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
 149         /* add the backref first */
 150         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 151                                  objectid, BTRFS_ROOT_BACKREF_KEY,
 152                                  root->root_key.objectid,
 153                                  dir->i_ino, index, name, namelen);
 154
 155         BUG_ON(ret);
 156
 157         /* now add the forward ref */
 158         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 159                                  root->root_key.objectid, BTRFS_ROOT_REF_KEY,
 160                                  objectid,
 161                                  dir->i_ino, index, name, namelen);
 162
 163         BUG_ON(ret);
 164
 165         ret = btrfs_commit_transaction(trans, root);
 166         if (ret)
 167                 goto fail_commit;
 168
 169         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 170         BUG_ON(!new_root);
 171
 172         trans = btrfs_start_transaction(new_root, 1);
 173         BUG_ON(!trans);
 174
 175         ret = btrfs_create_subvol_root(new_root, dentry, trans, new_dirid,
 176                                        BTRFS_I(dir)->block_group);
 177         if (ret)
 178                 goto fail;
 179
 180 fail:
 181         nr = trans->blocks_used;
 182         err = btrfs_commit_transaction(trans, new_root);
 183         if (err && !ret)
 184                 ret = err;
 185 fail_commit:
 186         btrfs_btree_balance_dirty(root, nr);
 187         return ret;
 188 }
 189
 190 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 191                            char *name, int namelen)
 192 {
 193         struct btrfs_pending_snapshot *pending_snapshot;
 194         struct btrfs_trans_handle *trans;
 195         int ret = 0;
 196         int err;
 197         unsigned long nr = 0;
 198
 199         if (!root->ref_cows)
 200                 return -EINVAL;
 201
 202         ret = btrfs_check_free_space(root, 1, 0);
 203         if (ret)
 204                 goto fail_unlock;
 205
 206         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 207         if (!pending_snapshot) {
 208                 ret = -ENOMEM;
 209                 goto fail_unlock;
 210         }
 211         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 212         if (!pending_snapshot->name) {
 213                 ret = -ENOMEM;
 214                 kfree(pending_snapshot);
 215                 goto fail_unlock;
 216         }
 217         memcpy(pending_snapshot->name, name, namelen);
 218         pending_snapshot->name[namelen] = '\0';
 219         pending_snapshot->dentry = dentry;
 220         trans = btrfs_start_transaction(root, 1);
 221         BUG_ON(!trans);
 222         pending_snapshot->root = root;
 223         list_add(&pending_snapshot->list,
 224                  &trans->transaction->pending_snapshots);
 225         err = btrfs_commit_transaction(trans, root);
 226
 227 fail_unlock:
 228         btrfs_btree_balance_dirty(root, nr);
 229         return ret;
 230 }
 231
 232 /* copy of may_create in fs/namei.c() */
 233 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 234 {
 235         if (child->d_inode)
 236                 return -EEXIST;
 237         if (IS_DEADDIR(dir))
 238                 return -ENOENT;
 239         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 240 }
 241
 242 /*
 243  * Create a new subvolume below @parent.  This is largely modeled after
 244  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 245  * inside this filesystem so it's quite a bit simpler.
 246  */
 247 static noinline int btrfs_mksubvol(struct path *parent, char *name,
 248                                    int mode, int namelen,
 249                                    struct btrfs_root *snap_src)
 250 {
 251         struct dentry *dentry;
 252         int error;
 253
 254         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 255
 256         dentry = lookup_one_len(name, parent->dentry, namelen);
 257         error = PTR_ERR(dentry);
 258         if (IS_ERR(dentry))
 259                 goto out_unlock;
 260
 261         error = -EEXIST;
 262         if (dentry->d_inode)
 263                 goto out_dput;
 264
 265         if (!IS_POSIXACL(parent->dentry->d_inode))
 266                 mode &= ~current->fs->umask;
 267
 268         error = mnt_want_write(parent->mnt);
 269         if (error)
 270                 goto out_dput;
 271
 272         error = btrfs_may_create(parent->dentry->d_inode, dentry);
 273         if (error)
 274                 goto out_drop_write;
 275
 276         /*
 277          * Actually perform the low-level subvolume creation after all
 278          * this VFS fuzz.
 279          *
 280          * Eventually we want to pass in an inode under which we create this
 281          * subvolume, but for now all are under the filesystem root.
 282          *
 283          * Also we should pass on the mode eventually to allow creating new
 284          * subvolume with specific mode bits.
