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 const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
  58                                       char nbuf[16])
  59 {
  60         char *errstr = NULL;
  61
  62         switch (errno) {
  63         case -EIO:
  64                 errstr = "IO failure";
  65                 break;
  66         case -ENOMEM:
  67                 errstr = "Out of memory";
  68                 break;
  69         case -EROFS:
  70                 errstr = "Readonly filesystem";
  71                 break;
  72         default:
  73                 if (nbuf) {
  74                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  75                                 errstr = nbuf;
  76                 }
  77                 break;
  78         }
  79
  80         return errstr;
  81 }
  82
  83 static void __save_error_info(struct btrfs_fs_info *fs_info)
  84 {
  85         /*
  86          * today we only save the error info into ram.  Long term we'll
  87          * also send it down to the disk
  88          */
  89         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
  90 }
  91
  92 /* NOTE:
  93  *      We move write_super stuff at umount in order to avoid deadlock
  94  *      for umount hold all lock.
  95  */
  96 static void save_error_info(struct btrfs_fs_info *fs_info)
  97 {
  98         __save_error_info(fs_info);
  99 }
 100
 101 /* btrfs handle error by forcing the filesystem readonly */
 102 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 103 {
 104         struct super_block *sb = fs_info->sb;
 105
 106         if (sb->s_flags & MS_RDONLY)
 107                 return;
 108
 109         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 110                 sb->s_flags |= MS_RDONLY;
 111                 printk(KERN_INFO "btrfs is forced readonly\n");
 112         }
 113 }
 114
 115 /*
 116  * __btrfs_std_error decodes expected errors from the caller and
 117  * invokes the approciate error response.
 118  */
 119 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 120                      unsigned int line, int errno)
 121 {
 122         struct super_block *sb = fs_info->sb;
 123         char nbuf[16];
 124         const char *errstr;
 125
 126         /*
 127          * Special case: if the error is EROFS, and we're already
 128          * under MS_RDONLY, then it is safe here.
 129          */
 130         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 131                 return;
 132
 133         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 134         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 135                 sb->s_id, function, line, errstr);
 136         save_error_info(fs_info);
 137
 138         btrfs_handle_error(fs_info);
 139 }
 140
 141 static void btrfs_put_super(struct super_block *sb)
 142 {
 143         struct btrfs_root *root = btrfs_sb(sb);
 144         int ret;
 145
 146         ret = close_ctree(root);
 147         sb->s_fs_info = NULL;
 148
 149         (void)ret; /* FIXME: need to fix VFS to return error? */
 150 }
 151
 152 enum {
 153         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 154         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 155         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 156         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 157         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 158         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed, Opt_err,
 159 };
 160
 161 static match_table_t tokens = {
 162         {Opt_degraded, "degraded"},
 163         {Opt_subvol, "subvol=%s"},
 164         {Opt_subvolid, "subvolid=%d"},
 165         {Opt_device, "device=%s"},
 166         {Opt_nodatasum, "nodatasum"},
 167         {Opt_nodatacow, "nodatacow"},
 168         {Opt_nobarrier, "nobarrier"},
 169         {Opt_max_inline, "max_inline=%s"},
 170         {Opt_alloc_start, "alloc_start=%s"},
 171         {Opt_thread_pool, "thread_pool=%d"},
 172         {Opt_compress, "compress"},
 173         {Opt_compress_type, "compress=%s"},
 174         {Opt_compress_force, "compress-force"},
 175         {Opt_compress_force_type, "compress-force=%s"},
 176         {Opt_ssd, "ssd"},
 177         {Opt_ssd_spread, "ssd_spread"},
 178         {Opt_nossd, "nossd"},
 179         {Opt_noacl, "noacl"},
 180         {Opt_notreelog, "notreelog"},
 181         {Opt_flushoncommit, "flushoncommit"},
 182         {Opt_ratio, "metadata_ratio=%d"},
 183         {Opt_discard, "discard"},
 184         {Opt_space_cache, "space_cache"},
 185         {Opt_clear_cache, "clear_cache"},
 186         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 187         {Opt_err, NULL},
 188 };
 189
 190 /*
 191  * Regular mount options parser.  Everything that is needed only when
 192  * reading in a new superblock is parsed here.
