git.oblomov.eu Git - linux-2.6/blob - 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/smp_lock.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/mount.h>
  33 #include <linux/mpage.h>
  34 #include <linux/namei.h>
  35 #include <linux/swap.h>
  36 #include <linux/writeback.h>
  37 #include <linux/statfs.h>
  38 #include <linux/compat.h>
  39 #include <linux/bit_spinlock.h>
  40 #include <linux/security.h>
  41 #include <linux/xattr.h>
  42 #include <linux/vmalloc.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 = BTRFS_I(dir)->flags;
 116
 117         if (S_ISREG(inode->i_mode))
 118                 flags &= ~BTRFS_INODE_DIRSYNC;
 119         else if (!S_ISDIR(inode->i_mode))
 120                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
 121
 122         BTRFS_I(inode)->flags = flags;
 123         btrfs_update_iflags(inode);
 124 }
 125
 126 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 127 {
 128         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 129         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 130
 131         if (copy_to_user(arg, &flags, sizeof(flags)))
 132                 return -EFAULT;
 133         return 0;
 134 }
 135
 136 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 137 {
 138         struct inode *inode = file->f_path.dentry->d_inode;
 139         struct btrfs_inode *ip = BTRFS_I(inode);
 140         struct btrfs_root *root = ip->root;
 141         struct btrfs_trans_handle *trans;
 142         unsigned int flags, oldflags;
 143         int ret;
 144
 145         if (copy_from_user(&flags, arg, sizeof(flags)))
 146                 return -EFAULT;
 147
 148         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 149                       FS_NOATIME_FL | FS_NODUMP_FL | \
 150                       FS_SYNC_FL | FS_DIRSYNC_FL))
 151                 return -EOPNOTSUPP;
 152
 153         if (!is_owner_or_cap(inode))
 154                 return -EACCES;
 155
 156         mutex_lock(&inode->i_mutex);
 157
 158         flags = btrfs_mask_flags(inode->i_mode, flags);
 159         oldflags = btrfs_flags_to_ioctl(ip->flags);
 160         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 161                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 162                         ret = -EPERM;
 163                         goto out_unlock;
 164                 }
 165         }
 166
 167         ret = mnt_want_write(file->f_path.mnt);
 168         if (ret)
 169                 goto out_unlock;
 170
 171         if (flags & FS_SYNC_FL)
 172                 ip->flags |= BTRFS_INODE_SYNC;
 173         else
 174                 ip->flags &= ~BTRFS_INODE_SYNC;
 175         if (flags & FS_IMMUTABLE_FL)
 176                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 177         else
 178                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 179         if (flags & FS_APPEND_FL)
 180                 ip->flags |= BTRFS_INODE_APPEND;
 181         else
 182                 ip->flags &= ~BTRFS_INODE_APPEND;
 183         if (flags & FS_NODUMP_FL)
 184                 ip->flags |= BTRFS_INODE_NODUMP;
 185         else
 186                 ip->flags &= ~BTRFS_INODE_NODUMP;
 187         if (flags & FS_NOATIME_FL)
 188                 ip->flags |= BTRFS_INODE_NOATIME;
 189         else
 190                 ip->flags &= ~BTRFS_INODE_NOATIME;
 191         if (flags & FS_DIRSYNC_FL)
 192                 ip->flags |= BTRFS_INODE_DIRSYNC;
 193         else
 194                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 195
 196
 197         trans = btrfs_join_transaction(root, 1);
 198         BUG_ON(!trans);
 199
 200         ret = btrfs_update_inode(trans, root, inode);
 201         BUG_ON(ret);
 202
 203         btrfs_update_iflags(inode);
 204         inode->i_ctime = CURRENT_TIME;
 205         btrfs_end_transaction(trans, root);
 206
 207         mnt_drop_write(file->f_path.mnt);
 208  out_unlock:
 209         mutex_unlock(&inode->i_mutex);
 210         return 0;
 211 }
 212
 213 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 214 {
 215         struct inode *inode = file->f_path.dentry->d_inode;
 216
 217         return put_user(inode->i_generation, arg);
 218 }
 219
 220 static noinline int create_subvol(struct btrfs_root *root,
 221                                   struct dentry *dentry,
 222                                   char *name, int namelen)
 223 {
 224         struct btrfs_trans_handle *trans;
 225         struct btrfs_key key;
 226         struct btrfs_root_item root_item;
 227         struct btrfs_inode_item *inode_item;
 228         struct extent_buffer *leaf;
 229         struct btrfs_root *new_root = root;
 230         struct inode *dir;
 231         int ret;
 232         int err;
 233         u64 objectid;
 234         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 235         u64 index = 0;
 236         unsigned long nr = 1;
 237
 238         ret = btrfs_check_metadata_free_space(root);
 239         if (ret)
 240                 goto fail_commit;
 241
 242         trans = btrfs_start_transaction(root, 1);
 243         BUG_ON(!trans);
 244
 245         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
 246                                        0, &objectid);
 247         if (ret)
 248                 goto fail;
 249
 250         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 251                                       0, objectid, NULL, 0, 0, 0);
 252         if (IS_ERR(leaf)) {
 253                 ret = PTR_ERR(leaf);
 254                 goto fail;
 255         }
 256
 257         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 258         btrfs_set_header_bytenr(leaf, leaf->start);
 259         btrfs_set_header_generation(leaf, trans->transid);
 260         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 261         btrfs_set_header_owner(leaf, objectid);
 262