 285          */
 286         if (snap_src) {
 287                 error = create_snapshot(snap_src, dentry, name, namelen);
 288         } else {
 289                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
 290                                       dentry, name, namelen);
 291         }
 292         if (error)
 293                 goto out_drop_write;
 294
 295         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 296 out_drop_write:
 297         mnt_drop_write(parent->mnt);
 298 out_dput:
 299         dput(dentry);
 300 out_unlock:
 301         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 302         return error;
 303 }
 304
 305
 306 int btrfs_defrag_file(struct file *file)
 307 {
 308         struct inode *inode = fdentry(file)->d_inode;
 309         struct btrfs_root *root = BTRFS_I(inode)->root;
 310         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 311         struct btrfs_ordered_extent *ordered;
 312         struct page *page;
 313         unsigned long last_index;
 314         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 315         unsigned long total_read = 0;
 316         u64 page_start;
 317         u64 page_end;
 318         unsigned long i;
 319         int ret;
 320
 321         ret = btrfs_check_free_space(root, inode->i_size, 0);
 322         if (ret)
 323                 return -ENOSPC;
 324
 325         mutex_lock(&inode->i_mutex);
 326         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 327         for (i = 0; i <= last_index; i++) {
 328                 if (total_read % ra_pages == 0) {
 329                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 330                                        min(last_index, i + ra_pages - 1));
 331                 }
 332                 total_read++;
 333 again:
 334                 page = grab_cache_page(inode->i_mapping, i);
 335                 if (!page)
 336                         goto out_unlock;
 337                 if (!PageUptodate(page)) {
 338                         btrfs_readpage(NULL, page);
 339                         lock_page(page);
 340                         if (!PageUptodate(page)) {
 341                                 unlock_page(page);
 342                                 page_cache_release(page);
 343                                 goto out_unlock;
 344                         }
 345                 }
 346
 347                 wait_on_page_writeback(page);
 348
 349                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 350                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 351                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 352
 353                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 354                 if (ordered) {
 355                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 356                         unlock_page(page);
 357                         page_cache_release(page);
 358                         btrfs_start_ordered_extent(inode, ordered, 1);
 359                         btrfs_put_ordered_extent(ordered);
 360                         goto again;
 361                 }
 362                 set_page_extent_mapped(page);
 363
 364                 /*
 365                  * this makes sure page_mkwrite is called on the
 366                  * page if it is dirtied again later
 367                  */
 368                 clear_page_dirty_for_io(page);
 369
 370                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 371
 372                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 373                 set_page_dirty(page);
 374                 unlock_page(page);
 375                 page_cache_release(page);
 376                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 377         }
 378
 379 out_unlock:
 380         mutex_unlock(&inode->i_mutex);
 381         return 0;
 382 }
 383
 384 /*
 385  * Called inside transaction, so use GFP_NOFS
 386  */
 387
 388 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 389 {
 390         u64 new_size;
 391         u64 old_size;
 392         u64 devid = 1;
 393         struct btrfs_ioctl_vol_args *vol_args;
 394         struct btrfs_trans_handle *trans;
 395         struct btrfs_device *device = NULL;
 396         char *sizestr;
 397         char *devstr = NULL;
 398         int ret = 0;
 399         int namelen;
 400         int mod = 0;
 401
 402         if (root->fs_info->sb->s_flags & MS_RDONLY)
 403                 return -EROFS;
 404
 405         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 406
 407         if (!vol_args)
 408                 return -ENOMEM;
 409
 410         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 411                 ret = -EFAULT;
 412                 goto out;
 413         }
 414
 415         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 416         namelen = strlen(vol_args->name);
 417
 418         mutex_lock(&root->fs_info->volume_mutex);
 419         sizestr = vol_args->name;
 420         devstr = strchr(sizestr, ':');
 421         if (devstr) {
 422                 char *end;
 423                 sizestr = devstr + 1;
 424                 *devstr = '\0';
 425                 devstr = vol_args->name;
 426                 devid = simple_strtoull(devstr, &end, 10);
 427                 printk(KERN_INFO "resizing devid %llu\n", devid);
 428         }
 429         device = btrfs_find_device(root, devid, NULL, NULL);
 430         if (!device) {
 431                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 432                 ret = -EINVAL;
 433                 goto out_unlock;
 434         }
 435         if (!strcmp(sizestr, "max"))
