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 "compat.h"
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "ioctl.h"
  50 #include "print-tree.h"
  51 #include "volumes.h"
  52 #include "locking.h"
  53
  54
  55
  56 static noinline int create_subvol(struct btrfs_root *root,
  57                                   struct dentry *dentry,
  58                                   char *name, int namelen)
  59 {
  60         struct btrfs_trans_handle *trans;
  61         struct btrfs_key key;
  62         struct btrfs_root_item root_item;
  63         struct btrfs_inode_item *inode_item;
  64         struct extent_buffer *leaf;
  65         struct btrfs_root *new_root = root;
  66         struct inode *dir;
  67         int ret;
  68         int err;
  69         u64 objectid;
  70         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
  71         u64 index = 0;
  72         unsigned long nr = 1;
  73
  74         ret = btrfs_check_free_space(root, 1, 0);
  75         if (ret)
  76                 goto fail_commit;
  77
  78         trans = btrfs_start_transaction(root, 1);
  79         BUG_ON(!trans);
  80
  81         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
  82                                        0, &objectid);
  83         if (ret)
  84                 goto fail;
  85
  86         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
  87                                       objectid, trans->transid, 0, 0, 0);
  88         if (IS_ERR(leaf)) {
  89                 ret = PTR_ERR(leaf);
  90                 goto fail;
  91         }
  92
  93         btrfs_set_header_nritems(leaf, 0);
  94         btrfs_set_header_level(leaf, 0);
  95         btrfs_set_header_bytenr(leaf, leaf->start);
  96         btrfs_set_header_generation(leaf, trans->transid);
  97         btrfs_set_header_owner(leaf, objectid);
  98
  99         write_extent_buffer(leaf, root->fs_info->fsid,
 100                             (unsigned long)btrfs_header_fsid(leaf),
 101                             BTRFS_FSID_SIZE);
 102         btrfs_mark_buffer_dirty(leaf);
 103
 104         inode_item = &root_item.inode;
 105         memset(inode_item, 0, sizeof(*inode_item));
 106         inode_item->generation = cpu_to_le64(1);
 107         inode_item->size = cpu_to_le64(3);
 108         inode_item->nlink = cpu_to_le32(1);
 109         inode_item->nbytes = cpu_to_le64(root->leafsize);
 110         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 111
 112         btrfs_set_root_bytenr(&root_item, leaf->start);
 113         btrfs_set_root_generation(&root_item, trans->transid);
 114         btrfs_set_root_level(&root_item, 0);
 115         btrfs_set_root_refs(&root_item, 1);
 116         btrfs_set_root_used(&root_item, 0);
 117         btrfs_set_root_last_snapshot(&root_item, 0);
 118
 119         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 120         root_item.drop_level = 0;
 121
 122         btrfs_tree_unlock(leaf);
 123         free_extent_buffer(leaf);
 124         leaf = NULL;
 125
 126         btrfs_set_root_dirid(&root_item, new_dirid);
 127
 128         key.objectid = objectid;
 129         key.offset = 1;
 130         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 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
 150         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 151         ret = btrfs_update_inode(trans, root, dir);
 152         BUG_ON(ret);
 153
 154         /* add the backref first */
 155         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 156                                  objectid, BTRFS_ROOT_BACKREF_KEY,
 157                                  root->root_key.objectid,
 158                                  dir->i_ino, index, name, namelen);
 159
 160         BUG_ON(ret);
 161
 162         /* now add the forward ref */
 163         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 164                                  root->root_key.objectid, BTRFS_ROOT_REF_KEY,
 165                                  objectid,
 166                                  dir->i_ino, index, name, namelen);
 167
 168         BUG_ON(ret);
 169
 170         ret = btrfs_commit_transaction(trans, root);
 171         if (ret)
 172                 goto fail_commit;
 173
 174         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 175         BUG_ON(!new_root);
 176
 177         trans = btrfs_start_transaction(new_root, 1);
 178         BUG_ON(!trans);
 179
 180         ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
 181                                        BTRFS_I(dir)->block_group);
 182         if (ret)
 183                 goto fail;
 184
 185 fail:
 186         nr = trans->blocks_used;
 187         err = btrfs_commit_transaction(trans, new_root);
 188         if (err && !ret)
 189                 ret = err;
