fs/btrfs/super.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/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/time.h>
  26 #include <linux/init.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mount.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/parser.h>
  37 #include <linux/ctype.h>
  38 #include <linux/namei.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/magic.h>
  41 #include <linux/slab.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "xattr.h"
  50 #include "volumes.h"
  51 #include "version.h"
  52 #include "export.h"
  53 #include "compression.h"
  54
  55 static const struct super_operations btrfs_super_ops;
  56
  57 static void btrfs_put_super(struct super_block *sb)
  58 {
  59         struct btrfs_root *root = btrfs_sb(sb);
  60         int ret;
  61
  62         ret = close_ctree(root);
  63         sb->s_fs_info = NULL;
  64
  65         (void)ret; /* FIXME: need to fix VFS to return error? */
  66 }
  67
  68 enum {
  69         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
  70         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
  71         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
  72         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
  73         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
  74         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed, Opt_err,
  75 };
  76
  77 static match_table_t tokens = {
  78         {Opt_degraded, "degraded"},
  79         {Opt_subvol, "subvol=%s"},
  80         {Opt_subvolid, "subvolid=%d"},
  81         {Opt_device, "device=%s"},
  82         {Opt_nodatasum, "nodatasum"},
  83         {Opt_nodatacow, "nodatacow"},
  84         {Opt_nobarrier, "nobarrier"},
  85         {Opt_max_inline, "max_inline=%s"},
  86         {Opt_alloc_start, "alloc_start=%s"},
  87         {Opt_thread_pool, "thread_pool=%d"},
  88         {Opt_compress, "compress"},
  89         {Opt_compress_type, "compress=%s"},
  90         {Opt_compress_force, "compress-force"},
  91         {Opt_compress_force_type, "compress-force=%s"},
  92         {Opt_ssd, "ssd"},
  93         {Opt_ssd_spread, "ssd_spread"},
  94         {Opt_nossd, "nossd"},
  95         {Opt_noacl, "noacl"},
  96         {Opt_notreelog, "notreelog"},
  97         {Opt_flushoncommit, "flushoncommit"},
  98         {Opt_ratio, "metadata_ratio=%d"},
  99         {Opt_discard, "discard"},
 100         {Opt_space_cache, "space_cache"},
 101         {Opt_clear_cache, "clear_cache"},
 102         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 103         {Opt_err, NULL},
 104 };
 105
 106 /*
 107  * Regular mount options parser.  Everything that is needed only when
 108  * reading in a new superblock is parsed here.
 109  */
 110 int btrfs_parse_options(struct btrfs_root *root, char *options)
 111 {
 112         struct btrfs_fs_info *info = root->fs_info;
 113         substring_t args[MAX_OPT_ARGS];
 114         char *p, *num, *orig;
 115         int intarg;
 116         int ret = 0;
 117         char *compress_type;
 118         bool compress_force = false;
 119
 120         if (!options)
 121                 return 0;
 122
 123         /*
 124          * strsep changes the string, duplicate it because parse_options
 125          * gets called twice
 126          */
 127         options = kstrdup(options, GFP_NOFS);
 128         if (!options)
 129                 return -ENOMEM;
 130
 131         orig = options;
 132
 133         while ((p = strsep(&options, ",")) != NULL) {
 134                 int token;
 135                 if (!*p)
 136                         continue;
 137
 138                 token = match_token(p, tokens, args);
 139                 switch (token) {
 140                 case Opt_degraded:
 141                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 142                         btrfs_set_opt(info->mount_opt, DEGRADED);
 143                         break;
 144                 case Opt_subvol:
 145                 case Opt_subvolid:
 146                 case Opt_device:
 147                         /*
 148                          * These are parsed by btrfs_parse_early_options
 149                          * and can be happily ignored here.