 263         write_extent_buffer(leaf, root->fs_info->fsid,
 264                             (unsigned long)btrfs_header_fsid(leaf),
 265                             BTRFS_FSID_SIZE);
 266         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 267                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 268                             BTRFS_UUID_SIZE);
 269         btrfs_mark_buffer_dirty(leaf);
 270
 271         inode_item = &root_item.inode;
 272         memset(inode_item, 0, sizeof(*inode_item));
 273         inode_item->generation = cpu_to_le64(1);
 274         inode_item->size = cpu_to_le64(3);
 275         inode_item->nlink = cpu_to_le32(1);
 276         inode_item->nbytes = cpu_to_le64(root->leafsize);
 277         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 278
 279         btrfs_set_root_bytenr(&root_item, leaf->start);
 280         btrfs_set_root_generation(&root_item, trans->transid);
 281         btrfs_set_root_level(&root_item, 0);
 282         btrfs_set_root_refs(&root_item, 1);
 283         btrfs_set_root_used(&root_item, 0);
 284         btrfs_set_root_last_snapshot(&root_item, 0);
 285
 286         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 287         root_item.drop_level = 0;
 288
 289         btrfs_tree_unlock(leaf);
 290         free_extent_buffer(leaf);
 291         leaf = NULL;
 292
 293         btrfs_set_root_dirid(&root_item, new_dirid);
 294
 295         key.objectid = objectid;
 296         key.offset = 0;
 297         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 298         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 299                                 &root_item);
 300         if (ret)
 301                 goto fail;
 302
 303         /*
 304          * insert the directory item
 305          */
 306         key.offset = (u64)-1;
 307         dir = dentry->d_parent->d_inode;
 308         ret = btrfs_set_inode_index(dir, &index);
 309         BUG_ON(ret);
 310
 311         ret = btrfs_insert_dir_item(trans, root,
 312                                     name, namelen, dir->i_ino, &key,
 313                                     BTRFS_FT_DIR, index);
 314         if (ret)
 315                 goto fail;
 316
 317         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 318         ret = btrfs_update_inode(trans, root, dir);
 319         BUG_ON(ret);
 320
 321         /* add the backref first */
 322         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 323                                  objectid, BTRFS_ROOT_BACKREF_KEY,
 324                                  root->root_key.objectid,
 325                                  dir->i_ino, index, name, namelen);
 326
 327         BUG_ON(ret);
 328
 329         /* now add the forward ref */
 330         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 331                                  root->root_key.objectid, BTRFS_ROOT_REF_KEY,
 332                                  objectid,
 333                                  dir->i_ino, index, name, namelen);
 334
 335         BUG_ON(ret);
 336
 337         ret = btrfs_commit_transaction(trans, root);
 338         if (ret)
 339                 goto fail_commit;
 340
 341         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 342         BUG_ON(!new_root);
 343
 344         trans = btrfs_start_transaction(new_root, 1);
 345         BUG_ON(!trans);
 346
 347         ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
 348                                        BTRFS_I(dir)->block_group);
 349         if (ret)
 350                 goto fail;
 351
 352 fail:
 353         nr = trans->blocks_used;
 354         err = btrfs_commit_transaction(trans, new_root);
 355         if (err && !ret)
 356                 ret = err;
 357 fail_commit:
 358         btrfs_btree_balance_dirty(root, nr);
 359         return ret;
 360 }
 361
 362 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 363                            char *name, int namelen)
 364 {
 365         struct btrfs_pending_snapshot *pending_snapshot;
 366         struct btrfs_trans_handle *trans;
 367         int ret = 0;
 368         int err;
 369         unsigned long nr = 0;
 370
 371         if (!root->ref_cows)
 372                 return -EINVAL;
 373
 374         ret = btrfs_check_metadata_free_space(root);
 375         if (ret)
 376                 goto fail_unlock;
 377
 378         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 379         if (!pending_snapshot) {
 380                 ret = -ENOMEM;
 381                 goto fail_unlock;
 382         }
 383         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 384         if (!pending_snapshot->name) {
 385                 ret = -ENOMEM;
 386                 kfree(pending_snapshot);
 387                 goto fail_unlock;
 388         }
 389         memcpy(pending_snapshot->name, name, namelen);
 390         pending_snapshot->name[namelen] = '\0';
 391         pending_snapshot->dentry = dentry;
 392         trans = btrfs_start_transaction(root, 1);
 393         BUG_ON(!trans);
 394         pending_snapshot->root = root;
 395         list_add(&pending_snapshot->list,
 396                  &trans->transaction->pending_snapshots);
 397         err = btrfs_commit_transaction(trans, root);
 398
 399 fail_unlock:
 400         btrfs_btree_balance_dirty(root, nr);
 401         return ret;
 402 }
 403
 404 /* copy of may_create in fs/namei.c() */
 405 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 406 {
 407         if (child->d_inode)
 408                 return -EEXIST;
 409         if (IS_DEADDIR(dir))
 410                 return -ENOENT;
 411         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 412 }
 413
 414 /*
 415  * Create a new subvolume below @parent.  This is largely modeled after
 416  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 417  * inside this filesystem so it's quite a bit simpler.