 190 fail_commit:
 191         btrfs_btree_balance_dirty(root, nr);
 192         return ret;
 193 }
 194
 195 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 196                            char *name, int namelen)
 197 {
 198         struct btrfs_pending_snapshot *pending_snapshot;
 199         struct btrfs_trans_handle *trans;
 200         int ret = 0;
 201         int err;
 202         unsigned long nr = 0;
 203
 204         if (!root->ref_cows)
 205                 return -EINVAL;
 206
 207         ret = btrfs_check_free_space(root, 1, 0);
 208         if (ret)
 209                 goto fail_unlock;
 210
 211         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 212         if (!pending_snapshot) {
 213                 ret = -ENOMEM;
 214                 goto fail_unlock;
 215         }
 216         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
 217         if (!pending_snapshot->name) {
 218                 ret = -ENOMEM;
 219                 kfree(pending_snapshot);
 220                 goto fail_unlock;
 221         }
 222         memcpy(pending_snapshot->name, name, namelen);
 223         pending_snapshot->name[namelen] = '\0';
 224         pending_snapshot->dentry = dentry;
 225         trans = btrfs_start_transaction(root, 1);
 226         BUG_ON(!trans);
 227         pending_snapshot->root = root;
 228         list_add(&pending_snapshot->list,
 229                  &trans->transaction->pending_snapshots);
 230         err = btrfs_commit_transaction(trans, root);
 231
 232 fail_unlock:
 233         btrfs_btree_balance_dirty(root, nr);
 234         return ret;
 235 }
 236
 237 /* copy of may_create in fs/namei.c() */
 238 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 239 {
 240         if (child->d_inode)
 241                 return -EEXIST;
 242         if (IS_DEADDIR(dir))
 243                 return -ENOENT;
 244         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 245 }
 246
 247 /*
 248  * Create a new subvolume below @parent.  This is largely modeled after
 249  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 250  * inside this filesystem so it's quite a bit simpler.
 251  */
 252 static noinline int btrfs_mksubvol(struct path *parent, char *name,
 253                                    int mode, int namelen,
 254                                    struct btrfs_root *snap_src)
 255 {
 256         struct dentry *dentry;
 257         int error;
 258
 259         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 260
 261         dentry = lookup_one_len(name, parent->dentry, namelen);
 262         error = PTR_ERR(dentry);
 263         if (IS_ERR(dentry))
 264                 goto out_unlock;
 265
 266         error = -EEXIST;
 267         if (dentry->d_inode)
 268                 goto out_dput;
 269
 270         if (!IS_POSIXACL(parent->dentry->d_inode))
 271                 mode &= ~current->fs->umask;
 272
 273         error = mnt_want_write(parent->mnt);
 274         if (error)
 275                 goto out_dput;
 276
 277         error = btrfs_may_create(parent->dentry->d_inode, dentry);
 278         if (error)
 279                 goto out_drop_write;
 280
 281         /*
 282          * Actually perform the low-level subvolume creation after all
 283          * this VFS fuzz.
 284          *
 285          * Eventually we want to pass in an inode under which we create this
 286          * subvolume, but for now all are under the filesystem root.
 287          *
 288          * Also we should pass on the mode eventually to allow creating new
 289          * subvolume with specific mode bits.
 290          */
 291         if (snap_src) {
 292                 struct dentry *dir = dentry->d_parent;
 293                 struct dentry *test = dir->d_parent;
 294                 struct btrfs_path *path = btrfs_alloc_path();
 295                 int ret;
 296                 u64 test_oid;
 297                 u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
 298
 299                 test_oid = snap_src->root_key.objectid;
 300
 301                 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
 302                                           path, parent_oid, test_oid);
 303                 if (ret == 0)
 304                         goto create;
 305                 btrfs_release_path(snap_src->fs_info->tree_root, path);
 306
 307                 /* we need to make sure we aren't creating a directory loop
 308                  * by taking a snapshot of something that has our current
 309                  * subvol in its directory tree.  So, this loops through
 310                  * the dentries and checks the forward refs for each subvolume
 311                  * to see if is references the subvolume where we are
 312                  * placing this new snapshot.
