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 "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "volumes.h"
  50 #include "locking.h"
  51
  52 /* Mask out flags that are inappropriate for the given type of inode. */
  53 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  54 {
  55         if (S_ISDIR(mode))
  56                 return flags;
  57         else if (S_ISREG(mode))
  58                 return flags & ~FS_DIRSYNC_FL;
  59         else
  60                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  61 }
  62
  63 /*
  64  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  65  */
  66 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  67 {
  68         unsigned int iflags = 0;
  69
  70         if (flags & BTRFS_INODE_SYNC)
  71                 iflags |= FS_SYNC_FL;
  72         if (flags & BTRFS_INODE_IMMUTABLE)
  73                 iflags |= FS_IMMUTABLE_FL;
  74         if (flags & BTRFS_INODE_APPEND)
  75                 iflags |= FS_APPEND_FL;
  76         if (flags & BTRFS_INODE_NODUMP)
  77                 iflags |= FS_NODUMP_FL;
  78         if (flags & BTRFS_INODE_NOATIME)
  79                 iflags |= FS_NOATIME_FL;
  80         if (flags & BTRFS_INODE_DIRSYNC)
  81                 iflags |= FS_DIRSYNC_FL;
  82
  83         return iflags;
  84 }
  85
  86 /*
  87  * Update inode->i_flags based on the btrfs internal flags.
  88  */
  89 void btrfs_update_iflags(struct inode *inode)
  90 {
  91         struct btrfs_inode *ip = BTRFS_I(inode);
  92
  93         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
  94
  95         if (ip->flags & BTRFS_INODE_SYNC)
  96                 inode->i_flags |= S_SYNC;
  97         if (ip->flags & BTRFS_INODE_IMMUTABLE)
  98                 inode->i_flags |= S_IMMUTABLE;
  99         if (ip->flags & BTRFS_INODE_APPEND)
 100                 inode->i_flags |= S_APPEND;
 101         if (ip->flags & BTRFS_INODE_NOATIME)
 102                 inode->i_flags |= S_NOATIME;
 103         if (ip->flags & BTRFS_INODE_DIRSYNC)
 104                 inode->i_flags |= S_DIRSYNC;
 105 }
 106
 107 /*
 108  * Inherit flags from the parent inode.
 109  *
 110  * Unlike extN we don't have any flags we don't want to inherit currently.
 111  */
 112 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 113 {
 114         unsigned int flags;
 115
 116         if (!dir)
 117                 return;
 118
 119         flags = BTRFS_I(dir)->flags;
 120
 121         if (S_ISREG(inode->i_mode))
 122                 flags &= ~BTRFS_INODE_DIRSYNC;
 123         else if (!S_ISDIR(inode->i_mode))
 124                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
 125
 126         BTRFS_I(inode)->flags = flags;
 127         btrfs_update_iflags(inode);
 128 }
 129
 130 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 131 {
 132         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 133         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 134
 135         if (copy_to_user(arg, &flags, sizeof(flags)))
 136                 return -EFAULT;
 137         return 0;
 138 }
 139
 140 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 141 {
 142         struct inode *inode = file->f_path.dentry->d_inode;
 143         struct btrfs_inode *ip = BTRFS_I(inode);
 144         struct btrfs_root *root = ip->root;
 145         struct btrfs_trans_handle *trans;
 146         unsigned int flags, oldflags;
 147         int ret;
 148
 149         if (copy_from_user(&flags, arg, sizeof(flags)))
 150                 return -EFAULT;
 151
 152         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 153                       FS_NOATIME_FL | FS_NODUMP_FL | \
 154                       FS_SYNC_FL | FS_DIRSYNC_FL))
 155                 return -EOPNOTSUPP;
 156
 157         if (!is_owner_or_cap(inode))
 158                 return -EACCES;
 159
 160         mutex_lock(&inode->i_mutex);
 161
 162         flags = btrfs_mask_flags(inode->i_mode, flags);
 163         oldflags = btrfs_flags_to_ioctl(ip->flags);
 164         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 165                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 166                         ret = -EPERM;
 167                         goto out_unlock;
 168                 }
 169         }
 170
 171         ret = mnt_want_write(file->f_path.mnt);
 172         if (ret)
 173                 goto out_unlock;
 174
 175         if (flags & FS_SYNC_FL)
 176                 ip->flags |= BTRFS_INODE_SYNC;
 177         else
 178                 ip->flags &= ~BTRFS_INODE_SYNC;
 179         if (flags & FS_IMMUTABLE_FL)
 180                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 181         else
 182                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 183         if (flags & FS_APPEND_FL)
 184                 ip->flags |= BTRFS_INODE_APPEND;
 185         else
 186                 ip->flags &= ~BTRFS_INODE_APPEND;
 187         if (flags & FS_NODUMP_FL)
 188                 ip->flags |= BTRFS_INODE_NODUMP;
 189         else
 190                 ip->flags &= ~BTRFS_INODE_NODUMP;
 191         if (flags & FS_NOATIME_FL)
 192                 ip->flags |= BTRFS_INODE_NOATIME;
 193         else
 194                 ip->flags &= ~BTRFS_INODE_NOATIME;
 195         if (flags & FS_DIRSYNC_FL)
