SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/mount.h>
  33 #include <linux/mpage.h>
  34 #include <linux/namei.h>
  35 #include <linux/swap.h>
  36 #include <linux/writeback.h>
  37 #include <linux/statfs.h>
  38 #include <linux/compat.h>
  39 #include <linux/bit_spinlock.h>
  40 #include <linux/security.h>
  41 #include <linux/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 printk("inserting root objectid %Lu\n", objectid);
 131         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 132                                 &root_item);
 133         if (ret)
 134                 goto fail;
 135
 136         /*
 137          * insert the directory item
 138          */
 139         key.offset = (u64)-1;
 140         dir = dentry->d_parent->d_inode;
 141         ret = btrfs_set_inode_index(dir, &index);
 142         BUG_ON(ret);
 143
 144         ret = btrfs_insert_dir_item(trans, root,
 145                                     name, namelen, dir->i_ino, &key,
 146                                     BTRFS_FT_DIR, index);
 147         if (ret)
 148                 goto fail;
 149 #if 0
 150         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 151                              name, namelen, objectid,
 152                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
 153         if (ret)
 154                 goto fail;
 155 #endif
 156         ret = btrfs_commit_transaction(trans, root);
 157         if (ret)
 158                 goto fail_commit;
 159
 160         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 161         BUG_ON(!new_root);
 162
 163         trans = btrfs_start_transaction(new_root, 1);
 164         BUG_ON(!trans);
 165
 166         ret = btrfs_create_subvol_root(new_root, dentry, trans, new_dirid,
 167                                        BTRFS_I(dir)->block_group);
 168         if (ret)
 169                 goto fail;
 170
 171 fail:
 172         nr = trans->blocks_used;
 173         err = btrfs_commit_transaction(trans, new_root);
 174         if (err && !ret)
 175                 ret = err;
 176 fail_commit:
 177         btrfs_btree_balance_dirty(root, nr);
 178 printk("all done ret %d\n", ret);
 179         return ret;
 180 }
 181
 182 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 183                            char *name, int namelen)
 184 {
 185         struct btrfs_pending_snapshot *pending_snapshot;
 186         struct btrfs_trans_handle *trans;
 187         int ret = 0;
 188         int err;
 189         unsigned long nr = 0;
 190
 191         if (!root->ref_cows)
 192                 return -EINVAL;
 193
 194         ret = btrfs_check_free_space(root, 1, 0);
 195         if (ret)
 196                 goto fail_unlock;
 197
 198         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 199         if (!pending_snapshot) {
 200                 ret = -ENOMEM;
 201                 goto fail_unlock;
 202         }
 203         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 204         if (!pending_snapshot->name) {
 205                 ret = -ENOMEM;
 206                 kfree(pending_snapshot);
 207                 goto fail_unlock;
 208         }
 209         memcpy(pending_snapshot->name, name, namelen);
 210         pending_snapshot->name[namelen] = '\0';
 211         pending_snapshot->dentry = dentry;
 212         trans = btrfs_start_transaction(root, 1);
 213         BUG_ON(!trans);
 214         pending_snapshot->root = root;
 215         list_add(&pending_snapshot->list,
 216                  &trans->transaction->pending_snapshots);
 217         err = btrfs_commit_transaction(trans, root);
 218
 219 fail_unlock:
 220         btrfs_btree_balance_dirty(root, nr);
 221         return ret;
 222 }
 223
 224 /* copy of may_create in fs/namei.c() */
 225 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 226 {
 227         if (child->d_inode)
 228                 return -EEXIST;
 229         if (IS_DEADDIR(dir))
 230                 return -ENOENT;
 231         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 232 }
 233
 234 /*
 235  * Create a new subvolume below @parent.  This is largely modeled after
 236  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 237  * inside this filesystem so it's quite a bit simpler.