 193  */
 194 int btrfs_parse_options(struct btrfs_root *root, char *options)
 195 {
 196         struct btrfs_fs_info *info = root->fs_info;
 197         substring_t args[MAX_OPT_ARGS];
 198         char *p, *num, *orig;
 199         int intarg;
 200         int ret = 0;
 201         char *compress_type;
 202         bool compress_force = false;
 203
 204         if (!options)
 205                 return 0;
 206
 207         /*
 208          * strsep changes the string, duplicate it because parse_options
 209          * gets called twice
 210          */
 211         options = kstrdup(options, GFP_NOFS);
 212         if (!options)
 213                 return -ENOMEM;
 214
 215         orig = options;
 216
 217         while ((p = strsep(&options, ",")) != NULL) {
 218                 int token;
 219                 if (!*p)
 220                         continue;
 221
 222                 token = match_token(p, tokens, args);
 223                 switch (token) {
 224                 case Opt_degraded:
 225                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 226                         btrfs_set_opt(info->mount_opt, DEGRADED);
 227                         break;
 228                 case Opt_subvol:
 229                 case Opt_subvolid:
 230                 case Opt_device:
 231                         /*
 232                          * These are parsed by btrfs_parse_early_options
 233                          * and can be happily ignored here.
 234                          */
 235                         break;
 236                 case Opt_nodatasum:
 237                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 238                         btrfs_set_opt(info->mount_opt, NODATASUM);
 239                         break;
 240                 case Opt_nodatacow:
 241                         printk(KERN_INFO "btrfs: setting nodatacow\n");
 242                         btrfs_set_opt(info->mount_opt, NODATACOW);
 243                         btrfs_set_opt(info->mount_opt, NODATASUM);
 244                         break;
 245                 case Opt_compress_force:
 246                 case Opt_compress_force_type:
 247                         compress_force = true;
 248                 case Opt_compress:
 249                 case Opt_compress_type:
 250                         if (token == Opt_compress ||
 251                             token == Opt_compress_force ||
 252                             strcmp(args[0].from, "zlib") == 0) {
 253                                 compress_type = "zlib";
 254                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 255                         } else if (strcmp(args[0].from, "lzo") == 0) {
 256                                 compress_type = "lzo";
 257                                 info->compress_type = BTRFS_COMPRESS_LZO;
 258                         } else {
 259                                 ret = -EINVAL;
 260                                 goto out;
 261                         }
 262
 263                         btrfs_set_opt(info->mount_opt, COMPRESS);
 264                         if (compress_force) {
 265                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 266                                 pr_info("btrfs: force %s compression\n",
 267                                         compress_type);
 268                         } else
 269                                 pr_info("btrfs: use %s compression\n",
 270                                         compress_type);
 271                         break;
 272                 case Opt_ssd:
 273                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 274                         btrfs_set_opt(info->mount_opt, SSD);
 275                         break;
 276                 case Opt_ssd_spread:
 277                         printk(KERN_INFO "btrfs: use spread ssd "
 278                                "allocation scheme\n");
 279                         btrfs_set_opt(info->mount_opt, SSD);
 280                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 281                         break;
 282                 case Opt_nossd:
 283                         printk(KERN_INFO "btrfs: not using ssd allocation "
 284                                "scheme\n");
 285                         btrfs_set_opt(info->mount_opt, NOSSD);
 286                         btrfs_clear_opt(info->mount_opt, SSD);
 287                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 288                         break;
 289                 case Opt_nobarrier:
 290                         printk(KERN_INFO "btrfs: turning off barriers\n");
 291                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 292                         break;
 293                 case Opt_thread_pool:
 294                         intarg = 0;
 295                         match_int(&args[0], &intarg);
 296                         if (intarg) {
 297                                 info->thread_pool_size = intarg;
 298                                 printk(KERN_INFO "btrfs: thread pool %d\n",
 299                                        info->thread_pool_size);
 300                         }
 301                         break;
 302                 case Opt_max_inline:
 303                         num = match_strdup(&args[0]);
 304                         if (num) {
 305                                 info->max_inline = memparse(num, NULL);
 306                                 kfree(num);
 307
 308                                 if (info->max_inline) {
 309                                         info->max_inline = max_t(u64,
 310                                                 info->max_inline,
 311                                                 root->sectorsize);
 312                                 }
 313                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 314                                         (unsigned long long)info->max_inline);
 315                         }
 316                         break;
 317                 case Opt_alloc_start:
 318                         num = match_strdup(&args[0]);
 319                         if (num) {
 320                                 info->alloc_start = memparse(num, NULL);
 321                                 kfree(num);
 322                                 printk(KERN_INFO
 323                                         "btrfs: allocations start at %llu\n",
 324                                         (unsigned long long)info->alloc_start);
 325                         }
 326                         break;
 327                 case Opt_noacl:
 328                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 329                         break;
 330                 case Opt_notreelog:
 331                         printk(KERN_INFO "btrfs: disabling tree log\n");
 332                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 333                         break;
 334                 case Opt_flushoncommit:
 335                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 336                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 337                         break;
 338                 case Opt_ratio:
 339                         intarg = 0;
 340                         match_int(&args[0], &intarg);
 341                         if (intarg) {
 342                                 info->metadata_ratio = intarg;
 343                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 344                                        info->metadata_ratio);
 345                         }
 346                         break;
 347                 case Opt_discard:
 348                         btrfs_set_opt(info->mount_opt, DISCARD);
 349                         break;
 350                 case Opt_space_cache:
 351                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
 352                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 353                         break;
 354                 case Opt_clear_cache:
 355                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 356                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 357                         break;
 358                 case Opt_user_subvol_rm_allowed:
 359                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 360                         break;
 361                 case Opt_err:
 362                         printk(KERN_INFO "btrfs: unrecognized mount option "
 363                                "'%s'\n", p);
 364                         ret = -EINVAL;
 365                         goto out;
 366                 default:
 367                         break;
 368                 }
 369         }
 370 out:
 371         kfree(orig);
 372         return ret;
 373 }
 374
 375 /*
 376  * Parse mount options that are required early in the mount process.
