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/version.h>
  42 #include <linux/xattr.h>
  43 #include <linux/vmalloc.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_free_space(root, 1, 0);
  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 #if 0
 149         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 150                              name, namelen, objectid,
 151                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 152         if (ret)
 153                 goto fail;
 154 #endif
 155         ret = btrfs_commit_transaction(trans, root);
 156         if (ret)
 157                 goto fail_commit;
 158
 159         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 160         BUG_ON(!new_root);
 161
 162         trans = btrfs_start_transaction(new_root, 1);
 163         BUG_ON(!trans);
 164
 165         ret = btrfs_create_subvol_root(new_root, dentry, trans, new_dirid,
 166                                        BTRFS_I(dir)->block_group);
 167         if (ret)
 168                 goto fail;
 169
 170 fail:
 171         nr = trans->blocks_used;
 172         err = btrfs_commit_transaction(trans, new_root);
 173         if (err && !ret)
 174                 ret = err;
 175 fail_commit:
 176         btrfs_btree_balance_dirty(root, nr);
 177         return ret;
 178 }
 179
 180 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 181                            char *name, int namelen)
 182 {
 183         struct btrfs_pending_snapshot *pending_snapshot;
 184         struct btrfs_trans_handle *trans;
 185         int ret = 0;
 186         int err;
 187         unsigned long nr = 0;
 188
 189         if (!root->ref_cows)
 190                 return -EINVAL;
 191
 192         ret = btrfs_check_free_space(root, 1, 0);
 193         if (ret)
 194                 goto fail_unlock;
 195
 196         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 197         if (!pending_snapshot) {
 198                 ret = -ENOMEM;
 199                 goto fail_unlock;
 200         }
 201         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 202         if (!pending_snapshot->name) {
 203                 ret = -ENOMEM;
 204                 kfree(pending_snapshot);
 205                 goto fail_unlock;
 206         }
 207         memcpy(pending_snapshot->name, name, namelen);
 208         pending_snapshot->name[namelen] = '\0';
 209         pending_snapshot->dentry = dentry;
 210         trans = btrfs_start_transaction(root, 1);
 211         BUG_ON(!trans);
 212         pending_snapshot->root = root;
 213         list_add(&pending_snapshot->list,
 214                  &trans->transaction->pending_snapshots);
 215         err = btrfs_commit_transaction(trans, root);
 216
 217 fail_unlock:
 218         btrfs_btree_balance_dirty(root, nr);
 219         return ret;
 220 }
 221
 222 /* copy of may_create in fs/namei.c() */
 223 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 224 {
 225         if (child->d_inode)
 226                 return -EEXIST;
 227         if (IS_DEADDIR(dir))
 228                 return -ENOENT;
 229         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 230 }
 231
 232 /*
 233  * Create a new subvolume below @parent.  This is largely modeled after
 234  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 235  * inside this filesystem so it's quite a bit simpler.
 236  */
 237 static noinline int btrfs_mksubvol(struct path *parent, char *name,
 238                                    int mode, int namelen,
 239                                    struct btrfs_root *snap_src)
 240 {
 241         struct dentry *dentry;
 242         int error;
 243
 244         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 245
 246         dentry = lookup_one_len(name, parent->dentry, namelen);
 247         error = PTR_ERR(dentry);
 248         if (IS_ERR(dentry))
 249                 goto out_unlock;
 250
 251         error = -EEXIST;
 252         if (dentry->d_inode)
 253                 goto out_dput;
 254
 255         if (!IS_POSIXACL(parent->dentry->d_inode))
 256                 mode &= ~current->fs->umask;
 257
 258         error = mnt_want_write(parent->mnt);
 259         if (error)
 260                 goto out_dput;
 261
 262         error = btrfs_may_create(parent->dentry->d_inode, dentry);
 263         if (error)
 264                 goto out_drop_write;
 265
 266         /*
 267          * Actually perform the low-level subvolume creation after all
 268          * this VFS fuzz.