 238  */
 239 static noinline int btrfs_mksubvol(struct path *parent, char *name,
 240                                    int mode, int namelen,
 241                                    struct btrfs_root *snap_src)
 242 {
 243         struct dentry *dentry;
 244         int error;
 245
 246         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 247
 248         dentry = lookup_one_len(name, parent->dentry, namelen);
 249         error = PTR_ERR(dentry);
 250         if (IS_ERR(dentry))
 251                 goto out_unlock;
 252
 253         error = -EEXIST;
 254         if (dentry->d_inode)
 255                 goto out_dput;
 256
 257         if (!IS_POSIXACL(parent->dentry->d_inode))
 258                 mode &= ~current->fs->umask;
 259
 260         error = mnt_want_write(parent->mnt);
 261         if (error)
 262                 goto out_dput;
 263
 264         error = btrfs_may_create(parent->dentry->d_inode, dentry);
 265         if (error)
 266                 goto out_drop_write;
 267
 268         /*
 269          * Actually perform the low-level subvolume creation after all
 270          * this VFS fuzz.
 271          *
 272          * Eventually we want to pass in an inode under which we create this
 273          * subvolume, but for now all are under the filesystem root.
 274          *
 275          * Also we should pass on the mode eventually to allow creating new
 276          * subvolume with specific mode bits.
 277          */
 278         if (snap_src) {
 279                 error = create_snapshot(snap_src, dentry, name, namelen);
 280         } else {
 281                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
 282                                       dentry, name, namelen);
 283         }
 284         if (error)
 285                 goto out_drop_write;
 286
 287         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 288 out_drop_write:
 289         mnt_drop_write(parent->mnt);
 290 out_dput:
 291         dput(dentry);
 292 out_unlock:
 293         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 294         return error;
 295 }
 296
 297
 298 int btrfs_defrag_file(struct file *file)
 299 {
 300         struct inode *inode = fdentry(file)->d_inode;
 301         struct btrfs_root *root = BTRFS_I(inode)->root;
 302         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 303         struct btrfs_ordered_extent *ordered;
 304         struct page *page;
 305         unsigned long last_index;
 306         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 307         unsigned long total_read = 0;
 308         u64 page_start;
 309         u64 page_end;
 310         unsigned long i;
 311         int ret;
 312
 313         ret = btrfs_check_free_space(root, inode->i_size, 0);
 314         if (ret)
 315                 return -ENOSPC;
 316
 317         mutex_lock(&inode->i_mutex);
 318         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 319         for (i = 0; i <= last_index; i++) {
 320                 if (total_read % ra_pages == 0) {
 321                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 322                                        min(last_index, i + ra_pages - 1));
 323                 }
 324                 total_read++;
 325 again:
 326                 page = grab_cache_page(inode->i_mapping, i);
 327                 if (!page)
 328                         goto out_unlock;
 329                 if (!PageUptodate(page)) {
 330                         btrfs_readpage(NULL, page);
 331                         lock_page(page);
 332                         if (!PageUptodate(page)) {
 333                                 unlock_page(page);
 334                                 page_cache_release(page);
 335                                 goto out_unlock;
 336                         }
 337                 }
 338
 339                 wait_on_page_writeback(page);
 340
 341                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 342                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 343                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 344
 345                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 346                 if (ordered) {
 347                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 348                         unlock_page(page);
 349                         page_cache_release(page);
 350                         btrfs_start_ordered_extent(inode, ordered, 1);
 351                         btrfs_put_ordered_extent(ordered);
 352                         goto again;
 353                 }
 354                 set_page_extent_mapped(page);
 355
 356                 /*
 357                  * this makes sure page_mkwrite is called on the
 358                  * page if it is dirtied again later
 359                  */
 360                 clear_page_dirty_for_io(page);
 361
 362                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 363
 364                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 365                 set_page_dirty(page);