 377  *
 378  * All other options will be parsed on much later in the mount process and
 379  * only when we need to allocate a new super block.
 380  */
 381 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 382                 void *holder, char **subvol_name, u64 *subvol_objectid,
 383                 struct btrfs_fs_devices **fs_devices)
 384 {
 385         substring_t args[MAX_OPT_ARGS];
 386         char *opts, *orig, *p;
 387         int error = 0;
 388         int intarg;
 389
 390         if (!options)
 391                 goto out;
 392
 393         /*
 394          * strsep changes the string, duplicate it because parse_options
 395          * gets called twice
 396          */
 397         opts = kstrdup(options, GFP_KERNEL);
 398         if (!opts)
 399                 return -ENOMEM;
 400         orig = opts;
 401
 402         while ((p = strsep(&opts, ",")) != NULL) {
 403                 int token;
 404                 if (!*p)
 405                         continue;
 406
 407                 token = match_token(p, tokens, args);
 408                 switch (token) {
 409                 case Opt_subvol:
 410                         *subvol_name = match_strdup(&args[0]);
 411                         break;
 412                 case Opt_subvolid:
 413                         intarg = 0;
 414                         error = match_int(&args[0], &intarg);
 415                         if (!error) {
 416                                 /* we want the original fs_tree */
 417                                 if (!intarg)
 418                                         *subvol_objectid =
 419                                                 BTRFS_FS_TREE_OBJECTID;
 420                                 else
 421                                         *subvol_objectid = intarg;
 422                         }
 423                         break;
 424                 case Opt_device:
 425                         error = btrfs_scan_one_device(match_strdup(&args[0]),
 426                                         flags, holder, fs_devices);
 427                         if (error)
 428                                 goto out_free_opts;
 429                         break;
 430                 default:
 431                         break;
 432                 }
 433         }
 434
 435  out_free_opts:
 436         kfree(orig);
 437  out:
 438         /*
 439          * If no subvolume name is specified we use the default one.  Allocate
 440          * a copy of the string "." here so that code later in the
 441          * mount path doesn't care if it's the default volume or another one.
 442          */
 443         if (!*subvol_name) {
 444                 *subvol_name = kstrdup(".", GFP_KERNEL);
 445                 if (!*subvol_name)
 446                         return -ENOMEM;
 447         }
 448         return error;
 449 }
 450
 451 static struct dentry *get_default_root(struct super_block *sb,
 452                                        u64 subvol_objectid)
 453 {
 454         struct btrfs_root *root = sb->s_fs_info;
 455         struct btrfs_root *new_root;
 456         struct btrfs_dir_item *di;
 457         struct btrfs_path *path;
 458         struct btrfs_key location;
 459         struct inode *inode;
 460         struct dentry *dentry;
 461         u64 dir_id;
 462         int new = 0;
 463
 464         /*
 465          * We have a specific subvol we want to mount, just setup location and
 466          * go look up the root.
 467          */
 468         if (subvol_objectid) {
 469                 location.objectid = subvol_objectid;
 470                 location.type = BTRFS_ROOT_ITEM_KEY;
 471                 location.offset = (u64)-1;
 472                 goto find_root;
 473         }
 474
 475         path = btrfs_alloc_path();
 476         if (!path)
 477                 return ERR_PTR(-ENOMEM);
 478         path->leave_spinning = 1;
 479
 480         /*
 481          * Find the "default" dir item which points to the root item that we
 482          * will mount by default if we haven't been given a specific subvolume
 483          * to mount.
 484          */
 485         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
 486         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 487         if (IS_ERR(di))
 488                 return ERR_CAST(di);
 489         if (!di) {
 490                 /*
 491                  * Ok the default dir item isn't there.  This is weird since
 492                  * it's always been there, but don't freak out, just try and
 493                  * mount to root most subvolume.