 269          *
 270          * Eventually we want to pass in an inode under which we create this
 271          * subvolume, but for now all are under the filesystem root.
 272          *
 273          * Also we should pass on the mode eventually to allow creating new
 274          * subvolume with specific mode bits.
 275          */
 276         if (snap_src) {
 277                 error = create_snapshot(snap_src, dentry, name, namelen);
 278         } else {
 279                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
 280                                       dentry, name, namelen);
 281         }
 282         if (error)
 283                 goto out_drop_write;
 284
 285         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 286 out_drop_write:
 287         mnt_drop_write(parent->mnt);
 288 out_dput:
 289         dput(dentry);
 290 out_unlock:
 291         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 292         return error;
 293 }
 294
 295
 296 int btrfs_defrag_file(struct file *file)
 297 {
 298         struct inode *inode = fdentry(file)->d_inode;
 299         struct btrfs_root *root = BTRFS_I(inode)->root;
 300         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 301         struct btrfs_ordered_extent *ordered;
 302         struct page *page;
 303         unsigned long last_index;
 304         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 305         unsigned long total_read = 0;
 306         u64 page_start;
 307         u64 page_end;
 308         unsigned long i;
 309         int ret;
 310
 311         ret = btrfs_check_free_space(root, inode->i_size, 0);
 312         if (ret)
 313                 return -ENOSPC;
 314
 315         mutex_lock(&inode->i_mutex);
 316         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 317         for (i = 0; i <= last_index; i++) {
 318                 if (total_read % ra_pages == 0) {
 319                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 320                                        min(last_index, i + ra_pages - 1));
 321                 }
 322                 total_read++;
 323 again:
 324                 page = grab_cache_page(inode->i_mapping, i);
 325                 if (!page)
 326                         goto out_unlock;
 327                 if (!PageUptodate(page)) {
 328                         btrfs_readpage(NULL, page);
 329                         lock_page(page);
 330                         if (!PageUptodate(page)) {
 331                                 unlock_page(page);
 332                                 page_cache_release(page);
 333                                 goto out_unlock;
 334                         }
 335                 }
 336
 337                 wait_on_page_writeback(page);
 338
 339                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 340                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 341                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 342
 343                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 344                 if (ordered) {
 345                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 346                         unlock_page(page);
 347                         page_cache_release(page);
 348                         btrfs_start_ordered_extent(inode, ordered, 1);
 349                         btrfs_put_ordered_extent(ordered);
 350                         goto again;
 351                 }
 352                 set_page_extent_mapped(page);
 353
 354                 /*
 355                  * this makes sure page_mkwrite is called on the
 356                  * page if it is dirtied again later
 357                  */
 358                 clear_page_dirty_for_io(page);
 359
 360                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 361
 362                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 363                 set_page_dirty(page);
 364                 unlock_page(page);
 365                 page_cache_release(page);
 366                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 367         }
 368
 369 out_unlock:
 370         mutex_unlock(&inode->i_mutex);
 371         return 0;
 372 }
 373
 374 /*
 375  * Called inside transaction, so use GFP_NOFS
 376  */
 377
 378 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 379 {
 380         u64 new_size;
 381         u64 old_size;
 382         u64 devid = 1;
 383         struct btrfs_ioctl_vol_args *vol_args;
 384         struct btrfs_trans_handle *trans;
 385         struct btrfs_device *device = NULL;
 386         char *sizestr;
 387         char *devstr = NULL;
 388         int ret = 0;
 389         int namelen;
 390         int mod = 0;
