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         unsigned long nr = 1;
  71
  72         ret = btrfs_check_free_space(root, 1, 0);
  73         if (ret)
  74                 goto fail_commit;
  75
  76         trans = btrfs_start_transaction(root, 1);
  77         BUG_ON(!trans);
  78
  79         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
  80                                        0, &objectid);
  81         if (ret)
  82                 goto fail;
  83
  84         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
  85                                       objectid, trans->transid, 0, 0, 0);
  86         if (IS_ERR(leaf)) {
  87                 ret = PTR_ERR(leaf);
  88                 goto fail;
  89         }
  90
  91         btrfs_set_header_nritems(leaf, 0);
  92         btrfs_set_header_level(leaf, 0);
  93         btrfs_set_header_bytenr(leaf, leaf->start);
  94         btrfs_set_header_generation(leaf, trans->transid);
  95         btrfs_set_header_owner(leaf, objectid);
  96
  97         write_extent_buffer(leaf, root->fs_info->fsid,
  98                             (unsigned long)btrfs_header_fsid(leaf),
  99                             BTRFS_FSID_SIZE);
 100         btrfs_mark_buffer_dirty(leaf);
 101
 102         inode_item = &root_item.inode;
 103         memset(inode_item, 0, sizeof(*inode_item));
 104         inode_item->generation = cpu_to_le64(1);
 105         inode_item->size = cpu_to_le64(3);
 106         inode_item->nlink = cpu_to_le32(1);
 107         inode_item->nbytes = cpu_to_le64(root->leafsize);
 108         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 109
 110         btrfs_set_root_bytenr(&root_item, leaf->start);
 111         btrfs_set_root_generation(&root_item, trans->transid);
 112         btrfs_set_root_level(&root_item, 0);
 113         btrfs_set_root_refs(&root_item, 1);
 114         btrfs_set_root_used(&root_item, 0);
 115         btrfs_set_root_last_snapshot(&root_item, 0);
 116
 117         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 118         root_item.drop_level = 0;
 119
 120         btrfs_tree_unlock(leaf);
 121         free_extent_buffer(leaf);
 122         leaf = NULL;
 123
 124         btrfs_set_root_dirid(&root_item, new_dirid);
 125
 126         key.objectid = objectid;
 127         key.offset = 1;
 128         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 129         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 130                                 &root_item);
 131         if (ret)
 132                 goto fail;
 133
 134         /*
 135          * insert the directory item
 136          */
 137         key.offset = (u64)-1;
 138         dir = root->fs_info->sb->s_root->d_inode;
 139         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 140                                     name, namelen, dir->i_ino, &key,
 141                                     BTRFS_FT_DIR, 0);
 142         if (ret)
 143                 goto fail;
 144
 145         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 146                              name, namelen, objectid,
 147                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 148         if (ret)
 149                 goto fail;
 150
 151         ret = btrfs_commit_transaction(trans, root);
 152         if (ret)
 153                 goto fail_commit;
 154
 155         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
 156         BUG_ON(!new_root);
 157
 158         trans = btrfs_start_transaction(new_root, 1);
 159         BUG_ON(!trans);
 160
 161         ret = btrfs_create_subvol_root(new_root, dentry, trans, new_dirid,
 162                                        BTRFS_I(dir)->block_group);
 163         if (ret)
 164                 goto fail;
 165
 166 fail:
 167         nr = trans->blocks_used;
 168         err = btrfs_commit_transaction(trans, new_root);
 169         if (err && !ret)
 170                 ret = err;
 171 fail_commit:
 172         btrfs_btree_balance_dirty(root, nr);
 173         return ret;
 174 }
 175
 176 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
 177 {
 178         struct btrfs_pending_snapshot *pending_snapshot;
 179         struct btrfs_trans_handle *trans;
 180         int ret;
 181         int err;
 182         unsigned long nr = 0;
 183
 184         if (!root->ref_cows)
 185                 return -EINVAL;
 186
 187         ret = btrfs_check_free_space(root, 1, 0);
 188         if (ret)
 189                 goto fail_unlock;
 190
 191         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
 192         if (!pending_snapshot) {
 193                 ret = -ENOMEM;
 194                 goto fail_unlock;
 195         }
 196         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 197         if (!pending_snapshot->name) {
 198                 ret = -ENOMEM;
 199                 kfree(pending_snapshot);
 200                 goto fail_unlock;
 201         }
 202         memcpy(pending_snapshot->name, name, namelen);
 203         pending_snapshot->name[namelen] = '\0';