 494                  */
 495                 btrfs_free_path(path);
 496                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 497                 new_root = root->fs_info->fs_root;
 498                 goto setup_root;
 499         }
 500
 501         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 502         btrfs_free_path(path);
 503
 504 find_root:
 505         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
 506         if (IS_ERR(new_root))
 507                 return ERR_CAST(new_root);
 508
 509         if (btrfs_root_refs(&new_root->root_item) == 0)
 510                 return ERR_PTR(-ENOENT);
 511
 512         dir_id = btrfs_root_dirid(&new_root->root_item);
 513 setup_root:
 514         location.objectid = dir_id;
 515         location.type = BTRFS_INODE_ITEM_KEY;
 516         location.offset = 0;
 517
 518         inode = btrfs_iget(sb, &location, new_root, &new);
 519         if (IS_ERR(inode))
 520                 return ERR_CAST(inode);
 521
 522         /*
 523          * If we're just mounting the root most subvol put the inode and return
 524          * a reference to the dentry.  We will have already gotten a reference
 525          * to the inode in btrfs_fill_super so we're good to go.
 526          */
 527         if (!new && sb->s_root->d_inode == inode) {
 528                 iput(inode);
 529                 return dget(sb->s_root);
 530         }
 531
 532         if (new) {
 533                 const struct qstr name = { .name = "/", .len = 1 };
 534
 535                 /*
 536                  * New inode, we need to make the dentry a sibling of s_root so
 537                  * everything gets cleaned up properly on unmount.
 538                  */
 539                 dentry = d_alloc(sb->s_root, &name);
 540                 if (!dentry) {
 541                         iput(inode);
 542                         return ERR_PTR(-ENOMEM);
 543                 }
 544                 d_splice_alias(inode, dentry);
 545         } else {
 546                 /*
 547                  * We found the inode in cache, just find a dentry for it and
 548                  * put the reference to the inode we just got.
 549                  */
 550                 dentry = d_find_alias(inode);
 551                 iput(inode);
 552         }
 553
 554         return dentry;
 555 }
 556
 557 static int btrfs_fill_super(struct super_block *sb,
 558                             struct btrfs_fs_devices *fs_devices,
 559                             void *data, int silent)
 560 {
 561         struct inode *inode;
 562         struct dentry *root_dentry;
 563         struct btrfs_root *tree_root;
 564         struct btrfs_key key;
 565         int err;
 566
 567         sb->s_maxbytes = MAX_LFS_FILESIZE;
 568         sb->s_magic = BTRFS_SUPER_MAGIC;
 569         sb->s_op = &btrfs_super_ops;
 570         sb->s_export_op = &btrfs_export_ops;
 571         sb->s_xattr = btrfs_xattr_handlers;
 572         sb->s_time_gran = 1;
 573 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 574         sb->s_flags |= MS_POSIXACL;
 575 #endif
 576
 577         tree_root = open_ctree(sb, fs_devices, (char *)data);
 578
 579         if (IS_ERR(tree_root)) {
 580                 printk("btrfs: open_ctree failed\n");
 581                 return PTR_ERR(tree_root);
 582         }
 583         sb->s_fs_info = tree_root;
 584
 585         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 586         key.type = BTRFS_INODE_ITEM_KEY;
 587         key.offset = 0;
 588         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
 589         if (IS_ERR(inode)) {
 590                 err = PTR_ERR(inode);
 591                 goto fail_close;
 592         }
 593
 594         root_dentry = d_alloc_root(inode);
 595         if (!root_dentry) {
 596                 iput(inode);
 597                 err = -ENOMEM;
 598                 goto fail_close;
 599         }
 600
 601         sb->s_root = root_dentry;
 602
 603         save_mount_options(sb, data);
 604         return 0;
 605
 606 fail_close:
 607         close_ctree(tree_root);
 608         return err;
 609 }
 610
 611 int btrfs_sync_fs(struct super_block *sb, int wait)
 612 {
 613         struct btrfs_trans_handle *trans;
 614         struct btrfs_root *root = btrfs_sb(sb);
 615         int ret;
 616
 617         if (!wait) {
 618                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 619                 return 0;
 620         }
 621
 622         btrfs_start_delalloc_inodes(root, 0);
 623         btrfs_wait_ordered_extents(root, 0, 0);
 624
 625         trans = btrfs_start_transaction(root, 0);
 626         if (IS_ERR(trans))
 627                 return PTR_ERR(trans);