 366                 unlock_page(page);
 367                 page_cache_release(page);
 368                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 369         }
 370
 371 out_unlock:
 372         mutex_unlock(&inode->i_mutex);
 373         return 0;
 374 }
 375
 376 /*
 377  * Called inside transaction, so use GFP_NOFS
 378  */
 379
 380 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 381 {
 382         u64 new_size;
 383         u64 old_size;
 384         u64 devid = 1;
 385         struct btrfs_ioctl_vol_args *vol_args;
 386         struct btrfs_trans_handle *trans;
 387         struct btrfs_device *device = NULL;
 388         char *sizestr;
 389         char *devstr = NULL;
 390         int ret = 0;
 391         int namelen;
 392         int mod = 0;
 393
 394         if (root->fs_info->sb->s_flags & MS_RDONLY)
 395                 return -EROFS;
 396
 397         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 398
 399         if (!vol_args)
 400                 return -ENOMEM;
 401
 402         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 403                 ret = -EFAULT;
 404                 goto out;
 405         }
 406
 407         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 408         namelen = strlen(vol_args->name);
 409
 410         mutex_lock(&root->fs_info->volume_mutex);
 411         sizestr = vol_args->name;
 412         devstr = strchr(sizestr, ':');
 413         if (devstr) {
 414                 char *end;
 415                 sizestr = devstr + 1;
 416                 *devstr = '\0';
 417                 devstr = vol_args->name;
 418                 devid = simple_strtoull(devstr, &end, 10);
 419                 printk(KERN_INFO "resizing devid %llu\n", devid);
 420         }
 421         device = btrfs_find_device(root, devid, NULL, NULL);
 422         if (!device) {
 423                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 424                 ret = -EINVAL;
 425                 goto out_unlock;
 426         }
 427         if (!strcmp(sizestr, "max"))
 428                 new_size = device->bdev->bd_inode->i_size;
 429         else {
 430                 if (sizestr[0] == '-') {
 431                         mod = -1;
 432                         sizestr++;
 433                 } else if (sizestr[0] == '+') {
 434                         mod = 1;
 435                         sizestr++;
 436                 }
 437                 new_size = btrfs_parse_size(sizestr);
 438                 if (new_size == 0) {
 439                         ret = -EINVAL;
 440                         goto out_unlock;
 441                 }
 442         }
 443
 444         old_size = device->total_bytes;
 445
 446         if (mod < 0) {
 447                 if (new_size > old_size) {
 448                         ret = -EINVAL;
 449                         goto out_unlock;
 450                 }
 451                 new_size = old_size - new_size;
 452         } else if (mod > 0) {
 453                 new_size = old_size + new_size;
 454         }
 455
 456         if (new_size < 256 * 1024 * 1024) {
 457                 ret = -EINVAL;
 458                 goto out_unlock;
 459         }
 460         if (new_size > device->bdev->bd_inode->i_size) {
 461                 ret = -EFBIG;
 462                 goto out_unlock;
 463         }
 464
 465         do_div(new_size, root->sectorsize);
 466         new_size *= root->sectorsize;
 467
 468         printk(KERN_INFO "new size for %s is %llu\n",
 469                 device->name, (unsigned long long)new_size);
 470
 471         if (new_size > old_size) {
 472                 trans = btrfs_start_transaction(root, 1);
 473                 ret = btrfs_grow_device(trans, device, new_size);
 474                 btrfs_commit_transaction(trans, root);
 475         } else {
 476                 ret = btrfs_shrink_device(device, new_size);
 477         }
 478
 479 out_unlock:
 480         mutex_unlock(&root->fs_info->volume_mutex);
 481 out:
 482         kfree(vol_args);
 483         return ret;
 484 }
 485
 486 static noinline int btrfs_ioctl_snap_create(struct file *file,
 487                                             void __user *arg, int subvol)
 488 {
 489         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 490         struct btrfs_ioctl_vol_args *vol_args;
 491         struct btrfs_dir_item *di;
 492         struct btrfs_path *path;
 493         struct file *src_file;
 494         u64 root_dirid;
 495         int namelen;
 496         int ret = 0;
 497
 498         if (root->fs_info->sb->s_flags & MS_RDONLY)
 499                 return -EROFS;
 500
 501         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 502
 503         if (!vol_args)
 504                 return -ENOMEM;
 505
 506         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 507                 ret = -EFAULT;
 508                 goto out;
 509         }
 510
 511         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 512         namelen = strlen(vol_args->name);
 513         if (strchr(vol_args->name, '/')) {