 204         trans = btrfs_start_transaction(root, 1);
 205         BUG_ON(!trans);
 206         pending_snapshot->root = root;
 207         list_add(&pending_snapshot->list,
 208                  &trans->transaction->pending_snapshots);
 209         ret = btrfs_update_inode(trans, root, root->inode);
 210         err = btrfs_commit_transaction(trans, root);
 211
 212 fail_unlock:
 213         btrfs_btree_balance_dirty(root, nr);
 214         return ret;
 215 }
 216
 217 /* copy of may_create in fs/namei.c() */
 218 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 219 {
 220         if (child->d_inode)
 221                 return -EEXIST;
 222         if (IS_DEADDIR(dir))
 223                 return -ENOENT;
 224         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 225 }
 226
 227 /*
 228  * Create a new subvolume below @parent.  This is largely modeled after
 229  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 230  * inside this filesystem so it's quite a bit simpler.
 231  */
 232 static noinline int btrfs_mksubvol(struct path *parent, char *name,
 233                                    int mode, int namelen)
 234 {
 235         struct dentry *dentry;
 236         int error;
 237
 238         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 239
 240         dentry = lookup_one_len(name, parent->dentry, namelen);
 241         error = PTR_ERR(dentry);
 242         if (IS_ERR(dentry))
 243                 goto out_unlock;
 244
 245         error = -EEXIST;
 246         if (dentry->d_inode)
 247                 goto out_dput;
 248
 249         if (!IS_POSIXACL(parent->dentry->d_inode))
 250                 mode &= ~current->fs->umask;
 251         error = mnt_want_write(parent->mnt);
 252         if (error)
 253                 goto out_dput;
 254
 255         error = btrfs_may_create(parent->dentry->d_inode, dentry);
 256         if (error)
 257                 goto out_drop_write;
 258
 259         /*
 260          * Actually perform the low-level subvolume creation after all
 261          * this VFS fuzz.
 262          *
 263          * Eventually we want to pass in an inode under which we create this
 264          * subvolume, but for now all are under the filesystem root.
 265          *
 266          * Also we should pass on the mode eventually to allow creating new
 267          * subvolume with specific mode bits.
 268          */
 269         error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root, dentry,
 270                               name, namelen);
 271         if (error)
 272                 goto out_drop_write;
 273
 274         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 275 out_drop_write:
 276         mnt_drop_write(parent->mnt);
 277 out_dput:
 278         dput(dentry);
 279 out_unlock:
 280         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 281         return error;
 282 }
 283
 284
 285 int btrfs_defrag_file(struct file *file)
 286 {
 287         struct inode *inode = fdentry(file)->d_inode;
 288         struct btrfs_root *root = BTRFS_I(inode)->root;
 289         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 290         struct btrfs_ordered_extent *ordered;
 291         struct page *page;
 292         unsigned long last_index;
 293         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 294         unsigned long total_read = 0;
 295         u64 page_start;
 296         u64 page_end;
 297         unsigned long i;
 298         int ret;
 299
 300         ret = btrfs_check_free_space(root, inode->i_size, 0);
 301         if (ret)
 302                 return -ENOSPC;
 303
 304         mutex_lock(&inode->i_mutex);
 305         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 306         for (i = 0; i <= last_index; i++) {
 307                 if (total_read % ra_pages == 0) {
 308                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 309                                        min(last_index, i + ra_pages - 1));
 310                 }
 311                 total_read++;
 312 again:
 313                 page = grab_cache_page(inode->i_mapping, i);
 314                 if (!page)
 315                         goto out_unlock;
 316                 if (!PageUptodate(page)) {
 317                         btrfs_readpage(NULL, page);
 318                         lock_page(page);
 319                         if (!PageUptodate(page)) {
 320                                 unlock_page(page);
 321                                 page_cache_release(page);
 322                                 goto out_unlock;
 323                         }
 324                 }
 325
 326                 wait_on_page_writeback(page);
 327
 328                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 329                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 330                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 331
 332                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 333                 if (ordered) {
 334                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 335                         unlock_page(page);