 313                  */
 314                 while(1) {
 315                         if (!test ||
 316                             dir == snap_src->fs_info->sb->s_root ||
 317                             test == snap_src->fs_info->sb->s_root ||
 318                             test->d_inode->i_sb != snap_src->fs_info->sb) {
 319                                 break;
 320                         }
 321                         if (S_ISLNK(test->d_inode->i_mode)) {
 322                                 printk("Symlink in snapshot path, failed\n");
 323                                 error = -EMLINK;
 324                                 btrfs_free_path(path);
 325                                 goto out_drop_write;
 326                         }
 327                         test_oid =
 328                                 BTRFS_I(test->d_inode)->root->root_key.objectid;
 329                         ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
 330                                   path, test_oid, parent_oid);
 331                         if (ret == 0) {
 332                                 printk("Snapshot creation failed, looping\n");
 333                                 error = -EMLINK;
 334                                 btrfs_free_path(path);
 335                                 goto out_drop_write;
 336                         }
 337                         btrfs_release_path(snap_src->fs_info->tree_root, path);
 338                         test = test->d_parent;
 339                 }
 340 create:
 341                 btrfs_free_path(path);
 342                 error = create_snapshot(snap_src, dentry, name, namelen);
 343         } else {
 344                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
 345                                       dentry, name, namelen);
 346         }
 347         if (error)
 348                 goto out_drop_write;
 349
 350         fsnotify_mkdir(parent->dentry->d_inode, dentry);
 351 out_drop_write:
 352         mnt_drop_write(parent->mnt);
 353 out_dput:
 354         dput(dentry);
 355 out_unlock:
 356         mutex_unlock(&parent->dentry->d_inode->i_mutex);
 357         return error;
 358 }
 359
 360
 361 static int btrfs_defrag_file(struct file *file)
 362 {
 363         struct inode *inode = fdentry(file)->d_inode;
 364         struct btrfs_root *root = BTRFS_I(inode)->root;
 365         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 366         struct btrfs_ordered_extent *ordered;
 367         struct page *page;
 368         unsigned long last_index;
 369         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
 370         unsigned long total_read = 0;
 371         u64 page_start;
 372         u64 page_end;
 373         unsigned long i;
 374         int ret;
 375
 376         ret = btrfs_check_free_space(root, inode->i_size, 0);
 377         if (ret)
 378                 return -ENOSPC;
 379
 380         mutex_lock(&inode->i_mutex);
 381         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
 382         for (i = 0; i <= last_index; i++) {
 383                 if (total_read % ra_pages == 0) {
 384                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
 385                                        min(last_index, i + ra_pages - 1));
 386                 }
 387                 total_read++;
 388 again:
 389                 page = grab_cache_page(inode->i_mapping, i);
 390                 if (!page)
 391                         goto out_unlock;
 392                 if (!PageUptodate(page)) {
 393                         btrfs_readpage(NULL, page);
 394                         lock_page(page);
 395                         if (!PageUptodate(page)) {
 396                                 unlock_page(page);
 397                                 page_cache_release(page);
 398                                 goto out_unlock;
 399                         }
 400                 }
 401
 402                 wait_on_page_writeback(page);
 403
 404                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 405                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 406                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 407
 408                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
 409                 if (ordered) {
 410                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 411                         unlock_page(page);
 412                         page_cache_release(page);
 413                         btrfs_start_ordered_extent(inode, ordered, 1);
 414                         btrfs_put_ordered_extent(ordered);
 415                         goto again;
 416                 }
 417                 set_page_extent_mapped(page);
 418
 419                 /*
 420                  * this makes sure page_mkwrite is called on the
 421                  * page if it is dirtied again later
 422                  */
 423                 clear_page_dirty_for_io(page);
 424
 425                 btrfs_set_extent_delalloc(inode, page_start, page_end);
 426
 427                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 428                 set_page_dirty(page);
 429                 unlock_page(page);
 430                 page_cache_release(page);
 431                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
 432         }
 433
