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                 struct dentry *dir = dentry->d_parent;
 288                 struct dentry *test = dir->d_parent;
 289                 struct btrfs_path *path = btrfs_alloc_path();
 290                 int ret;
 291                 u64 test_oid;
 292                 u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
 293
 294                 test_oid = snap_src->root_key.objectid;
 295
 296                 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
 297                                           path, parent_oid, test_oid);
 298                 if (ret == 0)
 299                         goto create;
 300                 btrfs_release_path(snap_src->fs_info->tree_root, path);
 301
 302                 /* we need to make sure we aren't creating a directory loop
 303                  * by taking a snapshot of something that has our current
 304                  * subvol in its directory tree.  So, this loops through
 305                  * the dentries and checks the forward refs for each subvolume
 306                  * to see if is references the subvolume where we are
 307                  * placing this new snapshot.
 308                  */
 309                 while(1) {
 310                         if (!test ||
 311                             dir == snap_src->fs_info->sb->s_root ||
 312                             test == snap_src->fs_info->sb->s_root ||
 313                             test->d_inode->i_sb != snap_src->fs_info->sb) {
 314                                 break;
 315                         }
 316                         if (S_ISLNK(test->d_inode->i_mode)) {
 317                                 printk("Symlink in snapshot path, failed\n");
 318                                 error = -EMLINK;
 319                                 btrfs_free_path(path);
 320                                 goto out_drop_write;
 321                         }
 322                         test_oid =
 323                                 BTRFS_I(test->d_inode)->root->root_key.objectid;
 324                         ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
 325                                   path, test_oid, parent_oid);
 326                         if (ret == 0) {
 327                                 printk("Snapshot creation failed, looping\n");
 328                                 error = -EMLINK;
 329                                 btrfs_free_path(path);
 330                                 goto out_drop_write;
 331                         }
 332                         btrfs_release_path(snap_src->fs_info->tree_root, path);
 333                         test = test->d_parent;
 334                 }
 335 create:
 336                 btrfs_free_path(path);
 337                 error = create_snapshot(snap_src, dentry, name, namelen);
 338         } else {
 339                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
 340                                       dentry, name, namelen);
 341         }
 342         if (error)
 343                 goto out_drop_write;
 344
 345         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 346 out_drop_write:
 347         mnt_drop_write(parent->mnt);
 348 out_dput:
 349         dput(dentry);
 350 out_unlock:
 351         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 352         return error;
 353 }
 354
 355
 356 int btrfs_defrag_file(struct file *file)
 357 {
 358         struct inode *inode = fdentry(file)->d_inode;
 359         struct btrfs_root *root = BTRFS_I(inode)->root;
 360         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 361         struct btrfs_ordered_extent *ordered;
 362         struct page *page;
 363         unsigned long last_index;
 364         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 365         unsigned long total_read = 0;
 366         u64 page_start;
 367         u64 page_end;
 368         unsigned long i;
 369         int ret;
 370
 371         ret = btrfs_check_free_space(root, inode->i_size, 0);
 372         if (ret)
 373                 return -ENOSPC;
 374
 375         mutex_lock(&inode->i_mutex);
 376         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 377         for (i = 0; i <= last_index; i++) {
 378                 if (total_read % ra_pages == 0) {
 379                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 380                                        min(last_index, i + ra_pages - 1));
 381                 }
 382                 total_read++;
 383 again:
 384                 page = grab_cache_page(inode->i_mapping, i);
 385                 if (!page)
 386                         goto out_unlock;
 387                 if (!PageUptodate(page)) {
 388                         btrfs_readpage(NULL, page);
 389                         lock_page(page);
 390                         if (!PageUptodate(page)) {
 391                                 unlock_page(page);
 392                                 page_cache_release(page);
 393                                 goto out_unlock;
 394                         }
 395                 }
 396
 397                 wait_on_page_writeback(page);
 398
 399                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 400                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 401                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 402
 403                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 404                 if (ordered) {