 436                 new_size = device->bdev->bd_inode->i_size;
 437         else {
 438                 if (sizestr[0] == '-') {
 439                         mod = -1;
 440                         sizestr++;
 441                 } else if (sizestr[0] == '+') {
 442                         mod = 1;
 443                         sizestr++;
 444                 }
 445                 new_size = btrfs_parse_size(sizestr);
 446                 if (new_size == 0) {
 447                         ret = -EINVAL;
 448                         goto out_unlock;
 449                 }
 450         }
 451
 452         old_size = device->total_bytes;
 453
 454         if (mod < 0) {
 455                 if (new_size > old_size) {
 456                         ret = -EINVAL;
 457                         goto out_unlock;
 458                 }
 459                 new_size = old_size - new_size;
 460         } else if (mod > 0) {
 461                 new_size = old_size + new_size;
 462         }
 463
 464         if (new_size < 256 * 1024 * 1024) {
 465                 ret = -EINVAL;
 466                 goto out_unlock;
 467         }
 468         if (new_size > device->bdev->bd_inode->i_size) {
 469                 ret = -EFBIG;
 470                 goto out_unlock;
 471         }
 472
 473         do_div(new_size, root->sectorsize);
 474         new_size *= root->sectorsize;
 475
 476         printk(KERN_INFO "new size for %s is %llu\n",
 477                 device->name, (unsigned long long)new_size);
 478
 479         if (new_size > old_size) {
 480                 trans = btrfs_start_transaction(root, 1);
 481                 ret = btrfs_grow_device(trans, device, new_size);
 482                 btrfs_commit_transaction(trans, root);
 483         } else {
 484                 ret = btrfs_shrink_device(device, new_size);
 485         }
 486
 487 out_unlock:
 488         mutex_unlock(&root->fs_info->volume_mutex);
 489 out:
 490         kfree(vol_args);
 491         return ret;
 492 }
 493
 494 static noinline int btrfs_ioctl_snap_create(struct file *file,
 495                                             void __user *arg, int subvol)
 496 {
 497         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 498         struct btrfs_ioctl_vol_args *vol_args;
 499         struct btrfs_dir_item *di;
 500         struct btrfs_path *path;
 501         struct file *src_file;
 502         u64 root_dirid;
 503         int namelen;
 504         int ret = 0;
 505
 506         if (root->fs_info->sb->s_flags & MS_RDONLY)
 507                 return -EROFS;
 508
 509         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 510
 511         if (!vol_args)
 512                 return -ENOMEM;
 513
 514         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 515                 ret = -EFAULT;
 516                 goto out;
 517         }
 518
 519         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 520         namelen = strlen(vol_args->name);
 521         if (strchr(vol_args->name, '/')) {
 522                 ret = -EINVAL;
 523                 goto out;
 524         }
 525
 526         path = btrfs_alloc_path();
 527         if (!path) {
 528                 ret = -ENOMEM;
 529                 goto out;
 530         }
 531
 532         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 533         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 534                             path, root_dirid,
 535                             vol_args->name, namelen, 0);
 536         btrfs_free_path(path);
 537
 538         if (di && !IS_ERR(di)) {
 539                 ret = -EEXIST;
 540                 goto out;
 541         }
 542
 543         if (IS_ERR(di)) {
 544                 ret = PTR_ERR(di);
 545                 goto out;
 546         }
 547
 548         if (subvol) {
 549                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 550                                      file->f_path.dentry->d_inode->i_mode,
 551                                      namelen, NULL);
 552         } else {
 553                 struct inode *src_inode;
 554                 src_file = fget(vol_args->fd);
 555                 if (!src_file) {
 556                         ret = -EINVAL;
 557                         goto out;
 558                 }
 559
 560                 src_inode = src_file->f_path.dentry->d_inode;
 561                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 562                         printk("btrfs: Snapshot src from another FS\n");
 563                         ret = -EINVAL;
 564                         fput(src_file);
 565                         goto out;
 566                 }
 567                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 568                              file->f_path.dentry->d_inode->i_mode,
 569                              namelen, BTRFS_I(src_inode)->root);
 570                 fput(src_file);
 571         }
 572
 573 out:
 574         kfree(vol_args);
 575         return ret;
 576 }
 577
 578 static int btrfs_ioctl_defrag(struct file *file)
 579 {
 580         struct inode *inode = fdentry(file)->d_inode;
 581         struct btrfs_root *root = BTRFS_I(inode)->root;
 582         int ret;
 583
 584         ret = mnt_want_write(file->f_path.mnt);
 585         if (ret)
 586                 return ret;
 587
 588         switch (inode->i_mode & S_IFMT) {
 589         case S_IFDIR:
 590                 btrfs_defrag_root(root, 0);
 591                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 592                 break;
 593         case S_IFREG:
 594                 btrfs_defrag_file(file);
 595                 break;
 596         }
 597
 598         return 0;
 599 }
 600
 601 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 602 {