 150                          */
 151                         break;
 152                 case Opt_nodatasum:
 153                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 154                         btrfs_set_opt(info->mount_opt, NODATASUM);
 155                         break;
 156                 case Opt_nodatacow:
 157                         printk(KERN_INFO "btrfs: setting nodatacow\n");
 158                         btrfs_set_opt(info->mount_opt, NODATACOW);
 159                         btrfs_set_opt(info->mount_opt, NODATASUM);
 160                         break;
 161                 case Opt_compress_force:
 162                 case Opt_compress_force_type:
 163                         compress_force = true;
 164                 case Opt_compress:
 165                 case Opt_compress_type:
 166                         if (token == Opt_compress ||
 167                             token == Opt_compress_force ||
 168                             strcmp(args[0].from, "zlib") == 0) {
 169                                 compress_type = "zlib";
 170                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 171                         } else if (strcmp(args[0].from, "lzo") == 0) {
 172                                 compress_type = "lzo";
 173                                 info->compress_type = BTRFS_COMPRESS_LZO;
 174                         } else {
 175                                 ret = -EINVAL;
 176                                 goto out;
 177                         }
 178
 179                         btrfs_set_opt(info->mount_opt, COMPRESS);
 180                         if (compress_force) {
 181                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 182                                 pr_info("btrfs: force %s compression\n",
 183                                         compress_type);
 184                         } else
 185                                 pr_info("btrfs: use %s compression\n",
 186                                         compress_type);
 187                         break;
 188                 case Opt_ssd:
 189                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 190                         btrfs_set_opt(info->mount_opt, SSD);
 191                         break;
 192                 case Opt_ssd_spread:
 193                         printk(KERN_INFO "btrfs: use spread ssd "
 194                                "allocation scheme\n");
 195                         btrfs_set_opt(info->mount_opt, SSD);
 196                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 197                         break;
 198                 case Opt_nossd:
 199                         printk(KERN_INFO "btrfs: not using ssd allocation "
 200                                "scheme\n");
 201                         btrfs_set_opt(info->mount_opt, NOSSD);
 202                         btrfs_clear_opt(info->mount_opt, SSD);
 203                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 204                         break;
 205                 case Opt_nobarrier:
 206                         printk(KERN_INFO "btrfs: turning off barriers\n");
 207                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 208                         break;
 209                 case Opt_thread_pool:
 210                         intarg = 0;
 211                         match_int(&args[0], &intarg);
 212                         if (intarg) {
 213                                 info->thread_pool_size = intarg;
 214                                 printk(KERN_INFO "btrfs: thread pool %d\n",
 215                                        info->thread_pool_size);
 216                         }
 217                         break;
 218                 case Opt_max_inline:
 219                         num = match_strdup(&args[0]);
 220                         if (num) {
 221                                 info->max_inline = memparse(num, NULL);
 222                                 kfree(num);
 223
 224                                 if (info->max_inline) {
 225                                         info->max_inline = max_t(u64,
 226                                                 info->max_inline,
 227                                                 root->sectorsize);
 228                                 }
 229                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 230                                         (unsigned long long)info->max_inline);
 231                         }
 232                         break;
 233                 case Opt_alloc_start:
 234                         num = match_strdup(&args[0]);
 235                         if (num) {
 236                                 info->alloc_start = memparse(num, NULL);
 237                                 kfree(num);
 238                                 printk(KERN_INFO
 239                                         "btrfs: allocations start at %llu\n",
 240                                         (unsigned long long)info->alloc_start);
 241                         }
 242                         break;
 243                 case Opt_noacl:
 244                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 245                         break;
 246                 case Opt_notreelog:
 247                         printk(KERN_INFO "btrfs: disabling tree log\n");
 248                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 249                         break;
 250                 case Opt_flushoncommit:
 251                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 252                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 253                         break;
 254                 case Opt_ratio:
 255                         intarg = 0;
 256                         match_int(&args[0], &intarg);
 257                         if (intarg) {
 258                                 info->metadata_ratio = intarg;
 259                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 260                                        info->metadata_ratio);
 261                         }
 262                         break;
 263                 case Opt_discard:
 264                         btrfs_set_opt(info->mount_opt, DISCARD);
 265                         break;
 266                 case Opt_space_cache:
 267                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
 268                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 269                         break;
 270                 case Opt_clear_cache:
 271                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 272                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 273                         break;
 274                 case Opt_user_subvol_rm_allowed:
 275                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 276                         break;
 277                 case Opt_err:
 278                         printk(KERN_INFO "btrfs: unrecognized mount option "
 279                                "'%s'\n", p);
 280                         ret = -EINVAL;
 281                         goto out;
 282                 default:
 283                         break;
 284                 }
 285         }
 286 out:
 287         kfree(orig);
 288         return ret;
 289 }
 290
 291 /*
 292  * Parse mount options that are required early in the mount process.
 293  *
 294  * All other options will be parsed on much later in the mount process and
 295  * only when we need to allocate a new super block.