 405                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 406                         unlock_page(page);
 407                         page_cache_release(page);
 408                         btrfs_start_ordered_extent(inode, ordered, 1);
 409                         btrfs_put_ordered_extent(ordered);
 410                         goto again;
 411                 }
 412                 set_page_extent_mapped(page);
 413
 414                 /*
 415                  * this makes sure page_mkwrite is called on the
 416                  * page if it is dirtied again later
 417                  */
 418                 clear_page_dirty_for_io(page);
 419
 420                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 421
 422                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 423                 set_page_dirty(page);
 424                 unlock_page(page);
 425                 page_cache_release(page);
 426                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 427         }
 428
 429 out_unlock:
 430         mutex_unlock(&inode->i_mutex);
 431         return 0;
 432 }
 433
 434 /*
 435  * Called inside transaction, so use GFP_NOFS
 436  */
 437
 438 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 439 {
 440         u64 new_size;
 441         u64 old_size;
 442         u64 devid = 1;
 443         struct btrfs_ioctl_vol_args *vol_args;
 444         struct btrfs_trans_handle *trans;
 445         struct btrfs_device *device = NULL;
 446         char *sizestr;
 447         char *devstr = NULL;
 448         int ret = 0;
 449         int namelen;
 450         int mod = 0;
 451
 452         if (root->fs_info->sb->s_flags & MS_RDONLY)
 453                 return -EROFS;
 454
 455         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 456
 457         if (!vol_args)
 458                 return -ENOMEM;
 459
 460         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 461                 ret = -EFAULT;
 462                 goto out;
 463         }
 464
 465         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 466         namelen = strlen(vol_args->name);
 467
 468         mutex_lock(&root->fs_info->volume_mutex);
 469         sizestr = vol_args->name;
 470         devstr = strchr(sizestr, ':');
 471         if (devstr) {
 472                 char *end;
 473                 sizestr = devstr + 1;
 474                 *devstr = '\0';
 475                 devstr = vol_args->name;
 476                 devid = simple_strtoull(devstr, &end, 10);
 477                 printk(KERN_INFO "resizing devid %llu\n", devid);
 478         }
 479         device = btrfs_find_device(root, devid, NULL, NULL);
 480         if (!device) {
 481                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 482                 ret = -EINVAL;
 483                 goto out_unlock;
 484         }
 485         if (!strcmp(sizestr, "max"))
 486                 new_size = device->bdev->bd_inode->i_size;
 487         else {
 488                 if (sizestr[0] == '-') {
 489                         mod = -1;
 490                         sizestr++;
 491                 } else if (sizestr[0] == '+') {
 492                         mod = 1;
 493                         sizestr++;
 494                 }
 495                 new_size = btrfs_parse_size(sizestr);
 496                 if (new_size == 0) {
 497                         ret = -EINVAL;
 498                         goto out_unlock;
 499                 }
 500         }
 501
 502         old_size = device->total_bytes;
 503
 504         if (mod < 0) {
 505                 if (new_size > old_size) {
 506                         ret = -EINVAL;
 507                         goto out_unlock;
 508                 }
 509                 new_size = old_size - new_size;
 510         } else if (mod > 0) {
 511                 new_size = old_size + new_size;
 512         }
 513
 514         if (new_size < 256 * 1024 * 1024) {
 515                 ret = -EINVAL;
 516                 goto out_unlock;
 517         }
 518         if (new_size > device->bdev->bd_inode->i_size) {
 519                 ret = -EFBIG;
 520                 goto out_unlock;
 521         }
 522
 523         do_div(new_size, root->sectorsize);
 524         new_size *= root->sectorsize;
 525
 526         printk(KERN_INFO "new size for %s is %llu\n",
 527                 device->name, (unsigned long long)new_size);
 528
 529         if (new_size > old_size) {
 530                 trans = btrfs_start_transaction(root, 1);
 531                 ret = btrfs_grow_device(trans, device, new_size);
 532                 btrfs_commit_transaction(trans, root);
 533         } else {
 534                 ret = btrfs_shrink_device(device, new_size);
 535         }
 536
 537 out_unlock:
 538         mutex_unlock(&root->fs_info->volume_mutex);
 539 out:
 540         kfree(vol_args);
 541         return ret;
 542 }
 543
 544 static noinline int btrfs_ioctl_snap_create(struct file *file,
 545                                             void __user *arg, int subvol)
 546 {
 547         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 548         struct btrfs_ioctl_vol_args *vol_args;
 549         struct btrfs_dir_item *di;
 550         struct btrfs_path *path;
 551         struct file *src_file;
 552         u64 root_dirid;
 553         int namelen;
 554         int ret = 0;
 555
 556         if (root->fs_info->sb->s_flags & MS_RDONLY)
 557                 return -EROFS;
 558