 603         struct btrfs_ioctl_vol_args *vol_args;
 604         int ret;
 605
 606         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 607
 608         if (!vol_args)
 609                 return -ENOMEM;
 610
 611         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 612                 ret = -EFAULT;
 613                 goto out;
 614         }
 615         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 616         ret = btrfs_init_new_device(root, vol_args->name);
 617
 618 out:
 619         kfree(vol_args);
 620         return ret;
 621 }
 622
 623 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 624 {
 625         struct btrfs_ioctl_vol_args *vol_args;
 626         int ret;
 627
 628         if (root->fs_info->sb->s_flags & MS_RDONLY)
 629                 return -EROFS;
 630
 631         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 632
 633         if (!vol_args)
 634                 return -ENOMEM;
 635
 636         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 637                 ret = -EFAULT;
 638                 goto out;
 639         }
 640         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 641         ret = btrfs_rm_device(root, vol_args->name);
 642
 643 out:
 644         kfree(vol_args);
 645         return ret;
 646 }
 647
 648 long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, u64 off,
 649                        u64 olen, u64 destoff)
 650 {
 651         struct inode *inode = fdentry(file)->d_inode;
 652         struct btrfs_root *root = BTRFS_I(inode)->root;
 653         struct file *src_file;
 654         struct inode *src;
 655         struct btrfs_trans_handle *trans;
 656         struct btrfs_path *path;
 657         struct extent_buffer *leaf;
 658         char *buf;
 659         struct btrfs_key key;
 660         u32 nritems;
 661         int slot;
 662         int ret;
 663         u64 len = olen;
 664         u64 bs = root->fs_info->sb->s_blocksize;
 665         u64 hint_byte;
 666
 667         /*
 668          * TODO:
 669          * - split compressed inline extents.  annoying: we need to
 670          *   decompress into destination's address_space (the file offset
 671          *   may change, so source mapping won't do), then recompress (or
 672          *   otherwise reinsert) a subrange.
 673          * - allow ranges within the same file to be cloned (provided
 674          *   they don't overlap)?
 675          */
 676
 677         ret = mnt_want_write(file->f_path.mnt);
 678         if (ret)
 679                 return ret;
 680
 681         src_file = fget(srcfd);
 682         if (!src_file)
 683                 return -EBADF;
 684         src = src_file->f_dentry->d_inode;
 685
 686         ret = -EINVAL;
 687         if (src == inode)
 688                 goto out_fput;
 689
 690         ret = -EISDIR;
 691         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 692                 goto out_fput;
 693
 694         ret = -EXDEV;
 695         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 696                 goto out_fput;
 697
 698         ret = -ENOMEM;
 699         buf = vmalloc(btrfs_level_size(root, 0));
 700         if (!buf)
 701                 goto out_fput;
 702
 703         path = btrfs_alloc_path();
 704         if (!path) {
 705                 vfree(buf);
 706                 goto out_fput;
 707         }
 708         path->reada = 2;
 709
 710         if (inode < src) {
 711                 mutex_lock(&inode->i_mutex);
 712                 mutex_lock(&src->i_mutex);
 713         } else {
 714                 mutex_lock(&src->i_mutex);
 715                 mutex_lock(&inode->i_mutex);
 716         }
 717
 718         /* determine range to clone */
 719         ret = -EINVAL;
 720         if (off >= src->i_size || off + len > src->i_size)
 721                 goto out_unlock;
 722         if (len == 0)
 723                 olen = len = src->i_size - off;
 724         /* if we extend to eof, continue to block boundary */
 725         if (off + len == src->i_size)
 726                 len = ((src->i_size + bs-1) & ~(bs-1))
 727                         - off;
 728
 729         /* verify the end result is block aligned */
 730         if ((off & (bs-1)) ||
 731             ((off + len) & (bs-1)))
 732                 goto out_unlock;
 733
 734         printk("final src extent is %llu~%llu\n", off, len);
 735         printk("final dst extent is %llu~%llu\n", destoff, len);
 736
 737         /* do any pending delalloc/csum calc on src, one way or
 738            another, and lock file content */
 739         while (1) {
 740                 struct btrfs_ordered_extent *ordered;
 741                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 742                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
 743                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 744                         break;
 745                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 746                 if (ordered)
 747                         btrfs_put_ordered_extent(ordered);
 748                 btrfs_wait_ordered_range(src, off, off+len);
 749         }
 750
 751         trans = btrfs_start_transaction(root, 1);
 752         BUG_ON(!trans);
 753
 754         /* punch hole in destination first */
 755         btrfs_drop_extents(trans, root, inode, off, off+len, 0, &hint_byte);
 756
 757         /* clone data */
 758         key.objectid = src->i_ino;
 759         key.type = BTRFS_EXTENT_DATA_KEY;
 760         key.offset = 0;
 761
 762         while (1) {
 763                 /*
 764                  * note the key will change type as we walk through the
 765                  * tree.
