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, *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
 401         while ((p = strsep(&opts, ",")) != NULL) {
 402                 int token;
 403                 if (!*p)
 404                         continue;
 405
 406                 token = match_token(p, tokens, args);
 407                 switch (token) {
 408                 case Opt_subvol:
 409                         *subvol_name = match_strdup(&args[0]);
 410                         break;
 411                 case Opt_subvolid:
 412                         intarg = 0;
 413                         error = match_int(&args[0], &intarg);
 414                         if (!error) {
 415                                 /* we want the original fs_tree */
 416                                 if (!intarg)
 417                                         *subvol_objectid =
 418                                                 BTRFS_FS_TREE_OBJECTID;
 419                                 else
 420                                         *subvol_objectid = intarg;
 421                         }
 422                         break;
 423                 case Opt_device:
 424                         error = btrfs_scan_one_device(match_strdup(&args[0]),
 425                                         flags, holder, fs_devices);
 426                         if (error)
 427                                 goto out_free_opts;
 428                         break;
 429                 default:
 430                         break;
 431                 }
 432         }
 433
 434  out_free_opts:
 435         kfree(opts);
 436  out:
 437         /*
 438          * If no subvolume name is specified we use the default one.  Allocate
 439          * a copy of the string "." here so that code later in the
 440          * mount path doesn't care if it's the default volume or another one.
 441          */
 442         if (!*subvol_name) {
 443                 *subvol_name = kstrdup(".", GFP_KERNEL);
 444                 if (!*subvol_name)
 445                         return -ENOMEM;
 446         }
 447         return error;
 448 }
 449
 450 static struct dentry *get_default_root(struct super_block *sb,
 451                                        u64 subvol_objectid)
 452 {
 453         struct btrfs_root *root = sb->s_fs_info;
 454         struct btrfs_root *new_root;
 455         struct btrfs_dir_item *di;
 456         struct btrfs_path *path;
 457         struct btrfs_key location;
 458         struct inode *inode;
 459         struct dentry *dentry;
 460         u64 dir_id;
 461         int new = 0;
 462
 463         /*
 464          * We have a specific subvol we want to mount, just setup location and
 465          * go look up the root.
 466          */
 467         if (subvol_objectid) {
 468                 location.objectid = subvol_objectid;
 469                 location.type = BTRFS_ROOT_ITEM_KEY;
 470                 location.offset = (u64)-1;
 471                 goto find_root;
 472         }
 473
 474         path = btrfs_alloc_path();
 475         if (!path)
 476                 return ERR_PTR(-ENOMEM);
 477         path->leave_spinning = 1;
 478
 479         /*
 480          * Find the "default" dir item which points to the root item that we
 481          * will mount by default if we haven't been given a specific subvolume
 482          * to mount.
 483          */
 484         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
 485         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 486         if (IS_ERR(di))
 487                 return ERR_CAST(di);
 488         if (!di) {
 489                 /*
 490                  * Ok the default dir item isn't there.  This is weird since
 491                  * it's always been there, but don't freak out, just try and
 492                  * mount to root most subvolume.
 493                  */
 494                 btrfs_free_path(path);
 495                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 496                 new_root = root->fs_info->fs_root;
 497                 goto setup_root;
 498         }
 499
 500         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 501         btrfs_free_path(path);
 502
 503 find_root:
 504         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
 505         if (IS_ERR(new_root))
 506                 return ERR_CAST(new_root);
 507
 508         if (btrfs_root_refs(&new_root->root_item) == 0)
 509                 return ERR_PTR(-ENOENT);
 510
 511         dir_id = btrfs_root_dirid(&new_root->root_item);
 512 setup_root:
 513         location.objectid = dir_id;
 514         location.type = BTRFS_INODE_ITEM_KEY;
 515         location.offset = 0;
 516
 517         inode = btrfs_iget(sb, &location, new_root, &new);
 518         if (IS_ERR(inode))
 519                 return ERR_CAST(inode);
 520
 521         /*
 522          * If we're just mounting the root most subvol put the inode and return
 523          * a reference to the dentry.  We will have already gotten a reference
 524          * to the inode in btrfs_fill_super so we're good to go.
 525          */
 526         if (!new && sb->s_root->d_inode == inode) {
 527                 iput(inode);
 528                 return dget(sb->s_root);
 529         }
 530
 531         if (new) {
 532                 const struct qstr name = { .name = "/", .len = 1 };
 533
 534                 /*
 535                  * New inode, we need to make the dentry a sibling of s_root so
 536                  * everything gets cleaned up properly on unmount.
 537                  */
 538                 dentry = d_alloc(sb->s_root, &name);
 539                 if (!dentry) {
 540                         iput(inode);
 541                         return ERR_PTR(-ENOMEM);
 542                 }
 543                 d_splice_alias(inode, dentry);
 544         } else {
 545                 /*
 546                  * We found the inode in cache, just find a dentry for it and
 547                  * put the reference to the inode we just got.