 559         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 560
 561         if (!vol_args)
 562                 return -ENOMEM;
 563
 564         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 565                 ret = -EFAULT;
 566                 goto out;
 567         }
 568
 569         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 570         namelen = strlen(vol_args->name);
 571         if (strchr(vol_args->name, '/')) {
 572                 ret = -EINVAL;
 573                 goto out;
 574         }
 575
 576         path = btrfs_alloc_path();
 577         if (!path) {
 578                 ret = -ENOMEM;
 579                 goto out;
 580         }
 581
 582         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 583         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 584                             path, root_dirid,
 585                             vol_args->name, namelen, 0);
 586         btrfs_free_path(path);
 587
 588         if (di && !IS_ERR(di)) {
 589                 ret = -EEXIST;
 590                 goto out;
 591         }
 592
 593         if (IS_ERR(di)) {
 594                 ret = PTR_ERR(di);
 595                 goto out;
 596         }
 597
 598         if (subvol) {
 599                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 600                                      file->f_path.dentry->d_inode->i_mode,
 601                                      namelen, NULL);
 602         } else {
 603                 struct inode *src_inode;
 604                 src_file = fget(vol_args->fd);
 605                 if (!src_file) {
 606                         ret = -EINVAL;
 607                         goto out;
 608                 }
 609
 610                 src_inode = src_file->f_path.dentry->d_inode;
 611                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 612                         printk("btrfs: Snapshot src from another FS\n");
 613                         ret = -EINVAL;
 614                         fput(src_file);
 615                         goto out;
 616                 }
 617                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 618                              file->f_path.dentry->d_inode->i_mode,
 619                              namelen, BTRFS_I(src_inode)->root);
 620                 fput(src_file);
 621         }
 622
 623 out:
 624         kfree(vol_args);
 625         return ret;
 626 }
 627
 628 static int btrfs_ioctl_defrag(struct file *file)
 629 {
 630         struct inode *inode = fdentry(file)->d_inode;
 631         struct btrfs_root *root = BTRFS_I(inode)->root;
 632         int ret;
 633
 634         ret = mnt_want_write(file->f_path.mnt);
 635         if (ret)
 636                 return ret;
 637
 638         switch (inode->i_mode & S_IFMT) {
 639         case S_IFDIR:
 640                 btrfs_defrag_root(root, 0);
 641                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 642                 break;
 643         case S_IFREG:
 644                 btrfs_defrag_file(file);
 645                 break;
 646         }
 647
 648         return 0;
 649 }
 650
 651 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 652 {
 653         struct btrfs_ioctl_vol_args *vol_args;
 654         int ret;
 655
 656         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 657
 658         if (!vol_args)
 659                 return -ENOMEM;
 660
 661         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 662                 ret = -EFAULT;
 663                 goto out;
 664         }
 665         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 666         ret = btrfs_init_new_device(root, vol_args->name);
 667
 668 out:
 669         kfree(vol_args);
 670         return ret;
 671 }
 672
 673 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 674 {
 675         struct btrfs_ioctl_vol_args *vol_args;
 676         int ret;
 677
 678         if (root->fs_info->sb->s_flags & MS_RDONLY)
 679                 return -EROFS;
 680
 681         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 682
 683         if (!vol_args)
 684                 return -ENOMEM;
 685
 686         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 687                 ret = -EFAULT;
 688                 goto out;
 689         }
 690         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 691         ret = btrfs_rm_device(root, vol_args->name);
 692
 693 out:
 694         kfree(vol_args);
 695         return ret;
 696 }
 697
 698 long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, u64 off,
 699                        u64 olen, u64 destoff)
 700 {
 701         struct inode *inode = fdentry(file)->d_inode;
 702         struct btrfs_root *root = BTRFS_I(inode)->root;
 703         struct file *src_file;
 704         struct inode *src;
 705         struct btrfs_trans_handle *trans;
 706         struct btrfs_path *path;
 707         struct extent_buffer *leaf;
 708         char *buf;
 709         struct btrfs_key key;
 710         u32 nritems;
 711         int slot;
 712         int ret;
 713         u64 len = olen;
 714         u64 bs = root->fs_info->sb->s_blocksize;
 715         u64 hint_byte;
 716
 717         /*
 718          * TODO:
 719          * - split compressed inline extents.  annoying: we need to
 720          *   decompress into destination's address_space (the file offset
 721          *   may change, so source mapping won't do), then recompress (or
 722          *   otherwise reinsert) a subrange.
