                 return -EPERM;
3445
3446         ret = mnt_want_write_file(file);
3447         if (ret)
3448                 return ret;
3449
3450         mutex_lock(&fs_info->volume_mutex);
3451         mutex_lock(&fs_info->balance_mutex);
3452
3453         if (arg) {
3454                 bargs = memdup_user(arg, sizeof(*bargs));
3455                 if (IS_ERR(bargs)) {
3456                         ret = PTR_ERR(bargs);
3457                         goto out;
3458                 }
3459
3460                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3461                         if (!fs_info->balance_ctl) {
3462                                 ret = -ENOTCONN;
3463                                 goto out_bargs;
3464                         }
3465
3466                         bctl = fs_info->balance_ctl;
3467                         spin_lock(&fs_info->balance_lock);
3468                         bctl->flags |= BTRFS_BALANCE_RESUME;
3469                         spin_unlock(&fs_info->balance_lock);
3470
3471                         goto do_balance;
3472                 }
3473         } else {
3474                 bargs = NULL;
3475         }
3476
3477         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
3478                         1)) {
3479                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3480                 ret = -EINPROGRESS;
3481                 goto out_bargs;
3482         }
3483         need_to_clear_lock = 1;
3484
3485         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3486         if (!bctl) {
3487                 ret = -ENOMEM;
3488                 goto out_bargs;
3489         }
3490
3491         bctl->fs_info = fs_info;
3492         if (arg) {
3493                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3494                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3495                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3496
3497                 bctl->flags = bargs->flags;
3498         } else {
3499                 /* balance everything - no filters */
3500                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3501         }
3502
3503 do_balance:
3504         ret = btrfs_balance(bctl, bargs);
3505         /*
3506          * bctl is freed in __cancel_balance or in free_fs_info if
3507          * restriper was paused all the way until unmount
3508          */
3509         if (arg) {
3510                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3511                         ret = -EFAULT;
3512         }
3513
3514 out_bargs:
3515         kfree(bargs);
3516 out:
3517         if (need_to_clear_lock)
3518                 atomic_set(&root->fs_info->mutually_exclusive_operation_running,
3519                            0);
3520         mutex_unlock(&fs_info->balance_mutex);
3521         mutex_unlock(&fs_info->volume_mutex);
3522         mnt_drop_write_file(file);
3523         return ret;
3524 }
3525
3526 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3527 {
3528         if (!capable(CAP_SYS_ADMIN))
3529                 return -EPERM;
3530
3531         switch (cmd) {
3532         case BTRFS_BALANCE_CTL_PAUSE:
3533                 return btrfs_pause_balance(root->fs_info);
3534         case BTRFS_BALANCE_CTL_CANCEL:
3535                 return btrfs_cancel_balance(root->fs_info);
3536         }
3537
3538         return -EINVAL;
3539 }
3540
3541 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3542                                          void __user *arg)
3543 {
3544         struct btrfs_fs_info *fs_info = root->fs_info;
3545         struct btrfs_ioctl_balance_args *bargs;
3546         int ret = 0;
3547
3548         if (!capable(CAP_SYS_ADMIN))
3549                 return -EPERM;
3550
3551         mutex_lock(&fs_info->balance_mutex);
3552         if (!fs_info->balance_ctl) {
3553                 ret = -ENOTCONN;
3554                 goto out;
3555         }
3556
3557         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3558         if (!bargs) {
3559                 ret = -ENOMEM;
3560                 goto out;
3561         }
3562
3563         update_ioctl_balance_args(fs_info, 1, bargs);
3564
3565         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3566                 ret = -EFAULT;
3567
3568         kfree(bargs);
3569 out:
3570         mutex_unlock(&fs_info->balance_mutex);
3571         return ret;
3572 }
3573
3574 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
3575 {
3576         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3577         struct btrfs_ioctl_quota_ctl_args *sa;
3578         struct btrfs_trans_handle *trans = NULL;
3579         int ret;