 196                 ip->flags |= BTRFS_INODE_DIRSYNC;
 197         else
 198                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 199
 200
 201         trans = btrfs_join_transaction(root, 1);
 202         BUG_ON(!trans);
 203
 204         ret = btrfs_update_inode(trans, root, inode);
 205         BUG_ON(ret);
 206
 207         btrfs_update_iflags(inode);
 208         inode->i_ctime = CURRENT_TIME;
 209         btrfs_end_transaction(trans, root);
 210
 211         mnt_drop_write(file->f_path.mnt);
 212  out_unlock:
 213         mutex_unlock(&inode->i_mutex);
 214         return 0;
 215 }
 216
 217 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 218 {
 219         struct inode *inode = file->f_path.dentry->d_inode;
 220
 221         return put_user(inode->i_generation, arg);
 222 }
 223
 224 static noinline int create_subvol(struct btrfs_root *root,
 225                                   struct dentry *dentry,
 226                                   char *name, int namelen)
 227 {
 228         struct btrfs_trans_handle *trans;
 229         struct btrfs_key key;
 230         struct btrfs_root_item root_item;
 231         struct btrfs_inode_item *inode_item;
 232         struct extent_buffer *leaf;
 233         struct btrfs_root *new_root = root;
 234         struct inode *dir;
 235         int ret;
 236         int err;
 237         u64 objectid;
 238         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 239         u64 index = 0;
 240         unsigned long nr = 1;
 241
 242         ret = btrfs_check_metadata_free_space(root);
 243         if (ret)
 244                 goto fail_commit;
 245
 246         trans = btrfs_start_transaction(root, 1);
 247         BUG_ON(!trans);
 248
 249         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
 250                                        0, &objectid);
 251         if (ret)
 252                 goto fail;
 253
 254         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 255                                       0, objectid, NULL, 0, 0, 0);
 256         if (IS_ERR(leaf)) {
 257                 ret = PTR_ERR(leaf);
 258                 goto fail;
 259         }
 260
 261         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 262         btrfs_set_header_bytenr(leaf, leaf->start);
 263         btrfs_set_header_generation(leaf, trans->transid);
 264         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 265         btrfs_set_header_owner(leaf, objectid);
 266
 267         write_extent_buffer(leaf, root->fs_info->fsid,
 268                             (unsigned long)btrfs_header_fsid(leaf),
 269                             BTRFS_FSID_SIZE);
 270         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 271                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 272                             BTRFS_UUID_SIZE);
 273         btrfs_mark_buffer_dirty(leaf);
 274
 275         inode_item = &root_item.inode;
 276         memset(inode_item, 0, sizeof(*inode_item));
 277         inode_item->generation = cpu_to_le64(1);
 278         inode_item->size = cpu_to_le64(3);
 279         inode_item->nlink = cpu_to_le32(1);
 280         inode_item->nbytes = cpu_to_le64(root->leafsize);
 281         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 282
 283         btrfs_set_root_bytenr(&root_item, leaf->start);
 284         btrfs_set_root_generation(&root_item, trans->transid);
 285         btrfs_set_root_level(&root_item, 0);
 286         btrfs_set_root_refs(&root_item, 1);
 287         btrfs_set_root_used(&root_item, 0);
 288         btrfs_set_root_last_snapshot(&root_item, 0);
 289
 290         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 291         root_item.drop_level = 0;
 292
 293         btrfs_tree_unlock(leaf);
 294         free_extent_buffer(leaf);
 295         leaf = NULL;
 296
 297         btrfs_set_root_dirid(&root_item, new_dirid);
 298
 299         key.objectid = objectid;
 300         key.offset = 0;
 301         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 302         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 303                                 &root_item);
 304         if (ret)
 305                 goto fail;
 306
 307         /*
 308          * insert the directory item
 309          */
 310         key.offset = (u64)-1;
 311         dir = dentry->d_parent->d_inode;
 312         ret = btrfs_set_inode_index(dir, &index);
 313         BUG_ON(ret);
 314
 315         ret = btrfs_insert_dir_item(trans, root,
 316                                     name, namelen, dir->i_ino, &key,
 317                                     BTRFS_FT_DIR, index);
 318         if (ret)
 319                 goto fail;
 320
 321         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 322         ret = btrfs_update_inode(trans, root, dir);
 323         BUG_ON(ret);
 324
 325         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 326                                  objectid, root->root_key.objectid,
 327                                  dir->i_ino, index, name, namelen);
 328         BUG_ON(ret);
 329
 330         ret = btrfs_commit_transaction(trans, root);
 331         if (ret)
 332                 goto fail_commit;
 333
 334         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 335         BUG_ON(!new_root);