 418  */
 419 static noinline int btrfs_mksubvol(struct path *parent, char *name,
 420                                    int mode, int namelen,
 421                                    struct btrfs_root *snap_src)
 422 {
 423         struct dentry *dentry;
 424         int error;
 425
 426         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 427
 428         dentry = lookup_one_len(name, parent->dentry, namelen);
 429         error = PTR_ERR(dentry);
 430         if (IS_ERR(dentry))
 431                 goto out_unlock;
 432
 433         error = -EEXIST;
 434         if (dentry->d_inode)
 435                 goto out_dput;
 436
 437         if (!IS_POSIXACL(parent->dentry->d_inode))
 438                 mode &= ~current_umask();
 439
 440         error = mnt_want_write(parent->mnt);
 441         if (error)
 442                 goto out_dput;
 443
 444         error = btrfs_may_create(parent->dentry->d_inode, dentry);
 445         if (error)
 446                 goto out_drop_write;
 447
 448         /*
 449          * Actually perform the low-level subvolume creation after all
 450          * this VFS fuzz.
 451          *
 452          * Eventually we want to pass in an inode under which we create this
 453          * subvolume, but for now all are under the filesystem root.
 454          *
 455          * Also we should pass on the mode eventually to allow creating new
 456          * subvolume with specific mode bits.
 457          */
 458         if (snap_src) {
 459                 struct dentry *dir = dentry->d_parent;
 460                 struct dentry *test = dir->d_parent;
 461                 struct btrfs_path *path = btrfs_alloc_path();
 462                 int ret;
 463                 u64 test_oid;
 464                 u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
 465
 466                 test_oid = snap_src->root_key.objectid;
 467
 468                 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
 469                                           path, parent_oid, test_oid);
 470                 if (ret == 0)
 471                         goto create;
 472                 btrfs_release_path(snap_src->fs_info->tree_root, path);
 473
 474                 /* we need to make sure we aren't creating a directory loop
 475                  * by taking a snapshot of something that has our current
 476                  * subvol in its directory tree.  So, this loops through
 477                  * the dentries and checks the forward refs for each subvolume
 478                  * to see if is references the subvolume where we are
 479                  * placing this new snapshot.
 480                  */
 481                 while (1) {
 482                         if (!test ||
 483                             dir == snap_src->fs_info->sb->s_root ||
 484                             test == snap_src->fs_info->sb->s_root ||
 485                             test->d_inode->i_sb != snap_src->fs_info->sb) {
 486                                 break;
 487                         }
 488                         if (S_ISLNK(test->d_inode->i_mode)) {
 489                                 printk(KERN_INFO "Btrfs symlink in snapshot "
 490                                        "path, failed\n");
 491                                 error = -EMLINK;
 492                                 btrfs_free_path(path);
 493                                 goto out_drop_write;
 494                         }
 495                         test_oid =
 496                                 BTRFS_I(test->d_inode)->root->root_key.objectid;
 497                         ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
 498                                   path, test_oid, parent_oid);
 499                         if (ret == 0) {
 500                                 printk(KERN_INFO "Btrfs snapshot creation "
 501                                        "failed, looping\n");
 502                                 error = -EMLINK;
 503                                 btrfs_free_path(path);
 504                                 goto out_drop_write;
 505                         }
 506                         btrfs_release_path(snap_src->fs_info->tree_root, path);
 507                         test = test->d_parent;
 508                 }
 509 create:
 510                 btrfs_free_path(path);
 511                 error = create_snapshot(snap_src, dentry, name, namelen);
 512         } else {
 513                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
 514                                       dentry, name, namelen);
 515         }
 516         if (error)
 517                 goto out_drop_write;
 518
 519         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 520 out_drop_write:
 521         mnt_drop_write(parent->mnt);
 522 out_dput:
 523         dput(dentry);
 524 out_unlock:
 525         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 526         return error;
 527 }
 528
 529
 530 static int btrfs_defrag_file(struct file *file)
 531 {
 532         struct inode *inode = fdentry(file)->d_inode;
 533         struct btrfs_root *root = BTRFS_I(inode)->root;
 534         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 535         struct btrfs_ordered_extent *ordered;
 536         struct page *page;
 537         unsigned long last_index;
 538         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 539         unsigned long total_read = 0;
 540         u64 page_start;
 541         u64 page_end;
 542         unsigned long i;
 543         int ret;
 544
 545         ret = btrfs_check_data_free_space(root, inode, inode->i_size);
 546         if (ret)
 547                 return -ENOSPC;
 548
 549         mutex_lock(&inode->i_mutex);
 550         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 551         for (i = 0; i <= last_index; i++) {
 552                 if (total_read % ra_pages == 0) {
 553                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 554                                        min(last_index, i + ra_pages - 1));