 296  */
 297 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 298                 void *holder, char **subvol_name, u64 *subvol_objectid,
 299                 struct btrfs_fs_devices **fs_devices)
 300 {
 301         substring_t args[MAX_OPT_ARGS];
 302         char *opts, *p;
 303         int error = 0;
 304         int intarg;
 305
 306         if (!options)
 307                 goto out;
 308
 309         /*
 310          * strsep changes the string, duplicate it because parse_options
 311          * gets called twice
 312          */
 313         opts = kstrdup(options, GFP_KERNEL);
 314         if (!opts)
 315                 return -ENOMEM;
 316
 317         while ((p = strsep(&opts, ",")) != NULL) {
 318                 int token;
 319                 if (!*p)
 320                         continue;
 321
 322                 token = match_token(p, tokens, args);
 323                 switch (token) {
 324                 case Opt_subvol:
 325                         *subvol_name = match_strdup(&args[0]);
 326                         break;
 327                 case Opt_subvolid:
 328                         intarg = 0;
 329                         error = match_int(&args[0], &intarg);
 330                         if (!error) {
 331                                 /* we want the original fs_tree */
 332                                 if (!intarg)
 333                                         *subvol_objectid =
 334                                                 BTRFS_FS_TREE_OBJECTID;
 335                                 else
 336                                         *subvol_objectid = intarg;
 337                         }
 338                         break;
 339                 case Opt_device:
 340                         error = btrfs_scan_one_device(match_strdup(&args[0]),
 341                                         flags, holder, fs_devices);
 342                         if (error)
 343                                 goto out_free_opts;
 344                         break;
 345                 default:
 346                         break;
 347                 }
 348         }
 349
 350  out_free_opts:
 351         kfree(opts);
 352  out:
 353         /*
 354          * If no subvolume name is specified we use the default one.  Allocate
 355          * a copy of the string "." here so that code later in the
 356          * mount path doesn't care if it's the default volume or another one.
 357          */
 358         if (!*subvol_name) {
 359                 *subvol_name = kstrdup(".", GFP_KERNEL);
 360                 if (!*subvol_name)
 361                         return -ENOMEM;
 362         }
 363         return error;
 364 }
 365
 366 static struct dentry *get_default_root(struct super_block *sb,
 367                                        u64 subvol_objectid)
 368 {
 369         struct btrfs_root *root = sb->s_fs_info;
 370         struct btrfs_root *new_root;
 371         struct btrfs_dir_item *di;
 372         struct btrfs_path *path;
 373         struct btrfs_key location;
 374         struct inode *inode;
 375         struct dentry *dentry;
 376         u64 dir_id;
 377         int new = 0;
 378
 379         /*
 380          * We have a specific subvol we want to mount, just setup location and
 381          * go look up the root.
 382          */
 383         if (subvol_objectid) {
 384                 location.objectid = subvol_objectid;
 385                 location.type = BTRFS_ROOT_ITEM_KEY;
 386                 location.offset = (u64)-1;
 387                 goto find_root;
 388         }
 389
 390         path = btrfs_alloc_path();
 391         if (!path)
 392                 return ERR_PTR(-ENOMEM);
 393         path->leave_spinning = 1;
 394
 395         /*
 396          * Find the "default" dir item which points to the root item that we
 397          * will mount by default if we haven't been given a specific subvolume
 398          * to mount.
 399          */
 400         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
 401         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 402         if (IS_ERR(di))
 403                 return ERR_CAST(di);
 404         if (!di) {
 405                 /*
 406                  * Ok the default dir item isn't there.  This is weird since
 407                  * it's always been there, but don't freak out, just try and
 408                  * mount to root most subvolume.
 409                  */
 410                 btrfs_free_path(path);
 411                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 412                 new_root = root->fs_info->fs_root;
 413                 goto setup_root;
 414         }
 415
 416         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 417         btrfs_free_path(path);
 418
 419 find_root:
 420         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
 421         if (IS_ERR(new_root))
 422                 return ERR_CAST(new_root);
 423
 424         if (btrfs_root_refs(&new_root->root_item) == 0)
 425                 return ERR_PTR(-ENOENT);
 426
 427         dir_id = btrfs_root_dirid(&new_root->root_item);
 428 setup_root:
 429         location.objectid = dir_id;
 430         location.type = BTRFS_INODE_ITEM_KEY;
 431         location.offset = 0;
 432
 433         inode = btrfs_iget(sb, &location, new_root, &new);
 434         if (IS_ERR(inode))
 435                 return ERR_CAST(inode);
 436
 437         /*
 438          * If we're just mounting the root most subvol put the inode and return
 439          * a reference to the dentry.  We will have already gotten a reference
 440          * to the inode in btrfs_fill_super so we're good to go.