 336
 337         trans = btrfs_start_transaction(new_root, 1);
 338         BUG_ON(!trans);
 339
 340         ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
 341                                        BTRFS_I(dir)->block_group);
 342         if (ret)
 343                 goto fail;
 344
 345 fail:
 346         nr = trans->blocks_used;
 347         err = btrfs_commit_transaction(trans, new_root);
 348         if (err && !ret)
 349                 ret = err;
 350 fail_commit:
 351         btrfs_btree_balance_dirty(root, nr);
 352         return ret;
 353 }
 354
 355 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 356                            char *name, int namelen)
 357 {
 358         struct btrfs_pending_snapshot *pending_snapshot;
 359         struct btrfs_trans_handle *trans;
 360         int ret = 0;
 361         int err;
 362         unsigned long nr = 0;
 363
 364         if (!root->ref_cows)
 365                 return -EINVAL;
 366
 367         ret = btrfs_check_metadata_free_space(root);
 368         if (ret)
 369                 goto fail_unlock;
 370
 371         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 372         if (!pending_snapshot) {
 373                 ret = -ENOMEM;
 374                 goto fail_unlock;
 375         }
 376         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 377         if (!pending_snapshot->name) {
 378                 ret = -ENOMEM;
 379                 kfree(pending_snapshot);
 380                 goto fail_unlock;
 381         }
 382         memcpy(pending_snapshot->name, name, namelen);
 383         pending_snapshot->name[namelen] = '\0';
 384         pending_snapshot->dentry = dentry;
 385         trans = btrfs_start_transaction(root, 1);
 386         BUG_ON(!trans);
 387         pending_snapshot->root = root;
 388         list_add(&pending_snapshot->list,
 389                  &trans->transaction->pending_snapshots);
 390         err = btrfs_commit_transaction(trans, root);
 391
 392 fail_unlock:
 393         btrfs_btree_balance_dirty(root, nr);
 394         return ret;
 395 }
 396
 397 /* copy of may_create in fs/namei.c() */
 398 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 399 {
 400         if (child->d_inode)
 401                 return -EEXIST;
 402         if (IS_DEADDIR(dir))
 403                 return -ENOENT;
 404         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 405 }
 406
 407 /*
 408  * Create a new subvolume below @parent.  This is largely modeled after
 409  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 410  * inside this filesystem so it's quite a bit simpler.
 411  */
 412 static noinline int btrfs_mksubvol(struct path *parent, char *name,
 413                                    int mode, int namelen,
 414                                    struct btrfs_root *snap_src)
 415 {
 416         struct dentry *dentry;
 417         int error;
 418
 419         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 420
 421         dentry = lookup_one_len(name, parent->dentry, namelen);
 422         error = PTR_ERR(dentry);
 423         if (IS_ERR(dentry))
 424                 goto out_unlock;
 425
 426         error = -EEXIST;
 427         if (dentry->d_inode)
 428                 goto out_dput;
 429
 430         if (!IS_POSIXACL(parent->dentry->d_inode))
 431                 mode &= ~current_umask();
 432
 433         error = mnt_want_write(parent->mnt);
 434         if (error)
 435                 goto out_dput;
 436
 437         error = btrfs_may_create(parent->dentry->d_inode, dentry);
 438         if (error)
 439                 goto out_drop_write;
 440
 441         /*
 442          * Actually perform the low-level subvolume creation after all
 443          * this VFS fuzz.
 444          *
 445          * Eventually we want to pass in an inode under which we create this
 446          * subvolume, but for now all are under the filesystem root.
 447          *
 448          * Also we should pass on the mode eventually to allow creating new
 449          * subvolume with specific mode bits.
 450          */
 451         if (snap_src) {
 452                 struct dentry *dir = dentry->d_parent;
 453                 struct dentry *test = dir->d_parent;
 454                 struct btrfs_path *path = btrfs_alloc_path();
 455                 int ret;
 456                 u64 test_oid;
 457                 u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
 458
 459                 test_oid = snap_src->root_key.objectid;
 460
 461                 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
 462                                           path, parent_oid, test_oid);
 463                 if (ret == 0)
 464                         goto create;
 465                 btrfs_release_path(snap_src->fs_info->tree_root, path);
 466
 467                 /* we need to make sure we aren't creating a directory loop
 468                  * by taking a snapshot of something that has our current
 469                  * subvol in its directory tree.  So, this loops through
 470                  * the dentries and checks the forward refs for each subvolume
 471                  * to see if is references the subvolume where we are
 472                  * placing this new snapshot.