 723          * - allow ranges within the same file to be cloned (provided
 724          *   they don't overlap)?
 725          */
 726
 727         ret = mnt_want_write(file->f_path.mnt);
 728         if (ret)
 729                 return ret;
 730
 731         src_file = fget(srcfd);
 732         if (!src_file)
 733                 return -EBADF;
 734         src = src_file->f_dentry->d_inode;
 735
 736         ret = -EINVAL;
 737         if (src == inode)
 738                 goto out_fput;
 739
 740         ret = -EISDIR;
 741         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 742                 goto out_fput;
 743
 744         ret = -EXDEV;
 745         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 746                 goto out_fput;
 747
 748         ret = -ENOMEM;
 749         buf = vmalloc(btrfs_level_size(root, 0));
 750         if (!buf)
 751                 goto out_fput;
 752
 753         path = btrfs_alloc_path();
 754         if (!path) {
 755                 vfree(buf);
 756                 goto out_fput;
 757         }
 758         path->reada = 2;
 759
 760         if (inode < src) {
 761                 mutex_lock(&inode->i_mutex);
 762                 mutex_lock(&src->i_mutex);
 763         } else {
 764                 mutex_lock(&src->i_mutex);
 765                 mutex_lock(&inode->i_mutex);
 766         }
 767
 768         /* determine range to clone */
 769         ret = -EINVAL;
 770         if (off >= src->i_size || off + len > src->i_size)
 771                 goto out_unlock;
 772         if (len == 0)
 773                 olen = len = src->i_size - off;
 774         /* if we extend to eof, continue to block boundary */
 775         if (off + len == src->i_size)
 776                 len = ((src->i_size + bs-1) & ~(bs-1))
 777                         - off;
 778
 779         /* verify the end result is block aligned */
 780         if ((off & (bs-1)) ||
 781             ((off + len) & (bs-1)))
 782                 goto out_unlock;
 783
 784         printk("final src extent is %llu~%llu\n", off, len);
 785         printk("final dst extent is %llu~%llu\n", destoff, len);
 786
 787         /* do any pending delalloc/csum calc on src, one way or
 788            another, and lock file content */
 789         while (1) {
 790                 struct btrfs_ordered_extent *ordered;
 791                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 792                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
 793                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 794                         break;
 795                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 796                 if (ordered)
 797                         btrfs_put_ordered_extent(ordered);
 798                 btrfs_wait_ordered_range(src, off, off+len);
 799         }
 800
 801         trans = btrfs_start_transaction(root, 1);
 802         BUG_ON(!trans);
 803
 804         /* punch hole in destination first */
 805         btrfs_drop_extents(trans, root, inode, off, off+len, 0, &hint_byte);
 806
 807         /* clone data */
 808         key.objectid = src->i_ino;
 809         key.type = BTRFS_EXTENT_DATA_KEY;
 810         key.offset = 0;
 811
 812         while (1) {
 813                 /*
 814                  * note the key will change type as we walk through the
 815                  * tree.