 766                  */
 767                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 768                 if (ret < 0)
 769                         goto out;
 770
 771                 nritems = btrfs_header_nritems(path->nodes[0]);
 772                 if (path->slots[0] >= nritems) {
 773                         ret = btrfs_next_leaf(root, path);
 774                         if (ret < 0)
 775                                 goto out;
 776                         if (ret > 0)
 777                                 break;
 778                         nritems = btrfs_header_nritems(path->nodes[0]);
 779                 }
 780                 leaf = path->nodes[0];
 781                 slot = path->slots[0];
 782
 783                 btrfs_item_key_to_cpu(leaf, &key, slot);
 784                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
 785                     key.objectid != src->i_ino)
 786                         break;
 787
 788                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 789                         struct btrfs_file_extent_item *extent;
 790                         int type;
 791                         u32 size;
 792                         struct btrfs_key new_key;
 793                         u64 disko = 0, diskl = 0;
 794                         u64 datao = 0, datal = 0;
 795                         u8 comp;
 796
 797                         size = btrfs_item_size_nr(leaf, slot);
 798                         read_extent_buffer(leaf, buf,
 799                                            btrfs_item_ptr_offset(leaf, slot),
 800                                            size);
 801
 802                         extent = btrfs_item_ptr(leaf, slot,
 803                                                 struct btrfs_file_extent_item);
 804                         comp = btrfs_file_extent_compression(leaf, extent);
 805                         type = btrfs_file_extent_type(leaf, extent);
 806                         if (type == BTRFS_FILE_EXTENT_REG) {
 807                                 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
 808                                 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
 809                                 datao = btrfs_file_extent_offset(leaf, extent);
 810                                 datal = btrfs_file_extent_num_bytes(leaf, extent);
 811                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 812                                 /* take upper bound, may be compressed */
 813                                 datal = btrfs_file_extent_ram_bytes(leaf,
 814                                                                     extent);
 815                         }
 816                         btrfs_release_path(root, path);
 817
 818                         if (key.offset + datal < off ||
 819                             key.offset >= off+len)
 820                                 goto next;
 821
 822                         memcpy(&new_key, &key, sizeof(new_key));
 823                         new_key.objectid = inode->i_ino;
 824                         new_key.offset = key.offset + destoff - off;
 825
 826                         if (type == BTRFS_FILE_EXTENT_REG) {
 827                                 ret = btrfs_insert_empty_item(trans, root, path,
 828                                                               &new_key, size);
 829                                 if (ret)
 830                                         goto out;
 831
 832                                 leaf = path->nodes[0];
 833                                 slot = path->slots[0];
 834                                 write_extent_buffer(leaf, buf,
 835                                             btrfs_item_ptr_offset(leaf, slot),
 836                                             size);
 837
 838                                 extent = btrfs_item_ptr(leaf, slot,
 839                                                 struct btrfs_file_extent_item);
 840                                 printk("  orig disk %llu~%llu data %llu~%llu\n",
 841                                        disko, diskl, datao, datal);
 842
 843                                 if (off > key.offset) {
 844                                         datao += off - key.offset;
 845                                         datal -= off - key.offset;
 846                                 }
 847                                 if (key.offset + datao + datal + key.offset >
 848                                     off + len)
 849                                         datal = off + len - key.offset - datao;
 850                                 /* disko == 0 means it's a hole */
 851                                 if (!disko)
 852                                         datao = 0;
 853                                 printk(" final disk %llu~%llu data %llu~%llu\n",
 854                                        disko, diskl, datao, datal);
 855
 856                                 btrfs_set_file_extent_offset(leaf, extent,
 857                                                              datao);
 858                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 859                                                                 datal);