 336                         page_cache_release(page);
 337                         btrfs_start_ordered_extent(inode, ordered, 1);
 338                         btrfs_put_ordered_extent(ordered);
 339                         goto again;
 340                 }
 341                 set_page_extent_mapped(page);
 342
 343                 /*
 344                  * this makes sure page_mkwrite is called on the
 345                  * page if it is dirtied again later
 346                  */
 347                 clear_page_dirty_for_io(page);
 348
 349                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 350
 351                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 352                 set_page_dirty(page);
 353                 unlock_page(page);
 354                 page_cache_release(page);
 355                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 356         }
 357
 358 out_unlock:
 359         mutex_unlock(&inode->i_mutex);
 360         return 0;
 361 }
 362
 363 /*
 364  * Called inside transaction, so use GFP_NOFS
 365  */
 366
 367 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 368 {
 369         u64 new_size;
 370         u64 old_size;
 371         u64 devid = 1;
 372         struct btrfs_ioctl_vol_args *vol_args;
 373         struct btrfs_trans_handle *trans;
 374         struct btrfs_device *device = NULL;
 375         char *sizestr;
 376         char *devstr = NULL;
 377         int ret = 0;
 378         int namelen;
 379         int mod = 0;
 380
 381         if (root->fs_info->sb->s_flags & MS_RDONLY)
 382                 return -EROFS;
 383
 384         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 385
 386         if (!vol_args)
 387                 return -ENOMEM;
 388
 389         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 390                 ret = -EFAULT;
 391                 goto out;
 392         }
 393
 394         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 395         namelen = strlen(vol_args->name);
 396
 397         mutex_lock(&root->fs_info->volume_mutex);
 398         sizestr = vol_args->name;
 399         devstr = strchr(sizestr, ':');
 400         if (devstr) {
 401                 char *end;
 402                 sizestr = devstr + 1;
 403                 *devstr = '\0';
 404                 devstr = vol_args->name;
 405                 devid = simple_strtoull(devstr, &end, 10);
 406                 printk(KERN_INFO "resizing devid %llu\n", devid);
 407         }
 408         device = btrfs_find_device(root, devid, NULL, NULL);
 409         if (!device) {
 410                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 411                 ret = -EINVAL;
 412                 goto out_unlock;
 413         }
 414         if (!strcmp(sizestr, "max"))
 415                 new_size = device->bdev->bd_inode->i_size;
 416         else {
 417                 if (sizestr[0] == '-') {
 418                         mod = -1;
 419                         sizestr++;
 420                 } else if (sizestr[0] == '+') {
 421                         mod = 1;
 422                         sizestr++;
 423                 }
 424                 new_size = btrfs_parse_size(sizestr);
 425                 if (new_size == 0) {
 426                         ret = -EINVAL;
 427                         goto out_unlock;
 428                 }
 429         }
 430
 431         old_size = device->total_bytes;
 432
 433         if (mod < 0) {
 434                 if (new_size > old_size) {
 435                         ret = -EINVAL;
 436                         goto out_unlock;
 437                 }
 438                 new_size = old_size - new_size;
 439         } else if (mod > 0) {
 440                 new_size = old_size + new_size;
 441         }
 442
 443         if (new_size < 256 * 1024 * 1024) {
 444                 ret = -EINVAL;
 445                 goto out_unlock;
 446         }
 447         if (new_size > device->bdev->bd_inode->i_size) {
 448                 ret = -EFBIG;
 449                 goto out_unlock;
 450         }
 451
 452         do_div(new_size, root->sectorsize);
 453         new_size *= root->sectorsize;
 454
 455         printk(KERN_INFO "new size for %s is %llu\n",
 456                 device->name, (unsigned long long)new_size);
 457
 458         if (new_size > old_size) {
 459                 trans = btrfs_start_transaction(root, 1);
 460                 ret = btrfs_grow_device(trans, device, new_size);
 461                 btrfs_commit_transaction(trans, root);
 462         } else {
 463                 ret = btrfs_shrink_device(device, new_size);
 464         }
 465
 466 out_unlock:
 467         mutex_unlock(&root->fs_info->volume_mutex);
 468 out:
 469         kfree(vol_args);
 470         return ret;
 471 }
 472
 473 static noinline int btrfs_ioctl_snap_create(struct file *file,
 474                                             void __user *arg)
 475 {
 476         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 477         struct btrfs_ioctl_vol_args *vol_args;