 441          */
 442         if (!new && sb->s_root->d_inode == inode) {
 443                 iput(inode);
 444                 return dget(sb->s_root);
 445         }
 446
 447         if (new) {
 448                 const struct qstr name = { .name = "/", .len = 1 };
 449
 450                 /*
 451                  * New inode, we need to make the dentry a sibling of s_root so
 452                  * everything gets cleaned up properly on unmount.
 453                  */
 454                 dentry = d_alloc(sb->s_root, &name);
 455                 if (!dentry) {
 456                         iput(inode);
 457                         return ERR_PTR(-ENOMEM);
 458                 }
 459                 d_splice_alias(inode, dentry);
 460         } else {
 461                 /*
 462                  * We found the inode in cache, just find a dentry for it and
 463                  * put the reference to the inode we just got.
 464                  */
 465                 dentry = d_find_alias(inode);
 466                 iput(inode);
 467         }
 468
 469         return dentry;
 470 }
 471
 472 static int btrfs_fill_super(struct super_block *sb,
 473                             struct btrfs_fs_devices *fs_devices,
 474                             void *data, int silent)
 475 {
 476         struct inode *inode;
 477         struct dentry *root_dentry;
 478         struct btrfs_root *tree_root;
 479         struct btrfs_key key;
 480         int err;
 481
 482         sb->s_maxbytes = MAX_LFS_FILESIZE;
 483         sb->s_magic = BTRFS_SUPER_MAGIC;
 484         sb->s_op = &btrfs_super_ops;
 485         sb->s_export_op = &btrfs_export_ops;
 486         sb->s_xattr = btrfs_xattr_handlers;
 487         sb->s_time_gran = 1;
 488 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 489         sb->s_flags |= MS_POSIXACL;
 490 #endif
 491
 492         tree_root = open_ctree(sb, fs_devices, (char *)data);
 493
 494         if (IS_ERR(tree_root)) {
 495                 printk("btrfs: open_ctree failed\n");
 496                 return PTR_ERR(tree_root);
 497         }
 498         sb->s_fs_info = tree_root;
 499
 500         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 501         key.type = BTRFS_INODE_ITEM_KEY;
 502         key.offset = 0;
 503         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
 504         if (IS_ERR(inode)) {
 505                 err = PTR_ERR(inode);
 506                 goto fail_close;
 507         }
 508
 509         root_dentry = d_alloc_root(inode);
 510         if (!root_dentry) {
 511                 iput(inode);
 512                 err = -ENOMEM;
 513                 goto fail_close;
 514         }
 515
 516         sb->s_root = root_dentry;
 517
 518         save_mount_options(sb, data);
 519         return 0;
 520
 521 fail_close:
 522         close_ctree(tree_root);
 523         return err;
 524 }
 525
 526 int btrfs_sync_fs(struct super_block *sb, int wait)
 527 {
 528         struct btrfs_trans_handle *trans;
 529         struct btrfs_root *root = btrfs_sb(sb);
 530         int ret;
 531
 532         if (!wait) {
 533                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 534                 return 0;
 535         }
 536
 537         btrfs_start_delalloc_inodes(root, 0);
 538         btrfs_wait_ordered_extents(root, 0, 0);
 539
 540         trans = btrfs_start_transaction(root, 0);
 541         ret = btrfs_commit_transaction(trans, root);
 542         return ret;
 543 }
 544
 545 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 546 {
 547         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
 548         struct btrfs_fs_info *info = root->fs_info;
 549
 550         if (btrfs_test_opt(root, DEGRADED))
 551                 seq_puts(seq, ",degraded");
 552         if (btrfs_test_opt(root, NODATASUM))
 553                 seq_puts(seq, ",nodatasum");
 554         if (btrfs_test_opt(root, NODATACOW))
 555                 seq_puts(seq, ",nodatacow");
 556         if (btrfs_test_opt(root, NOBARRIER))
 557                 seq_puts(seq, ",nobarrier");
 558         if (info->max_inline != 8192 * 1024)
 559                 seq_printf(seq, ",max_inline=%llu",
 560                            (unsigned long long)info->max_inline);
 561         if (info->alloc_start != 0)
 562                 seq_printf(seq, ",alloc_start=%llu",
 563                            (unsigned long long)info->alloc_start);
 564         if (info->thread_pool_size !=  min_t(unsigned long,
 565                                              num_online_cpus() + 2, 8))
 566                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 567         if (btrfs_test_opt(root, COMPRESS))
 568                 seq_puts(seq, ",compress");
 569         if (btrfs_test_opt(root, NOSSD))
 570                 seq_puts(seq, ",nossd");
 571         if (btrfs_test_opt(root, SSD_SPREAD))
 572                 seq_puts(seq, ",ssd_spread");
 573         else if (btrfs_test_opt(root, SSD))
 574                 seq_puts(seq, ",ssd");
 575         if (btrfs_test_opt(root, NOTREELOG))
 576                 seq_puts(seq, ",notreelog");
 577         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 578                 seq_puts(seq, ",flushoncommit");
 579         if (btrfs_test_opt(root, DISCARD))
 580                 seq_puts(seq, ",discard");
 581         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 582                 seq_puts(seq, ",noacl");
 583         return 0;
 584 }
 585
 586 static int btrfs_test_super(struct super_block *s, void *data)
 587 {
 588         struct btrfs_root *test_root = data;
 589         struct btrfs_root *root = btrfs_sb(s);
 590
 591         /*
 592          * If this super block is going away, return false as it
 593          * can't match as an existing super block.