 473                  */
 474                 while (1) {
 475                         if (!test ||
 476                             dir == snap_src->fs_info->sb->s_root ||
 477                             test == snap_src->fs_info->sb->s_root ||
 478                             test->d_inode->i_sb != snap_src->fs_info->sb) {
 479                                 break;
 480                         }
 481                         if (S_ISLNK(test->d_inode->i_mode)) {
 482                                 printk(KERN_INFO "Btrfs symlink in snapshot "
 483                                        "path, failed\n");
 484                                 error = -EMLINK;
 485                                 btrfs_free_path(path);
 486                                 goto out_drop_write;
 487                         }
 488                         test_oid =
 489                                 BTRFS_I(test->d_inode)->root->root_key.objectid;
 490                         ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
 491                                   path, test_oid, parent_oid);
 492                         if (ret == 0) {
 493                                 printk(KERN_INFO "Btrfs snapshot creation "
 494                                        "failed, looping\n");
 495                                 error = -EMLINK;
 496                                 btrfs_free_path(path);
 497                                 goto out_drop_write;
 498                         }
 499                         btrfs_release_path(snap_src->fs_info->tree_root, path);
 500                         test = test->d_parent;
 501                 }
 502 create:
 503                 btrfs_free_path(path);
 504                 error = create_snapshot(snap_src, dentry, name, namelen);
 505         } else {
 506                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
 507                                       dentry, name, namelen);
 508         }
 509         if (error)
 510                 goto out_drop_write;
 511
 512         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 513 out_drop_write:
 514         mnt_drop_write(parent->mnt);
 515 out_dput:
 516         dput(dentry);
 517 out_unlock:
 518         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 519         return error;
 520 }
 521
 522
 523 static int btrfs_defrag_file(struct file *file)
 524 {
 525         struct inode *inode = fdentry(file)->d_inode;
 526         struct btrfs_root *root = BTRFS_I(inode)->root;
 527         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 528         struct btrfs_ordered_extent *ordered;
 529         struct page *page;
 530         unsigned long last_index;
 531         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 532         unsigned long total_read = 0;
 533         u64 page_start;
 534         u64 page_end;
 535         unsigned long i;
 536         int ret;
 537
 538         ret = btrfs_check_data_free_space(root, inode, inode->i_size);
 539         if (ret)
 540                 return -ENOSPC;
 541
 542         mutex_lock(&inode->i_mutex);
 543         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 544         for (i = 0; i <= last_index; i++) {
 545                 if (total_read % ra_pages == 0) {
 546                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 547                                        min(last_index, i + ra_pages - 1));
 548                 }
 549                 total_read++;
 550 again:
 551                 page = grab_cache_page(inode->i_mapping, i);
 552                 if (!page)
 553                         goto out_unlock;
 554                 if (!PageUptodate(page)) {
 555                         btrfs_readpage(NULL, page);
 556                         lock_page(page);
 557                         if (!PageUptodate(page)) {
 558                                 unlock_page(page);
 559                                 page_cache_release(page);
 560                                 goto out_unlock;
 561                         }
 562                 }
 563
 564                 wait_on_page_writeback(page);
 565
 566                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 567                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 568                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 569
 570                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 571                 if (ordered) {
 572                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 573                         unlock_page(page);
 574                         page_cache_release(page);
 575                         btrfs_start_ordered_extent(inode, ordered, 1);
 576                         btrfs_put_ordered_extent(ordered);
 577                         goto again;
 578                 }
 579                 set_page_extent_mapped(page);
 580
 581                 /*
 582                  * this makes sure page_mkwrite is called on the
 583                  * page if it is dirtied again later
 584                  */
 585                 clear_page_dirty_for_io(page);
 586
 587                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 588                 set_page_dirty(page);
 589                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 590                 unlock_page(page);
 591                 page_cache_release(page);
 592                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 593         }
 594
 595 out_unlock:
 596         mutex_unlock(&inode->i_mutex);
 597         return 0;
 598 }
 599
 600 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 601 {
 602         u64 new_size;
 603         u64 old_size;
 604         u64 devid = 1;
 605         struct btrfs_ioctl_vol_args *vol_args;
 606         struct btrfs_trans_handle *trans;
 607         struct btrfs_device *device = NULL;
 608         char *sizestr;
 609         char *devstr = NULL;
 610         int ret = 0;
 611         int namelen;
 612         int mod = 0;
 613
 614         if (root->fs_info->sb->s_flags & MS_RDONLY)
 615                 return -EROFS;
 616
 617         if (!capable(CAP_SYS_ADMIN))
 618                 return -EPERM;
 619
 620         vol_args = memdup_user(arg, sizeof(*vol_args));
 621         if (IS_ERR(vol_args))
 622                 return PTR_ERR(vol_args);
 623