 555                 }
 556                 total_read++;
 557 again:
 558                 page = grab_cache_page(inode->i_mapping, i);
 559                 if (!page)
 560                         goto out_unlock;
 561                 if (!PageUptodate(page)) {
 562                         btrfs_readpage(NULL, page);
 563                         lock_page(page);
 564                         if (!PageUptodate(page)) {
 565                                 unlock_page(page);
 566                                 page_cache_release(page);
 567                                 goto out_unlock;
 568                         }
 569                 }
 570
 571                 wait_on_page_writeback(page);
 572
 573                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 574                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 575                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 576
 577                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 578                 if (ordered) {
 579                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 580                         unlock_page(page);
 581                         page_cache_release(page);
 582                         btrfs_start_ordered_extent(inode, ordered, 1);
 583                         btrfs_put_ordered_extent(ordered);
 584                         goto again;
 585                 }
 586                 set_page_extent_mapped(page);
 587
 588                 /*
 589                  * this makes sure page_mkwrite is called on the
 590                  * page if it is dirtied again later
 591                  */
 592                 clear_page_dirty_for_io(page);
 593
 594                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 595
 596                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 597                 set_page_dirty(page);
 598                 unlock_page(page);
 599                 page_cache_release(page);
 600                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 601         }
 602
 603 out_unlock:
 604         mutex_unlock(&inode->i_mutex);
 605         return 0;
 606 }
 607
 608 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 609 {
 610         u64 new_size;
 611         u64 old_size;
 612         u64 devid = 1;
 613         struct btrfs_ioctl_vol_args *vol_args;
 614         struct btrfs_trans_handle *trans;
 615         struct btrfs_device *device = NULL;
 616         char *sizestr;
 617         char *devstr = NULL;
 618         int ret = 0;
 619         int namelen;
 620         int mod = 0;
 621
 622         if (root->fs_info->sb->s_flags & MS_RDONLY)
 623                 return -EROFS;
 624
 625         if (!capable(CAP_SYS_ADMIN))
 626                 return -EPERM;
 627
 628         vol_args = memdup_user(arg, sizeof(*vol_args));
 629         if (IS_ERR(vol_args))
 630                 return PTR_ERR(vol_args);
 631
 632         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 633         namelen = strlen(vol_args->name);
 634
 635         mutex_lock(&root->fs_info->volume_mutex);
 636         sizestr = vol_args->name;
 637         devstr = strchr(sizestr, ':');
 638         if (devstr) {
 639                 char *end;
 640                 sizestr = devstr + 1;
 641                 *devstr = '\0';
 642                 devstr = vol_args->name;
 643                 devid = simple_strtoull(devstr, &end, 10);
 644                 printk(KERN_INFO "resizing devid %llu\n",
 645                        (unsigned long long)devid);
 646         }
 647         device = btrfs_find_device(root, devid, NULL, NULL);
 648         if (!device) {
 649                 printk(KERN_INFO "resizer unable to find device %llu\n",
 650                        (unsigned long long)devid);
 651                 ret = -EINVAL;
 652                 goto out_unlock;
 653         }
 654         if (!strcmp(sizestr, "max"))
 655                 new_size = device->bdev->bd_inode->i_size;
 656         else {
 657                 if (sizestr[0] == '-') {
 658                         mod = -1;
 659                         sizestr++;
 660                 } else if (sizestr[0] == '+') {
 661                         mod = 1;
 662                         sizestr++;
 663                 }
 664                 new_size = btrfs_parse_size(sizestr);
 665                 if (new_size == 0) {
 666                         ret = -EINVAL;
 667                         goto out_unlock;
 668                 }
 669         }
 670
 671         old_size = device->total_bytes;
 672
 673         if (mod < 0) {
 674                 if (new_size > old_size) {
 675                         ret = -EINVAL;
 676                         goto out_unlock;
 677                 }
 678                 new_size = old_size - new_size;
 679         } else if (mod > 0) {
 680                 new_size = old_size + new_size;
 681         }
 682
 683         if (new_size < 256 * 1024 * 1024) {
 684                 ret = -EINVAL;
 685                 goto out_unlock;
 686         }
 687         if (new_size > device->bdev->bd_inode->i_size) {
 688                 ret = -EFBIG;
 689                 goto out_unlock;
 690         }
 691
 692         do_div(new_size, root->sectorsize);
 693         new_size *= root->sectorsize;
 694
 695         printk(KERN_INFO "new size for %s is %llu\n",
 696                 device->name, (unsigned long long)new_size);
 697
 698         if (new_size > old_size) {
 699                 trans = btrfs_start_transaction(root, 1);
 700                 ret = btrfs_grow_device(trans, device, new_size);
 701                 btrfs_commit_transaction(trans, root);
 702         } else {
 703                 ret = btrfs_shrink_device(device, new_size);
 704         }
 705
 706 out_unlock:
 707         mutex_unlock(&root->fs_info->volume_mutex);
 708         kfree(vol_args);
 709         return ret;
 710 }
 711
 712 static noinline int btrfs_ioctl_snap_create(struct file *file,
 713                                             void __user *arg, int subvol)
 714 {