 628         ret = btrfs_commit_transaction(trans, root);
 629         return ret;
 630 }
 631
 632 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 633 {
 634         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
 635         struct btrfs_fs_info *info = root->fs_info;
 636
 637         if (btrfs_test_opt(root, DEGRADED))
 638                 seq_puts(seq, ",degraded");
 639         if (btrfs_test_opt(root, NODATASUM))
 640                 seq_puts(seq, ",nodatasum");
 641         if (btrfs_test_opt(root, NODATACOW))
 642                 seq_puts(seq, ",nodatacow");
 643         if (btrfs_test_opt(root, NOBARRIER))
 644                 seq_puts(seq, ",nobarrier");
 645         if (info->max_inline != 8192 * 1024)
 646                 seq_printf(seq, ",max_inline=%llu",
 647                            (unsigned long long)info->max_inline);
 648         if (info->alloc_start != 0)
 649                 seq_printf(seq, ",alloc_start=%llu",
 650                            (unsigned long long)info->alloc_start);
 651         if (info->thread_pool_size !=  min_t(unsigned long,
 652                                              num_online_cpus() + 2, 8))
 653                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 654         if (btrfs_test_opt(root, COMPRESS))
 655                 seq_puts(seq, ",compress");
 656         if (btrfs_test_opt(root, NOSSD))
 657                 seq_puts(seq, ",nossd");
 658         if (btrfs_test_opt(root, SSD_SPREAD))
 659                 seq_puts(seq, ",ssd_spread");
 660         else if (btrfs_test_opt(root, SSD))
 661                 seq_puts(seq, ",ssd");
 662         if (btrfs_test_opt(root, NOTREELOG))
 663                 seq_puts(seq, ",notreelog");
 664         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 665                 seq_puts(seq, ",flushoncommit");
 666         if (btrfs_test_opt(root, DISCARD))
 667                 seq_puts(seq, ",discard");
 668         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 669                 seq_puts(seq, ",noacl");
 670         return 0;
 671 }
 672
 673 static int btrfs_test_super(struct super_block *s, void *data)
 674 {
 675         struct btrfs_root *test_root = data;
 676         struct btrfs_root *root = btrfs_sb(s);
 677
 678         /*
 679          * If this super block is going away, return false as it
 680          * can't match as an existing super block.
 681          */
 682         if (!atomic_read(&s->s_active))
 683                 return 0;
 684         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
 685 }
 686
 687 static int btrfs_set_super(struct super_block *s, void *data)
 688 {
 689         s->s_fs_info = data;
 690
 691         return set_anon_super(s, data);
 692 }
 693
 694
 695 /*
 696  * Find a superblock for the given device / mount point.
 697  *
 698  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
 699  *        for multiple device setup.  Make sure to keep it in sync.
 700  */
 701 static int btrfs_get_sb(struct file_system_type *fs_type, int flags,
 702                 const char *dev_name, void *data, struct vfsmount *mnt)
 703 {
 704         struct block_device *bdev = NULL;
 705         struct super_block *s;
 706         struct dentry *root;
 707         struct btrfs_fs_devices *fs_devices = NULL;
 708         struct btrfs_root *tree_root = NULL;
 709         struct btrfs_fs_info *fs_info = NULL;
 710         fmode_t mode = FMODE_READ;
 711         char *subvol_name = NULL;
 712         u64 subvol_objectid = 0;
 713         int error = 0;
 714
 715         if (!(flags & MS_RDONLY))
 716                 mode |= FMODE_WRITE;
 717
 718         error = btrfs_parse_early_options(data, mode, fs_type,
 719                                           &subvol_name, &subvol_objectid,
 720                                           &fs_devices);
 721         if (error)
 722                 return error;
 723
 724         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
 725         if (error)
 726                 goto error_free_subvol_name;
 727
 728         error = btrfs_open_devices(fs_devices, mode, fs_type);
 729         if (error)
 730                 goto error_free_subvol_name;
 731
 732         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
 733                 error = -EACCES;
 734                 goto error_close_devices;
 735         }
 736
 737         /*
 738          * Setup a dummy root and fs_info for test/set super.  This is because
 739          * we don't actually fill this stuff out until open_ctree, but we need
 740          * it for searching for existing supers, so this lets us do that and
 741          * then open_ctree will properly initialize everything later.