 624         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 625         namelen = strlen(vol_args->name);
 626
 627         mutex_lock(&root->fs_info->volume_mutex);
 628         sizestr = vol_args->name;
 629         devstr = strchr(sizestr, ':');
 630         if (devstr) {
 631                 char *end;
 632                 sizestr = devstr + 1;
 633                 *devstr = '\0';
 634                 devstr = vol_args->name;
 635                 devid = simple_strtoull(devstr, &end, 10);
 636                 printk(KERN_INFO "resizing devid %llu\n",
 637                        (unsigned long long)devid);
 638         }
 639         device = btrfs_find_device(root, devid, NULL, NULL);
 640         if (!device) {
 641                 printk(KERN_INFO "resizer unable to find device %llu\n",
 642                        (unsigned long long)devid);
 643                 ret = -EINVAL;
 644                 goto out_unlock;
 645         }
 646         if (!strcmp(sizestr, "max"))
 647                 new_size = device->bdev->bd_inode->i_size;
 648         else {
 649                 if (sizestr[0] == '-') {
 650                         mod = -1;
 651                         sizestr++;
 652                 } else if (sizestr[0] == '+') {
 653                         mod = 1;
 654                         sizestr++;
 655                 }
 656                 new_size = btrfs_parse_size(sizestr);
 657                 if (new_size == 0) {
 658                         ret = -EINVAL;
 659                         goto out_unlock;
 660                 }
 661         }
 662
 663         old_size = device->total_bytes;
 664
 665         if (mod < 0) {
 666                 if (new_size > old_size) {
 667                         ret = -EINVAL;
 668                         goto out_unlock;
 669                 }
 670                 new_size = old_size - new_size;
 671         } else if (mod > 0) {
 672                 new_size = old_size + new_size;
 673         }
 674
 675         if (new_size < 256 * 1024 * 1024) {
 676                 ret = -EINVAL;
 677                 goto out_unlock;
 678         }
 679         if (new_size > device->bdev->bd_inode->i_size) {
 680                 ret = -EFBIG;
 681                 goto out_unlock;
 682         }
 683
 684         do_div(new_size, root->sectorsize);
 685         new_size *= root->sectorsize;
 686
 687         printk(KERN_INFO "new size for %s is %llu\n",
 688                 device->name, (unsigned long long)new_size);
 689
 690         if (new_size > old_size) {
 691                 trans = btrfs_start_transaction(root, 1);
 692                 ret = btrfs_grow_device(trans, device, new_size);
 693                 btrfs_commit_transaction(trans, root);
 694         } else {
 695                 ret = btrfs_shrink_device(device, new_size);
 696         }
 697
 698 out_unlock:
 699         mutex_unlock(&root->fs_info->volume_mutex);
 700         kfree(vol_args);
 701         return ret;
 702 }
 703
 704 static noinline int btrfs_ioctl_snap_create(struct file *file,
 705                                             void __user *arg, int subvol)
 706 {
 707         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 708         struct btrfs_ioctl_vol_args *vol_args;
 709         struct btrfs_dir_item *di;
 710         struct btrfs_path *path;
 711         struct file *src_file;
 712         u64 root_dirid;
 713         int namelen;
 714         int ret = 0;
 715
 716         if (root->fs_info->sb->s_flags & MS_RDONLY)
 717                 return -EROFS;
 718
 719         vol_args = memdup_user(arg, sizeof(*vol_args));
 720         if (IS_ERR(vol_args))
 721                 return PTR_ERR(vol_args);
 722
 723         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 724         namelen = strlen(vol_args->name);
 725         if (strchr(vol_args->name, '/')) {
 726                 ret = -EINVAL;
 727                 goto out;
 728         }
 729
 730         path = btrfs_alloc_path();
 731         if (!path) {
 732                 ret = -ENOMEM;
 733                 goto out;
 734         }
 735
 736         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 737         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 738                             path, root_dirid,
 739                             vol_args->name, namelen, 0);
 740         btrfs_free_path(path);
 741
 742         if (di && !IS_ERR(di)) {
 743                 ret = -EEXIST;
 744                 goto out;
 745         }
 746
 747         if (IS_ERR(di)) {
 748                 ret = PTR_ERR(di);
 749                 goto out;
 750         }
 751
 752         if (subvol) {
 753                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 754                                      file->f_path.dentry->d_inode->i_mode,
 755                                      namelen, NULL);
 756         } else {
 757                 struct inode *src_inode;
 758                 src_file = fget(vol_args->fd);
 759                 if (!src_file) {
 760                         ret = -EINVAL;
 761                         goto out;
 762                 }
 763
 764                 src_inode = src_file->f_path.dentry->d_inode;
 765                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 766                         printk(KERN_INFO "btrfs: Snapshot src from "
 767                                "another FS\n");
 768                         ret = -EINVAL;
 769                         fput(src_file);
 770                         goto out;
 771                 }
 772                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 773                              file->f_path.dentry->d_inode->i_mode,
 774                              namelen, BTRFS_I(src_inode)->root);
 775                 fput(src_file);
 776         }
 777
 778 out:
 779         kfree(vol_args);
 780         return ret;
 781 }
 782
 783 static int btrfs_ioctl_defrag(struct file *file)
 784 {
 785         struct inode *inode = fdentry(file)->d_inode;
 786         struct btrfs_root *root = BTRFS_I(inode)->root;
 787         int ret;
 788
 789         ret = mnt_want_write(file->f_path.mnt);
 790         if (ret)
 791                 return ret;
 792
 793         switch (inode->i_mode & S_IFMT) {
 794         case S_IFDIR:
 795                 if (!capable(CAP_SYS_ADMIN)) {