 514                 ret = -EINVAL;
 515                 goto out;
 516         }
 517
 518         path = btrfs_alloc_path();
 519         if (!path) {
 520                 ret = -ENOMEM;
 521                 goto out;
 522         }
 523
 524         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 525         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 526                             path, root_dirid,
 527                             vol_args->name, namelen, 0);
 528         btrfs_free_path(path);
 529
 530         if (di && !IS_ERR(di)) {
 531                 ret = -EEXIST;
 532                 goto out;
 533         }
 534
 535         if (IS_ERR(di)) {
 536                 ret = PTR_ERR(di);
 537                 goto out;
 538         }
 539
 540         if (subvol) {
 541                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 542                                      file->f_path.dentry->d_inode->i_mode,
 543                                      namelen, NULL);
 544         } else {
 545                 struct inode *src_inode;
 546                 src_file = fget(vol_args->fd);
 547                 if (!src_file) {
 548                         ret = -EINVAL;
 549                         goto out;
 550                 }
 551
 552                 src_inode = src_file->f_path.dentry->d_inode;
 553                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 554                         printk("btrfs: Snapshot src from another FS\n");
 555                         ret = -EINVAL;
 556                         fput(src_file);
 557                         goto out;
 558                 }
 559                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 560                              file->f_path.dentry->d_inode->i_mode,
 561                              namelen, BTRFS_I(src_inode)->root);
 562                 fput(src_file);
 563         }
 564
 565 out:
 566         kfree(vol_args);
 567         return ret;
 568 }
 569
 570 static int btrfs_ioctl_defrag(struct file *file)
 571 {
 572         struct inode *inode = fdentry(file)->d_inode;
 573         struct btrfs_root *root = BTRFS_I(inode)->root;
 574         int ret;
 575
 576         ret = mnt_want_write(file->f_path.mnt);
 577         if (ret)
 578                 return ret;
 579
 580         switch (inode->i_mode & S_IFMT) {
 581         case S_IFDIR:
 582                 btrfs_defrag_root(root, 0);
 583                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 584                 break;
 585         case S_IFREG:
 586                 btrfs_defrag_file(file);
 587                 break;
 588         }
 589
 590         return 0;
 591 }
 592
 593 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 594 {
 595         struct btrfs_ioctl_vol_args *vol_args;
 596         int ret;
 597
 598         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 599
 600         if (!vol_args)
 601                 return -ENOMEM;
 602
 603         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 604                 ret = -EFAULT;
 605                 goto out;
 606         }
 607         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 608         ret = btrfs_init_new_device(root, vol_args->name);
 609
 610 out:
 611         kfree(vol_args);
 612         return ret;
 613 }
 614
 615 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 616 {
 617         struct btrfs_ioctl_vol_args *vol_args;
 618         int ret;
 619
 620         if (root->fs_info->sb->s_flags & MS_RDONLY)
 621                 return -EROFS;
 622
 623         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 624
 625         if (!vol_args)
 626                 return -ENOMEM;
 627
 628         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 629                 ret = -EFAULT;
 630                 goto out;
 631         }
 632         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 633         ret = btrfs_rm_device(root, vol_args->name);
 634
 635 out:
 636         kfree(vol_args);
 637         return ret;
 638 }
 639
 640 long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, u64 off,
 641                        u64 olen, u64 destoff)
 642 {
 643         struct inode *inode = fdentry(file)->d_inode;
 644         struct btrfs_root *root = BTRFS_I(inode)->root;
 645         struct file *src_file;
 646         struct inode *src;
 647         struct btrfs_trans_handle *trans;
 648         struct btrfs_path *path;
 649         struct extent_buffer *leaf;
 650         char *buf;
 651         struct btrfs_key key;
 652         u32 nritems;
 653         int slot;
 654         int ret;
 655         u64 len = olen;
 656         u64 bs = root->fs_info->sb->s_blocksize;
 657         u64 hint_byte;
 658
 659         /*
 660          * TODO:
 661          * - split compressed inline extents.  annoying: we need to
 662          *   decompress into destination's address_space (the file offset
 663          *   may change, so source mapping won't do), then recompress (or
 664          *   otherwise reinsert) a subrange.