 742          */
 743         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
 744         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
 745         if (!fs_info || !tree_root) {
 746                 error = -ENOMEM;
 747                 goto error_close_devices;
 748         }
 749         fs_info->tree_root = tree_root;
 750         fs_info->fs_devices = fs_devices;
 751         tree_root->fs_info = fs_info;
 752
 753         bdev = fs_devices->latest_bdev;
 754         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
 755         if (IS_ERR(s))
 756                 goto error_s;
 757
 758         if (s->s_root) {
 759                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 760                         deactivate_locked_super(s);
 761                         error = -EBUSY;
 762                         goto error_close_devices;
 763                 }
 764
 765                 btrfs_close_devices(fs_devices);
 766                 kfree(fs_info);
 767                 kfree(tree_root);
 768         } else {
 769                 char b[BDEVNAME_SIZE];
 770
 771                 s->s_flags = flags;
 772                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 773                 error = btrfs_fill_super(s, fs_devices, data,
 774                                          flags & MS_SILENT ? 1 : 0);
 775                 if (error) {
 776                         deactivate_locked_super(s);
 777                         goto error_free_subvol_name;
 778                 }
 779
 780                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
 781                 s->s_flags |= MS_ACTIVE;
 782         }
 783
 784         root = get_default_root(s, subvol_objectid);
 785         if (IS_ERR(root)) {
 786                 error = PTR_ERR(root);
 787                 deactivate_locked_super(s);
 788                 goto error_free_subvol_name;
 789         }
 790         /* if they gave us a subvolume name bind mount into that */
 791         if (strcmp(subvol_name, ".")) {
 792                 struct dentry *new_root;
 793                 mutex_lock(&root->d_inode->i_mutex);
 794                 new_root = lookup_one_len(subvol_name, root,
 795                                       strlen(subvol_name));
 796                 mutex_unlock(&root->d_inode->i_mutex);
 797
 798                 if (IS_ERR(new_root)) {
 799                         dput(root);
 800                         deactivate_locked_super(s);
 801                         error = PTR_ERR(new_root);
 802                         goto error_free_subvol_name;
 803                 }
 804                 if (!new_root->d_inode) {
 805                         dput(root);
 806                         dput(new_root);
 807                         deactivate_locked_super(s);
 808                         error = -ENXIO;
 809                         goto error_free_subvol_name;
 810                 }
 811                 dput(root);
 812                 root = new_root;
 813         }
 814
 815         mnt->mnt_sb = s;
 816         mnt->mnt_root = root;
 817
 818         kfree(subvol_name);
 819         return 0;
 820
 821 error_s:
 822         error = PTR_ERR(s);
 823 error_close_devices:
 824         btrfs_close_devices(fs_devices);
 825         kfree(fs_info);
 826         kfree(tree_root);
 827 error_free_subvol_name:
 828         kfree(subvol_name);
 829         return error;
 830 }
 831
 832 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 833 {
 834         struct btrfs_root *root = btrfs_sb(sb);
 835         int ret;
 836
 837         ret = btrfs_parse_options(root, data);
 838         if (ret)
 839                 return -EINVAL;
 840
 841         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 842                 return 0;
 843
 844         if (*flags & MS_RDONLY) {
 845                 sb->s_flags |= MS_RDONLY;
 846
 847                 ret =  btrfs_commit_super(root);
 848                 WARN_ON(ret);
 849         } else {
 850                 if (root->fs_info->fs_devices->rw_devices == 0)
 851                         return -EACCES;
 852
 853                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
 854                         return -EINVAL;
 855
 856                 ret = btrfs_cleanup_fs_roots(root->fs_info);
 857                 WARN_ON(ret);
 858
 859                 /* recover relocation */
 860                 ret = btrfs_recover_relocation(root);
 861                 WARN_ON(ret);
 862
 863                 sb->s_flags &= ~MS_RDONLY;
 864         }
 865
 866         return 0;
 867 }
 868
 869 /*
 870  * The helper to calc the free space on the devices that can be used to store
 871  * file data.
 872  */
 873 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
 874 {
 875         struct btrfs_fs_info *fs_info = root->fs_info;
 876         struct btrfs_device_info *devices_info;
 877         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 878         struct btrfs_device *device;
 879         u64 skip_space;
 880         u64 type;
 881         u64 avail_space;
 882         u64 used_space;
 883         u64 min_stripe_size;
 884         int min_stripes = 1;
 885         int i = 0, nr_devices;
 886         int ret;
 887
 888         nr_devices = fs_info->fs_devices->rw_devices;
 889         BUG_ON(!nr_devices);
 890
 891         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
 892                                GFP_NOFS);
 893         if (!devices_info)
 894                 return -ENOMEM;
 895
 896         /* calc min stripe number for data space alloction */
 897         type = btrfs_get_alloc_profile(root, 1);
 898         if (type & BTRFS_BLOCK_GROUP_RAID0)
 899                 min_stripes = 2;
 900         else if (type & BTRFS_BLOCK_GROUP_RAID1)
 901                 min_stripes = 2;
 902         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 903                 min_stripes = 4;
 904
 905         if (type & BTRFS_BLOCK_GROUP_DUP)
 906                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
 907         else
 908                 min_stripe_size = BTRFS_STRIPE_LEN;
 909
 910         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
 911                 if (!device->in_fs_metadata)
 912                         continue;
 913
 914                 avail_space = device->total_bytes - device->bytes_used;
 915
 916                 /* align with stripe_len */
 917                 do_div(avail_space, BTRFS_STRIPE_LEN);
 918                 avail_space *= BTRFS_STRIPE_LEN;
 919
 920                 /*
 921                  * In order to avoid overwritting the superblock on the drive,
 922                  * btrfs starts at an offset of at least 1MB when doing chunk
 923                  * allocation.