 391
 392         if (root->fs_info->sb->s_flags & MS_RDONLY)
 393                 return -EROFS;
 394
 395         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 396
 397         if (!vol_args)
 398                 return -ENOMEM;
 399
 400         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 401                 ret = -EFAULT;
 402                 goto out;
 403         }
 404
 405         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 406         namelen = strlen(vol_args->name);
 407
 408         mutex_lock(&root->fs_info->volume_mutex);
 409         sizestr = vol_args->name;
 410         devstr = strchr(sizestr, ':');
 411         if (devstr) {
 412                 char *end;
 413                 sizestr = devstr + 1;
 414                 *devstr = '\0';
 415                 devstr = vol_args->name;
 416                 devid = simple_strtoull(devstr, &end, 10);
 417                 printk(KERN_INFO "resizing devid %llu\n", devid);
 418         }
 419         device = btrfs_find_device(root, devid, NULL, NULL);
 420         if (!device) {
 421                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 422                 ret = -EINVAL;
 423                 goto out_unlock;
 424         }
 425         if (!strcmp(sizestr, "max"))
 426                 new_size = device->bdev->bd_inode->i_size;
 427         else {
 428                 if (sizestr[0] == '-') {
 429                         mod = -1;
 430                         sizestr++;
 431                 } else if (sizestr[0] == '+') {
 432                         mod = 1;
 433                         sizestr++;
 434                 }
 435                 new_size = btrfs_parse_size(sizestr);
 436                 if (new_size == 0) {
 437                         ret = -EINVAL;
 438                         goto out_unlock;
 439                 }
 440         }
 441
 442         old_size = device->total_bytes;
 443
 444         if (mod < 0) {
 445                 if (new_size > old_size) {
 446                         ret = -EINVAL;
 447                         goto out_unlock;
 448                 }
 449                 new_size = old_size - new_size;
 450         } else if (mod > 0) {
 451                 new_size = old_size + new_size;
 452         }
 453
 454         if (new_size < 256 * 1024 * 1024) {
 455                 ret = -EINVAL;
 456                 goto out_unlock;
 457         }
 458         if (new_size > device->bdev->bd_inode->i_size) {
 459                 ret = -EFBIG;
 460                 goto out_unlock;
 461         }
 462
 463         do_div(new_size, root->sectorsize);
 464         new_size *= root->sectorsize;
 465
 466         printk(KERN_INFO "new size for %s is %llu\n",
 467                 device->name, (unsigned long long)new_size);
 468
 469         if (new_size > old_size) {
 470                 trans = btrfs_start_transaction(root, 1);
 471                 ret = btrfs_grow_device(trans, device, new_size);
 472                 btrfs_commit_transaction(trans, root);
 473         } else {
 474                 ret = btrfs_shrink_device(device, new_size);
 475         }
 476
 477 out_unlock:
 478         mutex_unlock(&root->fs_info->volume_mutex);
 479 out:
 480         kfree(vol_args);
 481         return ret;
 482 }
 483
 484 static noinline int btrfs_ioctl_snap_create(struct file *file,
 485                                             void __user *arg, int subvol)
 486 {
 487         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 488         struct btrfs_ioctl_vol_args *vol_args;
 489         struct btrfs_dir_item *di;
 490         struct btrfs_path *path;
 491         struct file *src_file;
 492         u64 root_dirid;
 493         int namelen;
 494         int ret = 0;
 495
 496         if (root->fs_info->sb->s_flags & MS_RDONLY)
 497                 return -EROFS;
 498
 499         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 500
 501         if (!vol_args)
 502                 return -ENOMEM;
 503
 504         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 505                 ret = -EFAULT;
 506                 goto out;
 507         }
 508
 509         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 510         namelen = strlen(vol_args->name);
 511         if (strchr(vol_args->name, '/')) {
 512                 ret = -EINVAL;
 513                 goto out;
 514         }
 515
 516         path = btrfs_alloc_path();
 517         if (!path) {
 518                 ret = -ENOMEM;
 519                 goto out;
 520         }
 521
 522         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 523         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 524                             path, root_dirid,