 715         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 716         struct btrfs_ioctl_vol_args *vol_args;
 717         struct btrfs_dir_item *di;
 718         struct btrfs_path *path;
 719         struct file *src_file;
 720         u64 root_dirid;
 721         int namelen;
 722         int ret = 0;
 723
 724         if (root->fs_info->sb->s_flags & MS_RDONLY)
 725                 return -EROFS;
 726
 727         vol_args = memdup_user(arg, sizeof(*vol_args));
 728         if (IS_ERR(vol_args))
 729                 return PTR_ERR(vol_args);
 730
 731         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 732         namelen = strlen(vol_args->name);
 733         if (strchr(vol_args->name, '/')) {
 734                 ret = -EINVAL;
 735                 goto out;
 736         }
 737
 738         path = btrfs_alloc_path();
 739         if (!path) {
 740                 ret = -ENOMEM;
 741                 goto out;
 742         }
 743
 744         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 745         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 746                             path, root_dirid,
 747                             vol_args->name, namelen, 0);
 748         btrfs_free_path(path);
 749
 750         if (di && !IS_ERR(di)) {
 751                 ret = -EEXIST;
 752                 goto out;
 753         }
 754
 755         if (IS_ERR(di)) {
 756                 ret = PTR_ERR(di);
 757                 goto out;
 758         }
 759
 760         if (subvol) {
 761                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 762                                      file->f_path.dentry->d_inode->i_mode,
 763                                      namelen, NULL);
 764         } else {
 765                 struct inode *src_inode;
 766                 src_file = fget(vol_args->fd);
 767                 if (!src_file) {
 768                         ret = -EINVAL;
 769                         goto out;
 770                 }
 771
 772                 src_inode = src_file->f_path.dentry->d_inode;
 773                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 774                         printk(KERN_INFO "btrfs: Snapshot src from "
 775                                "another FS\n");
 776                         ret = -EINVAL;
 777                         fput(src_file);
 778                         goto out;
 779                 }
 780                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 781                              file->f_path.dentry->d_inode->i_mode,
 782                              namelen, BTRFS_I(src_inode)->root);
 783                 fput(src_file);
 784         }
 785
 786 out:
 787         kfree(vol_args);
 788         return ret;
 789 }
 790
 791 static int btrfs_ioctl_defrag(struct file *file)
 792 {
 793         struct inode *inode = fdentry(file)->d_inode;
 794         struct btrfs_root *root = BTRFS_I(inode)->root;
 795         int ret;
 796
 797         ret = mnt_want_write(file->f_path.mnt);
 798         if (ret)
 799                 return ret;
 800
 801         switch (inode->i_mode & S_IFMT) {
 802         case S_IFDIR:
 803                 if (!capable(CAP_SYS_ADMIN)) {
 804                         ret = -EPERM;
 805                         goto out;
 806                 }
 807                 btrfs_defrag_root(root, 0);
 808                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 809                 break;
 810         case S_IFREG:
 811                 if (!(file->f_mode & FMODE_WRITE)) {
 812                         ret = -EINVAL;
 813                         goto out;
 814                 }
 815                 btrfs_defrag_file(file);
 816                 break;
 817         }
 818 out:
 819         mnt_drop_write(file->f_path.mnt);
 820         return ret;
 821 }
 822
 823 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 824 {
 825         struct btrfs_ioctl_vol_args *vol_args;
 826         int ret;
 827
 828         if (!capable(CAP_SYS_ADMIN))
 829                 return -EPERM;
 830
 831         vol_args = memdup_user(arg, sizeof(*vol_args));
 832         if (IS_ERR(vol_args))
 833                 return PTR_ERR(vol_args);
 834
 835         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 836         ret = btrfs_init_new_device(root, vol_args->name);
 837
 838         kfree(vol_args);
 839         return ret;
 840 }
 841
 842 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 843 {
 844         struct btrfs_ioctl_vol_args *vol_args;
 845         int ret;
 846
 847         if (!capable(CAP_SYS_ADMIN))
 848                 return -EPERM;
 849
 850         if (root->fs_info->sb->s_flags & MS_RDONLY)
 851                 return -EROFS;
 852
 853         vol_args = memdup_user(arg, sizeof(*vol_args));
 854         if (IS_ERR(vol_args))
 855                 return PTR_ERR(vol_args);
 856
 857         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 858         ret = btrfs_rm_device(root, vol_args->name);
 859
 860         kfree(vol_args);
 861         return ret;
 862 }
 863
 864 static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 865                 u64 off, u64 olen, u64 destoff)
 866 {
 867         struct inode *inode = fdentry(file)->d_inode;
 868         struct btrfs_root *root = BTRFS_I(inode)->root;
 869         struct file *src_file;
 870         struct inode *src;
 871         struct btrfs_trans_handle *trans;
 872         struct btrfs_path *path;
 873         struct extent_buffer *leaf;
 874         char *buf;
 875         struct btrfs_key key;
 876         u32 nritems;
 877         int slot;
 878         int ret;
 879         u64 len = olen;
 880         u64 bs = root->fs_info->sb->s_blocksize;
 881         u64 hint_byte;
 882
 883         /*
 884          * TODO:
 885          * - split compressed inline extents.  annoying: we need to
 886          *   decompress into destination's address_space (the file offset
 887          *   may change, so source mapping won't do), then recompress (or
 888          *   otherwise reinsert) a subrange.