 434 out_unlock:
 435         mutex_unlock(&inode->i_mutex);
 436         return 0;
 437 }
 438
 439 /*
 440  * Called inside transaction, so use GFP_NOFS
 441  */
 442
 443 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
 444 {
 445         u64 new_size;
 446         u64 old_size;
 447         u64 devid = 1;
 448         struct btrfs_ioctl_vol_args *vol_args;
 449         struct btrfs_trans_handle *trans;
 450         struct btrfs_device *device = NULL;
 451         char *sizestr;
 452         char *devstr = NULL;
 453         int ret = 0;
 454         int namelen;
 455         int mod = 0;
 456
 457         if (root->fs_info->sb->s_flags & MS_RDONLY)
 458                 return -EROFS;
 459
 460         if (!capable(CAP_SYS_ADMIN))
 461                 return -EPERM;
 462
 463         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 464
 465         if (!vol_args)
 466                 return -ENOMEM;
 467
 468         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 469                 ret = -EFAULT;
 470                 goto out;
 471         }
 472
 473         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 474         namelen = strlen(vol_args->name);
 475
 476         mutex_lock(&root->fs_info->volume_mutex);
 477         sizestr = vol_args->name;
 478         devstr = strchr(sizestr, ':');
 479         if (devstr) {
 480                 char *end;
 481                 sizestr = devstr + 1;
 482                 *devstr = '\0';
 483                 devstr = vol_args->name;
 484                 devid = simple_strtoull(devstr, &end, 10);
 485                 printk(KERN_INFO "resizing devid %llu\n", devid);
 486         }
 487         device = btrfs_find_device(root, devid, NULL, NULL);
 488         if (!device) {
 489                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
 490                 ret = -EINVAL;
 491                 goto out_unlock;
 492         }
 493         if (!strcmp(sizestr, "max"))
 494                 new_size = device->bdev->bd_inode->i_size;
 495         else {
 496                 if (sizestr[0] == '-') {
 497                         mod = -1;
 498                         sizestr++;
 499                 } else if (sizestr[0] == '+') {
 500                         mod = 1;
 501                         sizestr++;
 502                 }
 503                 new_size = btrfs_parse_size(sizestr);
 504                 if (new_size == 0) {
 505                         ret = -EINVAL;
 506                         goto out_unlock;
 507                 }
 508         }
 509
 510         old_size = device->total_bytes;
 511
 512         if (mod < 0) {
 513                 if (new_size > old_size) {
 514                         ret = -EINVAL;
 515                         goto out_unlock;
 516                 }
 517                 new_size = old_size - new_size;
 518         } else if (mod > 0) {
 519                 new_size = old_size + new_size;
 520         }
 521
 522         if (new_size < 256 * 1024 * 1024) {
 523                 ret = -EINVAL;
 524                 goto out_unlock;
 525         }
 526         if (new_size > device->bdev->bd_inode->i_size) {
 527                 ret = -EFBIG;
 528                 goto out_unlock;
 529         }
 530
 531         do_div(new_size, root->sectorsize);
 532         new_size *= root->sectorsize;
 533
 534         printk(KERN_INFO "new size for %s is %llu\n",
 535                 device->name, (unsigned long long)new_size);
 536
 537         if (new_size > old_size) {
 538                 trans = btrfs_start_transaction(root, 1);
 539                 ret = btrfs_grow_device(trans, device, new_size);
 540                 btrfs_commit_transaction(trans, root);
 541         } else {
 542                 ret = btrfs_shrink_device(device, new_size);
 543         }
 544
 545 out_unlock:
 546         mutex_unlock(&root->fs_info->volume_mutex);
 547 out:
 548         kfree(vol_args);
 549         return ret;
 550 }
 551
 552 static noinline int btrfs_ioctl_snap_create(struct file *file,
 553                                             void __user *arg, int subvol)
 554 {
 555         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
 556         struct btrfs_ioctl_vol_args *vol_args;
 557         struct btrfs_dir_item *di;
 558         struct btrfs_path *path;
 559         struct file *src_file;
 560         u64 root_dirid;
 561         int namelen;
 562         int ret = 0;
 563
 564         if (root->fs_info->sb->s_flags & MS_RDONLY)
 565                 return -EROFS;
 566
 567         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 568
 569         if (!vol_args)
 570                 return -ENOMEM;
 571
 572         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 573                 ret = -EFAULT;
 574                 goto out;
 575         }
 576
 577         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 578         namelen = strlen(vol_args->name);
 579         if (strchr(vol_args->name, '/')) {
 580                 ret = -EINVAL;
 581                 goto out;
 582         }
 583
 584         path = btrfs_alloc_path();
 585         if (!path) {
 586                 ret = -ENOMEM;
 587                 goto out;
 588         }
 589
 590         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
 591         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
