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 "compat.h"
  44 #include "ctree.h"
  45 #include "disk-io.h"
  46 #include "transaction.h"
  47 #include "btrfs_inode.h"
  48 #include "ioctl.h"
  49 #include "print-tree.h"
  50 #include "volumes.h"
  51 #include "locking.h"
  52
  53 /* Mask out flags that are inappropriate for the given type of inode. */
  54 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  55 {
  56         if (S_ISDIR(mode))
  57                 return flags;
  58         else if (S_ISREG(mode))
  59                 return flags & ~FS_DIRSYNC_FL;
  60         else
  61                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  62 }
  63
  64 /*
  65  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  66  */
  67 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  68 {
  69         unsigned int iflags = 0;
  70
  71         if (flags & BTRFS_INODE_SYNC)
  72                 iflags |= FS_SYNC_FL;
  73         if (flags & BTRFS_INODE_IMMUTABLE)
  74                 iflags |= FS_IMMUTABLE_FL;
  75         if (flags & BTRFS_INODE_APPEND)
  76                 iflags |= FS_APPEND_FL;
  77         if (flags & BTRFS_INODE_NODUMP)
  78                 iflags |= FS_NODUMP_FL;
  79         if (flags & BTRFS_INODE_NOATIME)
  80                 iflags |= FS_NOATIME_FL;
  81         if (flags & BTRFS_INODE_DIRSYNC)
  82                 iflags |= FS_DIRSYNC_FL;
  83
  84         return iflags;
  85 }
  86
  87 /*
  88  * Update inode->i_flags based on the btrfs internal flags.
  89  */
  90 void btrfs_update_iflags(struct inode *inode)
  91 {
  92         struct btrfs_inode *ip = BTRFS_I(inode);
  93
  94         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
  95
  96         if (ip->flags & BTRFS_INODE_SYNC)
  97                 inode->i_flags |= S_SYNC;
  98         if (ip->flags & BTRFS_INODE_IMMUTABLE)
  99                 inode->i_flags |= S_IMMUTABLE;
 100         if (ip->flags & BTRFS_INODE_APPEND)
 101                 inode->i_flags |= S_APPEND;
 102         if (ip->flags & BTRFS_INODE_NOATIME)
 103                 inode->i_flags |= S_NOATIME;
 104         if (ip->flags & BTRFS_INODE_DIRSYNC)
 105                 inode->i_flags |= S_DIRSYNC;
 106 }
 107
 108 /*
 109  * Inherit flags from the parent inode.
 110  *
 111  * Unlike extN we don't have any flags we don't want to inherit currently.
 112  */
 113 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 114 {
 115         unsigned int flags;
 116
 117         if (!dir)
 118                 return;
 119
 120         flags = BTRFS_I(dir)->flags;
 121
 122         if (S_ISREG(inode->i_mode))
 123                 flags &= ~BTRFS_INODE_DIRSYNC;
 124         else if (!S_ISDIR(inode->i_mode))
 125                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
 126
 127         BTRFS_I(inode)->flags = flags;
 128         btrfs_update_iflags(inode);
 129 }
 130
 131 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 132 {
 133         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 134         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 135
 136         if (copy_to_user(arg, &flags, sizeof(flags)))
 137                 return -EFAULT;
 138         return 0;
 139 }
 140
 141 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 142 {
 143         struct inode *inode = file->f_path.dentry->d_inode;
 144         struct btrfs_inode *ip = BTRFS_I(inode);
 145         struct btrfs_root *root = ip->root;
 146         struct btrfs_trans_handle *trans;
 147         unsigned int flags, oldflags;
 148         int ret;
 149
 150         if (copy_from_user(&flags, arg, sizeof(flags)))
 151                 return -EFAULT;
 152
 153         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 154                       FS_NOATIME_FL | FS_NODUMP_FL | \
 155                       FS_SYNC_FL | FS_DIRSYNC_FL))
 156                 return -EOPNOTSUPP;
 157
 158         if (!is_owner_or_cap(inode))
 159                 return -EACCES;
 160
 161         mutex_lock(&inode->i_mutex);
 162
 163         flags = btrfs_mask_flags(inode->i_mode, flags);
 164         oldflags = btrfs_flags_to_ioctl(ip->flags);
 165         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 166                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 167                         ret = -EPERM;
 168                         goto out_unlock;
 169                 }
 170         }
 171
 172         ret = mnt_want_write(file->f_path.mnt);
 173         if (ret)
 174                 goto out_unlock;
 175
 176         if (flags & FS_SYNC_FL)
 177                 ip->flags |= BTRFS_INODE_SYNC;
 178         else
 179                 ip->flags &= ~BTRFS_INODE_SYNC;
 180         if (flags & FS_IMMUTABLE_FL)
 181                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 182         else
 183                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 184         if (flags & FS_APPEND_FL)
 185                 ip->flags |= BTRFS_INODE_APPEND;
 186         else
 187                 ip->flags &= ~BTRFS_INODE_APPEND;
 188         if (flags & FS_NODUMP_FL)
 189                 ip->flags |= BTRFS_INODE_NODUMP;
 190         else
 191                 ip->flags &= ~BTRFS_INODE_NODUMP;
 192         if (flags & FS_NOATIME_FL)
 193                 ip->flags |= BTRFS_INODE_NOATIME;
 194         else
 195                 ip->flags &= ~BTRFS_INODE_NOATIME;
 196         if (flags & FS_DIRSYNC_FL)
 197                 ip->flags |= BTRFS_INODE_DIRSYNC;
 198         else
 199                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 200
 201
 202         trans = btrfs_join_transaction(root, 1);
 203         BUG_ON(!trans);
 204
 205         ret = btrfs_update_inode(trans, root, inode);
 206         BUG_ON(ret);
 207
 208         btrfs_update_iflags(inode);
 209         inode->i_ctime = CURRENT_TIME;
 210         btrfs_end_transaction(trans, root);
 211
 212         mnt_drop_write(file->f_path.mnt);
 213  out_unlock:
 214         mutex_unlock(&inode->i_mutex);
 215         return 0;
 216 }
 217
 218 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 219 {
 220         struct inode *inode = file->f_path.dentry->d_inode;
 221
 222         return put_user(inode->i_generation, arg);
 223 }
 224
 225 static noinline int create_subvol(struct btrfs_root *root,
 226                                   struct dentry *dentry,
 227                                   char *name, int namelen,
 228                                   u64 *async_transid)
 229 {
 230         struct btrfs_trans_handle *trans;
 231         struct btrfs_key key;
 232         struct btrfs_root_item root_item;
 233         struct btrfs_inode_item *inode_item;
 234         struct extent_buffer *leaf;
 235         struct btrfs_root *new_root;
 236         struct inode *dir = dentry->d_parent->d_inode;
 237         int ret;
 238         int err;
 239         u64 objectid;
 240         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 241         u64 index = 0;
 242
 243         ret = btrfs_find_free_objectid(NULL, root->fs_info->tree_root,
 244                                        0, &objectid);
 245         if (ret)
 246                 return ret;
 247         /*
 248          * 1 - inode item
 249          * 2 - refs
 250          * 1 - root item
 251          * 2 - dir items
 252          */
 253         trans = btrfs_start_transaction(root, 6);
 254         if (IS_ERR(trans))
 255                 return PTR_ERR(trans);
 256
 257         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 258                                       0, objectid, NULL, 0, 0, 0);
 259         if (IS_ERR(leaf)) {
 260                 ret = PTR_ERR(leaf);
 261                 goto fail;
 262         }
 263
 264         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 265         btrfs_set_header_bytenr(leaf, leaf->start);
 266         btrfs_set_header_generation(leaf, trans->transid);
 267         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 268         btrfs_set_header_owner(leaf, objectid);
 269
 270         write_extent_buffer(leaf, root->fs_info->fsid,
 271                             (unsigned long)btrfs_header_fsid(leaf),
 272                             BTRFS_FSID_SIZE);
 273         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 274                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 275                             BTRFS_UUID_SIZE);
 276         btrfs_mark_buffer_dirty(leaf);
 277
 278         inode_item = &root_item.inode;
 279         memset(inode_item, 0, sizeof(*inode_item));
 280         inode_item->generation = cpu_to_le64(1);
 281         inode_item->size = cpu_to_le64(3);
 282         inode_item->nlink = cpu_to_le32(1);
 283         inode_item->nbytes = cpu_to_le64(root->leafsize);
 284         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 285
 286         btrfs_set_root_bytenr(&root_item, leaf->start);
 287         btrfs_set_root_generation(&root_item, trans->transid);
 288         btrfs_set_root_level(&root_item, 0);