 796                         ret = -EPERM;
 797                         goto out;
 798                 }
 799                 btrfs_defrag_root(root, 0);
 800                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 801                 break;
 802         case S_IFREG:
 803                 if (!(file->f_mode & FMODE_WRITE)) {
 804                         ret = -EINVAL;
 805                         goto out;
 806                 }
 807                 btrfs_defrag_file(file);
 808                 break;
 809         }
 810 out:
 811         mnt_drop_write(file->f_path.mnt);
 812         return ret;
 813 }
 814
 815 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 816 {
 817         struct btrfs_ioctl_vol_args *vol_args;
 818         int ret;
 819
 820         if (!capable(CAP_SYS_ADMIN))
 821                 return -EPERM;
 822
 823         vol_args = memdup_user(arg, sizeof(*vol_args));
 824         if (IS_ERR(vol_args))
 825                 return PTR_ERR(vol_args);
 826
 827         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 828         ret = btrfs_init_new_device(root, vol_args->name);
 829
 830         kfree(vol_args);
 831         return ret;
 832 }
 833
 834 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 835 {
 836         struct btrfs_ioctl_vol_args *vol_args;
 837         int ret;
 838
 839         if (!capable(CAP_SYS_ADMIN))
 840                 return -EPERM;
 841
 842         if (root->fs_info->sb->s_flags & MS_RDONLY)
 843                 return -EROFS;
 844
 845         vol_args = memdup_user(arg, sizeof(*vol_args));
 846         if (IS_ERR(vol_args))
 847                 return PTR_ERR(vol_args);
 848
 849         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 850         ret = btrfs_rm_device(root, vol_args->name);
 851
 852         kfree(vol_args);
 853         return ret;
 854 }
 855
 856 static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 857                 u64 off, u64 olen, u64 destoff)
 858 {
 859         struct inode *inode = fdentry(file)->d_inode;
 860         struct btrfs_root *root = BTRFS_I(inode)->root;
 861         struct file *src_file;
 862         struct inode *src;
 863         struct btrfs_trans_handle *trans;
 864         struct btrfs_path *path;
 865         struct extent_buffer *leaf;
 866         char *buf;
 867         struct btrfs_key key;
 868         u32 nritems;
 869         int slot;
 870         int ret;
 871         u64 len = olen;
 872         u64 bs = root->fs_info->sb->s_blocksize;
 873         u64 hint_byte;
 874
 875         /*
 876          * TODO:
 877          * - split compressed inline extents.  annoying: we need to
 878          *   decompress into destination's address_space (the file offset
 879          *   may change, so source mapping won't do), then recompress (or
 880          *   otherwise reinsert) a subrange.
 881          * - allow ranges within the same file to be cloned (provided
 882          *   they don't overlap)?
 883          */
 884
 885         /* the destination must be opened for writing */
 886         if (!(file->f_mode & FMODE_WRITE))
 887                 return -EINVAL;
 888
 889         ret = mnt_want_write(file->f_path.mnt);
 890         if (ret)
 891                 return ret;
 892
 893         src_file = fget(srcfd);
 894         if (!src_file) {
 895                 ret = -EBADF;
 896                 goto out_drop_write;
 897         }
 898         src = src_file->f_dentry->d_inode;
 899
 900         ret = -EINVAL;
 901         if (src == inode)
 902                 goto out_fput;
 903
 904         ret = -EISDIR;
 905         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 906                 goto out_fput;
 907
 908         ret = -EXDEV;
 909         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 910                 goto out_fput;
 911
 912         ret = -ENOMEM;
 913         buf = vmalloc(btrfs_level_size(root, 0));
 914         if (!buf)
 915                 goto out_fput;
 916
 917         path = btrfs_alloc_path();
 918         if (!path) {
 919                 vfree(buf);
 920                 goto out_fput;
 921         }
 922         path->reada = 2;
 923
 924         if (inode < src) {
 925                 mutex_lock(&inode->i_mutex);
 926                 mutex_lock(&src->i_mutex);
 927         } else {
 928                 mutex_lock(&src->i_mutex);
 929                 mutex_lock(&inode->i_mutex);
 930         }
 931
 932         /* determine range to clone */
 933         ret = -EINVAL;
 934         if (off >= src->i_size || off + len > src->i_size)
 935                 goto out_unlock;
 936         if (len == 0)
 937                 olen = len = src->i_size - off;
 938         /* if we extend to eof, continue to block boundary */
 939         if (off + len == src->i_size)
 940                 len = ((src->i_size + bs-1) & ~(bs-1))
 941                         - off;
 942
 943         /* verify the end result is block aligned */
 944         if ((off & (bs-1)) ||
 945             ((off + len) & (bs-1)))
 946                 goto out_unlock;
 947
 948         /* do any pending delalloc/csum calc on src, one way or
 949            another, and lock file content */
 950         while (1) {
 951                 struct btrfs_ordered_extent *ordered;
 952                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 953                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
 954                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 955                         break;
 956                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 957                 if (ordered)
 958                         btrfs_put_ordered_extent(ordered);
 959                 btrfs_wait_ordered_range(src, off, off+len);
 960         }
 961
 962         trans = btrfs_start_transaction(root, 1);
 963         BUG_ON(!trans);
 964
 965         /* punch hole in destination first */
 966         btrfs_drop_extents(trans, root, inode, off, off + len,
 967                            off + len, 0, &hint_byte, 1);
 968
 969         /* clone data */
 970         key.objectid = src->i_ino;
 971         key.type = BTRFS_EXTENT_DATA_KEY;
 972         key.offset = 0;
 973
 974         while (1) {
 975                 /*
 976                  * note the key will change type as we walk through the
 977                  * tree.