 478         struct btrfs_dir_item *di;
 479         struct btrfs_path *path;
 480         u64 root_dirid;
 481         int namelen;
 482         int ret;
 483
 484         if (root->fs_info->sb->s_flags & MS_RDONLY)
 485                 return -EROFS;
 486
 487         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 488
 489         if (!vol_args)
 490                 return -ENOMEM;
 491
 492         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 493                 ret = -EFAULT;
 494                 goto out;
 495         }
 496
 497         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 498         namelen = strlen(vol_args->name);
 499         if (strchr(vol_args->name, '/')) {
 500                 ret = -EINVAL;
 501                 goto out;
 502         }
 503
 504         path = btrfs_alloc_path();
 505         if (!path) {
 506                 ret = -ENOMEM;
 507                 goto out;
 508         }
 509
 510         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 511         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 512                             path, root_dirid,
 513                             vol_args->name, namelen, 0);
 514         btrfs_free_path(path);
 515
 516         if (di && !IS_ERR(di)) {
 517                 ret = -EEXIST;
 518                 goto out;
 519         }
 520
 521         if (IS_ERR(di)) {
 522                 ret = PTR_ERR(di);
 523                 goto out;
 524         }
 525
 526         if (root == root->fs_info->tree_root) {
 527                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 528                                      file->f_path.dentry->d_inode->i_mode,
 529                                      namelen);
 530         } else {
 531                 ret = create_snapshot(root, vol_args->name, namelen);
 532         }
 533
 534 out:
 535         kfree(vol_args);
 536         return ret;
 537 }
 538
 539 static int btrfs_ioctl_defrag(struct file *file)
 540 {
 541         struct inode *inode = fdentry(file)->d_inode;
 542         struct btrfs_root *root = BTRFS_I(inode)->root;
 543         int ret;
 544
 545         ret = mnt_want_write(file->f_path.mnt);
 546         if (ret)
 547                 return ret;
 548
 549         switch (inode->i_mode & S_IFMT) {
 550         case S_IFDIR:
 551                 btrfs_defrag_root(root, 0);
 552                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 553                 break;
 554         case S_IFREG:
 555                 btrfs_defrag_file(file);
 556                 break;
 557         }
 558
 559         return 0;
 560 }
 561
 562 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 563 {
 564         struct btrfs_ioctl_vol_args *vol_args;
 565         int ret;
 566
 567         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 568
 569         if (!vol_args)
 570                 return -ENOMEM;
 571
 572         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 573                 ret = -EFAULT;
 574                 goto out;
 575         }
 576         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 577         ret = btrfs_init_new_device(root, vol_args->name);
 578
 579 out:
 580         kfree(vol_args);
 581         return ret;
 582 }
 583
 584 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 585 {
 586         struct btrfs_ioctl_vol_args *vol_args;
 587         int ret;
 588
 589         if (root->fs_info->sb->s_flags & MS_RDONLY)
 590                 return -EROFS;
 591
 592         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 593
 594         if (!vol_args)
 595                 return -ENOMEM;
 596
 597         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 598                 ret = -EFAULT;
 599                 goto out;
 600         }
 601         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 602         ret = btrfs_rm_device(root, vol_args->name);
 603
 604 out:
 605         kfree(vol_args);
 606         return ret;
 607 }
 608
 609 long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, u64 off,
 610                        u64 olen, u64 destoff)
 611 {
 612         struct inode *inode = fdentry(file)->d_inode;
 613         struct btrfs_root *root = BTRFS_I(inode)->root;
 614         struct file *src_file;
 615         struct inode *src;
 616         struct btrfs_trans_handle *trans;
 617         struct btrfs_path *path;
 618         struct extent_buffer *leaf;
 619         char *buf;
 620         struct btrfs_key key;
 621         u32 nritems;
 622         int slot;
 623         int ret;
 624         u64 len = olen;
 625         u64 bs = root->fs_info->sb->s_blocksize;
 626         u64 hint_byte;
 627
 628         /*
 629          * TODO:
 630          * - split compressed inline extents.  annoying: we need to
 631          *   decompress into destination's address_space (the file offset
 632          *   may change, so source mapping won't do), then recompress (or
 633          *   otherwise reinsert) a subrange.