 594          */
 595         if (!atomic_read(&s->s_active))
 596                 return 0;
 597         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
 598 }
 599
 600 static int btrfs_set_super(struct super_block *s, void *data)
 601 {
 602         s->s_fs_info = data;
 603
 604         return set_anon_super(s, data);
 605 }
 606
 607
 608 /*
 609  * Find a superblock for the given device / mount point.
 610  *
 611  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
 612  *        for multiple device setup.  Make sure to keep it in sync.
 613  */
 614 static int btrfs_get_sb(struct file_system_type *fs_type, int flags,
 615                 const char *dev_name, void *data, struct vfsmount *mnt)
 616 {
 617         struct block_device *bdev = NULL;
 618         struct super_block *s;
 619         struct dentry *root;
 620         struct btrfs_fs_devices *fs_devices = NULL;
 621         struct btrfs_root *tree_root = NULL;
 622         struct btrfs_fs_info *fs_info = NULL;
 623         fmode_t mode = FMODE_READ;
 624         char *subvol_name = NULL;
 625         u64 subvol_objectid = 0;
 626         int error = 0;
 627
 628         if (!(flags & MS_RDONLY))
 629                 mode |= FMODE_WRITE;
 630
 631         error = btrfs_parse_early_options(data, mode, fs_type,
 632                                           &subvol_name, &subvol_objectid,
 633                                           &fs_devices);
 634         if (error)
 635                 return error;
 636
 637         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
 638         if (error)
 639                 goto error_free_subvol_name;
 640
 641         error = btrfs_open_devices(fs_devices, mode, fs_type);
 642         if (error)
 643                 goto error_free_subvol_name;
 644
 645         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
 646                 error = -EACCES;
 647                 goto error_close_devices;
 648         }
 649
 650         /*
 651          * Setup a dummy root and fs_info for test/set super.  This is because
 652          * we don't actually fill this stuff out until open_ctree, but we need
 653          * it for searching for existing supers, so this lets us do that and
 654          * then open_ctree will properly initialize everything later.
 655          */
 656         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
 657         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
 658         if (!fs_info || !tree_root) {
 659                 error = -ENOMEM;
 660                 goto error_close_devices;
 661         }
 662         fs_info->tree_root = tree_root;
 663         fs_info->fs_devices = fs_devices;
 664         tree_root->fs_info = fs_info;
 665
 666         bdev = fs_devices->latest_bdev;
 667         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
 668         if (IS_ERR(s))
 669                 goto error_s;
 670
 671         if (s->s_root) {
 672                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 673                         deactivate_locked_super(s);
 674                         error = -EBUSY;
 675                         goto error_close_devices;
 676                 }
 677
 678                 btrfs_close_devices(fs_devices);
 679         } else {
 680                 char b[BDEVNAME_SIZE];
 681
 682                 s->s_flags = flags;
 683                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 684                 error = btrfs_fill_super(s, fs_devices, data,
 685                                          flags & MS_SILENT ? 1 : 0);
 686                 if (error) {
 687                         deactivate_locked_super(s);
 688                         goto error_free_subvol_name;
 689                 }
 690
 691                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
 692                 s->s_flags |= MS_ACTIVE;
 693         }
 694
 695         root = get_default_root(s, subvol_objectid);
 696         if (IS_ERR(root)) {
 697                 error = PTR_ERR(root);
 698                 deactivate_locked_super(s);
 699                 goto error_free_subvol_name;
 700         }
 701         /* if they gave us a subvolume name bind mount into that */
 702         if (strcmp(subvol_name, ".")) {
 703                 struct dentry *new_root;
 704                 mutex_lock(&root->d_inode->i_mutex);
 705                 new_root = lookup_one_len(subvol_name, root,
 706                                       strlen(subvol_name));
 707                 mutex_unlock(&root->d_inode->i_mutex);
 708
 709                 if (IS_ERR(new_root)) {
 710                         dput(root);
 711                         deactivate_locked_super(s);
 712                         error = PTR_ERR(new_root);
 713                         goto error_free_subvol_name;