 924                  */
 925                 skip_space = 1024 * 1024;
 926
 927                 /* user can set the offset in fs_info->alloc_start. */
 928                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
 929                     device->total_bytes)
 930                         skip_space = max(fs_info->alloc_start, skip_space);
 931
 932                 /*
 933                  * btrfs can not use the free space in [0, skip_space - 1],
 934                  * we must subtract it from the total. In order to implement
 935                  * it, we account the used space in this range first.
 936                  */
 937                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
 938                                                      &used_space);
 939                 if (ret) {
 940                         kfree(devices_info);
 941                         return ret;
 942                 }
 943
 944                 /* calc the free space in [0, skip_space - 1] */
 945                 skip_space -= used_space;
 946
 947                 /*
 948                  * we can use the free space in [0, skip_space - 1], subtract
 949                  * it from the total.
 950                  */
 951                 if (avail_space && avail_space >= skip_space)
 952                         avail_space -= skip_space;
 953                 else
 954                         avail_space = 0;
 955
 956                 if (avail_space < min_stripe_size)
 957                         continue;
 958
 959                 devices_info[i].dev = device;
 960                 devices_info[i].max_avail = avail_space;
 961
 962                 i++;
 963         }
 964
 965         nr_devices = i;
 966
 967         btrfs_descending_sort_devices(devices_info, nr_devices);
 968
 969         i = nr_devices - 1;
 970         avail_space = 0;
 971         while (nr_devices >= min_stripes) {
 972                 if (devices_info[i].max_avail >= min_stripe_size) {
 973                         int j;
 974                         u64 alloc_size;
 975
 976                         avail_space += devices_info[i].max_avail * min_stripes;
 977                         alloc_size = devices_info[i].max_avail;
 978                         for (j = i + 1 - min_stripes; j <= i; j++)
 979                                 devices_info[j].max_avail -= alloc_size;
 980                 }
 981                 i--;
 982                 nr_devices--;
 983         }
 984
 985         kfree(devices_info);
 986         *free_bytes = avail_space;
 987         return 0;
 988 }
 989
 990 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 991 {
 992         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
 993         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
 994         struct list_head *head = &root->fs_info->space_info;
 995         struct btrfs_space_info *found;
 996         u64 total_used = 0;
 997         u64 total_free_data = 0;
 998         int bits = dentry->d_sb->s_blocksize_bits;
 999         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1000         int ret;
1001
1002         /* holding chunk_muext to avoid allocating new chunks */
1003         mutex_lock(&root->fs_info->chunk_mutex);
1004         rcu_read_lock();
1005         list_for_each_entry_rcu(found, head, list) {
1006                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1007                         total_free_data += found->disk_total - found->disk_used;
1008                         total_free_data -=
1009                                 btrfs_account_ro_block_groups_free_space(found);
1010                 }
1011
1012                 total_used += found->disk_used;
1013         }
1014         rcu_read_unlock();
1015
1016         buf->f_namelen = BTRFS_NAME_LEN;
1017         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1018         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1019         buf->f_bsize = dentry->d_sb->s_blocksize;
1020         buf->f_type = BTRFS_SUPER_MAGIC;
1021         buf->f_bavail = total_free_data;
1022         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1023         if (ret) {
1024                 mutex_unlock(&root->fs_info->chunk_mutex);
1025                 return ret;
1026         }
1027         buf->f_bavail += total_free_data;
1028         buf->f_bavail = buf->f_bavail >> bits;
1029         mutex_unlock(&root->fs_info->chunk_mutex);
1030
1031         /* We treat it as constant endianness (it doesn't matter _which_)
1032            because we want the fsid to come out the same whether mounted
1033            on a big-endian or little-endian host */
1034         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1035         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1036         /* Mask in the root object ID too, to disambiguate subvols */
1037         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1038         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1039
1040         return 0;