 634          * - allow ranges within the same file to be cloned (provided
 635          *   they don't overlap)?
 636          */
 637
 638         ret = mnt_want_write(file->f_path.mnt);
 639         if (ret)
 640                 return ret;
 641
 642         src_file = fget(srcfd);
 643         if (!src_file)
 644                 return -EBADF;
 645         src = src_file->f_dentry->d_inode;
 646
 647         ret = -EINVAL;
 648         if (src == inode)
 649                 goto out_fput;
 650
 651         ret = -EISDIR;
 652         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 653                 goto out_fput;
 654
 655         ret = -EXDEV;
 656         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 657                 goto out_fput;
 658
 659         ret = -ENOMEM;
 660         buf = vmalloc(btrfs_level_size(root, 0));
 661         if (!buf)
 662                 goto out_fput;
 663
 664         path = btrfs_alloc_path();
 665         if (!path) {
 666                 vfree(buf);
 667                 goto out_fput;
 668         }
 669         path->reada = 2;
 670
 671         if (inode < src) {
 672                 mutex_lock(&inode->i_mutex);
 673                 mutex_lock(&src->i_mutex);
 674         } else {
 675                 mutex_lock(&src->i_mutex);
 676                 mutex_lock(&inode->i_mutex);
 677         }
 678
 679         /* determine range to clone */
 680         ret = -EINVAL;
 681         if (off >= src->i_size || off + len > src->i_size)
 682                 goto out_unlock;
 683         if (len == 0)
 684                 olen = len = src->i_size - off;
 685         /* if we extend to eof, continue to block boundary */
 686         if (off + len == src->i_size)
 687                 len = ((src->i_size + bs-1) & ~(bs-1))
 688                         - off;
 689
 690         /* verify the end result is block aligned */
 691         if ((off & (bs-1)) ||
 692             ((off + len) & (bs-1)))
 693                 goto out_unlock;
 694
 695         printk("final src extent is %llu~%llu\n", off, len);
 696         printk("final dst extent is %llu~%llu\n", destoff, len);
 697
 698         /* do any pending delalloc/csum calc on src, one way or
 699            another, and lock file content */
 700         while (1) {
 701                 struct btrfs_ordered_extent *ordered;
 702                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 703                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
 704                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 705                         break;
 706                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 707                 if (ordered)
 708                         btrfs_put_ordered_extent(ordered);
 709                 btrfs_wait_ordered_range(src, off, off+len);
 710         }
 711
 712         trans = btrfs_start_transaction(root, 1);
 713         BUG_ON(!trans);
 714
 715         /* punch hole in destination first */
 716         btrfs_drop_extents(trans, root, inode, off, off+len, 0, &hint_byte);
 717
 718         /* clone data */
 719         key.objectid = src->i_ino;
 720         key.type = BTRFS_EXTENT_DATA_KEY;
 721         key.offset = 0;
 722
 723         while (1) {
 724                 /*
 725                  * note the key will change type as we walk through the
 726                  * tree.