 714                 }
 715                 if (!new_root->d_inode) {
 716                         dput(root);
 717                         dput(new_root);
 718                         deactivate_locked_super(s);
 719                         error = -ENXIO;
 720                         goto error_free_subvol_name;
 721                 }
 722                 dput(root);
 723                 root = new_root;
 724         }
 725
 726         mnt->mnt_sb = s;
 727         mnt->mnt_root = root;
 728
 729         kfree(subvol_name);
 730         return 0;
 731
 732 error_s:
 733         error = PTR_ERR(s);
 734 error_close_devices:
 735         btrfs_close_devices(fs_devices);
 736         kfree(fs_info);
 737         kfree(tree_root);
 738 error_free_subvol_name:
 739         kfree(subvol_name);
 740         return error;
 741 }
 742
 743 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 744 {
 745         struct btrfs_root *root = btrfs_sb(sb);
 746         int ret;
 747
 748         ret = btrfs_parse_options(root, data);
 749         if (ret)
 750                 return -EINVAL;
 751
 752         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 753                 return 0;
 754
 755         if (*flags & MS_RDONLY) {
 756                 sb->s_flags |= MS_RDONLY;
 757
 758                 ret =  btrfs_commit_super(root);
 759                 WARN_ON(ret);
 760         } else {
 761                 if (root->fs_info->fs_devices->rw_devices == 0)
 762                         return -EACCES;
 763
 764                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
 765                         return -EINVAL;
 766
 767                 ret = btrfs_cleanup_fs_roots(root->fs_info);
 768                 WARN_ON(ret);
 769
 770                 /* recover relocation */
 771                 ret = btrfs_recover_relocation(root);
 772                 WARN_ON(ret);
 773
 774                 sb->s_flags &= ~MS_RDONLY;
 775         }
 776
 777         return 0;
 778 }
 779
 780 /*
 781  * The helper to calc the free space on the devices that can be used to store
 782  * file data.
 783  */
 784 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
 785 {
 786         struct btrfs_fs_info *fs_info = root->fs_info;
 787         struct btrfs_device_info *devices_info;
 788         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 789         struct btrfs_device *device;
 790         u64 skip_space;
 791         u64 type;
 792         u64 avail_space;
 793         u64 used_space;
 794         u64 min_stripe_size;
 795         int min_stripes = 1;
 796         int i = 0, nr_devices;
 797         int ret;
 798
 799         nr_devices = fs_info->fs_devices->rw_devices;
 800         BUG_ON(!nr_devices);
 801
 802         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
 803                                GFP_NOFS);
 804         if (!devices_info)
 805                 return -ENOMEM;
 806
 807         /* calc min stripe number for data space alloction */
 808         type = btrfs_get_alloc_profile(root, 1);
 809         if (type & BTRFS_BLOCK_GROUP_RAID0)
 810                 min_stripes = 2;
 811         else if (type & BTRFS_BLOCK_GROUP_RAID1)
 812                 min_stripes = 2;
 813         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 814                 min_stripes = 4;
 815
 816         if (type & BTRFS_BLOCK_GROUP_DUP)
 817                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
 818         else
 819                 min_stripe_size = BTRFS_STRIPE_LEN;
 820
 821         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
 822                 if (!device->in_fs_metadata)
 823                         continue;
 824
 825                 avail_space = device->total_bytes - device->bytes_used;
 826
 827                 /* align with stripe_len */
 828                 do_div(avail_space, BTRFS_STRIPE_LEN);
 829                 avail_space *= BTRFS_STRIPE_LEN;
 830
 831                 /*
 832                  * In order to avoid overwritting the superblock on the drive,
 833                  * btrfs starts at an offset of at least 1MB when doing chunk
 834                  * allocation.
 835                  */
 836                 skip_space = 1024 * 1024;
 837
 838                 /* user can set the offset in fs_info->alloc_start. */
 839                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
 840                     device->total_bytes)
 841                         skip_space = max(fs_info->alloc_start, skip_space);
 842
 843                 /*
 844                  * btrfs can not use the free space in [0, skip_space - 1],
 845                  * we must subtract it from the total. In order to implement
 846                  * it, we account the used space in this range first.