 665          * - allow ranges within the same file to be cloned (provided
 666          *   they don't overlap)?
 667          */
 668
 669         ret = mnt_want_write(file->f_path.mnt);
 670         if (ret)
 671                 return ret;
 672
 673         src_file = fget(srcfd);
 674         if (!src_file)
 675                 return -EBADF;
 676         src = src_file->f_dentry->d_inode;
 677
 678         ret = -EINVAL;
 679         if (src == inode)
 680                 goto out_fput;
 681
 682         ret = -EISDIR;
 683         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 684                 goto out_fput;
 685
 686         ret = -EXDEV;
 687         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 688                 goto out_fput;
 689
 690         ret = -ENOMEM;
 691         buf = vmalloc(btrfs_level_size(root, 0));
 692         if (!buf)
 693                 goto out_fput;
 694
 695         path = btrfs_alloc_path();
 696         if (!path) {
 697                 vfree(buf);
 698                 goto out_fput;
 699         }
 700         path->reada = 2;
 701
 702         if (inode < src) {
 703                 mutex_lock(&inode->i_mutex);
 704                 mutex_lock(&src->i_mutex);
 705         } else {
 706                 mutex_lock(&src->i_mutex);
 707                 mutex_lock(&inode->i_mutex);
 708         }
 709
 710         /* determine range to clone */
 711         ret = -EINVAL;
 712         if (off >= src->i_size || off + len > src->i_size)
 713                 goto out_unlock;
 714         if (len == 0)
 715                 olen = len = src->i_size - off;
 716         /* if we extend to eof, continue to block boundary */
 717         if (off + len == src->i_size)
 718                 len = ((src->i_size + bs-1) & ~(bs-1))
 719                         - off;
 720
 721         /* verify the end result is block aligned */
 722         if ((off & (bs-1)) ||
 723             ((off + len) & (bs-1)))
 724                 goto out_unlock;
 725
 726         printk("final src extent is %llu~%llu\n", off, len);
 727         printk("final dst extent is %llu~%llu\n", destoff, len);
 728
 729         /* do any pending delalloc/csum calc on src, one way or
 730            another, and lock file content */
 731         while (1) {
 732                 struct btrfs_ordered_extent *ordered;
 733                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 734                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
 735                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 736                         break;
 737                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 738                 if (ordered)
 739                         btrfs_put_ordered_extent(ordered);
 740                 btrfs_wait_ordered_range(src, off, off+len);
 741         }
 742
 743         trans = btrfs_start_transaction(root, 1);
 744         BUG_ON(!trans);
 745
 746         /* punch hole in destination first */
 747         btrfs_drop_extents(trans, root, inode, off, off+len, 0, &hint_byte);
 748
 749         /* clone data */
 750         key.objectid = src->i_ino;
 751         key.type = BTRFS_EXTENT_DATA_KEY;
 752         key.offset = 0;
 753
 754         while (1) {
 755                 /*
 756                  * note the key will change type as we walk through the
 757                  * tree.