 889          * - allow ranges within the same file to be cloned (provided
 890          *   they don't overlap)?
 891          */
 892
 893         /* the destination must be opened for writing */
 894         if (!(file->f_mode & FMODE_WRITE))
 895                 return -EINVAL;
 896
 897         ret = mnt_want_write(file->f_path.mnt);
 898         if (ret)
 899                 return ret;
 900
 901         src_file = fget(srcfd);
 902         if (!src_file) {
 903                 ret = -EBADF;
 904                 goto out_drop_write;
 905         }
 906         src = src_file->f_dentry->d_inode;
 907
 908         ret = -EINVAL;
 909         if (src == inode)
 910                 goto out_fput;
 911
 912         ret = -EISDIR;
 913         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 914                 goto out_fput;
 915
 916         ret = -EXDEV;
 917         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 918                 goto out_fput;
 919
 920         ret = -ENOMEM;
 921         buf = vmalloc(btrfs_level_size(root, 0));
 922         if (!buf)
 923                 goto out_fput;
 924
 925         path = btrfs_alloc_path();
 926         if (!path) {
 927                 vfree(buf);
 928                 goto out_fput;
 929         }
 930         path->reada = 2;
 931
 932         if (inode < src) {
 933                 mutex_lock(&inode->i_mutex);
 934                 mutex_lock(&src->i_mutex);
 935         } else {
 936                 mutex_lock(&src->i_mutex);
 937                 mutex_lock(&inode->i_mutex);
 938         }
 939
 940         /* determine range to clone */
 941         ret = -EINVAL;
 942         if (off >= src->i_size || off + len > src->i_size)
 943                 goto out_unlock;
 944         if (len == 0)
 945                 olen = len = src->i_size - off;
 946         /* if we extend to eof, continue to block boundary */
 947         if (off + len == src->i_size)
 948                 len = ((src->i_size + bs-1) & ~(bs-1))
 949                         - off;
 950
 951         /* verify the end result is block aligned */
 952         if ((off & (bs-1)) ||
 953             ((off + len) & (bs-1)))
 954                 goto out_unlock;
 955
 956         /* do any pending delalloc/csum calc on src, one way or
 957            another, and lock file content */
 958         while (1) {
 959                 struct btrfs_ordered_extent *ordered;
 960                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 961                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
 962                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 963                         break;
 964                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 965                 if (ordered)
 966                         btrfs_put_ordered_extent(ordered);
 967                 btrfs_wait_ordered_range(src, off, off+len);
 968         }
 969
 970         trans = btrfs_start_transaction(root, 1);
 971         BUG_ON(!trans);
 972
 973         /* punch hole in destination first */
 974         btrfs_drop_extents(trans, root, inode, off, off + len,
 975                            off + len, 0, &hint_byte);
 976
 977         /* clone data */
 978         key.objectid = src->i_ino;
 979         key.type = BTRFS_EXTENT_DATA_KEY;
 980         key.offset = 0;
 981
 982         while (1) {
 983                 /*
 984                  * note the key will change type as we walk through the
 985                  * tree.