 592                             path, root_dirid,
 593                             vol_args->name, namelen, 0);
 594         btrfs_free_path(path);
 595
 596         if (di && !IS_ERR(di)) {
 597                 ret = -EEXIST;
 598                 goto out;
 599         }
 600
 601         if (IS_ERR(di)) {
 602                 ret = PTR_ERR(di);
 603                 goto out;
 604         }
 605
 606         if (subvol) {
 607                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 608                                      file->f_path.dentry->d_inode->i_mode,
 609                                      namelen, NULL);
 610         } else {
 611                 struct inode *src_inode;
 612                 src_file = fget(vol_args->fd);
 613                 if (!src_file) {
 614                         ret = -EINVAL;
 615                         goto out;
 616                 }
 617
 618                 src_inode = src_file->f_path.dentry->d_inode;
 619                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
 620                         printk("btrfs: Snapshot src from another FS\n");
 621                         ret = -EINVAL;
 622                         fput(src_file);
 623                         goto out;
 624                 }
 625                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
 626                              file->f_path.dentry->d_inode->i_mode,
 627                              namelen, BTRFS_I(src_inode)->root);
 628                 fput(src_file);
 629         }
 630
 631 out:
 632         kfree(vol_args);
 633         return ret;
 634 }
 635
 636 static int btrfs_ioctl_defrag(struct file *file)
 637 {
 638         struct inode *inode = fdentry(file)->d_inode;
 639         struct btrfs_root *root = BTRFS_I(inode)->root;
 640         int ret;
 641
 642         ret = mnt_want_write(file->f_path.mnt);
 643         if (ret)
 644                 return ret;
 645
 646         switch (inode->i_mode & S_IFMT) {
 647         case S_IFDIR:
 648                 if (!capable(CAP_SYS_ADMIN)) {
 649                         ret = -EPERM;
 650                         goto out;
 651                 }
 652                 btrfs_defrag_root(root, 0);
 653                 btrfs_defrag_root(root->fs_info->extent_root, 0);
 654                 break;
 655         case S_IFREG:
 656                 if (!(file->f_mode & FMODE_WRITE)) {
 657                         ret = -EINVAL;
 658                         goto out;
 659                 }
 660                 btrfs_defrag_file(file);
 661                 break;
 662         }
 663 out:
 664         mnt_drop_write(file->f_path.mnt);
 665         return ret;
 666 }
 667
 668 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
 669 {
 670         struct btrfs_ioctl_vol_args *vol_args;
 671         int ret;
 672
 673         if (!capable(CAP_SYS_ADMIN))
 674                 return -EPERM;
 675
 676         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 677
 678         if (!vol_args)
 679                 return -ENOMEM;
 680
 681         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 682                 ret = -EFAULT;
 683                 goto out;
 684         }
 685         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 686         ret = btrfs_init_new_device(root, vol_args->name);
 687
 688 out:
 689         kfree(vol_args);
 690         return ret;
 691 }
 692
 693 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
 694 {
 695         struct btrfs_ioctl_vol_args *vol_args;
 696         int ret;
 697
 698         if (!capable(CAP_SYS_ADMIN))
 699                 return -EPERM;
 700
 701         if (root->fs_info->sb->s_flags & MS_RDONLY)
 702                 return -EROFS;
 703
 704         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
 705
 706         if (!vol_args)
 707                 return -ENOMEM;
 708
 709         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
 710                 ret = -EFAULT;
 711                 goto out;
 712         }
 713         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
 714         ret = btrfs_rm_device(root, vol_args->name);
 715
 716 out:
 717         kfree(vol_args);
 718         return ret;
 719 }
 720
 721 static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 722                 u64 off, u64 olen, u64 destoff)
 723 {
 724         struct inode *inode = fdentry(file)->d_inode;
 725         struct btrfs_root *root = BTRFS_I(inode)->root;
 726         struct file *src_file;
 727         struct inode *src;
 728         struct btrfs_trans_handle *trans;
 729         struct btrfs_path *path;
 730         struct extent_buffer *leaf;
 731         char *buf;
 732         struct btrfs_key key;
 733         u32 nritems;
 734         int slot;
 735         int ret;
 736         u64 len = olen;
 737         u64 bs = root->fs_info->sb->s_blocksize;
 738         u64 hint_byte;
 739
 740         /*
 741          * TODO:
 742          * - split compressed inline extents.  annoying: we need to
 743          *   decompress into destination's address_space (the file offset
 744          *   may change, so source mapping won't do), then recompress (or
 745          *   otherwise reinsert) a subrange.
 746          * - allow ranges within the same file to be cloned (provided
 747          *   they don't overlap)?