 289         btrfs_set_root_refs(&root_item, 1);
 290         btrfs_set_root_used(&root_item, leaf->len);
 291         btrfs_set_root_last_snapshot(&root_item, 0);
 292
 293         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 294         root_item.drop_level = 0;
 295
 296         btrfs_tree_unlock(leaf);
 297         free_extent_buffer(leaf);
 298         leaf = NULL;
 299
 300         btrfs_set_root_dirid(&root_item, new_dirid);
 301
 302         key.objectid = objectid;
 303         key.offset = 0;
 304         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 305         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 306                                 &root_item);
 307         if (ret)
 308                 goto fail;
 309
 310         key.offset = (u64)-1;
 311         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 312         BUG_ON(IS_ERR(new_root));
 313
 314         btrfs_record_root_in_trans(trans, new_root);
 315
 316         ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
 317                                        BTRFS_I(dir)->block_group);
 318         /*
 319          * insert the directory item
 320          */
 321         ret = btrfs_set_inode_index(dir, &index);
 322         BUG_ON(ret);
 323
 324         ret = btrfs_insert_dir_item(trans, root,
 325                                     name, namelen, dir->i_ino, &key,
 326                                     BTRFS_FT_DIR, index);
 327         if (ret)
 328                 goto fail;
 329
 330         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 331         ret = btrfs_update_inode(trans, root, dir);
 332         BUG_ON(ret);
 333
 334         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 335                                  objectid, root->root_key.objectid,
 336                                  dir->i_ino, index, name, namelen);
 337
 338         BUG_ON(ret);
 339
 340         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 341 fail:
 342         if (async_transid) {
 343                 *async_transid = trans->transid;
 344                 err = btrfs_commit_transaction_async(trans, root, 1);
 345         } else {
 346                 err = btrfs_commit_transaction(trans, root);
 347         }
 348         if (err && !ret)
 349                 ret = err;
 350         return ret;
 351 }
 352
 353 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 354                            char *name, int namelen, u64 *async_transid)
 355 {
 356         struct inode *inode;
 357         struct btrfs_pending_snapshot *pending_snapshot;
 358         struct btrfs_trans_handle *trans;
 359         int ret;
 360
 361         if (!root->ref_cows)
 362                 return -EINVAL;
 363
 364         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 365         if (!pending_snapshot)
 366                 return -ENOMEM;
 367
 368         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
 369         pending_snapshot->dentry = dentry;
 370         pending_snapshot->root = root;
 371
 372         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
 373         if (IS_ERR(trans)) {
 374                 ret = PTR_ERR(trans);
 375                 goto fail;
 376         }
 377
 378         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 379         BUG_ON(ret);
 380
 381         list_add(&pending_snapshot->list,
 382                  &trans->transaction->pending_snapshots);
 383         if (async_transid) {
 384                 *async_transid = trans->transid;
 385                 ret = btrfs_commit_transaction_async(trans,
 386                                      root->fs_info->extent_root, 1);
 387         } else {
 388                 ret = btrfs_commit_transaction(trans,
 389                                                root->fs_info->extent_root);
 390         }
 391         BUG_ON(ret);
 392
 393         ret = pending_snapshot->error;
 394         if (ret)
 395                 goto fail;
 396
 397         btrfs_orphan_cleanup(pending_snapshot->snap);
 398
 399         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 400         if (IS_ERR(inode)) {
 401                 ret = PTR_ERR(inode);
 402                 goto fail;
 403         }
 404         BUG_ON(!inode);
 405         d_instantiate(dentry, inode);
 406         ret = 0;
 407 fail:
 408         kfree(pending_snapshot);
 409         return ret;
 410 }
 411
 412 /* copy of may_create in fs/namei.c() */
 413 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 414 {
 415         if (child->d_inode)
 416                 return -EEXIST;
 417         if (IS_DEADDIR(dir))
 418                 return -ENOENT;
 419         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 420 }
 421
 422 /*
 423  * Create a new subvolume below @parent.  This is largely modeled after
 424  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 425  * inside this filesystem so it's quite a bit simpler.
 426  */
 427 static noinline int btrfs_mksubvol(struct path *parent,
 428                                    char *name, int namelen,
 429                                    struct btrfs_root *snap_src,
 430                                    u64 *async_transid)
 431 {
 432         struct inode *dir  = parent->dentry->d_inode;
 433         struct dentry *dentry;
 434         int error;
 435
 436         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 437
 438         dentry = lookup_one_len(name, parent->dentry, namelen);
 439         error = PTR_ERR(dentry);
 440         if (IS_ERR(dentry))
 441                 goto out_unlock;
 442
 443         error = -EEXIST;
 444         if (dentry->d_inode)
 445                 goto out_dput;
 446
 447         error = mnt_want_write(parent->mnt);
 448         if (error)
 449                 goto out_dput;
 450
 451         error = btrfs_may_create(dir, dentry);
 452         if (error)
 453                 goto out_drop_write;
 454
 455         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 456
 457         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 458                 goto out_up_read;
 459
 460         if (snap_src) {
 461                 error = create_snapshot(snap_src, dentry,
 462                                         name, namelen, async_transid);
 463         } else {
 464                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 465                                       name, namelen, async_transid);
 466         }
 467         if (!error)
 468                 fsnotify_mkdir(dir, dentry);
 469 out_up_read:
 470         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 471 out_drop_write:
 472         mnt_drop_write(parent->mnt);
 473 out_dput:
 474         dput(dentry);
 475 out_unlock:
 476         mutex_unlock(&dir->i_mutex);
 477         return error;
 478 }
 479
 480 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
 481                                int thresh, u64 *last_len, u64 *skip,
 482                                u64 *defrag_end)
 483 {
 484         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 485         struct extent_map *em = NULL;
 486         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 487         int ret = 1;
 488
 489
 490         if (thresh == 0)
 491                 thresh = 256 * 1024;
 492
 493         /*
 494          * make sure that once we start defragging and extent, we keep on
 495          * defragging it
 496          */
 497         if (start < *defrag_end)
 498                 return 1;
 499
 500         *skip = 0;
 501
 502         /*
 503          * hopefully we have this extent in the tree already, try without
 504          * the full extent lock
 505          */
 506         read_lock(&em_tree->lock);
 507         em = lookup_extent_mapping(em_tree, start, len);
 508         read_unlock(&em_tree->lock);
 509
 510         if (!em) {
 511                 /* get the big lock and read metadata off disk */
 512                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 513                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 514                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
 515
 516                 if (IS_ERR(em))
 517                         return 0;
 518         }
 519
 520         /* this will cover holes, and inline extents */
 521         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
 522                 ret = 0;
 523
 524         /*
 525          * we hit a real extent, if it is big don't bother defragging it again
 526          */
 527         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
 528                 ret = 0;
 529
 530         /*
 531          * last_len ends up being a counter of how many bytes we've defragged.
 532          * every time we choose not to defrag an extent, we reset *last_len
 533          * so that the next tiny extent will force a defrag.
 534          *
 535          * The end result of this is that tiny extents before a single big
 536          * extent will force at least part of that big extent to be defragged.