 816                  */
 817                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 818                 if (ret < 0)
 819                         goto out;
 820
 821                 nritems = btrfs_header_nritems(path->nodes[0]);
 822                 if (path->slots[0] >= nritems) {
 823                         ret = btrfs_next_leaf(root, path);
 824                         if (ret < 0)
 825                                 goto out;
 826                         if (ret > 0)
 827                                 break;
 828                         nritems = btrfs_header_nritems(path->nodes[0]);
 829                 }
 830                 leaf = path->nodes[0];
 831                 slot = path->slots[0];
 832
 833                 btrfs_item_key_to_cpu(leaf, &key, slot);
 834                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
 835                     key.objectid != src->i_ino)
 836                         break;
 837
 838                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 839                         struct btrfs_file_extent_item *extent;
 840                         int type;
 841                         u32 size;
 842                         struct btrfs_key new_key;
 843                         u64 disko = 0, diskl = 0;
 844                         u64 datao = 0, datal = 0;
 845                         u8 comp;
 846
 847                         size = btrfs_item_size_nr(leaf, slot);
 848                         read_extent_buffer(leaf, buf,
 849                                            btrfs_item_ptr_offset(leaf, slot),
 850                                            size);
 851
 852                         extent = btrfs_item_ptr(leaf, slot,
 853                                                 struct btrfs_file_extent_item);
 854                         comp = btrfs_file_extent_compression(leaf, extent);
 855                         type = btrfs_file_extent_type(leaf, extent);
 856                         if (type == BTRFS_FILE_EXTENT_REG) {
 857                                 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
 858                                 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
 859                                 datao = btrfs_file_extent_offset(leaf, extent);
 860                                 datal = btrfs_file_extent_num_bytes(leaf, extent);
 861                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 862                                 /* take upper bound, may be compressed */
 863                                 datal = btrfs_file_extent_ram_bytes(leaf,
 864                                                                     extent);
 865                         }
 866                         btrfs_release_path(root, path);
 867
 868                         if (key.offset + datal < off ||
 869                             key.offset >= off+len)
 870                                 goto next;
 871
 872                         memcpy(&new_key, &key, sizeof(new_key));
 873                         new_key.objectid = inode->i_ino;
 874                         new_key.offset = key.offset + destoff - off;
 875
 876                         if (type == BTRFS_FILE_EXTENT_REG) {
 877                                 ret = btrfs_insert_empty_item(trans, root, path,
 878                                                               &new_key, size);
 879                                 if (ret)
 880                                         goto out;
 881
 882                                 leaf = path->nodes[0];
 883                                 slot = path->slots[0];
 884                                 write_extent_buffer(leaf, buf,
 885                                             btrfs_item_ptr_offset(leaf, slot),
 886                                             size);
 887
 888                                 extent = btrfs_item_ptr(leaf, slot,
 889                                                 struct btrfs_file_extent_item);
 890                                 printk("  orig disk %llu~%llu data %llu~%llu\n",
 891                                        disko, diskl, datao, datal);
 892
 893                                 if (off > key.offset) {
 894                                         datao += off - key.offset;
 895                                         datal -= off - key.offset;
 896                                 }
 897                                 if (key.offset + datao + datal + key.offset >
 898                                     off + len)
 899                                         datal = off + len - key.offset - datao;
 900                                 /* disko == 0 means it's a hole */
 901                                 if (!disko)
 902                                         datao = 0;
 903                                 printk(" final disk %llu~%llu data %llu~%llu\n",
 904                                        disko, diskl, datao, datal);
 905
 906                                 btrfs_set_file_extent_offset(leaf, extent,
 907                                                              datao);
 908                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 909                                                                 datal);
 910                                 if (disko) {
 911                                         inode_add_bytes(inode, datal);
 912                                         ret = btrfs_inc_extent_ref(trans, root,
 913                                                    disko, diskl, leaf->start,
 914                                                    root->root_key.objectid,
 915                                                    trans->transid,
 916                                                    inode->i_ino);
 917                                         BUG_ON(ret);
 918                                 }
 919                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 920                                 u64 skip = 0;
 921                                 u64 trim = 0;
 922                                 if (off > key.offset) {
 923                                         skip = off - key.offset;
 924                                         new_key.offset += skip;