 748          */
 749
 750         /* the destination must be opened for writing */
 751         if (!(file->f_mode & FMODE_WRITE))
 752                 return -EINVAL;
 753
 754         ret = mnt_want_write(file->f_path.mnt);
 755         if (ret)
 756                 return ret;
 757
 758         src_file = fget(srcfd);
 759         if (!src_file) {
 760                 ret = -EBADF;
 761                 goto out_drop_write;
 762         }
 763         src = src_file->f_dentry->d_inode;
 764
 765         ret = -EINVAL;
 766         if (src == inode)
 767                 goto out_fput;
 768
 769         ret = -EISDIR;
 770         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
 771                 goto out_fput;
 772
 773         ret = -EXDEV;
 774         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
 775                 goto out_fput;
 776
 777         ret = -ENOMEM;
 778         buf = vmalloc(btrfs_level_size(root, 0));
 779         if (!buf)
 780                 goto out_fput;
 781
 782         path = btrfs_alloc_path();
 783         if (!path) {
 784                 vfree(buf);
 785                 goto out_fput;
 786         }
 787         path->reada = 2;
 788
 789         if (inode < src) {
 790                 mutex_lock(&inode->i_mutex);
 791                 mutex_lock(&src->i_mutex);
 792         } else {
 793                 mutex_lock(&src->i_mutex);
 794                 mutex_lock(&inode->i_mutex);
 795         }
 796
 797         /* determine range to clone */
 798         ret = -EINVAL;
 799         if (off >= src->i_size || off + len > src->i_size)
 800                 goto out_unlock;
 801         if (len == 0)
 802                 olen = len = src->i_size - off;
 803         /* if we extend to eof, continue to block boundary */
 804         if (off + len == src->i_size)
 805                 len = ((src->i_size + bs-1) & ~(bs-1))
 806                         - off;
 807
 808         /* verify the end result is block aligned */
 809         if ((off & (bs-1)) ||
 810             ((off + len) & (bs-1)))
 811                 goto out_unlock;
 812
 813         printk("final src extent is %llu~%llu\n", off, len);
 814         printk("final dst extent is %llu~%llu\n", destoff, len);
 815
 816         /* do any pending delalloc/csum calc on src, one way or
 817            another, and lock file content */
 818         while (1) {
 819                 struct btrfs_ordered_extent *ordered;
 820                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 821                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
 822                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
 823                         break;
 824                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
 825                 if (ordered)
 826                         btrfs_put_ordered_extent(ordered);
 827                 btrfs_wait_ordered_range(src, off, off+len);
 828         }
 829
 830         trans = btrfs_start_transaction(root, 1);
 831         BUG_ON(!trans);
 832
 833         /* punch hole in destination first */
 834         btrfs_drop_extents(trans, root, inode, off, off+len, 0, &hint_byte);
 835
 836         /* clone data */
 837         key.objectid = src->i_ino;
 838         key.type = BTRFS_EXTENT_DATA_KEY;
 839         key.offset = 0;
 840
 841         while (1) {
 842                 /*
 843                  * note the key will change type as we walk through the
 844                  * tree.
 845                  */
 846                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 847                 if (ret < 0)
 848                         goto out;
 849
 850                 nritems = btrfs_header_nritems(path->nodes[0]);
 851                 if (path->slots[0] >= nritems) {
 852                         ret = btrfs_next_leaf(root, path);
 853                         if (ret < 0)
 854                                 goto out;
 855                         if (ret > 0)
 856                                 break;
 857                         nritems = btrfs_header_nritems(path->nodes[0]);
 858                 }
 859                 leaf = path->nodes[0];
 860                 slot = path->slots[0];
 861
 862                 btrfs_item_key_to_cpu(leaf, &key, slot);
 863                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 864                     key.objectid != src->i_ino)
 865                         break;
 866
 867                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 868                         struct btrfs_file_extent_item *extent;
 869                         int type;
 870                         u32 size;
 871                         struct btrfs_key new_key;
 872                         u64 disko = 0, diskl = 0;
 873                         u64 datao = 0, datal = 0;
 874                         u8 comp;
 875
 876                         size = btrfs_item_size_nr(leaf, slot);
 877                         read_extent_buffer(leaf, buf,
 878                                            btrfs_item_ptr_offset(leaf, slot),
 879                                            size);
 880
 881                         extent = btrfs_item_ptr(leaf, slot,
 882                                                 struct btrfs_file_extent_item);
 883                         comp = btrfs_file_extent_compression(leaf, extent);
 884                         type = btrfs_file_extent_type(leaf, extent);
 885                         if (type == BTRFS_FILE_EXTENT_REG) {