3580         int err;
3581
3582         if (!capable(CAP_SYS_ADMIN))
3583                 return -EPERM;
3584
3585         ret = mnt_want_write_file(file);
3586         if (ret)
3587                 return ret;
3588
3589         sa = memdup_user(arg, sizeof(*sa));
3590         if (IS_ERR(sa)) {
3591                 ret = PTR_ERR(sa);
3592                 goto drop_write;
3593         }
3594
3595         if (sa->cmd != BTRFS_QUOTA_CTL_RESCAN) {
3596                 trans = btrfs_start_transaction(root, 2);
3597                 if (IS_ERR(trans)) {
3598                         ret = PTR_ERR(trans);
3599                         goto out;
3600                 }
3601         }
3602
3603         switch (sa->cmd) {
3604         case BTRFS_QUOTA_CTL_ENABLE:
3605                 ret = btrfs_quota_enable(trans, root->fs_info);
3606                 break;
3607         case BTRFS_QUOTA_CTL_DISABLE:
3608                 ret = btrfs_quota_disable(trans, root->fs_info);
3609                 break;
3610         case BTRFS_QUOTA_CTL_RESCAN:
3611                 ret = btrfs_quota_rescan(root->fs_info);
3612                 break;
3613         default:
3614                 ret = -EINVAL;
3615                 break;
3616         }
3617
3618         if (copy_to_user(arg, sa, sizeof(*sa)))
3619                 ret = -EFAULT;
3620
3621         if (trans) {
3622                 err = btrfs_commit_transaction(trans, root);
3623                 if (err && !ret)
3624                         ret = err;
3625         }
3626 out:
3627         kfree(sa);
3628 drop_write:
3629         mnt_drop_write_file(file);
3630         return ret;
3631 }
3632
3633 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
3634 {
3635         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3636         struct btrfs_ioctl_qgroup_assign_args *sa;
3637         struct btrfs_trans_handle *trans;
3638         int ret;
3639         int err;
3640
3641         if (!capable(CAP_SYS_ADMIN))
3642                 return -EPERM;
3643
3644         ret = mnt_want_write_file(file);
3645         if (ret)
3646                 return ret;
3647
3648         sa = memdup_user(arg, sizeof(*sa));
3649         if (IS_ERR(sa)) {
3650                 ret = PTR_ERR(sa);
3651                 goto drop_write;
3652         }
3653
3654         trans = btrfs_join_transaction(root);
3655         if (IS_ERR(trans)) {
3656                 ret = PTR_ERR(trans);
3657                 goto out;
3658         }
3659
3660         /* FIXME: check if the IDs really exist */
3661         if (sa->assign) {
3662                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
3663                                                 sa->src, sa->dst);
3664         } else {
3665                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
3666                                                 sa->src, sa->dst);
3667         }
3668
3669         err = btrfs_end_transaction(trans, root);
3670         if (err && !ret)
3671                 ret = err;
3672
3673 out:
3674         kfree(sa);
3675 drop_write:
3676         mnt_drop_write_file(file);
3677         return ret;
3678 }
3679
3680 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
3681 {
3682         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3683         struct btrfs_ioctl_qgroup_create_args *sa;
3684         struct btrfs_trans_handle *trans;
3685         int ret;
3686         int err;
3687
3688         if (!capable(CAP_SYS_ADMIN))
3689                 return -EPERM;
3690
3691         ret = mnt_want_write_file(file);
3692         if (ret)
3693                 return ret;
3694
3695         sa = memdup_user(arg, sizeof(*sa));
3696         if (IS_ERR(sa)) {
3697                 ret = PTR_ERR(sa);
3698                 goto drop_write;
3699         }
3700
3701         trans = btrfs_join_transaction(root);
3702         if (IS_ERR(trans)) {
3703                 ret = PTR_ERR(trans);
3704                 goto out;
3705         }
3706
3707         /* FIXME: check if the IDs really exist */
3708         if (sa->create) {
3709                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
3710                                           NULL);
3711         } else {
3712                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
3713         }
3714
3715         err = btrfs_end_transaction(trans, root);
3716         if (err && !ret)
3717                 ret = err;
3718
3719 out:
3720         kfree(sa);
3721 drop_write:
3722         mnt_drop_write_file(file);
3723         return ret;
3724 }
3725
3726 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
3727 {