 525                             vol_args->name, namelen, 0);
 526         btrfs_free_path(path);
 527
 528         if (di && !IS_ERR(di)) {
 529                 ret = -EEXIST;
 530                 goto out;
 531         }
 532
 533         if (IS_ERR(di)) {
 534                 ret = PTR_ERR(di);
 535                 goto out;
 536         }
 537
 538         if (subvol) {
 539                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 540                                      file->f_path.dentry->d_inode->i_mode,
 541                                      namelen, NULL);
 542         } else {
 543                 struct inode *src_inode;
 544                 src_file = fget(vol_args->fd);
 545                 if (!src_file) {
 546                         ret = -EINVAL;
 547                         goto out;
 548                 }
 549
 550                 src_inode = src_file->f_path.dentry->d_inode;
 551                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 552                         printk("btrfs: Snapshot src from another FS\n");
 553                         ret = -EINVAL;
 554                         fput(src_file);
 555                         goto out;
 556                 }
 557                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 558                              file->f_path.dentry->d_inode->i_mode,
 559                              namelen, BTRFS_I(src_inode)->root);
 560                 fput(src_file);
 561         }
 562
 563 out:
 564         kfree(vol_args);
 565         return ret;
 566 }
 567
 568 static int btrfs_ioctl_defrag(struct file *file)
 569 {
 570         struct inode *inode = fdentry(file)->d_inode;
 571         struct btrfs_root *root = BTRFS_I(inode)->root;
 572         int ret;
 573
 574         ret = mnt_want_write(file->f_path.mnt);
 575         if (ret)
 576                 return ret;
 577
 578         switch (inode->i_mode & S_IFMT) {
 579         case S_IFDIR:
 580                 btrfs_defrag_root(root, 0);
 581                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 582                 break;
 583         case S_IFREG:
 584                 btrfs_defrag_file(file);
 585                 break;
 586         }
 587
 588         return 0;
 589 }
 590
 591 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 592 {
 593         struct btrfs_ioctl_vol_args *vol_args;
 594         int ret;
 595
 596         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 597
 598         if (!vol_args)
 599                 return -ENOMEM;
 600
 601         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 602                 ret = -EFAULT;
 603                 goto out;
 604         }
 605         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 606         ret = btrfs_init_new_device(root, vol_args->name);
 607
 608 out:
 609         kfree(vol_args);
 610         return ret;
 611 }
 612
 613 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 614 {
 615         struct btrfs_ioctl_vol_args *vol_args;
 616         int ret;
 617
 618         if (root->fs_info->sb->s_flags & MS_RDONLY)
 619                 return -EROFS;
 620
 621         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 622
 623         if (!vol_args)
 624                 return -ENOMEM;
 625
 626         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 627                 ret = -EFAULT;
 628                 goto out;
 629         }
 630         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 631         ret = btrfs_rm_device(root, vol_args->name);
 632
 633 out:
 634         kfree(vol_args);
 635         return ret;
 636 }
 637
 638 long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, u64 off,
 639                        u64 olen, u64 destoff)
 640 {
 641         struct inode *inode = fdentry(file)->d_inode;
 642         struct btrfs_root *root = BTRFS_I(inode)->root;
 643         struct file *src_file;
 644         struct inode *src;
 645         struct btrfs_trans_handle *trans;
 646         struct btrfs_path *path;
 647         struct extent_buffer *leaf;
 648         char *buf;
 649         struct btrfs_key key;
 650         u32 nritems;
 651         int slot;
 652         int ret;
 653         u64 len = olen;
 654         u64 bs = root->fs_info->sb->s_blocksize;
 655         u64 hint_byte;
 656
 657         /*
 658          * TODO:
 659          * - split compressed inline extents.  annoying: we need to
 660          *   decompress into destination's address_space (the file offset
 661          *   may change, so source mapping won't do), then recompress (or
 662          *   otherwise reinsert) a subrange.
 663          * - allow ranges within the same file to be cloned (provided
 664          *   they don't overlap)?