 537          */
 538         if (ret) {
 539                 *last_len += len;
 540                 *defrag_end = extent_map_end(em);
 541         } else {
 542                 *last_len = 0;
 543                 *skip = extent_map_end(em);
 544                 *defrag_end = 0;
 545         }
 546
 547         free_extent_map(em);
 548         return ret;
 549 }
 550
 551 static int btrfs_defrag_file(struct file *file,
 552                              struct btrfs_ioctl_defrag_range_args *range)
 553 {
 554         struct inode *inode = fdentry(file)->d_inode;
 555         struct btrfs_root *root = BTRFS_I(inode)->root;
 556         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 557         struct btrfs_ordered_extent *ordered;
 558         struct page *page;
 559         unsigned long last_index;
 560         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 561         unsigned long total_read = 0;
 562         u64 page_start;
 563         u64 page_end;
 564         u64 last_len = 0;
 565         u64 skip = 0;
 566         u64 defrag_end = 0;
 567         unsigned long i;
 568         int ret;
 569
 570         if (inode->i_size == 0)
 571                 return 0;
 572
 573         if (range->start + range->len > range->start) {
 574                 last_index = min_t(u64, inode->i_size - 1,
 575                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
 576         } else {
 577                 last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
 578         }
 579
 580         i = range->start >> PAGE_CACHE_SHIFT;
 581         while (i <= last_index) {
 582                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
 583                                         PAGE_CACHE_SIZE,
 584                                         range->extent_thresh,
 585                                         &last_len, &skip,
 586                                         &defrag_end)) {
 587                         unsigned long next;
 588                         /*
 589                          * the should_defrag function tells us how much to skip
 590                          * bump our counter by the suggested amount
 591                          */
 592                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 593                         i = max(i + 1, next);
 594                         continue;
 595                 }
 596
 597                 if (total_read % ra_pages == 0) {
 598                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 599                                        min(last_index, i + ra_pages - 1));
 600                 }
 601                 total_read++;
 602                 mutex_lock(&inode->i_mutex);
 603                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
 604                         BTRFS_I(inode)->force_compress = 1;
 605
 606                 ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
 607                 if (ret)
 608                         goto err_unlock;
 609 again:
 610                 if (inode->i_size == 0 ||
 611                     i > ((inode->i_size - 1) >> PAGE_CACHE_SHIFT)) {
 612                         ret = 0;
 613                         goto err_reservations;
 614                 }
 615
 616                 page = grab_cache_page(inode->i_mapping, i);
 617                 if (!page) {
 618                         ret = -ENOMEM;
 619                         goto err_reservations;
 620                 }
 621
 622                 if (!PageUptodate(page)) {
 623                         btrfs_readpage(NULL, page);
 624                         lock_page(page);
 625                         if (!PageUptodate(page)) {
 626                                 unlock_page(page);
 627                                 page_cache_release(page);
 628                                 ret = -EIO;
 629                                 goto err_reservations;
 630                         }
 631                 }
 632
 633                 if (page->mapping != inode->i_mapping) {
 634                         unlock_page(page);
 635                         page_cache_release(page);
 636                         goto again;
 637                 }
 638
 639                 wait_on_page_writeback(page);
 640
 641                 if (PageDirty(page)) {
 642                         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 643                         goto loop_unlock;
 644                 }
 645
 646                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 647                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 648                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 649
 650                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 651                 if (ordered) {
 652                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 653                         unlock_page(page);
 654                         page_cache_release(page);
 655                         btrfs_start_ordered_extent(inode, ordered, 1);
 656                         btrfs_put_ordered_extent(ordered);
 657                         goto again;
 658                 }
 659                 set_page_extent_mapped(page);
 660
 661                 /*
 662                  * this makes sure page_mkwrite is called on the
 663                  * page if it is dirtied again later
 664                  */
 665                 clear_page_dirty_for_io(page);
 666                 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start,
 667                                   page_end, EXTENT_DIRTY | EXTENT_DELALLOC |
 668                                   EXTENT_DO_ACCOUNTING, GFP_NOFS);
 669
 670                 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
 671                 ClearPageChecked(page);
 672                 set_page_dirty(page);
 673                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 674
 675 loop_unlock:
 676                 unlock_page(page);
 677                 page_cache_release(page);
 678                 mutex_unlock(&inode->i_mutex);
 679
 680                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 681                 i++;
 682         }
 683
 684         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
 685                 filemap_flush(inode->i_mapping);
 686
 687         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
 688                 /* the filemap_flush will queue IO into the worker threads, but
 689                  * we have to make sure the IO is actually started and that
 690                  * ordered extents get created before we return
 691                  */
 692                 atomic_inc(&root->fs_info->async_submit_draining);
 693                 while (atomic_read(&root->fs_info->nr_async_submits) ||
 694                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 695                         wait_event(root->fs_info->async_submit_wait,
 696                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
 697                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
 698                 }
 699                 atomic_dec(&root->fs_info->async_submit_draining);
 700
 701                 mutex_lock(&inode->i_mutex);
 702                 BTRFS_I(inode)->force_compress = 0;
 703                 mutex_unlock(&inode->i_mutex);
 704         }
 705
 706         return 0;
 707
 708 err_reservations:
 709         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
 710 err_unlock:
 711         mutex_unlock(&inode->i_mutex);
 712         return ret;
 713 }
 714
 715 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
 716                                         void __user *arg)
 717 {
 718         u64 new_size;
 719         u64 old_size;
 720         u64 devid = 1;
 721         struct btrfs_ioctl_vol_args *vol_args;
 722         struct btrfs_trans_handle *trans;
 723         struct btrfs_device *device = NULL;
 724         char *sizestr;
 725         char *devstr = NULL;
 726         int ret = 0;
 727         int mod = 0;
 728
 729         if (root->fs_info->sb->s_flags & MS_RDONLY)
 730                 return -EROFS;
 731
 732         if (!capable(CAP_SYS_ADMIN))
 733                 return -EPERM;
 734
 735         vol_args = memdup_user(arg, sizeof(*vol_args));
 736         if (IS_ERR(vol_args))
 737                 return PTR_ERR(vol_args);
 738
 739         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 740
 741         mutex_lock(&root->fs_info->volume_mutex);
 742         sizestr = vol_args->name;
 743         devstr = strchr(sizestr, ':');
 744         if (devstr) {
 745                 char *end;
 746                 sizestr = devstr + 1;
 747                 *devstr = '\0';
 748                 devstr = vol_args->name;
 749                 devid = simple_strtoull(devstr, &end, 10);
 750                 printk(KERN_INFO "resizing devid %llu\n",
 751                        (unsigned long long)devid);
 752         }
 753         device = btrfs_find_device(root, devid, NULL, NULL);
 754         if (!device) {
 755                 printk(KERN_INFO "resizer unable to find device %llu\n",
 756                        (unsigned long long)devid);
 757                 ret = -EINVAL;
 758                 goto out_unlock;
 759         }
 760         if (!strcmp(sizestr, "max"))
 761                 new_size = device->bdev->bd_inode->i_size;
 762         else {
 763                 if (sizestr[0] == '-') {
 764                         mod = -1;
 765                         sizestr++;
 766                 } else if (sizestr[0] == '+') {
 767                         mod = 1;
 768                         sizestr++;
 769                 }
 770                 new_size = memparse(sizestr, NULL);
 771                 if (new_size == 0) {
 772                         ret = -EINVAL;
 773                         goto out_unlock;
 774                 }
 775         }
 776
 777         old_size = device->total_bytes;
 778
 779         if (mod < 0) {
 780                 if (new_size > old_size) {
 781                         ret = -EINVAL;
 782                         goto out_unlock;
 783                 }
 784                 new_size = old_size - new_size;
 785         } else if (mod > 0) {
 786                 new_size = old_size + new_size;
 787         }
 788
 789         if (new_size < 256 * 1024 * 1024) {
 790                 ret = -EINVAL;
 791                 goto out_unlock;
 792         }
 793         if (new_size > device->bdev->bd_inode->i_size) {
 794                 ret = -EFBIG;
 795                 goto out_unlock;
 796         }
 797
 798         do_div(new_size, root->sectorsize);
 799         new_size *= root->sectorsize;
 800
 801         printk(KERN_INFO "new size for %s is %llu\n",
 802                 device->name, (unsigned long long)new_size);
 803
 804         if (new_size > old_size) {
 805                 trans = btrfs_start_transaction(root, 0);
 806                 ret = btrfs_grow_device(trans, device, new_size);
 807                 btrfs_commit_transaction(trans, root);
 808         } else {
 809                 ret = btrfs_shrink_device(device, new_size);
 810         }
 811
 812 out_unlock:
 813         mutex_unlock(&root->fs_info->volume_mutex);
 814         kfree(vol_args);
 815         return ret;
 816 }
 817
 818 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
 819                                                     char *name,
 820                                                     unsigned long fd,
 821                                                     int subvol,
 822                                                     u64 *transid)
 823 {