 727                  */
 728                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 729                 if (ret < 0)
 730                         goto out;
 731
 732                 nritems = btrfs_header_nritems(path->nodes[0]);
 733                 if (path->slots[0] >= nritems) {
 734                         ret = btrfs_next_leaf(root, path);
 735                         if (ret < 0)
 736                                 goto out;
 737                         if (ret > 0)
 738                                 break;
 739                         nritems = btrfs_header_nritems(path->nodes[0]);
 740                 }
 741                 leaf = path->nodes[0];
 742                 slot = path->slots[0];
 743
 744                 btrfs_item_key_to_cpu(leaf, &key, slot);
 745                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
 746                     key.objectid != src->i_ino)
 747                         break;
 748
 749                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 750                         struct btrfs_file_extent_item *extent;
 751                         int type;
 752                         u32 size;
 753                         struct btrfs_key new_key;
 754                         u64 disko = 0, diskl = 0;
 755                         u64 datao = 0, datal = 0;
 756                         u8 comp;
 757
 758                         size = btrfs_item_size_nr(leaf, slot);
 759                         read_extent_buffer(leaf, buf,
 760                                            btrfs_item_ptr_offset(leaf, slot),
 761                                            size);
 762
 763                         extent = btrfs_item_ptr(leaf, slot,
 764                                                 struct btrfs_file_extent_item);
 765                         comp = btrfs_file_extent_compression(leaf, extent);
 766                         type = btrfs_file_extent_type(leaf, extent);
 767                         if (type == BTRFS_FILE_EXTENT_REG) {
 768                                 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
 769                                 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
 770                                 datao = btrfs_file_extent_offset(leaf, extent);
 771                                 datal = btrfs_file_extent_num_bytes(leaf, extent);
 772                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 773                                 /* take upper bound, may be compressed */
 774                                 datal = btrfs_file_extent_ram_bytes(leaf,
 775                                                                     extent);
 776                         }
 777                         btrfs_release_path(root, path);
 778
 779                         if (key.offset + datal < off ||
 780                             key.offset >= off+len)
 781                                 goto next;
 782
 783                         memcpy(&new_key, &key, sizeof(new_key));
 784                         new_key.objectid = inode->i_ino;
 785                         new_key.offset = key.offset + destoff - off;
 786
 787                         if (type == BTRFS_FILE_EXTENT_REG) {
 788                                 ret = btrfs_insert_empty_item(trans, root, path,
 789                                                               &new_key, size);
 790                                 if (ret)
 791                                         goto out;
 792
 793                                 leaf = path->nodes[0];
 794                                 slot = path->slots[0];
 795                                 write_extent_buffer(leaf, buf,
 796                                             btrfs_item_ptr_offset(leaf, slot),
 797                                             size);
 798
 799                                 extent = btrfs_item_ptr(leaf, slot,
 800                                                 struct btrfs_file_extent_item);
 801                                 printk("  orig disk %llu~%llu data %llu~%llu\n",
 802                                        disko, diskl, datao, datal);
 803
 804                                 if (off > key.offset) {
 805                                         datao += off - key.offset;
 806                                         datal -= off - key.offset;
 807                                 }
 808                                 if (key.offset + datao + datal + key.offset >
 809                                     off + len)
 810                                         datal = off + len - key.offset - datao;
 811                                 /* disko == 0 means it's a hole */
 812                                 if (!disko)
 813                                         datao = 0;
 814                                 printk(" final disk %llu~%llu data %llu~%llu\n",
 815                                        disko, diskl, datao, datal);
 816
 817                                 btrfs_set_file_extent_offset(leaf, extent,
 818                                                              datao);
 819                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 820                                                                 datal);
 821                                 if (disko) {
 822                                         inode_add_bytes(inode, datal);
 823                                         ret = btrfs_inc_extent_ref(trans, root,
 824                                                    disko, diskl, leaf->start,
 825                                                    root->root_key.objectid,
 826                                                    trans->transid,
 827                                                    inode->i_ino);
 828                                         BUG_ON(ret);
 829                                 }
 830                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 831                                 u64 skip = 0;
 832                                 u64 trim = 0;
 833                                 if (off > key.offset) {
 834                                         skip = off - key.offset;
 835                                         new_key.offset += skip;