 665          */
 666
 667         ret = mnt_want_write(file->f_path.mnt);
 668         if (ret)
 669                 return ret;
 670
 671         src_file = fget(srcfd);
 672         if (!src_file)
 673                 return -EBADF;
 674         src = src_file->f_dentry->d_inode;
 675
 676         ret = -EINVAL;
 677         if (src == inode)
 678                 goto out_fput;
 679
 680         ret = -EISDIR;
 681         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 682                 goto out_fput;
 683
 684         ret = -EXDEV;
 685         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 686                 goto out_fput;
 687
 688         ret = -ENOMEM;
 689         buf = vmalloc(btrfs_level_size(root, 0));
 690         if (!buf)
 691                 goto out_fput;
 692
 693         path = btrfs_alloc_path();
 694         if (!path) {
 695                 vfree(buf);
 696                 goto out_fput;
 697         }
 698         path->reada = 2;
 699
 700         if (inode < src) {
 701                 mutex_lock(&inode->i_mutex);
 702                 mutex_lock(&src->i_mutex);
 703         } else {
 704                 mutex_lock(&src->i_mutex);
 705                 mutex_lock(&inode->i_mutex);
 706         }
 707
 708         /* determine range to clone */
 709         ret = -EINVAL;
 710         if (off >= src->i_size || off + len > src->i_size)
 711                 goto out_unlock;
 712         if (len == 0)
 713                 olen = len = src->i_size - off;
 714         /* if we extend to eof, continue to block boundary */
 715         if (off + len == src->i_size)
 716                 len = ((src->i_size + bs-1) & ~(bs-1))
 717                         - off;
 718
 719         /* verify the end result is block aligned */
 720         if ((off & (bs-1)) ||
 721             ((off + len) & (bs-1)))
 722                 goto out_unlock;
 723
 724         printk("final src extent is %llu~%llu\n", off, len);
 725         printk("final dst extent is %llu~%llu\n", destoff, len);
 726
 727         /* do any pending delalloc/csum calc on src, one way or
 728            another, and lock file content */
 729         while (1) {
 730                 struct btrfs_ordered_extent *ordered;
 731                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 732                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
 733                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 734                         break;
 735                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 736                 if (ordered)
 737                         btrfs_put_ordered_extent(ordered);
 738                 btrfs_wait_ordered_range(src, off, off+len);
 739         }
 740
 741         trans = btrfs_start_transaction(root, 1);
 742         BUG_ON(!trans);
 743
 744         /* punch hole in destination first */
 745         btrfs_drop_extents(trans, root, inode, off, off+len, 0, &hint_byte);
 746
 747         /* clone data */
 748         key.objectid = src->i_ino;
 749         key.type = BTRFS_EXTENT_DATA_KEY;
 750         key.offset = 0;
 751
 752         while (1) {
 753                 /*
 754                  * note the key will change type as we walk through the
 755                  * tree.