 847                  */
 848                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
 849                                                      &used_space);
 850                 if (ret) {
 851                         kfree(devices_info);
 852                         return ret;
 853                 }
 854
 855                 /* calc the free space in [0, skip_space - 1] */
 856                 skip_space -= used_space;
 857
 858                 /*
 859                  * we can use the free space in [0, skip_space - 1], subtract
 860                  * it from the total.
 861                  */
 862                 if (avail_space && avail_space >= skip_space)
 863                         avail_space -= skip_space;
 864                 else
 865                         avail_space = 0;
 866
 867                 if (avail_space < min_stripe_size)
 868                         continue;
 869
 870                 devices_info[i].dev = device;
 871                 devices_info[i].max_avail = avail_space;
 872
 873                 i++;
 874         }
 875
 876         nr_devices = i;
 877
 878         btrfs_descending_sort_devices(devices_info, nr_devices);
 879
 880         i = nr_devices - 1;
 881         avail_space = 0;
 882         while (nr_devices >= min_stripes) {
 883                 if (devices_info[i].max_avail >= min_stripe_size) {
 884                         int j;
 885                         u64 alloc_size;
 886
 887                         avail_space += devices_info[i].max_avail * min_stripes;
 888                         alloc_size = devices_info[i].max_avail;
 889                         for (j = i + 1 - min_stripes; j <= i; j++)
 890                                 devices_info[j].max_avail -= alloc_size;
 891                 }
 892                 i--;
 893                 nr_devices--;
 894         }
 895
 896         kfree(devices_info);
 897         *free_bytes = avail_space;
 898         return 0;
 899 }
 900
 901 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 902 {
 903         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
 904         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
 905         struct list_head *head = &root->fs_info->space_info;
 906         struct btrfs_space_info *found;
 907         u64 total_used = 0;
 908         u64 total_free_data = 0;
 909         int bits = dentry->d_sb->s_blocksize_bits;
 910         __be32 *fsid = (__be32 *)root->fs_info->fsid;
 911         int ret;
 912
 913         /* holding chunk_muext to avoid allocating new chunks */
 914         mutex_lock(&root->fs_info->chunk_mutex);
 915         rcu_read_lock();
 916         list_for_each_entry_rcu(found, head, list) {
 917                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
 918                         total_free_data += found->disk_total - found->disk_used;
 919                         total_free_data -=
 920                                 btrfs_account_ro_block_groups_free_space(found);
 921                 }
 922
 923                 total_used += found->disk_used;
 924         }
 925         rcu_read_unlock();
 926
 927         buf->f_namelen = BTRFS_NAME_LEN;
 928         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
 929         buf->f_bfree = buf->f_blocks - (total_used >> bits);
 930         buf->f_bsize = dentry->d_sb->s_blocksize;
 931         buf->f_type = BTRFS_SUPER_MAGIC;
 932         buf->f_bavail = total_free_data;
 933         ret = btrfs_calc_avail_data_space(root, &total_free_data);
 934         if (ret) {
 935                 mutex_unlock(&root->fs_info->chunk_mutex);
 936                 return ret;
 937         }
 938         buf->f_bavail += total_free_data;
 939         buf->f_bavail = buf->f_bavail >> bits;
 940         mutex_unlock(&root->fs_info->chunk_mutex);
 941
 942         /* We treat it as constant endianness (it doesn't matter _which_)
 943            because we want the fsid to come out the same whether mounted
 944            on a big-endian or little-endian host */
 945         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
 946         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
 947         /* Mask in the root object ID too, to disambiguate subvols */
 948         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
 949         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
 950
 951         return 0;
 952 }
 953
 954 static struct file_system_type btrfs_fs_type = {
 955         .owner          = THIS_MODULE,
 956         .name           = "btrfs",
 957         .get_sb         = btrfs_get_sb,
 958         .kill_sb        = kill_anon_super,
 959         .fs_flags       = FS_REQUIRES_DEV,
 960 };
 961