 836                                 }
 837                                 if (key.offset + datal > off+len)
 838                                         trim = key.offset + datal - (off+len);
 839                                 printk("len %lld skip %lld trim %lld\n",
 840                                        datal, skip, trim);
 841                                 if (comp && (skip || trim)) {
 842                                         printk("btrfs clone_range can't split compressed inline extents yet\n");
 843                                         ret = -EINVAL;
 844                                         goto out;
 845                                 }
 846                                 size -= skip + trim;
 847                                 datal -= skip + trim;
 848                                 ret = btrfs_insert_empty_item(trans, root, path,
 849                                                               &new_key, size);
 850                                 if (ret)
 851                                         goto out;
 852
 853                                 if (skip) {
 854                                         u32 start = btrfs_file_extent_calc_inline_size(0);
 855                                         memmove(buf+start, buf+start+skip,
 856                                                 datal);
 857                                 }
 858
 859                                 leaf = path->nodes[0];
 860                                 slot = path->slots[0];
 861                                 write_extent_buffer(leaf, buf,
 862                                             btrfs_item_ptr_offset(leaf, slot),
 863                                             size);
 864                                 inode_add_bytes(inode, datal);
 865                         }
 866
 867                         btrfs_mark_buffer_dirty(leaf);
 868                 }
 869
 870                 if (btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
 871                         u32 size;
 872                         struct btrfs_key new_key;
 873                         u64 coverslen;
 874                         int coff, clen;
 875
 876                         size = btrfs_item_size_nr(leaf, slot);
 877                         coverslen = (size / BTRFS_CRC32_SIZE) <<
 878                                 root->fs_info->sb->s_blocksize_bits;
 879                         printk("csums for %llu~%llu\n",
 880                                key.offset, coverslen);
 881                         if (key.offset + coverslen < off ||
 882                             key.offset >= off+len)
 883                                 goto next;
 884
 885                         read_extent_buffer(leaf, buf,
 886                                            btrfs_item_ptr_offset(leaf, slot),
 887                                            size);
 888                         btrfs_release_path(root, path);
 889
 890                         coff = 0;
 891                         if (off > key.offset)
 892                                 coff = ((off - key.offset) >>
 893                                         root->fs_info->sb->s_blocksize_bits) *
 894                                         BTRFS_CRC32_SIZE;
 895                         clen = size - coff;
 896                         if (key.offset + coverslen > off+len)
 897                                 clen -= ((key.offset+coverslen-off-len) >>
 898                                          root->fs_info->sb->s_blocksize_bits) *
 899                                         BTRFS_CRC32_SIZE;
 900                         printk(" will dup %d~%d of %d\n",
 901                                coff, clen, size);
 902
 903                         memcpy(&new_key, &key, sizeof(new_key));
 904                         new_key.objectid = inode->i_ino;
 905                         new_key.offset = key.offset + destoff - off;
 906
 907                         ret = btrfs_insert_empty_item(trans, root, path,
 908                                                       &new_key, clen);
 909                         if (ret)
 910                                 goto out;
 911
 912                         leaf = path->nodes[0];
 913                         slot = path->slots[0];
 914                         write_extent_buffer(leaf, buf + coff,
 915                                             btrfs_item_ptr_offset(leaf, slot),
 916                                             clen);
 917                         btrfs_mark_buffer_dirty(leaf);
 918                 }
 919
 920         next:
 921                 btrfs_release_path(root, path);
 922                 key.offset++;
 923         }
 924         ret = 0;
 925 out:
 926         btrfs_release_path(root, path);
 927         if (ret == 0) {
 928                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 929                 if (destoff + olen > inode->i_size)
 930                         btrfs_i_size_write(inode, destoff + olen);
 931                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
 932                 ret = btrfs_update_inode(trans, root, inode);
 933         }
 934         btrfs_end_transaction(trans, root);
 935         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 936         if (ret)
 937                 vmtruncate(inode, 0);
 938 out_unlock:
 939         mutex_unlock(&src->i_mutex);
 940         mutex_unlock(&inode->i_mutex);
 941         vfree(buf);
 942         btrfs_free_path(path);
 943 out_fput:
 944         fput(src_file);
 945         return ret;
 946 }
 947
 948 long btrfs_ioctl_clone_range(struct file *file, unsigned long argptr)
 949 {
 950         struct btrfs_ioctl_clone_range_args args;
 951
 952         if (copy_from_user(&args, (void *)argptr, sizeof(args)))
 953                 return -EFAULT;
 954         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
 955                                  args.src_length, args.dest_offset);
 956 }
 957
 958 /*
 959  * there are many ways the trans_start and trans_end ioctls can lead
 960  * to deadlocks.  They should only be used by applications that
 961  * basically own the machine, and have a very in depth understanding
 962  * of all the possible deadlocks and enospc problems.