 824         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 825         struct file *src_file;
 826         int namelen;
 827         int ret = 0;
 828
 829         if (root->fs_info->sb->s_flags & MS_RDONLY)
 830                 return -EROFS;
 831
 832         namelen = strlen(name);
 833         if (strchr(name, '/')) {
 834                 ret = -EINVAL;
 835                 goto out;
 836         }
 837
 838         if (subvol) {
 839                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
 840                                      NULL, transid);
 841         } else {
 842                 struct inode *src_inode;
 843                 src_file = fget(fd);
 844                 if (!src_file) {
 845                         ret = -EINVAL;
 846                         goto out;
 847                 }
 848
 849                 src_inode = src_file->f_path.dentry->d_inode;
 850                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 851                         printk(KERN_INFO "btrfs: Snapshot src from "
 852                                "another FS\n");
 853                         ret = -EINVAL;
 854                         fput(src_file);
 855                         goto out;
 856                 }
 857                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
 858                                      BTRFS_I(src_inode)->root,
 859                                      transid);
 860                 fput(src_file);
 861         }
 862 out:
 863         return ret;
 864 }
 865
 866 static noinline int btrfs_ioctl_snap_create(struct file *file,
 867                                             void __user *arg, int subvol,
 868                                             int async)
 869 {
 870         struct btrfs_ioctl_vol_args *vol_args = NULL;
 871         struct btrfs_ioctl_async_vol_args *async_vol_args = NULL;
 872         char *name;
 873         u64 fd;
 874         u64 transid = 0;
 875         int ret;
 876
 877         if (async) {
 878                 async_vol_args = memdup_user(arg, sizeof(*async_vol_args));
 879                 if (IS_ERR(async_vol_args))
 880                         return PTR_ERR(async_vol_args);
 881
 882                 name = async_vol_args->name;
 883                 fd = async_vol_args->fd;
 884                 async_vol_args->name[BTRFS_SNAPSHOT_NAME_MAX] = '\0';
 885         } else {
 886                 vol_args = memdup_user(arg, sizeof(*vol_args));
 887                 if (IS_ERR(vol_args))
 888                         return PTR_ERR(vol_args);
 889                 name = vol_args->name;
 890                 fd = vol_args->fd;
 891                 vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 892         }
 893
 894         ret = btrfs_ioctl_snap_create_transid(file, name, fd,
 895                                               subvol, &transid);
 896
 897         if (!ret && async) {
 898                 if (copy_to_user(arg +
 899                                 offsetof(struct btrfs_ioctl_async_vol_args,
 900                                 transid), &transid, sizeof(transid)))
 901                         return -EFAULT;
 902         }
 903
 904         kfree(vol_args);
 905         kfree(async_vol_args);
 906
 907         return ret;
 908 }
 909
 910 /*
 911  * helper to check if the subvolume references other subvolumes
 912  */
 913 static noinline int may_destroy_subvol(struct btrfs_root *root)
 914 {
 915         struct btrfs_path *path;
 916         struct btrfs_key key;
 917         int ret;
 918
 919         path = btrfs_alloc_path();
 920         if (!path)
 921                 return -ENOMEM;
 922
 923         key.objectid = root->root_key.objectid;
 924         key.type = BTRFS_ROOT_REF_KEY;
 925         key.offset = (u64)-1;
 926
 927         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
 928                                 &key, path, 0, 0);
 929         if (ret < 0)
 930                 goto out;
 931         BUG_ON(ret == 0);
 932
 933         ret = 0;
 934         if (path->slots[0] > 0) {
 935                 path->slots[0]--;
 936                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 937                 if (key.objectid == root->root_key.objectid &&
 938                     key.type == BTRFS_ROOT_REF_KEY)
 939                         ret = -ENOTEMPTY;
 940         }
 941 out:
 942         btrfs_free_path(path);
 943         return ret;
 944 }
 945
 946 static noinline int key_in_sk(struct btrfs_key *key,
 947                               struct btrfs_ioctl_search_key *sk)
 948 {
 949         struct btrfs_key test;
 950         int ret;
 951
 952         test.objectid = sk->min_objectid;
 953         test.type = sk->min_type;
 954         test.offset = sk->min_offset;
 955
 956         ret = btrfs_comp_cpu_keys(key, &test);
 957         if (ret < 0)
 958                 return 0;
 959
 960         test.objectid = sk->max_objectid;
 961         test.type = sk->max_type;
 962         test.offset = sk->max_offset;
 963
 964         ret = btrfs_comp_cpu_keys(key, &test);
 965         if (ret > 0)
 966                 return 0;
 967         return 1;
 968 }
 969
 970 static noinline int copy_to_sk(struct btrfs_root *root,
 971                                struct btrfs_path *path,
 972                                struct btrfs_key *key,
 973                                struct btrfs_ioctl_search_key *sk,
 974                                char *buf,
 975                                unsigned long *sk_offset,
 976                                int *num_found)
 977 {
 978         u64 found_transid;
 979         struct extent_buffer *leaf;
 980         struct btrfs_ioctl_search_header sh;
 981         unsigned long item_off;
 982         unsigned long item_len;
 983         int nritems;
 984         int i;
 985         int slot;
 986         int found = 0;
 987         int ret = 0;
 988
 989         leaf = path->nodes[0];
 990         slot = path->slots[0];
 991         nritems = btrfs_header_nritems(leaf);
 992
 993         if (btrfs_header_generation(leaf) > sk->max_transid) {
 994                 i = nritems;
 995                 goto advance_key;
 996         }
 997         found_transid = btrfs_header_generation(leaf);
 998
 999         for (i = slot; i < nritems; i++) {
1000                 item_off = btrfs_item_ptr_offset(leaf, i);
1001                 item_len = btrfs_item_size_nr(leaf, i);
1002
1003                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1004                         item_len = 0;
1005
1006                 if (sizeof(sh) + item_len + *sk_offset >
1007                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1008                         ret = 1;
1009                         goto overflow;
1010                 }
1011
1012                 btrfs_item_key_to_cpu(leaf, key, i);
1013                 if (!key_in_sk(key, sk))
1014                         continue;
1015
1016                 sh.objectid = key->objectid;
1017                 sh.offset = key->offset;
1018                 sh.type = key->type;
1019                 sh.len = item_len;
1020                 sh.transid = found_transid;
1021
1022                 /* copy search result header */
1023                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1024                 *sk_offset += sizeof(sh);
1025
1026                 if (item_len) {
1027                         char *p = buf + *sk_offset;
1028                         /* copy the item */
1029                         read_extent_buffer(leaf, p,
1030                                            item_off, item_len);
1031                         *sk_offset += item_len;
1032                 }
1033                 found++;
1034
1035                 if (*num_found >= sk->nr_items)
1036                         break;
1037         }
1038 advance_key:
1039         ret = 0;
1040         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1041                 key->offset++;
1042         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1043                 key->offset = 0;
1044                 key->type++;
1045         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1046                 key->offset = 0;
1047                 key->type = 0;
1048                 key->objectid++;
1049         } else
1050                 ret = 1;
1051 overflow:
1052         *num_found += found;
1053         return ret;
1054 }
1055
1056 static noinline int search_ioctl(struct inode *inode,
1057                                  struct btrfs_ioctl_search_args *args)
1058 {
1059         struct btrfs_root *root;
1060         struct btrfs_key key;
1061         struct btrfs_key max_key;
1062         struct btrfs_path *path;
1063         struct btrfs_ioctl_search_key *sk = &args->key;
1064         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1065         int ret;
1066         int num_found = 0;
1067         unsigned long sk_offset = 0;
1068
1069         path = btrfs_alloc_path();
1070         if (!path)
1071                 return -ENOMEM;
1072
1073         if (sk->tree_id == 0) {
1074                 /* search the root of the inode that was passed */
1075                 root = BTRFS_I(inode)->root;
1076         } else {
1077                 key.objectid = sk->tree_id;
1078                 key.type = BTRFS_ROOT_ITEM_KEY;
1079                 key.offset = (u64)-1;
1080                 root = btrfs_read_fs_root_no_name(info, &key);
1081                 if (IS_ERR(root)) {
1082                         printk(KERN_ERR "could not find root %llu\n",
1083                                sk->tree_id);
1084                         btrfs_free_path(path);
1085                         return -ENOENT;
1086                 }
1087         }
1088
1089         key.objectid = sk->min_objectid;
1090         key.type = sk->min_type;
1091         key.offset = sk->min_offset;
1092
1093         max_key.objectid = sk->max_objectid;
1094         max_key.type = sk->max_type;
1095         max_key.offset = sk->max_offset;
1096
1097         path->keep_locks = 1;
1098
1099         while(1) {
1100                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1101                                            sk->min_transid);
1102                 if (ret != 0) {
1103                         if (ret > 0)
1104                                 ret = 0;
1105                         goto err;
1106                 }
1107                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1108                                  &sk_offset, &num_found);
1109                 btrfs_release_path(root, path);
1110                 if (ret || num_found >= sk->nr_items)
1111                         break;
1112
1113         }
1114         ret = 0;
1115 err:
1116         sk->nr_items = num_found;
1117         btrfs_free_path(path);
1118         return ret;
1119 }
1120
1121 static noinline int btrfs_ioctl_tree_search(struct file *file,
1122                                            void __user *argp)
1123 {
1124          struct btrfs_ioctl_search_args *args;
1125          struct inode *inode;
1126          int ret;
1127
1128         if (!capable(CAP_SYS_ADMIN))
1129                 return -EPERM;
1130
1131         args = memdup_user(argp, sizeof(*args));
1132         if (IS_ERR(args))
1133                 return PTR_ERR(args);
1134
1135         inode = fdentry(file)->d_inode;
1136         ret = search_ioctl(inode, args);
1137         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1138                 ret = -EFAULT;
1139         kfree(args);
1140         return ret;
1141 }
1142
1143 /*
1144  * Search INODE_REFs to identify path name of 'dirid' directory
1145  * in a 'tree_id' tree. and sets path name to 'name'.