 756                  */
 757                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 758                 if (ret < 0)
 759                         goto out;
 760
 761                 nritems = btrfs_header_nritems(path->nodes[0]);
 762                 if (path->slots[0] >= nritems) {
 763                         ret = btrfs_next_leaf(root, path);
 764                         if (ret < 0)
 765                                 goto out;
 766                         if (ret > 0)
 767                                 break;
 768                         nritems = btrfs_header_nritems(path->nodes[0]);
 769                 }
 770                 leaf = path->nodes[0];
 771                 slot = path->slots[0];
 772
 773                 btrfs_item_key_to_cpu(leaf, &key, slot);
 774                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
 775                     key.objectid != src->i_ino)
 776                         break;
 777
 778                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 779                         struct btrfs_file_extent_item *extent;
 780                         int type;
 781                         u32 size;
 782                         struct btrfs_key new_key;
 783                         u64 disko = 0, diskl = 0;
 784                         u64 datao = 0, datal = 0;
 785                         u8 comp;
 786
 787                         size = btrfs_item_size_nr(leaf, slot);
 788                         read_extent_buffer(leaf, buf,
 789                                            btrfs_item_ptr_offset(leaf, slot),
 790                                            size);
 791
 792                         extent = btrfs_item_ptr(leaf, slot,
 793                                                 struct btrfs_file_extent_item);
 794                         comp = btrfs_file_extent_compression(leaf, extent);
 795                         type = btrfs_file_extent_type(leaf, extent);
 796                         if (type == BTRFS_FILE_EXTENT_REG) {
 797                                 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
 798                                 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
 799                                 datao = btrfs_file_extent_offset(leaf, extent);
 800                                 datal = btrfs_file_extent_num_bytes(leaf, extent);
 801                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 802                                 /* take upper bound, may be compressed */
 803                                 datal = btrfs_file_extent_ram_bytes(leaf,
 804                                                                     extent);
 805                         }
 806                         btrfs_release_path(root, path);
 807
 808                         if (key.offset + datal < off ||
 809                             key.offset >= off+len)
 810                                 goto next;
 811
 812                         memcpy(&new_key, &key, sizeof(new_key));
 813                         new_key.objectid = inode->i_ino;
 814                         new_key.offset = key.offset + destoff - off;
 815
 816                         if (type == BTRFS_FILE_EXTENT_REG) {
 817                                 ret = btrfs_insert_empty_item(trans, root, path,
 818                                                               &new_key, size);
 819                                 if (ret)
 820                                         goto out;
 821
 822                                 leaf = path->nodes[0];
 823                                 slot = path->slots[0];
 824                                 write_extent_buffer(leaf, buf,
 825                                             btrfs_item_ptr_offset(leaf, slot),
 826                                             size);
 827
 828                                 extent = btrfs_item_ptr(leaf, slot,
 829                                                 struct btrfs_file_extent_item);
 830                                 printk("  orig disk %llu~%llu data %llu~%llu\n",
 831                                        disko, diskl, datao, datal);
 832
 833                                 if (off > key.offset) {
 834                                         datao += off - key.offset;
 835                                         datal -= off - key.offset;
 836                                 }
 837                                 if (key.offset + datao + datal + key.offset >
 838                                     off + len)
 839                                         datal = off + len - key.offset - datao;
 840                                 /* disko == 0 means it's a hole */
 841                                 if (!disko)
 842                                         datao = 0;
 843                                 printk(" final disk %llu~%llu data %llu~%llu\n",
 844                                        disko, diskl, datao, datal);
 845
 846                                 btrfs_set_file_extent_offset(leaf, extent,
 847                                                              datao);
 848                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 849                                                                 datal);
 850                                 if (disko) {
 851                                         inode_add_bytes(inode, datal);
 852                                         ret = btrfs_inc_extent_ref(trans, root,
 853                                                    disko, diskl, leaf->start,
 854                                                    root->root_key.objectid,
 855                                                    trans->transid,
 856                                                    inode->i_ino);
 857                                         BUG_ON(ret);
 858                                 }
 859                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 860                                 u64 skip = 0;
 861                                 u64 trim = 0;
 862                                 if (off > key.offset) {
