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