 548                  */
 549                 dentry = d_find_alias(inode);
 550                 iput(inode);
 551         }
 552
 553         return dentry;
 554 }
 555
 556 static int btrfs_fill_super(struct super_block *sb,
 557                             struct btrfs_fs_devices *fs_devices,
 558                             void *data, int silent)
 559 {
 560         struct inode *inode;
 561         struct dentry *root_dentry;
 562         struct btrfs_root *tree_root;
 563         struct btrfs_key key;
 564         int err;
 565
 566         sb->s_maxbytes = MAX_LFS_FILESIZE;
 567         sb->s_magic = BTRFS_SUPER_MAGIC;
 568         sb->s_op = &btrfs_super_ops;
 569         sb->s_export_op = &btrfs_export_ops;
 570         sb->s_xattr = btrfs_xattr_handlers;
 571         sb->s_time_gran = 1;
 572 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 573         sb->s_flags |= MS_POSIXACL;
 574 #endif
 575
 576         tree_root = open_ctree(sb, fs_devices, (char *)data);
 577
 578         if (IS_ERR(tree_root)) {
 579                 printk("btrfs: open_ctree failed\n");
 580                 return PTR_ERR(tree_root);
 581         }
 582         sb->s_fs_info = tree_root;
 583
 584         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 585         key.type = BTRFS_INODE_ITEM_KEY;
 586         key.offset = 0;
 587         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
 588         if (IS_ERR(inode)) {
 589                 err = PTR_ERR(inode);
 590                 goto fail_close;
 591         }
 592
 593         root_dentry = d_alloc_root(inode);
 594         if (!root_dentry) {
 595                 iput(inode);
 596                 err = -ENOMEM;
 597                 goto fail_close;
 598         }
 599
 600         sb->s_root = root_dentry;
 601
 602         save_mount_options(sb, data);
 603         return 0;
 604
 605 fail_close:
 606         close_ctree(tree_root);
 607         return err;
 608 }
 609
 610 int btrfs_sync_fs(struct super_block *sb, int wait)
 611 {
 612         struct btrfs_trans_handle *trans;
 613         struct btrfs_root *root = btrfs_sb(sb);
 614         int ret;
 615
 616         if (!wait) {
 617                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 618                 return 0;
 619         }
 620
 621         btrfs_start_delalloc_inodes(root, 0);
 622         btrfs_wait_ordered_extents(root, 0, 0);
 623
 624         trans = btrfs_start_transaction(root, 0);
 625         ret = btrfs_commit_transaction(trans, root);
 626         return ret;
 627 }
 628
 629 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 630 {
 631         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
 632         struct btrfs_fs_info *info = root->fs_info;
 633
 634         if (btrfs_test_opt(root, DEGRADED))
 635                 seq_puts(seq, ",degraded");
 636         if (btrfs_test_opt(root, NODATASUM))
 637                 seq_puts(seq, ",nodatasum");
 638         if (btrfs_test_opt(root, NODATACOW))
 639                 seq_puts(seq, ",nodatacow");
 640         if (btrfs_test_opt(root, NOBARRIER))
 641                 seq_puts(seq, ",nobarrier");
 642         if (info->max_inline != 8192 * 1024)
 643                 seq_printf(seq, ",max_inline=%llu",
 644                            (unsigned long long)info->max_inline);
 645         if (info->alloc_start != 0)
 646                 seq_printf(seq, ",alloc_start=%llu",
 647                            (unsigned long long)info->alloc_start);
 648         if (info->thread_pool_size !=  min_t(unsigned long,
 649                                              num_online_cpus() + 2, 8))
 650                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 651         if (btrfs_test_opt(root, COMPRESS))
 652                 seq_puts(seq, ",compress");
 653         if (btrfs_test_opt(root, NOSSD))
 654                 seq_puts(seq, ",nossd");
 655         if (btrfs_test_opt(root, SSD_SPREAD))
 656                 seq_puts(seq, ",ssd_spread");
 657         else if (btrfs_test_opt(root, SSD))
 658                 seq_puts(seq, ",ssd");
 659         if (btrfs_test_opt(root, NOTREELOG))
 660                 seq_puts(seq, ",notreelog");
 661         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 662                 seq_puts(seq, ",flushoncommit");
 663         if (btrfs_test_opt(root, DISCARD))
 664                 seq_puts(seq, ",discard");
 665         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 666                 seq_puts(seq, ",noacl");
 667         return 0;
 668 }
 669
 670 static int btrfs_test_super(struct super_block *s, void *data)
 671 {
 672         struct btrfs_root *test_root = data;
 673         struct btrfs_root *root = btrfs_sb(s);
 674
 675         /*
 676          * If this super block is going away, return false as it
 677          * can't match as an existing super block.