3728         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3729         struct btrfs_ioctl_qgroup_limit_args *sa;
3730         struct btrfs_trans_handle *trans;
3731         int ret;
3732         int err;
3733         u64 qgroupid;
3734
3735         if (!capable(CAP_SYS_ADMIN))
3736                 return -EPERM;
3737
3738         ret = mnt_want_write_file(file);
3739         if (ret)
3740                 return ret;
3741
3742         sa = memdup_user(arg, sizeof(*sa));
3743         if (IS_ERR(sa)) {
3744                 ret = PTR_ERR(sa);
3745                 goto drop_write;
3746         }
3747
3748         trans = btrfs_join_transaction(root);
3749         if (IS_ERR(trans)) {
3750                 ret = PTR_ERR(trans);
3751                 goto out;
3752         }
3753
3754         qgroupid = sa->qgroupid;
3755         if (!qgroupid) {
3756                 /* take the current subvol as qgroup */
3757                 qgroupid = root->root_key.objectid;
3758         }
3759
3760         /* FIXME: check if the IDs really exist */
3761         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
3762
3763         err = btrfs_end_transaction(trans, root);
3764         if (err && !ret)
3765                 ret = err;
3766
3767 out:
3768         kfree(sa);
3769 drop_write:
3770         mnt_drop_write_file(file);
3771         return ret;
3772 }
3773
3774 static long btrfs_ioctl_set_received_subvol(struct file *file,
3775                                             void __user *arg)
3776 {
3777         struct btrfs_ioctl_received_subvol_args *sa = NULL;
3778         struct inode *inode = fdentry(file)->d_inode;
3779         struct btrfs_root *root = BTRFS_I(inode)->root;
3780         struct btrfs_root_item *root_item = &root->root_item;
3781         struct btrfs_trans_handle *trans;
3782         struct timespec ct = CURRENT_TIME;
3783         int ret = 0;
3784
3785         ret = mnt_want_write_file(file);
3786         if (ret < 0)
3787                 return ret;
3788
3789         down_write(&root->fs_info->subvol_sem);
3790
3791         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
3792                 ret = -EINVAL;
3793                 goto out;
3794         }
3795
3796         if (btrfs_root_readonly(root)) {
3797                 ret = -EROFS;
3798                 goto out;
3799         }
3800
3801         if (!inode_owner_or_capable(inode)) {
3802                 ret = -EACCES;
3803                 goto out;
3804         }
3805
3806         sa = memdup_user(arg, sizeof(*sa));
3807         if (IS_ERR(sa)) {
3808                 ret = PTR_ERR(sa);
3809                 sa = NULL;
3810                 goto out;
3811         }
3812
3813         trans = btrfs_start_transaction(root, 1);
3814         if (IS_ERR(trans)) {
3815                 ret = PTR_ERR(trans);
3816                 trans = NULL;
3817                 goto out;
3818         }
3819
3820         sa->rtransid = trans->transid;
3821         sa->rtime.sec = ct.tv_sec;
3822         sa->rtime.nsec = ct.tv_nsec;
3823
3824         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
3825         btrfs_set_root_stransid(root_item, sa->stransid);
3826         btrfs_set_root_rtransid(root_item, sa->rtransid);
3827         root_item->stime.sec = cpu_to_le64(sa->stime.sec);
3828         root_item->stime.nsec = cpu_to_le32(sa->stime.nsec);
3829         root_item->rtime.sec = cpu_to_le64(sa->rtime.sec);
3830         root_item->rtime.nsec = cpu_to_le32(sa->rtime.nsec);
3831
3832         ret = btrfs_update_root(trans, root->fs_info->tree_root,
3833                                 &root->root_key, &root->root_item);
3834         if (ret < 0) {
3835                 btrfs_end_transaction(trans, root);
3836                 trans = NULL;
3837                 goto out;
3838         } else {
3839                 ret = btrfs_commit_transaction(trans, root);
3840                 if (ret < 0)
3841                         goto out;
3842         }
3843
3844         ret = copy_to_user(arg, sa, sizeof(*sa));
3845         if (ret)
3846                 ret = -EFAULT;
3847
3848 out:
3849         kfree(sa);
3850         up_write(&root->fs_info->subvol_sem);
3851         mnt_drop_write_file(file);
3852         return ret;
3853 }
3854
3855 long btrfs_ioctl(struct file *file, unsigned int
3856                 cmd, unsigned long arg)
3857 {
3858         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3859         void __user *argp = (void __user *)arg;
3860
3861         switch (cmd) {
3862         case FS_IOC_GETFLAGS:
3863                 return btrfs_ioctl_getflags(file, argp);
3864         case FS_IOC_SETFLAGS:
3865                 return btrfs_ioctl_setflags(file, argp);
3866         case FS_IOC_GETVERSION:
3867                 return btrfs_ioctl_getversion(file, argp);
3868         case FITRIM:
3869                 return btrfs_ioctl_fitrim(file, argp);
3870         case BTRFS_IOC_SNAP_CREATE:
3871                 return btrfs_ioctl_snap_create(file, argp, 0);
3872         case BTRFS_IOC_SNAP_CREATE_V2:
3873                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3874         case BTRFS_IOC_SUBVOL_CREATE:
3875                 return btrfs_ioctl_snap_create(file, argp, 1);
3876         case BTRFS_IOC_SUBVOL_CREATE_V2:
3877                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
3878         case BTRFS_IOC_SNAP_DESTROY:
3879                 return btrfs_ioctl_snap_destroy(file, argp);
3880         case BTRFS_IOC_SUBVOL_GETFLAGS:
3881                 return btrfs_ioctl_subvol_getflags(file, argp);
3882         case BTRFS_IOC_SUBVOL_SETFLAGS:
3883                 return btrfs_ioctl_subvol_setflags(file, argp);
3884         case BTRFS_IOC_DEFAULT_SUBVOL:
3885                 return btrfs_ioctl_default_subvol(file, argp);
3886         case BTRFS_IOC_DEFRAG:
3887                 return btrfs_ioctl_defrag(file, NULL);
3888         case BTRFS_IOC_DEFRAG_RANGE:
3889                 return btrfs_ioctl_defrag(file, argp);
3890         case BTRFS_IOC_RESIZE:
3891                 return btrfs_ioctl_resize(file, argp);
3892         case BTRFS_IOC_ADD_DEV:
3893                 return btrfs_ioctl_add_dev(root, argp);
3894         case BTRFS_IOC_RM_DEV:
3895                 return btrfs_ioctl_rm_dev(file, argp);
3896         case BTRFS_IOC_FS_INFO:
3897                 return btrfs_ioctl_fs_info(root, argp);
3898         case BTRFS_IOC_DEV_INFO:
3899                 return btrfs_ioctl_dev_info(root, argp);
3900         case BTRFS_IOC_BALANCE:
3901                 return btrfs_ioctl_balance(file, NULL);
3902         case BTRFS_IOC_CLONE:
3903                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3904         case BTRFS_IOC_CLONE_RANGE:
3905                 return btrfs_ioctl_clone_range(file, argp);
3906         case BTRFS_IOC_TRANS_START:
3907                 return btrfs_ioctl_trans_start(file);
3908         case BTRFS_IOC_TRANS_END:
3909                 return btrfs_ioctl_trans_end(file);
3910         case BTRFS_IOC_TREE_SEARCH:
3911                 return btrfs_ioctl_tree_search(file, argp);
3912         case BTRFS_IOC_INO_LOOKUP:
3913                 return btrfs_ioctl_ino_lookup(file, argp);
3914         case BTRFS_IOC_INO_PATHS:
3915                 return btrfs_ioctl_ino_to_path(root, argp);
3916         case BTRFS_IOC_LOGICAL_INO:
3917                 return btrfs_ioctl_logical_to_ino(root, argp);
3918         case BTRFS_IOC_SPACE_INFO:
3919                 return btrfs_ioctl_space_info(root, argp);
3920         case BTRFS_IOC_SYNC:
3921                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3922                 return 0;
3923         case BTRFS_IOC_START_SYNC:
3924                 return btrfs_ioctl_start_sync(root, argp);
3925         case BTRFS_IOC_WAIT_SYNC:
3926                 return btrfs_ioctl_wait_sync(root, argp);
3927         case BTRFS_IOC_SCRUB:
3928                 return btrfs_ioctl_scrub(file, argp);
3929         case BTRFS_IOC_SCRUB_CANCEL:
3930                 return btrfs_ioctl_scrub_cancel(root, argp);
3931         case BTRFS_IOC_SCRUB_PROGRESS:
3932                 return btrfs_ioctl_scrub_progress(root, argp);
3933         case BTRFS_IOC_BALANCE_V2:
3934                 return btrfs_ioctl_balance(file, argp);
3935         case BTRFS_IOC_BALANCE_CTL:
3936                 return btrfs_ioctl_balance_ctl(root, arg);
3937         case BTRFS_IOC_BALANCE_PROGRESS:
3938                 return btrfs_ioctl_balance_progress(root, argp);
3939         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
3940                 return btrfs_ioctl_set_received_subvol(file, argp);
3941         case BTRFS_IOC_SEND:
3942                 return btrfs_ioctl_send(file, argp);
3943         case BTRFS_IOC_GET_DEV_STATS:
3944                 return btrfs_ioctl_get_dev_stats(root, argp);
3945         case BTRFS_IOC_QUOTA_CTL:
3946                 return btrfs_ioctl_quota_ctl(file, argp);
3947         case BTRFS_IOC_QGROUP_ASSIGN:
3948                 return btrfs_ioctl_qgroup_assign(file, argp);
3949         case BTRFS_IOC_QGROUP_CREATE:
3950                 return btrfs_ioctl_qgroup_create(file, argp);
3951         case BTRFS_IOC_QGROUP_LIMIT:
3952                 return btrfs_ioctl_qgroup_limit(file, argp);
3953         case BTRFS_IOC_DEV_REPLACE:
3954                 return btrfs_ioctl_dev_replace(root, argp);
3955         }
3956
3957         return -ENOTTY;
3958 }