1041 }
1042
1043 static struct file_system_type btrfs_fs_type = {
1044         .owner          = THIS_MODULE,
1045         .name           = "btrfs",
1046         .get_sb         = btrfs_get_sb,
1047         .kill_sb        = kill_anon_super,
1048         .fs_flags       = FS_REQUIRES_DEV,
1049 };
1050
1051 /*
1052  * used by btrfsctl to scan devices when no FS is mounted
1053  */
1054 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1055                                 unsigned long arg)
1056 {
1057         struct btrfs_ioctl_vol_args *vol;
1058         struct btrfs_fs_devices *fs_devices;
1059         int ret = -ENOTTY;
1060
1061         if (!capable(CAP_SYS_ADMIN))
1062                 return -EPERM;
1063
1064         vol = memdup_user((void __user *)arg, sizeof(*vol));
1065         if (IS_ERR(vol))
1066                 return PTR_ERR(vol);
1067
1068         switch (cmd) {
1069         case BTRFS_IOC_SCAN_DEV:
1070                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1071                                             &btrfs_fs_type, &fs_devices);
1072                 break;
1073         }
1074
1075         kfree(vol);
1076         return ret;
1077 }
1078
1079 static int btrfs_freeze(struct super_block *sb)
1080 {
1081         struct btrfs_root *root = btrfs_sb(sb);
1082         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1083         mutex_lock(&root->fs_info->cleaner_mutex);
1084         return 0;
1085 }
1086
1087 static int btrfs_unfreeze(struct super_block *sb)
1088 {
1089         struct btrfs_root *root = btrfs_sb(sb);
1090         mutex_unlock(&root->fs_info->cleaner_mutex);
1091         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1092         return 0;
1093 }
1094
1095 static const struct super_operations btrfs_super_ops = {
1096         .drop_inode     = btrfs_drop_inode,
1097         .evict_inode    = btrfs_evict_inode,
1098         .put_super      = btrfs_put_super,
1099         .sync_fs        = btrfs_sync_fs,
1100         .show_options   = btrfs_show_options,
1101         .write_inode    = btrfs_write_inode,
1102         .dirty_inode    = btrfs_dirty_inode,
1103         .alloc_inode    = btrfs_alloc_inode,
1104         .destroy_inode  = btrfs_destroy_inode,
1105         .statfs         = btrfs_statfs,
1106         .remount_fs     = btrfs_remount,
1107         .freeze_fs      = btrfs_freeze,
1108         .unfreeze_fs    = btrfs_unfreeze,
1109 };
1110
1111 static const struct file_operations btrfs_ctl_fops = {
1112         .unlocked_ioctl  = btrfs_control_ioctl,
1113         .compat_ioctl = btrfs_control_ioctl,
1114         .owner   = THIS_MODULE,
1115 };
1116
1117 static struct miscdevice btrfs_misc = {
1118         .minor          = BTRFS_MINOR,
1119         .name           = "btrfs-control",
1120         .fops           = &btrfs_ctl_fops
1121 };
1122
1123 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1124 MODULE_ALIAS("devname:btrfs-control");
1125
1126 static int btrfs_interface_init(void)
1127 {
1128         return misc_register(&btrfs_misc);
1129 }
1130
1131 static void btrfs_interface_exit(void)
1132 {
1133         if (misc_deregister(&btrfs_misc) < 0)
1134                 printk(KERN_INFO "misc_deregister failed for control device");
1135 }
1136
1137 static int __init init_btrfs_fs(void)
1138 {
1139         int err;
1140
1141         err = btrfs_init_sysfs();
1142         if (err)
1143                 return err;
1144
1145         err = btrfs_init_compress();
1146         if (err)
1147                 goto free_sysfs;
1148
1149         err = btrfs_init_cachep();
1150         if (err)
1151                 goto free_compress;
1152
1153         err = extent_io_init();
1154         if (err)
1155                 goto free_cachep;
1156
1157         err = extent_map_init();
1158         if (err)
1159                 goto free_extent_io;
1160
1161         err = btrfs_interface_init();
1162         if (err)
1163                 goto free_extent_map;
1164
1165         err = register_filesystem(&btrfs_fs_type);
1166         if (err)
1167                 goto unregister_ioctl;
1168
1169         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1170         return 0;
1171
1172 unregister_ioctl:
1173         btrfs_interface_exit();
1174 free_extent_map:
1175         extent_map_exit();
1176 free_extent_io:
1177         extent_io_exit();
1178 free_cachep:
1179         btrfs_destroy_cachep();
1180 free_compress:
1181         btrfs_exit_compress();
1182 free_sysfs:
1183         btrfs_exit_sysfs();
1184         return err;
1185 }
1186
1187 static void __exit exit_btrfs_fs(void)
1188 {
1189         btrfs_destroy_cachep();
1190         extent_map_exit();
1191         extent_io_exit();
1192         btrfs_interface_exit();
1193         unregister_filesystem(&btrfs_fs_type);
1194         btrfs_exit_sysfs();
1195         btrfs_cleanup_fs_uuids();
1196         btrfs_exit_compress();
1197 }
1198
1199 module_init(init_btrfs_fs)
1200 module_exit(exit_btrfs_fs)
1201
1202 MODULE_LICENSE("GPL");