 758                  */
 759                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 760                 if (ret < 0)
 761                         goto out;
 762
 763                 nritems = btrfs_header_nritems(path->nodes[0]);
 764                 if (path->slots[0] >= nritems) {
 765                         ret = btrfs_next_leaf(root, path);
 766                         if (ret < 0)
 767                                 goto out;
 768                         if (ret > 0)
 769                                 break;
 770                         nritems = btrfs_header_nritems(path->nodes[0]);
 771                 }
 772                 leaf = path->nodes[0];
 773                 slot = path->slots[0];
 774
 775                 btrfs_item_key_to_cpu(leaf, &key, slot);
 776                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
 777                     key.objectid != src->i_ino)
 778                         break;
 779
 780                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 781                         struct btrfs_file_extent_item *extent;
 782                         int type;
 783                         u32 size;
 784                         struct btrfs_key new_key;
 785                         u64 disko = 0, diskl = 0;
 786                         u64 datao = 0, datal = 0;
 787                         u8 comp;
 788
 789                         size = btrfs_item_size_nr(leaf, slot);
 790                         read_extent_buffer(leaf, buf,
 791                                            btrfs_item_ptr_offset(leaf, slot),
 792                                            size);
 793
 794                         extent = btrfs_item_ptr(leaf, slot,
 795                                                 struct btrfs_file_extent_item);
 796                         comp = btrfs_file_extent_compression(leaf, extent);
 797                         type = btrfs_file_extent_type(leaf, extent);
 798                         if (type == BTRFS_FILE_EXTENT_REG) {
 799                                 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
 800                                 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
 801                                 datao = btrfs_file_extent_offset(leaf, extent);
 802                                 datal = btrfs_file_extent_num_bytes(leaf, extent);
 803                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 804                                 /* take upper bound, may be compressed */
 805                                 datal = btrfs_file_extent_ram_bytes(leaf,
 806                                                                     extent);
 807                         }
 808                         btrfs_release_path(root, path);
 809
 810                         if (key.offset + datal < off ||
 811                             key.offset >= off+len)
 812                                 goto next;
 813
 814                         memcpy(&new_key, &key, sizeof(new_key));
 815                         new_key.objectid = inode->i_ino;
 816                         new_key.offset = key.offset + destoff - off;
 817
 818                         if (type == BTRFS_FILE_EXTENT_REG) {
 819                                 ret = btrfs_insert_empty_item(trans, root, path,
 820                                                               &new_key, size);
 821                                 if (ret)
 822                                         goto out;
 823
 824                                 leaf = path->nodes[0];
 825                                 slot = path->slots[0];
 826                                 write_extent_buffer(leaf, buf,
 827                                             btrfs_item_ptr_offset(leaf, slot),
 828                                             size);
 829
 830                                 extent = btrfs_item_ptr(leaf, slot,
 831                                                 struct btrfs_file_extent_item);
 832                                 printk("  orig disk %llu~%llu data %llu~%llu\n",
 833                                        disko, diskl, datao, datal);
 834
 835                                 if (off > key.offset) {
 836                                         datao += off - key.offset;
 837                                         datal -= off - key.offset;
 838                                 }
 839                                 if (key.offset + datao + datal + key.offset >
 840                                     off + len)
 841                                         datal = off + len - key.offset - datao;
 842                                 /* disko == 0 means it's a hole */
 843                                 if (!disko)
 844                                         datao = 0;
 845                                 printk(" final disk %llu~%llu data %llu~%llu\n",
 846                                        disko, diskl, datao, datal);
 847
 848                                 btrfs_set_file_extent_offset(leaf, extent,
 849                                                              datao);
 850                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 851                                                                 datal);
 852                                 if (disko) {
 853                                         inode_add_bytes(inode, datal);
 854                                         ret = btrfs_inc_extent_ref(trans, root,
 855                                                    disko, diskl, leaf->start,
 856                                                    root->root_key.objectid,
 857                                                    trans->transid,
 858                                                    inode->i_ino);
 859                                         BUG_ON(ret);
 860                                 }
 861                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 862                                 u64 skip = 0;
 863                                 u64 trim = 0;
 864                                 if (off > key.offset) {
 865                                         skip = off - key.offset;
 866                                         new_key.offset += skip;