 925                                 }
 926                                 if (key.offset + datal > off+len)
 927                                         trim = key.offset + datal - (off+len);
 928                                 printk("len %lld skip %lld trim %lld\n",
 929                                        datal, skip, trim);
 930                                 if (comp && (skip || trim)) {
 931                                         printk("btrfs clone_range can't split compressed inline extents yet\n");
 932                                         ret = -EINVAL;
 933                                         goto out;
 934                                 }
 935                                 size -= skip + trim;
 936                                 datal -= skip + trim;
 937                                 ret = btrfs_insert_empty_item(trans, root, path,
 938                                                               &new_key, size);
 939                                 if (ret)
 940                                         goto out;
 941
 942                                 if (skip) {
 943                                         u32 start = btrfs_file_extent_calc_inline_size(0);
 944                                         memmove(buf+start, buf+start+skip,
 945                                                 datal);
 946                                 }
 947
 948                                 leaf = path->nodes[0];
 949                                 slot = path->slots[0];
 950                                 write_extent_buffer(leaf, buf,
 951                                             btrfs_item_ptr_offset(leaf, slot),
 952                                             size);
 953                                 inode_add_bytes(inode, datal);
 954                         }
 955
 956                         btrfs_mark_buffer_dirty(leaf);
 957                 }
 958
 959                 if (btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
 960                         u32 size;
 961                         struct btrfs_key new_key;
 962                         u64 coverslen;
 963                         int coff, clen;
 964
 965                         size = btrfs_item_size_nr(leaf, slot);
 966                         coverslen = (size / BTRFS_CRC32_SIZE) <<
 967                                 root->fs_info->sb->s_blocksize_bits;
 968                         printk("csums for %llu~%llu\n",
 969                                key.offset, coverslen);
 970                         if (key.offset + coverslen < off ||
 971                             key.offset >= off+len)
 972                                 goto next;
 973
 974                         read_extent_buffer(leaf, buf,
 975                                            btrfs_item_ptr_offset(leaf, slot),
 976                                            size);
 977                         btrfs_release_path(root, path);
 978
 979                         coff = 0;
 980                         if (off > key.offset)
 981                                 coff = ((off - key.offset) >>
 982                                         root->fs_info->sb->s_blocksize_bits) *
 983                                         BTRFS_CRC32_SIZE;
 984                         clen = size - coff;
 985                         if (key.offset + coverslen > off+len)
 986                                 clen -= ((key.offset+coverslen-off-len) >>
 987                                          root->fs_info->sb->s_blocksize_bits) *
 988                                         BTRFS_CRC32_SIZE;
 989                         printk(" will dup %d~%d of %d\n",
 990                                coff, clen, size);
 991
 992                         memcpy(&new_key, &key, sizeof(new_key));
 993                         new_key.objectid = inode->i_ino;
 994                         new_key.offset = key.offset + destoff - off;
 995
 996                         ret = btrfs_insert_empty_item(trans, root, path,
 997                                                       &new_key, clen);
 998                         if (ret)
 999                                 goto out;
1000
1001                         leaf = path->nodes[0];
1002                         slot = path->slots[0];
1003                         write_extent_buffer(leaf, buf + coff,
1004                                             btrfs_item_ptr_offset(leaf, slot),
1005                                             clen);
1006                         btrfs_mark_buffer_dirty(leaf);
1007                 }
1008
1009         next:
1010                 btrfs_release_path(root, path);
1011                 key.offset++;
1012         }
1013         ret = 0;
1014 out:
1015         btrfs_release_path(root, path);
1016         if (ret == 0) {
1017                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1018                 if (destoff + olen > inode->i_size)
1019                         btrfs_i_size_write(inode, destoff + olen);
1020                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1021                 ret = btrfs_update_inode(trans, root, inode);
1022         }
1023         btrfs_end_transaction(trans, root);
1024         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1025         if (ret)
1026                 vmtruncate(inode, 0);
1027 out_unlock:
1028         mutex_unlock(&src->i_mutex);
1029         mutex_unlock(&inode->i_mutex);
1030         vfree(buf);
1031         btrfs_free_path(path);
1032 out_fput:
1033         fput(src_file);
1034         return ret;
1035 }
1036
1037 long btrfs_ioctl_clone_range(struct file *file, unsigned long argptr)
1038 {
1039         struct btrfs_ioctl_clone_range_args args;
1040
1041         if (copy_from_user(&args, (void *)argptr, sizeof(args)))
1042                 return -EFAULT;
1043         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1044                                  args.src_length, args.dest_offset);
1045 }
1046
1047 /*
1048  * there are many ways the trans_start and trans_end ioctls can lead
1049  * to deadlocks.  They should only be used by applications that
1050  * basically own the machine, and have a very in depth understanding
1051  * of all the possible deadlocks and enospc problems.