 886                                 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
 887                                 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
 888                                 datao = btrfs_file_extent_offset(leaf, extent);
 889                                 datal = btrfs_file_extent_num_bytes(leaf, extent);
 890                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 891                                 /* take upper bound, may be compressed */
 892                                 datal = btrfs_file_extent_ram_bytes(leaf,
 893                                                                     extent);
 894                         }
 895                         btrfs_release_path(root, path);
 896
 897                         if (key.offset + datal < off ||
 898                             key.offset >= off+len)
 899                                 goto next;
 900
 901                         memcpy(&new_key, &key, sizeof(new_key));
 902                         new_key.objectid = inode->i_ino;
 903                         new_key.offset = key.offset + destoff - off;
 904
 905                         if (type == BTRFS_FILE_EXTENT_REG) {
 906                                 ret = btrfs_insert_empty_item(trans, root, path,
 907                                                               &new_key, size);
 908                                 if (ret)
 909                                         goto out;
 910
 911                                 leaf = path->nodes[0];
 912                                 slot = path->slots[0];
 913                                 write_extent_buffer(leaf, buf,
 914                                             btrfs_item_ptr_offset(leaf, slot),
 915                                             size);
 916
 917                                 extent = btrfs_item_ptr(leaf, slot,
 918                                                 struct btrfs_file_extent_item);
 919                                 printk("  orig disk %llu~%llu data %llu~%llu\n",
 920                                        disko, diskl, datao, datal);
 921
 922                                 if (off > key.offset) {
 923                                         datao += off - key.offset;
 924                                         datal -= off - key.offset;
 925                                 }
 926                                 if (key.offset + datao + datal + key.offset >
 927                                     off + len)
 928                                         datal = off + len - key.offset - datao;
 929                                 /* disko == 0 means it's a hole */
 930                                 if (!disko)
 931                                         datao = 0;
 932                                 printk(" final disk %llu~%llu data %llu~%llu\n",
 933                                        disko, diskl, datao, datal);
 934
 935                                 btrfs_set_file_extent_offset(leaf, extent,
 936                                                              datao);
 937                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 938                                                                 datal);
 939                                 if (disko) {
 940                                         inode_add_bytes(inode, datal);
 941                                         ret = btrfs_inc_extent_ref(trans, root,
 942                                                    disko, diskl, leaf->start,
 943                                                    root->root_key.objectid,
 944                                                    trans->transid,
 945                                                    inode->i_ino);
 946                                         BUG_ON(ret);
 947                                 }
 948                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 949                                 u64 skip = 0;
 950                                 u64 trim = 0;
 951                                 if (off > key.offset) {
 952                                         skip = off - key.offset;
 953                                         new_key.offset += skip;
 954                                 }
 955                                 if (key.offset + datal > off+len)
 956                                         trim = key.offset + datal - (off+len);
 957                                 printk("len %lld skip %lld trim %lld\n",
 958                                        datal, skip, trim);
 959                                 if (comp && (skip || trim)) {
 960                                         printk("btrfs clone_range can't split compressed inline extents yet\n");
 961                                         ret = -EINVAL;
 962                                         goto out;
 963                                 }
 964                                 size -= skip + trim;
 965                                 datal -= skip + trim;
 966                                 ret = btrfs_insert_empty_item(trans, root, path,
 967                                                               &new_key, size);
 968                                 if (ret)
 969                                         goto out;
 970
 971                                 if (skip) {
 972                                         u32 start = btrfs_file_extent_calc_inline_size(0);
 973                                         memmove(buf+start, buf+start+skip,
 974                                                 datal);