 863                                         skip = off - key.offset;
 864                                         new_key.offset += skip;
 865                                 }
 866                                 if (key.offset + datal > off+len)
 867                                         trim = key.offset + datal - (off+len);
 868                                 printk("len %lld skip %lld trim %lld\n",
 869                                        datal, skip, trim);
 870                                 if (comp && (skip || trim)) {
 871                                         printk("btrfs clone_range can't split compressed inline extents yet\n");
 872                                         ret = -EINVAL;
 873                                         goto out;
 874                                 }
 875                                 size -= skip + trim;
 876                                 datal -= skip + trim;
 877                                 ret = btrfs_insert_empty_item(trans, root, path,
 878                                                               &new_key, size);
 879                                 if (ret)
 880                                         goto out;
 881
 882                                 if (skip) {
 883                                         u32 start = btrfs_file_extent_calc_inline_size(0);
 884                                         memmove(buf+start, buf+start+skip,
 885                                                 datal);
 886                                 }
 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                                 inode_add_bytes(inode, datal);
 894                         }
 895
 896                         btrfs_mark_buffer_dirty(leaf);
 897                 }
 898
 899                 if (btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
 900                         u32 size;
 901                         struct btrfs_key new_key;
 902                         u64 coverslen;
 903                         int coff, clen;
 904
 905                         size = btrfs_item_size_nr(leaf, slot);
 906                         coverslen = (size / BTRFS_CRC32_SIZE) <<
 907                                 root->fs_info->sb->s_blocksize_bits;
 908                         printk("csums for %llu~%llu\n",
 909                                key.offset, coverslen);
 910                         if (key.offset + coverslen < off ||
 911                             key.offset >= off+len)
 912                                 goto next;
 913
 914                         read_extent_buffer(leaf, buf,
 915                                            btrfs_item_ptr_offset(leaf, slot),
 916                                            size);
 917                         btrfs_release_path(root, path);
 918
 919                         coff = 0;
 920                         if (off > key.offset)
 921                                 coff = ((off - key.offset) >>
 922                                         root->fs_info->sb->s_blocksize_bits) *
 923                                         BTRFS_CRC32_SIZE;
 924                         clen = size - coff;
 925                         if (key.offset + coverslen > off+len)
 926                                 clen -= ((key.offset+coverslen-off-len) >>
 927                                          root->fs_info->sb->s_blocksize_bits) *
 928                                         BTRFS_CRC32_SIZE;
 929                         printk(" will dup %d~%d of %d\n",
 930                                coff, clen, size);
 931
 932                         memcpy(&new_key, &key, sizeof(new_key));
 933                         new_key.objectid = inode->i_ino;
 934                         new_key.offset = key.offset + destoff - off;
 935
 936                         ret = btrfs_insert_empty_item(trans, root, path,
 937                                                       &new_key, clen);
 938                         if (ret)
 939                                 goto out;
 940
 941                         leaf = path->nodes[0];
 942                         slot = path->slots[0];
 943                         write_extent_buffer(leaf, buf + coff,
 944                                             btrfs_item_ptr_offset(leaf, slot),
 945                                             clen);
 946                         btrfs_mark_buffer_dirty(leaf);
 947                 }
 948
 949         next:
 950                 btrfs_release_path(root, path);
 951                 key.offset++;
 952         }
 953         ret = 0;
 954 out:
 955         btrfs_release_path(root, path);
 956         if (ret == 0) {
 957                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 958                 if (destoff + olen > inode->i_size)
 959                         btrfs_i_size_write(inode, destoff + olen);
 960                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
 961                 ret = btrfs_update_inode(trans, root, inode);
 962         }
 963         btrfs_end_transaction(trans, root);
 964         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 965         if (ret)
 966                 vmtruncate(inode, 0);
 967 out_unlock:
 968         mutex_unlock(&src->i_mutex);
 969         mutex_unlock(&inode->i_mutex);
 970         vfree(buf);
 971         btrfs_free_path(path);
 972 out_fput:
 973         fput(src_file);
 974         return ret;
 975 }
 976
 977 long btrfs_ioctl_clone_range(struct file *file, unsigned long argptr)
 978 {
 979         struct btrfs_ioctl_clone_range_args args;
 980
 981         if (copy_from_user(&args, (void *)argptr, sizeof(args)))
 982                 return -EFAULT;
 983         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
 984                                  args.src_length, args.dest_offset);
 985 }
 986
 987 /*
 988  * there are many ways the trans_start and trans_end ioctls can lead
 989  * to deadlocks.  They should only be used by applications that
 990  * basically own the machine, and have a very in depth understanding
 991  * of all the possible deadlocks and enospc problems.