1052  */
1053 long btrfs_ioctl_trans_start(struct file *file)
1054 {
1055         struct inode *inode = fdentry(file)->d_inode;
1056         struct btrfs_root *root = BTRFS_I(inode)->root;
1057         struct btrfs_trans_handle *trans;
1058         int ret = 0;
1059
1060         if (!capable(CAP_SYS_ADMIN))
1061                 return -EPERM;
1062
1063         if (file->private_data) {
1064                 ret = -EINPROGRESS;
1065                 goto out;
1066         }
1067
1068         ret = mnt_want_write(file->f_path.mnt);
1069         if (ret)
1070                 goto out;
1071
1072         mutex_lock(&root->fs_info->trans_mutex);
1073         root->fs_info->open_ioctl_trans++;
1074         mutex_unlock(&root->fs_info->trans_mutex);
1075
1076         trans = btrfs_start_ioctl_transaction(root, 0);
1077         if (trans)
1078                 file->private_data = trans;
1079         else
1080                 ret = -ENOMEM;
1081         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1082 out:
1083         return ret;
1084 }
1085
1086 /*
1087  * there are many ways the trans_start and trans_end ioctls can lead
1088  * to deadlocks.  They should only be used by applications that
1089  * basically own the machine, and have a very in depth understanding
1090  * of all the possible deadlocks and enospc problems.
1091  */
1092 long btrfs_ioctl_trans_end(struct file *file)
1093 {
1094         struct inode *inode = fdentry(file)->d_inode;
1095         struct btrfs_root *root = BTRFS_I(inode)->root;
1096         struct btrfs_trans_handle *trans;
1097         int ret = 0;
1098
1099         trans = file->private_data;
1100         if (!trans) {
1101                 ret = -EINVAL;
1102                 goto out;
1103         }
1104         btrfs_end_transaction(trans, root);
1105         file->private_data = NULL;
1106
1107         mutex_lock(&root->fs_info->trans_mutex);
1108         root->fs_info->open_ioctl_trans--;
1109         mutex_unlock(&root->fs_info->trans_mutex);
1110
1111 out:
1112         return ret;
1113 }
1114
1115 long btrfs_ioctl(struct file *file, unsigned int
1116                 cmd, unsigned long arg)
1117 {
1118         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1119
1120         switch (cmd) {
1121         case BTRFS_IOC_SNAP_CREATE:
1122                 return btrfs_ioctl_snap_create(file, (void __user *)arg, 0);
1123         case BTRFS_IOC_SUBVOL_CREATE:
1124                 return btrfs_ioctl_snap_create(file, (void __user *)arg, 1);
1125         case BTRFS_IOC_DEFRAG:
1126                 return btrfs_ioctl_defrag(file);
1127         case BTRFS_IOC_RESIZE:
1128                 return btrfs_ioctl_resize(root, (void __user *)arg);
1129         case BTRFS_IOC_ADD_DEV:
1130                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
1131         case BTRFS_IOC_RM_DEV:
1132                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
1133         case BTRFS_IOC_BALANCE:
1134                 return btrfs_balance(root->fs_info->dev_root);
1135         case BTRFS_IOC_CLONE:
1136                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1137         case BTRFS_IOC_CLONE_RANGE:
1138                 return btrfs_ioctl_clone_range(file, arg);
1139         case BTRFS_IOC_TRANS_START:
1140                 return btrfs_ioctl_trans_start(file);
1141         case BTRFS_IOC_TRANS_END:
1142                 return btrfs_ioctl_trans_end(file);
1143         case BTRFS_IOC_SYNC:
1144                 btrfs_start_delalloc_inodes(root);
1145                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1146                 return 0;
1147         }
1148
1149         return -ENOTTY;
1150 }