 978                  */
 979                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 980                 if (ret < 0)
 981                         goto out;
 982
 983                 nritems = btrfs_header_nritems(path->nodes[0]);
 984                 if (path->slots[0] >= nritems) {
 985                         ret = btrfs_next_leaf(root, path);
 986                         if (ret < 0)
 987                                 goto out;
 988                         if (ret > 0)
 989                                 break;
 990                         nritems = btrfs_header_nritems(path->nodes[0]);
 991                 }
 992                 leaf = path->nodes[0];
 993                 slot = path->slots[0];
 994
 995                 btrfs_item_key_to_cpu(leaf, &key, slot);
 996                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 997                     key.objectid != src->i_ino)
 998                         break;
 999
1000                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1001                         struct btrfs_file_extent_item *extent;
1002                         int type;
1003                         u32 size;
1004                         struct btrfs_key new_key;
1005                         u64 disko = 0, diskl = 0;
1006                         u64 datao = 0, datal = 0;
1007                         u8 comp;
1008
1009                         size = btrfs_item_size_nr(leaf, slot);
1010                         read_extent_buffer(leaf, buf,
1011                                            btrfs_item_ptr_offset(leaf, slot),
1012                                            size);
1013
1014                         extent = btrfs_item_ptr(leaf, slot,
1015                                                 struct btrfs_file_extent_item);
1016                         comp = btrfs_file_extent_compression(leaf, extent);
1017                         type = btrfs_file_extent_type(leaf, extent);
1018                         if (type == BTRFS_FILE_EXTENT_REG ||
1019                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1020                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1021                                                                       extent);
1022                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1023                                                                  extent);
1024                                 datao = btrfs_file_extent_offset(leaf, extent);
1025                                 datal = btrfs_file_extent_num_bytes(leaf,
1026                                                                     extent);
1027                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1028                                 /* take upper bound, may be compressed */
1029                                 datal = btrfs_file_extent_ram_bytes(leaf,
1030                                                                     extent);
1031                         }
1032                         btrfs_release_path(root, path);
1033
1034                         if (key.offset + datal < off ||
1035                             key.offset >= off+len)
1036                                 goto next;
1037
1038                         memcpy(&new_key, &key, sizeof(new_key));
1039                         new_key.objectid = inode->i_ino;
1040                         new_key.offset = key.offset + destoff - off;
1041
1042                         if (type == BTRFS_FILE_EXTENT_REG ||
1043                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1044                                 ret = btrfs_insert_empty_item(trans, root, path,
1045                                                               &new_key, size);
1046                                 if (ret)
1047                                         goto out;
1048
1049                                 leaf = path->nodes[0];
1050                                 slot = path->slots[0];
1051                                 write_extent_buffer(leaf, buf,
1052                                             btrfs_item_ptr_offset(leaf, slot),
1053                                             size);
1054
1055                                 extent = btrfs_item_ptr(leaf, slot,
1056                                                 struct btrfs_file_extent_item);
1057
1058                                 if (off > key.offset) {
1059                                         datao += off - key.offset;
1060                                         datal -= off - key.offset;
1061                                 }
1062                                 if (key.offset + datao + datal + key.offset >
1063                                     off + len)
1064                                         datal = off + len - key.offset - datao;
1065                                 /* disko == 0 means it's a hole */
1066                                 if (!disko)
1067                                         datao = 0;
1068
1069                                 btrfs_set_file_extent_offset(leaf, extent,
1070                                                              datao);
1071                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1072                                                                 datal);
1073                                 if (disko) {
1074                                         inode_add_bytes(inode, datal);
1075                                         ret = btrfs_inc_extent_ref(trans, root,
1076                                                         disko, diskl, 0,
1077                                                         root->root_key.objectid,
1078                                                         inode->i_ino,
1079                                                         new_key.offset - datao);
1080                                         BUG_ON(ret);
1081                                 }
1082                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1083                                 u64 skip = 0;
1084                                 u64 trim = 0;
1085                                 if (off > key.offset) {
1086                                         skip = off - key.offset;
1087                                         new_key.offset += skip;
1088                                 }
1089
1090                                 if (key.offset + datal > off+len)
1091                                         trim = key.offset + datal - (off+len);
1092
1093                                 if (comp && (skip || trim)) {
1094                                         ret = -EINVAL;
1095                                         goto out;
1096                                 }
1097                                 size -= skip + trim;
1098                                 datal -= skip + trim;
1099                                 ret = btrfs_insert_empty_item(trans, root, path,
1100                                                               &new_key, size);
1101                                 if (ret)
1102                                         goto out;
1103
1104                                 if (skip) {
1105                                         u32 start =
1106                                           btrfs_file_extent_calc_inline_size(0);
1107                                         memmove(buf+start, buf+start+skip,
1108                                                 datal);
1109                                 }
1110
1111                                 leaf = path->nodes[0];
1112                                 slot = path->slots[0];
1113                                 write_extent_buffer(leaf, buf,
1114                                             btrfs_item_ptr_offset(leaf, slot),
1115                                             size);
1116                                 inode_add_bytes(inode, datal);
1117                         }
1118
1119                         btrfs_mark_buffer_dirty(leaf);
1120                 }
1121
1122 next:
1123                 btrfs_release_path(root, path);
1124                 key.offset++;
1125         }
1126         ret = 0;
1127 out:
1128         btrfs_release_path(root, path);
1129         if (ret == 0) {
1130                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1131                 if (destoff + olen > inode->i_size)
1132                         btrfs_i_size_write(inode, destoff + olen);
1133                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1134                 ret = btrfs_update_inode(trans, root, inode);
1135         }
1136         btrfs_end_transaction(trans, root);
1137         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1138         if (ret)
1139                 vmtruncate(inode, 0);
1140 out_unlock:
1141         mutex_unlock(&src->i_mutex);
1142         mutex_unlock(&inode->i_mutex);
1143         vfree(buf);
1144         btrfs_free_path(path);
1145 out_fput:
1146         fput(src_file);
1147 out_drop_write:
1148         mnt_drop_write(file->f_path.mnt);
1149         return ret;
1150 }
1151
1152 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1153 {
1154         struct btrfs_ioctl_clone_range_args args;
1155
1156         if (copy_from_user(&args, argp, sizeof(args)))
1157                 return -EFAULT;
1158         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1159                                  args.src_length, args.dest_offset);
1160 }
1161
1162 /*
1163  * there are many ways the trans_start and trans_end ioctls can lead
1164  * to deadlocks.  They should only be used by applications that
1165  * basically own the machine, and have a very in depth understanding
1166  * of all the possible deadlocks and enospc problems.