 986                  */
 987                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 988                 if (ret < 0)
 989                         goto out;
 990
 991                 nritems = btrfs_header_nritems(path->nodes[0]);
 992                 if (path->slots[0] >= nritems) {
 993                         ret = btrfs_next_leaf(root, path);
 994                         if (ret < 0)
 995                                 goto out;
 996                         if (ret > 0)
 997                                 break;
 998                         nritems = btrfs_header_nritems(path->nodes[0]);
 999                 }
1000                 leaf = path->nodes[0];
1001                 slot = path->slots[0];
1002
1003                 btrfs_item_key_to_cpu(leaf, &key, slot);
1004                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1005                     key.objectid != src->i_ino)
1006                         break;
1007
1008                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1009                         struct btrfs_file_extent_item *extent;
1010                         int type;
1011                         u32 size;
1012                         struct btrfs_key new_key;
1013                         u64 disko = 0, diskl = 0;
1014                         u64 datao = 0, datal = 0;
1015                         u8 comp;
1016
1017                         size = btrfs_item_size_nr(leaf, slot);
1018                         read_extent_buffer(leaf, buf,
1019                                            btrfs_item_ptr_offset(leaf, slot),
1020                                            size);
1021
1022                         extent = btrfs_item_ptr(leaf, slot,
1023                                                 struct btrfs_file_extent_item);
1024                         comp = btrfs_file_extent_compression(leaf, extent);
1025                         type = btrfs_file_extent_type(leaf, extent);
1026                         if (type == BTRFS_FILE_EXTENT_REG) {
1027                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1028                                                                       extent);
1029                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1030                                                                  extent);
1031                                 datao = btrfs_file_extent_offset(leaf, extent);
1032                                 datal = btrfs_file_extent_num_bytes(leaf,
1033                                                                     extent);
1034                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1035                                 /* take upper bound, may be compressed */
1036                                 datal = btrfs_file_extent_ram_bytes(leaf,
1037                                                                     extent);
1038                         }
1039                         btrfs_release_path(root, path);
1040
1041                         if (key.offset + datal < off ||
1042                             key.offset >= off+len)
1043                                 goto next;
1044
1045                         memcpy(&new_key, &key, sizeof(new_key));
1046                         new_key.objectid = inode->i_ino;
1047                         new_key.offset = key.offset + destoff - off;
1048
1049                         if (type == BTRFS_FILE_EXTENT_REG) {
1050                                 ret = btrfs_insert_empty_item(trans, root, path,
1051                                                               &new_key, size);
1052                                 if (ret)
1053                                         goto out;
1054
1055                                 leaf = path->nodes[0];
1056                                 slot = path->slots[0];
1057                                 write_extent_buffer(leaf, buf,
1058                                             btrfs_item_ptr_offset(leaf, slot),
1059                                             size);
1060
1061                                 extent = btrfs_item_ptr(leaf, slot,
1062                                                 struct btrfs_file_extent_item);
1063
1064                                 if (off > key.offset) {
1065                                         datao += off - key.offset;
1066                                         datal -= off - key.offset;
1067                                 }
1068                                 if (key.offset + datao + datal + key.offset >
1069                                     off + len)
1070                                         datal = off + len - key.offset - datao;
1071                                 /* disko == 0 means it's a hole */
1072                                 if (!disko)
1073                                         datao = 0;
1074
1075                                 btrfs_set_file_extent_offset(leaf, extent,
1076                                                              datao);
1077                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1078                                                                 datal);
1079                                 if (disko) {
1080                                         inode_add_bytes(inode, datal);
1081                                         ret = btrfs_inc_extent_ref(trans, root,
1082                                                         disko, diskl, 0,
1083                                                         root->root_key.objectid,
1084                                                         inode->i_ino,
1085                                                         new_key.offset - datao);
1086                                         BUG_ON(ret);
1087                                 }
1088                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1089                                 u64 skip = 0;
1090                                 u64 trim = 0;
1091                                 if (off > key.offset) {
1092                                         skip = off - key.offset;
1093                                         new_key.offset += skip;
1094                                 }
1095
1096                                 if (key.offset + datal > off+len)
1097                                         trim = key.offset + datal - (off+len);
1098
1099                                 if (comp && (skip || trim)) {
1100                                         ret = -EINVAL;
1101                                         goto out;
1102                                 }
1103                                 size -= skip + trim;
1104                                 datal -= skip + trim;
1105                                 ret = btrfs_insert_empty_item(trans, root, path,
1106                                                               &new_key, size);
1107                                 if (ret)
1108                                         goto out;
1109
1110                                 if (skip) {
1111                                         u32 start =
1112                                           btrfs_file_extent_calc_inline_size(0);
1113                                         memmove(buf+start, buf+start+skip,
1114                                                 datal);
1115                                 }
1116
1117                                 leaf = path->nodes[0];
1118                                 slot = path->slots[0];
1119                                 write_extent_buffer(leaf, buf,
1120                                             btrfs_item_ptr_offset(leaf, slot),
1121                                             size);
1122                                 inode_add_bytes(inode, datal);
1123                         }
1124
1125                         btrfs_mark_buffer_dirty(leaf);
1126                 }
1127
1128 next:
1129                 btrfs_release_path(root, path);
1130                 key.offset++;
1131         }
1132         ret = 0;
1133 out:
1134         btrfs_release_path(root, path);
1135         if (ret == 0) {
1136                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1137                 if (destoff + olen > inode->i_size)
1138                         btrfs_i_size_write(inode, destoff + olen);
1139                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1140                 ret = btrfs_update_inode(trans, root, inode);
1141         }
1142         btrfs_end_transaction(trans, root);
1143         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1144         if (ret)
1145                 vmtruncate(inode, 0);
1146 out_unlock:
1147         mutex_unlock(&src->i_mutex);
1148         mutex_unlock(&inode->i_mutex);
1149         vfree(buf);
1150         btrfs_free_path(path);
1151 out_fput:
1152         fput(src_file);
1153 out_drop_write:
1154         mnt_drop_write(file->f_path.mnt);
1155         return ret;
1156 }
1157
1158 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1159 {
1160         struct btrfs_ioctl_clone_range_args args;
1161
1162         if (copy_from_user(&args, argp, sizeof(args)))
1163                 return -EFAULT;
1164         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1165                                  args.src_length, args.dest_offset);
1166 }
1167
1168 /*
1169  * there are many ways the trans_start and trans_end ioctls can lead
1170  * to deadlocks.  They should only be used by applications that
1171  * basically own the machine, and have a very in depth understanding
1172  * of all the possible deadlocks and enospc problems.