 975                                 }
 976
 977                                 leaf = path->nodes[0];
 978                                 slot = path->slots[0];
 979                                 write_extent_buffer(leaf, buf,
 980                                             btrfs_item_ptr_offset(leaf, slot),
 981                                             size);
 982                                 inode_add_bytes(inode, datal);
 983                         }
 984
 985                         btrfs_mark_buffer_dirty(leaf);
 986                 }
 987
 988         next:
 989                 btrfs_release_path(root, path);
 990                 key.offset++;
 991         }
 992         ret = 0;
 993 out:
 994         btrfs_release_path(root, path);
 995         if (ret == 0) {
 996                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 997                 if (destoff + olen > inode->i_size)
 998                         btrfs_i_size_write(inode, destoff + olen);
 999                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1000                 ret = btrfs_update_inode(trans, root, inode);
1001         }
1002         btrfs_end_transaction(trans, root);
1003         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1004         if (ret)
1005                 vmtruncate(inode, 0);
1006 out_unlock:
1007         mutex_unlock(&src->i_mutex);
1008         mutex_unlock(&inode->i_mutex);
1009         vfree(buf);
1010         btrfs_free_path(path);
1011 out_fput:
1012         fput(src_file);
1013 out_drop_write:
1014         mnt_drop_write(file->f_path.mnt);
1015         return ret;
1016 }
1017
1018 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1019 {
1020         struct btrfs_ioctl_clone_range_args args;
1021
1022         if (copy_from_user(&args, argp, sizeof(args)))
1023                 return -EFAULT;
1024         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1025                                  args.src_length, args.dest_offset);
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 static long btrfs_ioctl_trans_start(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         if (!capable(CAP_SYS_ADMIN))
1042                 return -EPERM;
1043
1044         if (file->private_data) {
1045                 ret = -EINPROGRESS;
1046                 goto out;
1047         }
1048
1049         ret = mnt_want_write(file->f_path.mnt);
1050         if (ret)
1051                 goto out;
1052
1053         mutex_lock(&root->fs_info->trans_mutex);
1054         root->fs_info->open_ioctl_trans++;
1055         mutex_unlock(&root->fs_info->trans_mutex);
1056
1057         trans = btrfs_start_ioctl_transaction(root, 0);
1058         if (trans)
1059                 file->private_data = trans;
1060         else
1061                 ret = -ENOMEM;
1062         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1063 out:
1064         return ret;
1065 }
1066
1067 /*
1068  * there are many ways the trans_start and trans_end ioctls can lead
1069  * to deadlocks.  They should only be used by applications that
1070  * basically own the machine, and have a very in depth understanding
1071  * of all the possible deadlocks and enospc problems.
1072  */
1073 long btrfs_ioctl_trans_end(struct file *file)
1074 {
1075         struct inode *inode = fdentry(file)->d_inode;
1076         struct btrfs_root *root = BTRFS_I(inode)->root;
1077         struct btrfs_trans_handle *trans;
1078         int ret = 0;
1079
1080         trans = file->private_data;
1081         if (!trans) {
1082                 ret = -EINVAL;
1083                 goto out;
1084         }
1085         btrfs_end_transaction(trans, root);
1086         file->private_data = NULL;
1087
1088         mutex_lock(&root->fs_info->trans_mutex);
1089         root->fs_info->open_ioctl_trans--;
1090         mutex_unlock(&root->fs_info->trans_mutex);
1091
1092         mnt_drop_write(file->f_path.mnt);
1093
1094 out:
1095         return ret;
1096 }
1097
1098 long btrfs_ioctl(struct file *file, unsigned int
1099                 cmd, unsigned long arg)
1100 {
1101         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1102         void __user *argp = (void __user *)arg;
1103
1104         switch (cmd) {
1105         case BTRFS_IOC_SNAP_CREATE:
1106                 return btrfs_ioctl_snap_create(file, argp, 0);
1107         case BTRFS_IOC_SUBVOL_CREATE:
1108                 return btrfs_ioctl_snap_create(file, argp, 1);
1109         case BTRFS_IOC_DEFRAG:
1110                 return btrfs_ioctl_defrag(file);
1111         case BTRFS_IOC_RESIZE:
1112                 return btrfs_ioctl_resize(root, argp);
1113         case BTRFS_IOC_ADD_DEV:
1114                 return btrfs_ioctl_add_dev(root, argp);
1115         case BTRFS_IOC_RM_DEV:
1116                 return btrfs_ioctl_rm_dev(root, argp);
1117         case BTRFS_IOC_BALANCE:
1118                 return btrfs_balance(root->fs_info->dev_root);
1119         case BTRFS_IOC_CLONE:
1120                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1121         case BTRFS_IOC_CLONE_RANGE:
1122                 return btrfs_ioctl_clone_range(file, argp);
1123         case BTRFS_IOC_TRANS_START:
1124                 return btrfs_ioctl_trans_start(file);
1125         case BTRFS_IOC_TRANS_END:
1126                 return btrfs_ioctl_trans_end(file);
1127         case BTRFS_IOC_SYNC:
1128                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1129                 return 0;
1130         }
1131
1132         return -ENOTTY;
1133 }