1146  */
1147 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1148                                 u64 tree_id, u64 dirid, char *name)
1149 {
1150         struct btrfs_root *root;
1151         struct btrfs_key key;
1152         char *ptr;
1153         int ret = -1;
1154         int slot;
1155         int len;
1156         int total_len = 0;
1157         struct btrfs_inode_ref *iref;
1158         struct extent_buffer *l;
1159         struct btrfs_path *path;
1160
1161         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1162                 name[0]='\0';
1163                 return 0;
1164         }
1165
1166         path = btrfs_alloc_path();
1167         if (!path)
1168                 return -ENOMEM;
1169
1170         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1171
1172         key.objectid = tree_id;
1173         key.type = BTRFS_ROOT_ITEM_KEY;
1174         key.offset = (u64)-1;
1175         root = btrfs_read_fs_root_no_name(info, &key);
1176         if (IS_ERR(root)) {
1177                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1178                 ret = -ENOENT;
1179                 goto out;
1180         }
1181
1182         key.objectid = dirid;
1183         key.type = BTRFS_INODE_REF_KEY;
1184         key.offset = (u64)-1;
1185
1186         while(1) {
1187                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1188                 if (ret < 0)
1189                         goto out;
1190
1191                 l = path->nodes[0];
1192                 slot = path->slots[0];
1193                 if (ret > 0 && slot > 0)
1194                         slot--;
1195                 btrfs_item_key_to_cpu(l, &key, slot);
1196
1197                 if (ret > 0 && (key.objectid != dirid ||
1198                                 key.type != BTRFS_INODE_REF_KEY)) {
1199                         ret = -ENOENT;
1200                         goto out;
1201                 }
1202
1203                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1204                 len = btrfs_inode_ref_name_len(l, iref);
1205                 ptr -= len + 1;
1206                 total_len += len + 1;
1207                 if (ptr < name)
1208                         goto out;
1209
1210                 *(ptr + len) = '/';
1211                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1212
1213                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1214                         break;
1215
1216                 btrfs_release_path(root, path);
1217                 key.objectid = key.offset;
1218                 key.offset = (u64)-1;
1219                 dirid = key.objectid;
1220
1221         }
1222         if (ptr < name)
1223                 goto out;
1224         memcpy(name, ptr, total_len);
1225         name[total_len]='\0';
1226         ret = 0;
1227 out:
1228         btrfs_free_path(path);
1229         return ret;
1230 }
1231
1232 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1233                                            void __user *argp)
1234 {
1235          struct btrfs_ioctl_ino_lookup_args *args;
1236          struct inode *inode;
1237          int ret;
1238
1239         if (!capable(CAP_SYS_ADMIN))
1240                 return -EPERM;
1241
1242         args = memdup_user(argp, sizeof(*args));
1243         if (IS_ERR(args))
1244                 return PTR_ERR(args);
1245
1246         inode = fdentry(file)->d_inode;
1247
1248         if (args->treeid == 0)
1249                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1250
1251         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1252                                         args->treeid, args->objectid,
1253                                         args->name);
1254
1255         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1256                 ret = -EFAULT;
1257
1258         kfree(args);
1259         return ret;
1260 }
1261
1262 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1263                                              void __user *arg)
1264 {
1265         struct dentry *parent = fdentry(file);
1266         struct dentry *dentry;
1267         struct inode *dir = parent->d_inode;
1268         struct inode *inode;
1269         struct btrfs_root *root = BTRFS_I(dir)->root;
1270         struct btrfs_root *dest = NULL;
1271         struct btrfs_ioctl_vol_args *vol_args;
1272         struct btrfs_trans_handle *trans;
1273         int namelen;
1274         int ret;
1275         int err = 0;
1276
1277         if (!capable(CAP_SYS_ADMIN))
1278                 return -EPERM;
1279
1280         vol_args = memdup_user(arg, sizeof(*vol_args));
1281         if (IS_ERR(vol_args))
1282                 return PTR_ERR(vol_args);
1283
1284         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1285         namelen = strlen(vol_args->name);
1286         if (strchr(vol_args->name, '/') ||
1287             strncmp(vol_args->name, "..", namelen) == 0) {
1288                 err = -EINVAL;
1289                 goto out;
1290         }
1291
1292         err = mnt_want_write(file->f_path.mnt);
1293         if (err)
1294                 goto out;
1295
1296         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1297         dentry = lookup_one_len(vol_args->name, parent, namelen);
1298         if (IS_ERR(dentry)) {
1299                 err = PTR_ERR(dentry);
1300                 goto out_unlock_dir;
1301         }
1302
1303         if (!dentry->d_inode) {
1304                 err = -ENOENT;
1305                 goto out_dput;
1306         }
1307
1308         inode = dentry->d_inode;
1309         if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
1310                 err = -EINVAL;
1311                 goto out_dput;
1312         }
1313
1314         dest = BTRFS_I(inode)->root;
1315
1316         mutex_lock(&inode->i_mutex);
1317         err = d_invalidate(dentry);
1318         if (err)
1319                 goto out_unlock;
1320
1321         down_write(&root->fs_info->subvol_sem);
1322
1323         err = may_destroy_subvol(dest);
1324         if (err)
1325                 goto out_up_write;
1326
1327         trans = btrfs_start_transaction(root, 0);
1328         if (IS_ERR(trans)) {
1329                 err = PTR_ERR(trans);
1330                 goto out_up_write;
1331         }
1332         trans->block_rsv = &root->fs_info->global_block_rsv;
1333
1334         ret = btrfs_unlink_subvol(trans, root, dir,
1335                                 dest->root_key.objectid,
1336                                 dentry->d_name.name,
1337                                 dentry->d_name.len);
1338         BUG_ON(ret);
1339
1340         btrfs_record_root_in_trans(trans, dest);
1341
1342         memset(&dest->root_item.drop_progress, 0,
1343                 sizeof(dest->root_item.drop_progress));
1344         dest->root_item.drop_level = 0;
1345         btrfs_set_root_refs(&dest->root_item, 0);
1346
1347         if (!xchg(&dest->orphan_item_inserted, 1)) {
1348                 ret = btrfs_insert_orphan_item(trans,
1349                                         root->fs_info->tree_root,
1350                                         dest->root_key.objectid);
1351                 BUG_ON(ret);
1352         }
1353
1354         ret = btrfs_end_transaction(trans, root);
1355         BUG_ON(ret);
1356         inode->i_flags |= S_DEAD;
1357 out_up_write:
1358         up_write(&root->fs_info->subvol_sem);
1359 out_unlock:
1360         mutex_unlock(&inode->i_mutex);
1361         if (!err) {
1362                 shrink_dcache_sb(root->fs_info->sb);
1363                 btrfs_invalidate_inodes(dest);
1364                 d_delete(dentry);
1365         }
1366 out_dput:
1367         dput(dentry);
1368 out_unlock_dir:
1369         mutex_unlock(&dir->i_mutex);
1370         mnt_drop_write(file->f_path.mnt);
1371 out:
1372         kfree(vol_args);
1373         return err;
1374 }
1375
1376 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1377 {
1378         struct inode *inode = fdentry(file)->d_inode;
1379         struct btrfs_root *root = BTRFS_I(inode)->root;
1380         struct btrfs_ioctl_defrag_range_args *range;
1381         int ret;
1382
1383         ret = mnt_want_write(file->f_path.mnt);
1384         if (ret)
1385                 return ret;
1386
1387         switch (inode->i_mode & S_IFMT) {
1388         case S_IFDIR:
1389                 if (!capable(CAP_SYS_ADMIN)) {
1390                         ret = -EPERM;
1391                         goto out;
1392                 }
1393                 ret = btrfs_defrag_root(root, 0);
1394                 if (ret)
1395                         goto out;
1396                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
1397                 break;
1398         case S_IFREG:
1399                 if (!(file->f_mode & FMODE_WRITE)) {
1400                         ret = -EINVAL;
1401                         goto out;
1402                 }
1403
1404                 range = kzalloc(sizeof(*range), GFP_KERNEL);
1405                 if (!range) {
1406                         ret = -ENOMEM;
1407                         goto out;
1408                 }
1409
1410                 if (argp) {
1411                         if (copy_from_user(range, argp,
1412                                            sizeof(*range))) {
1413                                 ret = -EFAULT;
1414                                 kfree(range);
1415                                 goto out;
1416                         }
1417                         /* compression requires us to start the IO */
1418                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1419                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
1420                                 range->extent_thresh = (u32)-1;
1421                         }
1422                 } else {
1423                         /* the rest are all set to zero by kzalloc */
1424                         range->len = (u64)-1;
1425                 }
1426                 ret = btrfs_defrag_file(file, range);
1427                 kfree(range);
1428                 break;
1429         default:
1430                 ret = -EINVAL;
1431         }
1432 out:
1433         mnt_drop_write(file->f_path.mnt);
1434         return ret;
1435 }
1436
1437 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