 992  */
 993 long btrfs_ioctl_trans_start(struct file *file)
 994 {
 995         struct inode *inode = fdentry(file)->d_inode;
 996         struct btrfs_root *root = BTRFS_I(inode)->root;
 997         struct btrfs_trans_handle *trans;
 998         int ret = 0;
 999
1000         if (!capable(CAP_SYS_ADMIN))
1001                 return -EPERM;
1002
1003         if (file->private_data) {
1004                 ret = -EINPROGRESS;
1005                 goto out;
1006         }
1007
1008         ret = mnt_want_write(file->f_path.mnt);
1009         if (ret)
1010                 goto out;
1011
1012         mutex_lock(&root->fs_info->trans_mutex);
1013         root->fs_info->open_ioctl_trans++;
1014         mutex_unlock(&root->fs_info->trans_mutex);
1015
1016         trans = btrfs_start_ioctl_transaction(root, 0);
1017         if (trans)
1018                 file->private_data = trans;
1019         else
1020                 ret = -ENOMEM;
1021         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1022 out:
1023         return ret;
1024 }
1025
1026 /*
1027  * there are many ways the trans_start and trans_end ioctls can lead
1028  * to deadlocks.  They should only be used by applications that
1029  * basically own the machine, and have a very in depth understanding
1030  * of all the possible deadlocks and enospc problems.
1031  */
1032 long btrfs_ioctl_trans_end(struct file *file)
1033 {
1034         struct inode *inode = fdentry(file)->d_inode;
1035         struct btrfs_root *root = BTRFS_I(inode)->root;
1036         struct btrfs_trans_handle *trans;
1037         int ret = 0;
1038
1039         trans = file->private_data;
1040         if (!trans) {
1041                 ret = -EINVAL;
1042                 goto out;
1043         }
1044         btrfs_end_transaction(trans, root);
1045         file->private_data = NULL;
1046
1047         mutex_lock(&root->fs_info->trans_mutex);
1048         root->fs_info->open_ioctl_trans--;
1049         mutex_unlock(&root->fs_info->trans_mutex);
1050
1051 out:
1052         return ret;
1053 }
1054
1055 long btrfs_ioctl(struct file *file, unsigned int
1056                 cmd, unsigned long arg)
1057 {
1058         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1059
1060         switch (cmd) {
1061         case BTRFS_IOC_SNAP_CREATE:
1062                 return btrfs_ioctl_snap_create(file, (void __user *)arg, 0);
1063         case BTRFS_IOC_SUBVOL_CREATE:
1064                 return btrfs_ioctl_snap_create(file, (void __user *)arg, 1);
1065         case BTRFS_IOC_DEFRAG:
1066                 return btrfs_ioctl_defrag(file);
1067         case BTRFS_IOC_RESIZE:
1068                 return btrfs_ioctl_resize(root, (void __user *)arg);
1069         case BTRFS_IOC_ADD_DEV:
1070                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
1071         case BTRFS_IOC_RM_DEV:
1072                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
1073         case BTRFS_IOC_BALANCE:
1074                 return btrfs_balance(root->fs_info->dev_root);
1075         case BTRFS_IOC_CLONE:
1076                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1077         case BTRFS_IOC_CLONE_RANGE:
1078                 return btrfs_ioctl_clone_range(file, arg);
1079         case BTRFS_IOC_TRANS_START:
1080                 return btrfs_ioctl_trans_start(file);
1081         case BTRFS_IOC_TRANS_END:
1082                 return btrfs_ioctl_trans_end(file);
1083         case BTRFS_IOC_SYNC:
1084                 btrfs_start_delalloc_inodes(root);
1085                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1086                 return 0;
1087         }
1088
1089         return -ENOTTY;
1090 }