 963  */
 964 long btrfs_ioctl_trans_start(struct file *file)
 965 {
 966         struct inode *inode = fdentry(file)->d_inode;
 967         struct btrfs_root *root = BTRFS_I(inode)->root;
 968         struct btrfs_trans_handle *trans;
 969         int ret = 0;
 970
 971         if (!capable(CAP_SYS_ADMIN))
 972                 return -EPERM;
 973
 974         if (file->private_data) {
 975                 ret = -EINPROGRESS;
 976                 goto out;
 977         }
 978
 979         ret = mnt_want_write(file->f_path.mnt);
 980         if (ret)
 981                 goto out;
 982
 983         mutex_lock(&root->fs_info->trans_mutex);
 984         root->fs_info->open_ioctl_trans++;
 985         mutex_unlock(&root->fs_info->trans_mutex);
 986
 987         trans = btrfs_start_ioctl_transaction(root, 0);
 988         if (trans)
 989                 file->private_data = trans;
 990         else
 991                 ret = -ENOMEM;
 992         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
 993 out:
 994         return ret;
 995 }
 996
 997 /*
 998  * there are many ways the trans_start and trans_end ioctls can lead
 999  * to deadlocks.  They should only be used by applications that
1000  * basically own the machine, and have a very in depth understanding
1001  * of all the possible deadlocks and enospc problems.
1002  */
1003 long btrfs_ioctl_trans_end(struct file *file)
1004 {
1005         struct inode *inode = fdentry(file)->d_inode;
1006         struct btrfs_root *root = BTRFS_I(inode)->root;
1007         struct btrfs_trans_handle *trans;
1008         int ret = 0;
1009
1010         trans = file->private_data;
1011         if (!trans) {
1012                 ret = -EINVAL;
1013                 goto out;
1014         }
1015         btrfs_end_transaction(trans, root);
1016         file->private_data = NULL;
1017
1018         mutex_lock(&root->fs_info->trans_mutex);
1019         root->fs_info->open_ioctl_trans--;
1020         mutex_unlock(&root->fs_info->trans_mutex);
1021
1022 out:
1023         return ret;
1024 }
1025
1026 long btrfs_ioctl(struct file *file, unsigned int
1027                 cmd, unsigned long arg)
1028 {
1029         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1030
1031         switch (cmd) {
1032         case BTRFS_IOC_SNAP_CREATE:
1033                 return btrfs_ioctl_snap_create(file, (void __user *)arg);
1034         case BTRFS_IOC_DEFRAG:
1035                 return btrfs_ioctl_defrag(file);
1036         case BTRFS_IOC_RESIZE:
1037                 return btrfs_ioctl_resize(root, (void __user *)arg);
1038         case BTRFS_IOC_ADD_DEV:
1039                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
1040         case BTRFS_IOC_RM_DEV:
1041                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
1042         case BTRFS_IOC_BALANCE:
1043                 return btrfs_balance(root->fs_info->dev_root);
1044         case BTRFS_IOC_CLONE:
1045                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1046         case BTRFS_IOC_CLONE_RANGE:
1047                 return btrfs_ioctl_clone_range(file, arg);
1048         case BTRFS_IOC_TRANS_START:
1049                 return btrfs_ioctl_trans_start(file);
1050         case BTRFS_IOC_TRANS_END:
1051                 return btrfs_ioctl_trans_end(file);
1052         case BTRFS_IOC_SYNC:
1053                 btrfs_start_delalloc_inodes(root);
1054                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1055                 return 0;
1056         }
1057
1058         return -ENOTTY;
1059 }