1438 {
1439         struct btrfs_ioctl_vol_args *vol_args;
1440         int ret;
1441
1442         if (!capable(CAP_SYS_ADMIN))
1443                 return -EPERM;
1444
1445         vol_args = memdup_user(arg, sizeof(*vol_args));
1446         if (IS_ERR(vol_args))
1447                 return PTR_ERR(vol_args);
1448
1449         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1450         ret = btrfs_init_new_device(root, vol_args->name);
1451
1452         kfree(vol_args);
1453         return ret;
1454 }
1455
1456 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
1457 {
1458         struct btrfs_ioctl_vol_args *vol_args;
1459         int ret;
1460
1461         if (!capable(CAP_SYS_ADMIN))
1462                 return -EPERM;
1463
1464         if (root->fs_info->sb->s_flags & MS_RDONLY)
1465                 return -EROFS;
1466
1467         vol_args = memdup_user(arg, sizeof(*vol_args));
1468         if (IS_ERR(vol_args))
1469                 return PTR_ERR(vol_args);
1470
1471         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1472         ret = btrfs_rm_device(root, vol_args->name);
1473
1474         kfree(vol_args);
1475         return ret;
1476 }
1477
1478 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
1479                                        u64 off, u64 olen, u64 destoff)
1480 {
1481         struct inode *inode = fdentry(file)->d_inode;
1482         struct btrfs_root *root = BTRFS_I(inode)->root;
1483         struct file *src_file;
1484         struct inode *src;
1485         struct btrfs_trans_handle *trans;
1486         struct btrfs_path *path;
1487         struct extent_buffer *leaf;
1488         char *buf;
1489         struct btrfs_key key;
1490         u32 nritems;
1491         int slot;
1492         int ret;
1493         u64 len = olen;
1494         u64 bs = root->fs_info->sb->s_blocksize;
1495         u64 hint_byte;
1496
1497         /*
1498          * TODO:
1499          * - split compressed inline extents.  annoying: we need to
1500          *   decompress into destination's address_space (the file offset
1501          *   may change, so source mapping won't do), then recompress (or
1502          *   otherwise reinsert) a subrange.
1503          * - allow ranges within the same file to be cloned (provided
1504          *   they don't overlap)?
1505          */
1506
1507         /* the destination must be opened for writing */
1508         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
1509                 return -EINVAL;
1510
1511         ret = mnt_want_write(file->f_path.mnt);
1512         if (ret)
1513                 return ret;
1514
1515         src_file = fget(srcfd);
1516         if (!src_file) {
1517                 ret = -EBADF;
1518                 goto out_drop_write;
1519         }
1520
1521         src = src_file->f_dentry->d_inode;
1522
1523         ret = -EINVAL;
1524         if (src == inode)
1525                 goto out_fput;
1526
1527         /* the src must be open for reading */
1528         if (!(src_file->f_mode & FMODE_READ))
1529                 goto out_fput;
1530
1531         ret = -EISDIR;
1532         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
1533                 goto out_fput;
1534
1535         ret = -EXDEV;
1536         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
1537                 goto out_fput;
1538
1539         ret = -ENOMEM;
1540         buf = vmalloc(btrfs_level_size(root, 0));
1541         if (!buf)
1542                 goto out_fput;
1543
1544         path = btrfs_alloc_path();
1545         if (!path) {
1546                 vfree(buf);
1547                 goto out_fput;
1548         }
1549         path->reada = 2;
1550
1551         if (inode < src) {
1552                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
1553                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
1554         } else {
1555                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
1556                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1557         }
1558
1559         /* determine range to clone */
1560         ret = -EINVAL;
1561         if (off + len > src->i_size || off + len < off)
1562                 goto out_unlock;
1563         if (len == 0)
1564                 olen = len = src->i_size - off;
1565         /* if we extend to eof, continue to block boundary */
1566         if (off + len == src->i_size)
1567                 len = ((src->i_size + bs-1) & ~(bs-1))
1568                         - off;
1569
1570         /* verify the end result is block aligned */
1571         if ((off & (bs-1)) ||
1572             ((off + len) & (bs-1)))
1573                 goto out_unlock;
1574
1575         /* do any pending delalloc/csum calc on src, one way or
1576            another, and lock file content */
1577         while (1) {
1578                 struct btrfs_ordered_extent *ordered;
1579                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1580                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
1581                 if (!ordered &&
1582                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
1583                                    EXTENT_DELALLOC, 0, NULL))
1584                         break;
1585                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1586                 if (ordered)
1587                         btrfs_put_ordered_extent(ordered);
1588                 btrfs_wait_ordered_range(src, off, len);
1589         }
1590
1591         /* clone data */
1592         key.objectid = src->i_ino;
1593         key.type = BTRFS_EXTENT_DATA_KEY;
1594         key.offset = 0;
1595
1596         while (1) {
1597                 /*
1598                  * note the key will change type as we walk through the
1599                  * tree.
1600                  */
1601                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1602                 if (ret < 0)
1603                         goto out;
1604
1605                 nritems = btrfs_header_nritems(path->nodes[0]);
1606                 if (path->slots[0] >= nritems) {
1607                         ret = btrfs_next_leaf(root, path);
1608                         if (ret < 0)
1609                                 goto out;
1610                         if (ret > 0)
1611                                 break;
1612                         nritems = btrfs_header_nritems(path->nodes[0]);
1613                 }
1614                 leaf = path->nodes[0];
1615                 slot = path->slots[0];
1616
1617                 btrfs_item_key_to_cpu(leaf, &key, slot);
1618                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1619                     key.objectid != src->i_ino)
1620                         break;
1621
1622                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1623                         struct btrfs_file_extent_item *extent;
1624                         int type;
1625                         u32 size;
1626                         struct btrfs_key new_key;
1627                         u64 disko = 0, diskl = 0;
1628                         u64 datao = 0, datal = 0;
1629                         u8 comp;
1630                         u64 endoff;
1631
1632                         size = btrfs_item_size_nr(leaf, slot);
1633                         read_extent_buffer(leaf, buf,
1634                                            btrfs_item_ptr_offset(leaf, slot),
1635                                            size);
1636
1637                         extent = btrfs_item_ptr(leaf, slot,
1638                                                 struct btrfs_file_extent_item);
1639                         comp = btrfs_file_extent_compression(leaf, extent);
1640                         type = btrfs_file_extent_type(leaf, extent);
1641                         if (type == BTRFS_FILE_EXTENT_REG ||
1642                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1643                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1644                                                                       extent);
1645                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1646                                                                  extent);
1647                                 datao = btrfs_file_extent_offset(leaf, extent);
1648                                 datal = btrfs_file_extent_num_bytes(leaf,
1649                                                                     extent);
1650                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1651                                 /* take upper bound, may be compressed */
1652                                 datal = btrfs_file_extent_ram_bytes(leaf,
1653                                                                     extent);
1654                         }
1655                         btrfs_release_path(root, path);
1656
1657                         if (key.offset + datal <= off ||
1658                             key.offset >= off+len)
1659                                 goto next;
1660
1661                         memcpy(&new_key, &key, sizeof(new_key));
1662                         new_key.objectid = inode->i_ino;
1663                         new_key.offset = key.offset + destoff - off;
1664
1665                         trans = btrfs_start_transaction(root, 1);
1666                         if (IS_ERR(trans)) {
1667                                 ret = PTR_ERR(trans);
1668                                 goto out;
1669                         }
1670
1671                         if (type == BTRFS_FILE_EXTENT_REG ||
1672                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1673                                 if (off > key.offset) {
1674                                         datao += off - key.offset;
1675                                         datal -= off - key.offset;
1676                                 }
1677
1678                                 if (key.offset + datal > off + len)
1679                                         datal = off + len - key.offset;
1680
1681                                 ret = btrfs_drop_extents(trans, inode,
1682                                                          new_key.offset,
1683                                                          new_key.offset + datal,