 867                                 }
 868                                 if (key.offset + datal > off+len)
 869                                         trim = key.offset + datal - (off+len);
 870                                 printk("len %lld skip %lld trim %lld\n",
 871                                        datal, skip, trim);
 872                                 if (comp && (skip || trim)) {
 873                                         printk("btrfs clone_range can't split compressed inline extents yet\n");
 874                                         ret = -EINVAL;
 875                                         goto out;
 876                                 }
 877                                 size -= skip + trim;
 878                                 datal -= skip + trim;
 879                                 ret = btrfs_insert_empty_item(trans, root, path,
 880                                                               &new_key, size);
 881                                 if (ret)
 882                                         goto out;
 883
 884                                 if (skip) {
 885                                         u32 start = btrfs_file_extent_calc_inline_size(0);
 886                                         memmove(buf+start, buf+start+skip,
 887                                                 datal);
 888                                 }
 889
 890                                 leaf = path->nodes[0];
 891                                 slot = path->slots[0];
 892                                 write_extent_buffer(leaf, buf,
 893                                             btrfs_item_ptr_offset(leaf, slot),
 894                                             size);
 895                                 inode_add_bytes(inode, datal);
 896                         }
 897
 898                         btrfs_mark_buffer_dirty(leaf);
 899                 }
 900
 901                 if (btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
 902                         u32 size;
 903                         struct btrfs_key new_key;
 904                         u64 coverslen;
 905                         int coff, clen;
 906
 907                         size = btrfs_item_size_nr(leaf, slot);
 908                         coverslen = (size / BTRFS_CRC32_SIZE) <<
 909                                 root->fs_info->sb->s_blocksize_bits;
 910                         printk("csums for %llu~%llu\n",
 911                                key.offset, coverslen);
 912                         if (key.offset + coverslen < off ||
 913                             key.offset >= off+len)
 914                                 goto next;
 915
 916                         read_extent_buffer(leaf, buf,
 917                                            btrfs_item_ptr_offset(leaf, slot),
 918                                            size);
 919                         btrfs_release_path(root, path);
 920
 921                         coff = 0;
 922                         if (off > key.offset)
 923                                 coff = ((off - key.offset) >>
 924                                         root->fs_info->sb->s_blocksize_bits) *
 925                                         BTRFS_CRC32_SIZE;
 926                         clen = size - coff;
 927                         if (key.offset + coverslen > off+len)
 928                                 clen -= ((key.offset+coverslen-off-len) >>
 929                                          root->fs_info->sb->s_blocksize_bits) *
 930                                         BTRFS_CRC32_SIZE;
 931                         printk(" will dup %d~%d of %d\n",
 932                                coff, clen, size);
 933
 934                         memcpy(&new_key, &key, sizeof(new_key));
 935                         new_key.objectid = inode->i_ino;
 936                         new_key.offset = key.offset + destoff - off;
 937
 938                         ret = btrfs_insert_empty_item(trans, root, path,
 939                                                       &new_key, clen);
 940                         if (ret)
 941                                 goto out;
 942
 943                         leaf = path->nodes[0];
 944                         slot = path->slots[0];
 945                         write_extent_buffer(leaf, buf + coff,
 946                                             btrfs_item_ptr_offset(leaf, slot),
 947                                             clen);
 948                         btrfs_mark_buffer_dirty(leaf);
 949                 }
 950
 951         next:
 952                 btrfs_release_path(root, path);
 953                 key.offset++;
 954         }
 955         ret = 0;
 956 out:
 957         btrfs_release_path(root, path);
 958         if (ret == 0) {
 959                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 960                 if (destoff + olen > inode->i_size)
 961                         btrfs_i_size_write(inode, destoff + olen);
 962                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
 963                 ret = btrfs_update_inode(trans, root, inode);
 964         }
 965         btrfs_end_transaction(trans, root);
 966         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 967         if (ret)
 968                 vmtruncate(inode, 0);
 969 out_unlock:
 970         mutex_unlock(&src->i_mutex);
 971         mutex_unlock(&inode->i_mutex);
 972         vfree(buf);
 973         btrfs_free_path(path);
 974 out_fput:
 975         fput(src_file);
 976         return ret;
 977 }
 978
 979 long btrfs_ioctl_clone_range(struct file *file, unsigned long argptr)
 980 {
 981         struct btrfs_ioctl_clone_range_args args;
 982
 983         if (copy_from_user(&args, (void *)argptr, sizeof(args)))
 984                 return -EFAULT;
 985         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
 986                                  args.src_length, args.dest_offset);
 987 }
 988
 989 /*
 990  * there are many ways the trans_start and trans_end ioctls can lead
 991  * to deadlocks.  They should only be used by applications that
 992  * basically own the machine, and have a very in depth understanding
 993  * of all the possible deadlocks and enospc problems.