 962 /*
 963  * used by btrfsctl to scan devices when no FS is mounted
 964  */
 965 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
 966                                 unsigned long arg)
 967 {
 968         struct btrfs_ioctl_vol_args *vol;
 969         struct btrfs_fs_devices *fs_devices;
 970         int ret = -ENOTTY;
 971
 972         if (!capable(CAP_SYS_ADMIN))
 973                 return -EPERM;
 974
 975         vol = memdup_user((void __user *)arg, sizeof(*vol));
 976         if (IS_ERR(vol))
 977                 return PTR_ERR(vol);
 978
 979         switch (cmd) {
 980         case BTRFS_IOC_SCAN_DEV:
 981                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
 982                                             &btrfs_fs_type, &fs_devices);
 983                 break;
 984         }
 985
 986         kfree(vol);
 987         return ret;
 988 }
 989
 990 static int btrfs_freeze(struct super_block *sb)
 991 {
 992         struct btrfs_root *root = btrfs_sb(sb);
 993         mutex_lock(&root->fs_info->transaction_kthread_mutex);
 994         mutex_lock(&root->fs_info->cleaner_mutex);
 995         return 0;
 996 }
 997
 998 static int btrfs_unfreeze(struct super_block *sb)
 999 {
1000         struct btrfs_root *root = btrfs_sb(sb);
1001         mutex_unlock(&root->fs_info->cleaner_mutex);
1002         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1003         return 0;
1004 }
1005
1006 static const struct super_operations btrfs_super_ops = {
1007         .drop_inode     = btrfs_drop_inode,
1008         .evict_inode    = btrfs_evict_inode,
1009         .put_super      = btrfs_put_super,
1010         .sync_fs        = btrfs_sync_fs,
1011         .show_options   = btrfs_show_options,
1012         .write_inode    = btrfs_write_inode,
1013         .dirty_inode    = btrfs_dirty_inode,
1014         .alloc_inode    = btrfs_alloc_inode,
1015         .destroy_inode  = btrfs_destroy_inode,
1016         .statfs         = btrfs_statfs,
1017         .remount_fs     = btrfs_remount,
1018         .freeze_fs      = btrfs_freeze,
1019         .unfreeze_fs    = btrfs_unfreeze,
1020 };
1021
1022 static const struct file_operations btrfs_ctl_fops = {
1023         .unlocked_ioctl  = btrfs_control_ioctl,
1024         .compat_ioctl = btrfs_control_ioctl,
1025         .owner   = THIS_MODULE,
1026 };
1027
1028 static struct miscdevice btrfs_misc = {
1029         .minor          = BTRFS_MINOR,
1030         .name           = "btrfs-control",
1031         .fops           = &btrfs_ctl_fops
1032 };
1033
1034 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1035 MODULE_ALIAS("devname:btrfs-control");
1036
1037 static int btrfs_interface_init(void)
1038 {
1039         return misc_register(&btrfs_misc);
1040 }
1041
1042 static void btrfs_interface_exit(void)
1043 {
1044         if (misc_deregister(&btrfs_misc) < 0)
1045                 printk(KERN_INFO "misc_deregister failed for control device");
1046 }
1047
1048 static int __init init_btrfs_fs(void)
1049 {
1050         int err;
1051
1052         err = btrfs_init_sysfs();
1053         if (err)
1054                 return err;
1055
1056         err = btrfs_init_compress();
1057         if (err)
1058                 goto free_sysfs;
1059
1060         err = btrfs_init_cachep();
1061         if (err)
1062                 goto free_compress;
1063
1064         err = extent_io_init();
1065         if (err)
1066                 goto free_cachep;
1067
1068         err = extent_map_init();
1069         if (err)
1070                 goto free_extent_io;
1071
1072         err = btrfs_interface_init();
1073         if (err)
1074                 goto free_extent_map;
1075
1076         err = register_filesystem(&btrfs_fs_type);
1077         if (err)
1078                 goto unregister_ioctl;
1079
1080         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1081         return 0;
1082
1083 unregister_ioctl:
1084         btrfs_interface_exit();
1085 free_extent_map:
1086         extent_map_exit();
1087 free_extent_io:
1088         extent_io_exit();
1089 free_cachep:
1090         btrfs_destroy_cachep();
1091 free_compress:
1092         btrfs_exit_compress();
1093 free_sysfs:
1094         btrfs_exit_sysfs();
1095         return err;
1096 }
1097
1098 static void __exit exit_btrfs_fs(void)
1099 {
1100         btrfs_destroy_cachep();
1101         extent_map_exit();
1102         extent_io_exit();
1103         btrfs_interface_exit();
1104         unregister_filesystem(&btrfs_fs_type);
1105         btrfs_exit_sysfs();
1106         btrfs_cleanup_fs_uuids();
1107         btrfs_exit_compress();
1108 }
1109
1110 module_init(init_btrfs_fs)
1111 module_exit(exit_btrfs_fs)
1112
1113 MODULE_LICENSE("GPL");