1684                                                          &hint_byte, 1);
1685                                 BUG_ON(ret);
1686
1687                                 ret = btrfs_insert_empty_item(trans, root, path,
1688                                                               &new_key, size);
1689                                 BUG_ON(ret);
1690
1691                                 leaf = path->nodes[0];
1692                                 slot = path->slots[0];
1693                                 write_extent_buffer(leaf, buf,
1694                                             btrfs_item_ptr_offset(leaf, slot),
1695                                             size);
1696
1697                                 extent = btrfs_item_ptr(leaf, slot,
1698                                                 struct btrfs_file_extent_item);
1699
1700                                 /* disko == 0 means it's a hole */
1701                                 if (!disko)
1702                                         datao = 0;
1703
1704                                 btrfs_set_file_extent_offset(leaf, extent,
1705                                                              datao);
1706                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1707                                                                 datal);
1708                                 if (disko) {
1709                                         inode_add_bytes(inode, datal);
1710                                         ret = btrfs_inc_extent_ref(trans, root,
1711                                                         disko, diskl, 0,
1712                                                         root->root_key.objectid,
1713                                                         inode->i_ino,
1714                                                         new_key.offset - datao);
1715                                         BUG_ON(ret);
1716                                 }
1717                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1718                                 u64 skip = 0;
1719                                 u64 trim = 0;
1720                                 if (off > key.offset) {
1721                                         skip = off - key.offset;
1722                                         new_key.offset += skip;
1723                                 }
1724
1725                                 if (key.offset + datal > off+len)
1726                                         trim = key.offset + datal - (off+len);
1727
1728                                 if (comp && (skip || trim)) {
1729                                         ret = -EINVAL;
1730                                         btrfs_end_transaction(trans, root);
1731                                         goto out;
1732                                 }
1733                                 size -= skip + trim;
1734                                 datal -= skip + trim;
1735
1736                                 ret = btrfs_drop_extents(trans, inode,
1737                                                          new_key.offset,
1738                                                          new_key.offset + datal,
1739                                                          &hint_byte, 1);
1740                                 BUG_ON(ret);
1741
1742                                 ret = btrfs_insert_empty_item(trans, root, path,
1743                                                               &new_key, size);
1744                                 BUG_ON(ret);
1745
1746                                 if (skip) {
1747                                         u32 start =
1748                                           btrfs_file_extent_calc_inline_size(0);
1749                                         memmove(buf+start, buf+start+skip,
1750                                                 datal);
1751                                 }
1752
1753                                 leaf = path->nodes[0];
1754                                 slot = path->slots[0];
1755                                 write_extent_buffer(leaf, buf,
1756                                             btrfs_item_ptr_offset(leaf, slot),
1757                                             size);
1758                                 inode_add_bytes(inode, datal);
1759                         }
1760
1761                         btrfs_mark_buffer_dirty(leaf);
1762                         btrfs_release_path(root, path);
1763
1764                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1765
1766                         /*
1767                          * we round up to the block size at eof when
1768                          * determining which extents to clone above,
1769                          * but shouldn't round up the file size
1770                          */
1771                         endoff = new_key.offset + datal;
1772                         if (endoff > off+olen)
1773                                 endoff = off+olen;
1774                         if (endoff > inode->i_size)
1775                                 btrfs_i_size_write(inode, endoff);
1776
1777                         BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1778                         ret = btrfs_update_inode(trans, root, inode);
1779                         BUG_ON(ret);
1780                         btrfs_end_transaction(trans, root);
1781                 }
1782 next:
1783                 btrfs_release_path(root, path);
1784                 key.offset++;
1785         }
1786         ret = 0;
1787 out:
1788         btrfs_release_path(root, path);
1789         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1790 out_unlock:
1791         mutex_unlock(&src->i_mutex);
1792         mutex_unlock(&inode->i_mutex);
1793         vfree(buf);
1794         btrfs_free_path(path);
1795 out_fput:
1796         fput(src_file);
1797 out_drop_write:
1798         mnt_drop_write(file->f_path.mnt);
1799         return ret;
1800 }
1801
1802 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1803 {
1804         struct btrfs_ioctl_clone_range_args args;
1805
1806         if (copy_from_user(&args, argp, sizeof(args)))
1807                 return -EFAULT;
1808         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1809                                  args.src_length, args.dest_offset);
1810 }
1811
1812 /*
1813  * there are many ways the trans_start and trans_end ioctls can lead
1814  * to deadlocks.  They should only be used by applications that
1815  * basically own the machine, and have a very in depth understanding
1816  * of all the possible deadlocks and enospc problems.
1817  */
1818 static long btrfs_ioctl_trans_start(struct file *file)
1819 {
1820         struct inode *inode = fdentry(file)->d_inode;
1821         struct btrfs_root *root = BTRFS_I(inode)->root;
1822         struct btrfs_trans_handle *trans;
1823         int ret;
1824
1825         ret = -EPERM;
1826         if (!capable(CAP_SYS_ADMIN))
1827                 goto out;
1828
1829         ret = -EINPROGRESS;
1830         if (file->private_data)
1831                 goto out;
1832
1833         ret = mnt_want_write(file->f_path.mnt);
1834         if (ret)
1835                 goto out;
1836
1837         mutex_lock(&root->fs_info->trans_mutex);
1838         root->fs_info->open_ioctl_trans++;
1839         mutex_unlock(&root->fs_info->trans_mutex);
1840
1841         ret = -ENOMEM;
1842         trans = btrfs_start_ioctl_transaction(root, 0);
1843         if (!trans)
1844                 goto out_drop;
1845
1846         file->private_data = trans;
1847         return 0;
1848
1849 out_drop:
1850         mutex_lock(&root->fs_info->trans_mutex);
1851         root->fs_info->open_ioctl_trans--;
1852         mutex_unlock(&root->fs_info->trans_mutex);
1853         mnt_drop_write(file->f_path.mnt);
1854 out:
1855         return ret;
1856 }
1857
1858 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
1859 {
1860         struct inode *inode = fdentry(file)->d_inode;
1861         struct btrfs_root *root = BTRFS_I(inode)->root;
1862         struct btrfs_root *new_root;
1863         struct btrfs_dir_item *di;
1864         struct btrfs_trans_handle *trans;
1865         struct btrfs_path *path;
1866         struct btrfs_key location;
1867         struct btrfs_disk_key disk_key;
1868         struct btrfs_super_block *disk_super;
1869         u64 features;
1870         u64 objectid = 0;
1871         u64 dir_id;
1872
1873         if (!capable(CAP_SYS_ADMIN))
1874                 return -EPERM;
1875
1876         if (copy_from_user(&objectid, argp, sizeof(objectid)))
1877                 return -EFAULT;
1878
1879         if (!objectid)
1880                 objectid = root->root_key.objectid;
1881
1882         location.objectid = objectid;
1883         location.type = BTRFS_ROOT_ITEM_KEY;
1884         location.offset = (u64)-1;
1885
1886         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
1887         if (IS_ERR(new_root))
1888                 return PTR_ERR(new_root);
1889
1890         if (btrfs_root_refs(&new_root->root_item) == 0)
1891                 return -ENOENT;
1892
1893         path = btrfs_alloc_path();
1894         if (!path)
1895                 return -ENOMEM;
1896         path->leave_spinning = 1;
1897
1898         trans = btrfs_start_transaction(root, 1);
1899         if (!trans) {
1900                 btrfs_free_path(path);
1901                 return -ENOMEM;
1902         }
1903
1904         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
1905         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
1906                                    dir_id, "default", 7, 1);
1907         if (IS_ERR_OR_NULL(di)) {
1908                 btrfs_free_path(path);
1909                 btrfs_end_transaction(trans, root);
1910                 printk(KERN_ERR "Umm, you don't have the default dir item, "
1911                        "this isn't going to work\n");
1912                 return -ENOENT;
1913         }
1914
1915         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
1916         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
1917         btrfs_mark_buffer_dirty(path->nodes[0]);
1918         btrfs_free_path(path);
1919
1920         disk_super = &root->fs_info->super_copy;