 860                                 if (disko) {
 861                                         inode_add_bytes(inode, datal);
 862                                         ret = btrfs_inc_extent_ref(trans, root,
 863                                                    disko, diskl, leaf->start,
 864                                                    root->root_key.objectid,
 865                                                    trans->transid,
 866                                                    inode->i_ino);
 867                                         BUG_ON(ret);
 868                                 }
 869                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 870                                 u64 skip = 0;
 871                                 u64 trim = 0;
 872                                 if (off > key.offset) {
 873                                         skip = off - key.offset;
 874                                         new_key.offset += skip;
 875                                 }
 876                                 if (key.offset + datal > off+len)
 877                                         trim = key.offset + datal - (off+len);
 878                                 printk("len %lld skip %lld trim %lld\n",
 879                                        datal, skip, trim);
 880                                 if (comp && (skip || trim)) {
 881                                         printk("btrfs clone_range can't split compressed inline extents yet\n");
 882                                         ret = -EINVAL;
 883                                         goto out;
 884                                 }
 885                                 size -= skip + trim;
 886                                 datal -= skip + trim;
 887                                 ret = btrfs_insert_empty_item(trans, root, path,
 888                                                               &new_key, size);
 889                                 if (ret)
 890                                         goto out;
 891
 892                                 if (skip) {
 893                                         u32 start = btrfs_file_extent_calc_inline_size(0);
 894                                         memmove(buf+start, buf+start+skip,
 895                                                 datal);
 896                                 }
 897
 898                                 leaf = path->nodes[0];
 899                                 slot = path->slots[0];
 900                                 write_extent_buffer(leaf, buf,
 901                                             btrfs_item_ptr_offset(leaf, slot),
 902                                             size);
 903                                 inode_add_bytes(inode, datal);
 904                         }
 905
 906                         btrfs_mark_buffer_dirty(leaf);
 907                 }
 908
 909                 if (btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
 910                         u32 size;
 911                         struct btrfs_key new_key;
 912                         u64 coverslen;
 913                         int coff, clen;
 914
 915                         size = btrfs_item_size_nr(leaf, slot);
 916                         coverslen = (size / BTRFS_CRC32_SIZE) <<
 917                                 root->fs_info->sb->s_blocksize_bits;
 918                         printk("csums for %llu~%llu\n",
 919                                key.offset, coverslen);
 920                         if (key.offset + coverslen < off ||
 921                             key.offset >= off+len)
 922                                 goto next;
 923
 924                         read_extent_buffer(leaf, buf,
 925                                            btrfs_item_ptr_offset(leaf, slot),
 926                                            size);
 927                         btrfs_release_path(root, path);
 928
 929                         coff = 0;
 930                         if (off > key.offset)
 931                                 coff = ((off - key.offset) >>
 932                                         root->fs_info->sb->s_blocksize_bits) *
 933                                         BTRFS_CRC32_SIZE;
 934                         clen = size - coff;
 935                         if (key.offset + coverslen > off+len)
 936                                 clen -= ((key.offset+coverslen-off-len) >>
 937                                          root->fs_info->sb->s_blocksize_bits) *
 938                                         BTRFS_CRC32_SIZE;
 939                         printk(" will dup %d~%d of %d\n",
 940                                coff, clen, size);
 941
 942                         memcpy(&new_key, &key, sizeof(new_key));
 943                         new_key.objectid = inode->i_ino;
 944                         new_key.offset = key.offset + destoff - off;
 945
 946                         ret = btrfs_insert_empty_item(trans, root, path,
 947                                                       &new_key, clen);
 948                         if (ret)
 949                                 goto out;
 950
 951                         leaf = path->nodes[0];