1167  */
1168 static long btrfs_ioctl_trans_start(struct file *file)
1169 {
1170         struct inode *inode = fdentry(file)->d_inode;
1171         struct btrfs_root *root = BTRFS_I(inode)->root;
1172         struct btrfs_trans_handle *trans;
1173         int ret = 0;
1174
1175         if (!capable(CAP_SYS_ADMIN))
1176                 return -EPERM;
1177
1178         if (file->private_data) {
1179                 ret = -EINPROGRESS;
1180                 goto out;
1181         }
1182
1183         ret = mnt_want_write(file->f_path.mnt);
1184         if (ret)
1185                 goto out;
1186
1187         mutex_lock(&root->fs_info->trans_mutex);
1188         root->fs_info->open_ioctl_trans++;
1189         mutex_unlock(&root->fs_info->trans_mutex);
1190
1191         trans = btrfs_start_ioctl_transaction(root, 0);
1192         if (trans)
1193                 file->private_data = trans;
1194         else
1195                 ret = -ENOMEM;
1196         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1197 out:
1198         return ret;
1199 }
1200
1201 /*
1202  * there are many ways the trans_start and trans_end ioctls can lead
1203  * to deadlocks.  They should only be used by applications that
1204  * basically own the machine, and have a very in depth understanding
1205  * of all the possible deadlocks and enospc problems.
1206  */
1207 long btrfs_ioctl_trans_end(struct file *file)
1208 {
1209         struct inode *inode = fdentry(file)->d_inode;
1210         struct btrfs_root *root = BTRFS_I(inode)->root;
1211         struct btrfs_trans_handle *trans;
1212         int ret = 0;
1213
1214         trans = file->private_data;
1215         if (!trans) {
1216                 ret = -EINVAL;
1217                 goto out;
1218         }
1219         btrfs_end_transaction(trans, root);
1220         file->private_data = NULL;
1221
1222         mutex_lock(&root->fs_info->trans_mutex);
1223         root->fs_info->open_ioctl_trans--;
1224         mutex_unlock(&root->fs_info->trans_mutex);
1225
1226         mnt_drop_write(file->f_path.mnt);
1227
1228 out:
1229         return ret;
1230 }
1231
1232 long btrfs_ioctl(struct file *file, unsigned int
1233                 cmd, unsigned long arg)
1234 {
1235         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1236         void __user *argp = (void __user *)arg;
1237
1238         switch (cmd) {
1239         case FS_IOC_GETFLAGS:
1240                 return btrfs_ioctl_getflags(file, argp);
1241         case FS_IOC_SETFLAGS:
1242                 return btrfs_ioctl_setflags(file, argp);
1243         case FS_IOC_GETVERSION:
1244                 return btrfs_ioctl_getversion(file, argp);
1245         case BTRFS_IOC_SNAP_CREATE:
1246                 return btrfs_ioctl_snap_create(file, argp, 0);
1247         case BTRFS_IOC_SUBVOL_CREATE:
1248                 return btrfs_ioctl_snap_create(file, argp, 1);
1249         case BTRFS_IOC_DEFRAG:
1250                 return btrfs_ioctl_defrag(file);
1251         case BTRFS_IOC_RESIZE:
1252                 return btrfs_ioctl_resize(root, argp);
1253         case BTRFS_IOC_ADD_DEV:
1254                 return btrfs_ioctl_add_dev(root, argp);
1255         case BTRFS_IOC_RM_DEV:
1256                 return btrfs_ioctl_rm_dev(root, argp);
1257         case BTRFS_IOC_BALANCE:
1258                 return btrfs_balance(root->fs_info->dev_root);
1259         case BTRFS_IOC_CLONE:
1260                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1261         case BTRFS_IOC_CLONE_RANGE:
1262                 return btrfs_ioctl_clone_range(file, argp);
1263         case BTRFS_IOC_TRANS_START:
1264                 return btrfs_ioctl_trans_start(file);
1265         case BTRFS_IOC_TRANS_END:
1266                 return btrfs_ioctl_trans_end(file);
1267         case BTRFS_IOC_SYNC:
1268                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1269                 return 0;
1270         }
1271
1272         return -ENOTTY;
1273 }