 994  */
 995 long btrfs_ioctl_trans_start(struct file *file)
 996 {
 997         struct inode *inode = fdentry(file)->d_inode;
 998         struct btrfs_root *root = BTRFS_I(inode)->root;
 999         struct btrfs_trans_handle *trans;
1000         int ret = 0;
1001
1002         if (!capable(CAP_SYS_ADMIN))
1003                 return -EPERM;
1004
1005         if (file->private_data) {
1006                 ret = -EINPROGRESS;
1007                 goto out;
1008         }
1009
1010         ret = mnt_want_write(file->f_path.mnt);
1011         if (ret)
1012                 goto out;
1013
1014         mutex_lock(&root->fs_info->trans_mutex);
1015         root->fs_info->open_ioctl_trans++;
1016         mutex_unlock(&root->fs_info->trans_mutex);
1017
1018         trans = btrfs_start_ioctl_transaction(root, 0);
1019         if (trans)
1020                 file->private_data = trans;
1021         else
1022                 ret = -ENOMEM;
1023         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1024 out:
1025         return ret;
1026 }
1027
1028 /*
1029  * there are many ways the trans_start and trans_end ioctls can lead
1030  * to deadlocks.  They should only be used by applications that
1031  * basically own the machine, and have a very in depth understanding
1032  * of all the possible deadlocks and enospc problems.
1033  */
1034 long btrfs_ioctl_trans_end(struct file *file)
1035 {
1036         struct inode *inode = fdentry(file)->d_inode;
1037         struct btrfs_root *root = BTRFS_I(inode)->root;
1038         struct btrfs_trans_handle *trans;
1039         int ret = 0;
1040
1041         trans = file->private_data;
1042         if (!trans) {
1043                 ret = -EINVAL;
1044                 goto out;
1045         }
1046         btrfs_end_transaction(trans, root);
1047         file->private_data = NULL;
1048
1049         mutex_lock(&root->fs_info->trans_mutex);
1050         root->fs_info->open_ioctl_trans--;
1051         mutex_unlock(&root->fs_info->trans_mutex);
1052
1053 out:
1054         return ret;
1055 }
1056
1057 long btrfs_ioctl(struct file *file, unsigned int
1058                 cmd, unsigned long arg)
1059 {
1060         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1061
1062         switch (cmd) {
1063         case BTRFS_IOC_SNAP_CREATE:
1064                 return btrfs_ioctl_snap_create(file, (void __user *)arg, 0);
1065         case BTRFS_IOC_SUBVOL_CREATE:
1066                 return btrfs_ioctl_snap_create(file, (void __user *)arg, 1);
1067         case BTRFS_IOC_DEFRAG:
1068                 return btrfs_ioctl_defrag(file);
1069         case BTRFS_IOC_RESIZE:
1070                 return btrfs_ioctl_resize(root, (void __user *)arg);
1071         case BTRFS_IOC_ADD_DEV:
1072                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
1073         case BTRFS_IOC_RM_DEV:
1074                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
1075         case BTRFS_IOC_BALANCE:
1076                 return btrfs_balance(root->fs_info->dev_root);
1077         case BTRFS_IOC_CLONE:
1078                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1079         case BTRFS_IOC_CLONE_RANGE:
1080                 return btrfs_ioctl_clone_range(file, arg);
1081         case BTRFS_IOC_TRANS_START:
1082                 return btrfs_ioctl_trans_start(file);
1083         case BTRFS_IOC_TRANS_END:
1084                 return btrfs_ioctl_trans_end(file);
1085         case BTRFS_IOC_SYNC:
1086                 btrfs_start_delalloc_inodes(root);
1087                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1088                 return 0;
1089         }
1090
1091         return -ENOTTY;
1092 }