 952                         slot = path->slots[0];
 953                         write_extent_buffer(leaf, buf + coff,
 954                                             btrfs_item_ptr_offset(leaf, slot),
 955                                             clen);
 956                         btrfs_mark_buffer_dirty(leaf);
 957                 }
 958
 959         next:
 960                 btrfs_release_path(root, path);
 961                 key.offset++;
 962         }
 963         ret = 0;
 964 out:
 965         btrfs_release_path(root, path);
 966         if (ret == 0) {
 967                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 968                 if (destoff + olen > inode->i_size)
 969                         btrfs_i_size_write(inode, destoff + olen);
 970                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
 971                 ret = btrfs_update_inode(trans, root, inode);
 972         }
 973         btrfs_end_transaction(trans, root);
 974         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 975         if (ret)
 976                 vmtruncate(inode, 0);
 977 out_unlock:
 978         mutex_unlock(&src->i_mutex);
 979         mutex_unlock(&inode->i_mutex);
 980         vfree(buf);
 981         btrfs_free_path(path);
 982 out_fput:
 983         fput(src_file);
 984         return ret;
 985 }
 986
 987 long btrfs_ioctl_clone_range(struct file *file, unsigned long argptr)
 988 {
 989         struct btrfs_ioctl_clone_range_args args;
 990
 991         if (copy_from_user(&args, (void *)argptr, sizeof(args)))
 992                 return -EFAULT;
 993         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
 994                                  args.src_length, args.dest_offset);
 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_start(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         if (!capable(CAP_SYS_ADMIN))
1011                 return -EPERM;
1012
1013         if (file->private_data) {
1014                 ret = -EINPROGRESS;
1015                 goto out;
1016         }
1017
1018         ret = mnt_want_write(file->f_path.mnt);
1019         if (ret)
1020                 goto out;
1021
1022         mutex_lock(&root->fs_info->trans_mutex);
1023         root->fs_info->open_ioctl_trans++;
1024         mutex_unlock(&root->fs_info->trans_mutex);
1025
1026         trans = btrfs_start_ioctl_transaction(root, 0);
1027         if (trans)
1028                 file->private_data = trans;
1029         else
1030                 ret = -ENOMEM;
1031         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1032 out:
1033         return ret;
1034 }
1035
1036 /*
1037  * there are many ways the trans_start and trans_end ioctls can lead
1038  * to deadlocks.  They should only be used by applications that
1039  * basically own the machine, and have a very in depth understanding
1040  * of all the possible deadlocks and enospc problems.
1041  */
1042 long btrfs_ioctl_trans_end(struct file *file)
1043 {
1044         struct inode *inode = fdentry(file)->d_inode;
1045         struct btrfs_root *root = BTRFS_I(inode)->root;
1046         struct btrfs_trans_handle *trans;
1047         int ret = 0;
1048
1049         trans = file->private_data;
1050         if (!trans) {
1051                 ret = -EINVAL;
1052                 goto out;
1053         }
1054         btrfs_end_transaction(trans, root);
1055         file->private_data = NULL;
1056
1057         mutex_lock(&root->fs_info->trans_mutex);
1058         root->fs_info->open_ioctl_trans--;
1059         mutex_unlock(&root->fs_info->trans_mutex);
1060
1061 out:
1062         return ret;
1063 }
1064
1065 long btrfs_ioctl(struct file *file, unsigned int
1066                 cmd, unsigned long arg)
1067 {
1068         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1069
1070         switch (cmd) {
1071         case BTRFS_IOC_SNAP_CREATE:
1072                 return btrfs_ioctl_snap_create(file, (void __user *)arg, 0);
1073         case BTRFS_IOC_SUBVOL_CREATE:
1074                 return btrfs_ioctl_snap_create(file, (void __user *)arg, 1);
1075         case BTRFS_IOC_DEFRAG:
1076                 return btrfs_ioctl_defrag(file);
1077         case BTRFS_IOC_RESIZE:
1078                 return btrfs_ioctl_resize(root, (void __user *)arg);
1079         case BTRFS_IOC_ADD_DEV:
1080                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
1081         case BTRFS_IOC_RM_DEV:
1082                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
1083         case BTRFS_IOC_BALANCE:
1084                 return btrfs_balance(root->fs_info->dev_root);
1085         case BTRFS_IOC_CLONE:
1086                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1087         case BTRFS_IOC_CLONE_RANGE:
1088                 return btrfs_ioctl_clone_range(file, arg);
1089         case BTRFS_IOC_TRANS_START:
1090                 return btrfs_ioctl_trans_start(file);
1091         case BTRFS_IOC_TRANS_END:
1092                 return btrfs_ioctl_trans_end(file);
1093         case BTRFS_IOC_SYNC:
1094                 btrfs_start_delalloc_inodes(root);
1095                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1096                 return 0;
1097         }
1098
1099         return -ENOTTY;
1100 }