1921         features = btrfs_super_incompat_flags(disk_super);
1922         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
1923                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
1924                 btrfs_set_super_incompat_flags(disk_super, features);
1925         }
1926         btrfs_end_transaction(trans, root);
1927
1928         return 0;
1929 }
1930
1931 static void get_block_group_info(struct list_head *groups_list,
1932                                  struct btrfs_ioctl_space_info *space)
1933 {
1934         struct btrfs_block_group_cache *block_group;
1935
1936         space->total_bytes = 0;
1937         space->used_bytes = 0;
1938         space->flags = 0;
1939         list_for_each_entry(block_group, groups_list, list) {
1940                 space->flags = block_group->flags;
1941                 space->total_bytes += block_group->key.offset;
1942                 space->used_bytes +=
1943                         btrfs_block_group_used(&block_group->item);
1944         }
1945 }
1946
1947 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
1948 {
1949         struct btrfs_ioctl_space_args space_args;
1950         struct btrfs_ioctl_space_info space;
1951         struct btrfs_ioctl_space_info *dest;
1952         struct btrfs_ioctl_space_info *dest_orig;
1953         struct btrfs_ioctl_space_info *user_dest;
1954         struct btrfs_space_info *info;
1955         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
1956                        BTRFS_BLOCK_GROUP_SYSTEM,
1957                        BTRFS_BLOCK_GROUP_METADATA,
1958                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
1959         int num_types = 4;
1960         int alloc_size;
1961         int ret = 0;
1962         int slot_count = 0;
1963         int i, c;
1964
1965         if (copy_from_user(&space_args,
1966                            (struct btrfs_ioctl_space_args __user *)arg,
1967                            sizeof(space_args)))
1968                 return -EFAULT;
1969
1970         for (i = 0; i < num_types; i++) {
1971                 struct btrfs_space_info *tmp;
1972
1973                 info = NULL;
1974                 rcu_read_lock();
1975                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
1976                                         list) {
1977                         if (tmp->flags == types[i]) {
1978                                 info = tmp;
1979                                 break;
1980                         }
1981                 }
1982                 rcu_read_unlock();
1983
1984                 if (!info)
1985                         continue;
1986
1987                 down_read(&info->groups_sem);
1988                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
1989                         if (!list_empty(&info->block_groups[c]))
1990                                 slot_count++;
1991                 }
1992                 up_read(&info->groups_sem);
1993         }
1994
1995         /* space_slots == 0 means they are asking for a count */
1996         if (space_args.space_slots == 0) {
1997                 space_args.total_spaces = slot_count;
1998                 goto out;
1999         }
2000
2001         slot_count = min_t(int, space_args.space_slots, slot_count);
2002
2003         alloc_size = sizeof(*dest) * slot_count;
2004
2005         /* we generally have at most 6 or so space infos, one for each raid
2006          * level.  So, a whole page should be more than enough for everyone
2007          */
2008         if (alloc_size > PAGE_CACHE_SIZE)
2009                 return -ENOMEM;
2010
2011         space_args.total_spaces = 0;
2012         dest = kmalloc(alloc_size, GFP_NOFS);
2013         if (!dest)
2014                 return -ENOMEM;
2015         dest_orig = dest;
2016
2017         /* now we have a buffer to copy into */
2018         for (i = 0; i < num_types; i++) {
2019                 struct btrfs_space_info *tmp;
2020
2021                 info = NULL;
2022                 rcu_read_lock();
2023                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2024                                         list) {
2025                         if (tmp->flags == types[i]) {
2026                                 info = tmp;
2027                                 break;
2028                         }
2029                 }
2030                 rcu_read_unlock();
2031
2032                 if (!info)
2033                         continue;
2034                 down_read(&info->groups_sem);
2035                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2036                         if (!list_empty(&info->block_groups[c])) {
2037                                 get_block_group_info(&info->block_groups[c],
2038                                                      &space);
2039                                 memcpy(dest, &space, sizeof(space));
2040                                 dest++;
2041                                 space_args.total_spaces++;
2042                         }
2043                 }
2044                 up_read(&info->groups_sem);
2045         }
2046
2047         user_dest = (struct btrfs_ioctl_space_info *)
2048                 (arg + sizeof(struct btrfs_ioctl_space_args));
2049
2050         if (copy_to_user(user_dest, dest_orig, alloc_size))
2051                 ret = -EFAULT;
2052
2053         kfree(dest_orig);
2054 out:
2055         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2056                 ret = -EFAULT;
2057
2058         return ret;
2059 }
2060
2061 /*
2062  * there are many ways the trans_start and trans_end ioctls can lead
2063  * to deadlocks.  They should only be used by applications that
2064  * basically own the machine, and have a very in depth understanding
2065  * of all the possible deadlocks and enospc problems.
2066  */
2067 long btrfs_ioctl_trans_end(struct file *file)
2068 {
2069         struct inode *inode = fdentry(file)->d_inode;
2070         struct btrfs_root *root = BTRFS_I(inode)->root;
2071         struct btrfs_trans_handle *trans;
2072
2073         trans = file->private_data;
2074         if (!trans)
2075                 return -EINVAL;
2076         file->private_data = NULL;
2077
2078         btrfs_end_transaction(trans, root);
2079
2080         mutex_lock(&root->fs_info->trans_mutex);
2081         root->fs_info->open_ioctl_trans--;
2082         mutex_unlock(&root->fs_info->trans_mutex);
2083
2084         mnt_drop_write(file->f_path.mnt);
2085         return 0;
2086 }
2087
2088 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2089 {
2090         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2091         struct btrfs_trans_handle *trans;
2092         u64 transid;
2093
2094         trans = btrfs_start_transaction(root, 0);
2095         transid = trans->transid;
2096         btrfs_commit_transaction_async(trans, root, 0);
2097
2098         if (argp)
2099                 if (copy_to_user(argp, &transid, sizeof(transid)))
2100                         return -EFAULT;
2101         return 0;
2102 }
2103
2104 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2105 {
2106         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2107         u64 transid;
2108
2109         if (argp) {
2110                 if (copy_from_user(&transid, argp, sizeof(transid)))
2111                         return -EFAULT;
2112         } else {
2113                 transid = 0;  /* current trans */
2114         }
2115         return btrfs_wait_for_commit(root, transid);
2116 }
2117
2118 long btrfs_ioctl(struct file *file, unsigned int
2119                 cmd, unsigned long arg)
2120 {
2121         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2122         void __user *argp = (void __user *)arg;
2123
2124         switch (cmd) {
2125         case FS_IOC_GETFLAGS:
2126                 return btrfs_ioctl_getflags(file, argp);
2127         case FS_IOC_SETFLAGS:
2128                 return btrfs_ioctl_setflags(file, argp);
2129         case FS_IOC_GETVERSION:
2130                 return btrfs_ioctl_getversion(file, argp);
2131         case BTRFS_IOC_SNAP_CREATE:
2132                 return btrfs_ioctl_snap_create(file, argp, 0, 0);
2133         case BTRFS_IOC_SNAP_CREATE_ASYNC:
2134                 return btrfs_ioctl_snap_create(file, argp, 0, 1);
2135         case BTRFS_IOC_SUBVOL_CREATE:
2136                 return btrfs_ioctl_snap_create(file, argp, 1, 0);
2137         case BTRFS_IOC_SNAP_DESTROY:
2138                 return btrfs_ioctl_snap_destroy(file, argp);
2139         case BTRFS_IOC_DEFAULT_SUBVOL:
2140                 return btrfs_ioctl_default_subvol(file, argp);
2141         case BTRFS_IOC_DEFRAG:
2142                 return btrfs_ioctl_defrag(file, NULL);
2143         case BTRFS_IOC_DEFRAG_RANGE:
2144                 return btrfs_ioctl_defrag(file, argp);
2145         case BTRFS_IOC_RESIZE:
2146                 return btrfs_ioctl_resize(root, argp);
2147         case BTRFS_IOC_ADD_DEV:
2148                 return btrfs_ioctl_add_dev(root, argp);
2149         case BTRFS_IOC_RM_DEV:
2150                 return btrfs_ioctl_rm_dev(root, argp);
2151         case BTRFS_IOC_BALANCE:
2152                 return btrfs_balance(root->fs_info->dev_root);
2153         case BTRFS_IOC_CLONE:
2154                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
2155         case BTRFS_IOC_CLONE_RANGE:
2156                 return btrfs_ioctl_clone_range(file, argp);
2157         case BTRFS_IOC_TRANS_START:
2158                 return btrfs_ioctl_trans_start(file);
2159         case BTRFS_IOC_TRANS_END:
2160                 return btrfs_ioctl_trans_end(file);
2161         case BTRFS_IOC_TREE_SEARCH:
2162                 return btrfs_ioctl_tree_search(file, argp);
2163         case BTRFS_IOC_INO_LOOKUP:
2164                 return btrfs_ioctl_ino_lookup(file, argp);
2165         case BTRFS_IOC_SPACE_INFO:
2166                 return btrfs_ioctl_space_info(root, argp);
2167         case BTRFS_IOC_SYNC:
2168                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
2169                 return 0;
2170         case BTRFS_IOC_START_SYNC:
2171                 return btrfs_ioctl_start_sync(file, argp);
2172         case BTRFS_IOC_WAIT_SYNC:
2173                 return btrfs_ioctl_wait_sync(file, argp);
2174         }
2175
2176         return -ENOTTY;
2177 }