1173  */
1174 static long btrfs_ioctl_trans_start(struct file *file)
1175 {
1176         struct inode *inode = fdentry(file)->d_inode;
1177         struct btrfs_root *root = BTRFS_I(inode)->root;
1178         struct btrfs_trans_handle *trans;
1179         int ret = 0;
1180
1181         if (!capable(CAP_SYS_ADMIN))
1182                 return -EPERM;
1183
1184         if (file->private_data) {
1185                 ret = -EINPROGRESS;
1186                 goto out;
1187         }
1188
1189         ret = mnt_want_write(file->f_path.mnt);
1190         if (ret)
1191                 goto out;
1192
1193         mutex_lock(&root->fs_info->trans_mutex);
1194         root->fs_info->open_ioctl_trans++;
1195         mutex_unlock(&root->fs_info->trans_mutex);
1196
1197         trans = btrfs_start_ioctl_transaction(root, 0);
1198         if (trans)
1199                 file->private_data = trans;
1200         else
1201                 ret = -ENOMEM;
1202         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1203 out:
1204         return ret;
1205 }
1206
1207 /*
1208  * there are many ways the trans_start and trans_end ioctls can lead
1209  * to deadlocks.  They should only be used by applications that
1210  * basically own the machine, and have a very in depth understanding
1211  * of all the possible deadlocks and enospc problems.
1212  */
1213 long btrfs_ioctl_trans_end(struct file *file)
1214 {
1215         struct inode *inode = fdentry(file)->d_inode;
1216         struct btrfs_root *root = BTRFS_I(inode)->root;
1217         struct btrfs_trans_handle *trans;
1218         int ret = 0;
1219
1220         trans = file->private_data;
1221         if (!trans) {
1222                 ret = -EINVAL;
1223                 goto out;
1224         }
1225         btrfs_end_transaction(trans, root);
1226         file->private_data = NULL;
1227
1228         mutex_lock(&root->fs_info->trans_mutex);
1229         root->fs_info->open_ioctl_trans--;
1230         mutex_unlock(&root->fs_info->trans_mutex);
1231
1232         mnt_drop_write(file->f_path.mnt);
1233
1234 out:
1235         return ret;
1236 }
1237
1238 long btrfs_ioctl(struct file *file, unsigned int
1239                 cmd, unsigned long arg)
1240 {
1241         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1242         void __user *argp = (void __user *)arg;
1243
1244         switch (cmd) {
1245         case FS_IOC_GETFLAGS:
1246                 return btrfs_ioctl_getflags(file, argp);
1247         case FS_IOC_SETFLAGS:
1248                 return btrfs_ioctl_setflags(file, argp);
1249         case FS_IOC_GETVERSION:
1250                 return btrfs_ioctl_getversion(file, argp);
1251         case BTRFS_IOC_SNAP_CREATE:
1252                 return btrfs_ioctl_snap_create(file, argp, 0);
1253         case BTRFS_IOC_SUBVOL_CREATE:
1254                 return btrfs_ioctl_snap_create(file, argp, 1);
1255         case BTRFS_IOC_DEFRAG:
1256                 return btrfs_ioctl_defrag(file);
1257         case BTRFS_IOC_RESIZE:
1258                 return btrfs_ioctl_resize(root, argp);
1259         case BTRFS_IOC_ADD_DEV:
1260                 return btrfs_ioctl_add_dev(root, argp);
1261         case BTRFS_IOC_RM_DEV:
1262                 return btrfs_ioctl_rm_dev(root, argp);
1263         case BTRFS_IOC_BALANCE:
1264                 return btrfs_balance(root->fs_info->dev_root);
1265         case BTRFS_IOC_CLONE:
1266                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1267         case BTRFS_IOC_CLONE_RANGE:
1268                 return btrfs_ioctl_clone_range(file, argp);
1269         case BTRFS_IOC_TRANS_START:
1270                 return btrfs_ioctl_trans_start(file);
1271         case BTRFS_IOC_TRANS_END:
1272                 return btrfs_ioctl_trans_end(file);
1273         case BTRFS_IOC_SYNC:
1274                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1275                 return 0;
1276         }
1277
1278         return -ENOTTY;
1279 }