 678          */
 679         if (!atomic_read(&s->s_active))
 680                 return 0;
 681         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
 682 }
 683
 684 static int btrfs_set_super(struct super_block *s, void *data)
 685 {
 686         s->s_fs_info = data;
 687
 688         return set_anon_super(s, data);
 689 }
 690
 691
 692 /*
 693  * Find a superblock for the given device / mount point.
 694  *
 695  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
 696  *        for multiple device setup.  Make sure to keep it in sync.
 697  */
 698 static int btrfs_get_sb(struct file_system_type *fs_type, int flags,
 699                 const char *dev_name, void *data, struct vfsmount *mnt)
 700 {
 701         struct block_device *bdev = NULL;
 702         struct super_block *s;
 703         struct dentry *root;
 704         struct btrfs_fs_devices *fs_devices = NULL;
 705         struct btrfs_root *tree_root = NULL;
 706         struct btrfs_fs_info *fs_info = NULL;
 707         fmode_t mode = FMODE_READ;
 708         char *subvol_name = NULL;
 709         u64 subvol_objectid = 0;
 710         int error = 0;
 711
 712         if (!(flags & MS_RDONLY))
 713                 mode |= FMODE_WRITE;
 714
 715         error = btrfs_parse_early_options(data, mode, fs_type,
 716                                           &subvol_name, &subvol_objectid,
 717                                           &fs_devices);
 718         if (error)
 719                 return error;
 720
 721         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
 722         if (error)
 723                 goto error_free_subvol_name;
 724
 725         error = btrfs_open_devices(fs_devices, mode, fs_type);
 726         if (error)
 727                 goto error_free_subvol_name;
 728
 729         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
 730                 error = -EACCES;
 731                 goto error_close_devices;
 732         }
 733
 734         /*
 735          * Setup a dummy root and fs_info for test/set super.  This is because
 736          * we don't actually fill this stuff out until open_ctree, but we need
 737          * it for searching for existing supers, so this lets us do that and
 738          * then open_ctree will properly initialize everything later.
 739          */
 740         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
 741         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
 742         if (!fs_info || !tree_root) {
 743                 error = -ENOMEM;
 744                 goto error_close_devices;
 745         }
 746         fs_info->tree_root = tree_root;
 747         fs_info->fs_devices = fs_devices;
 748         tree_root->fs_info = fs_info;
 749
 750         bdev = fs_devices->latest_bdev;
 751         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
 752         if (IS_ERR(s))
 753                 goto error_s;
 754
 755         if (s->s_root) {
 756                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 757                         deactivate_locked_super(s);
 758                         error = -EBUSY;
 759                         goto error_close_devices;
 760                 }
 761
 762                 btrfs_close_devices(fs_devices);
 763         } else {
 764                 char b[BDEVNAME_SIZE];
 765
 766                 s->s_flags = flags;
 767                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 768                 error = btrfs_fill_super(s, fs_devices, data,
 769                                          flags & MS_SILENT ? 1 : 0);
 770                 if (error) {
 771                         deactivate_locked_super(s);
 772                         goto error_free_subvol_name;
 773                 }
 774
 775                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
 776                 s->s_flags |= MS_ACTIVE;
 777         }
 778
 779         root = get_default_root(s, subvol_objectid);
 780         if (IS_ERR(root)) {
 781                 error = PTR_ERR(root);
 782                 deactivate_locked_super(s);
 783                 goto error_free_subvol_name;
 784         }
 785         /* if they gave us a subvolume name bind mount into that */
 786         if (strcmp(subvol_name, ".")) {
 787                 struct dentry *new_root;
 788                 mutex_lock(&root->d_inode->i_mutex);
 789                 new_root = lookup_one_len(subvol_name, root,
 790                                       strlen(subvol_name));
 791                 mutex_unlock(&root->d_inode->i_mutex);
 792
 793                 if (IS_ERR(new_root)) {
 794                         dput(root);
 795                         deactivate_locked_super(s);
 796                         error = PTR_ERR(new_root);
 797                         goto error_free_subvol_name;
 798                 }
 799                 if (!new_root->d_inode) {
 800                         dput(root);
 801                         dput(new_root);
 802                         deactivate_locked_super(s);
 803                         error = -ENXIO;
 804                         goto error_free_subvol_name;
 805                 }
 806                 dput(root);
 807                 root = new_root;
 808         }
 809
 810         mnt->mnt_sb = s;
 811         mnt->mnt_root = root;
 812
 813         kfree(subvol_name);
 814         return 0;
 815
 816 error_s:
 817         error = PTR_ERR(s);
 818 error_close_devices:
 819         btrfs_close_devices(fs_devices);
 820         kfree(fs_info);
 821         kfree(tree_root);
 822 error_free_subvol_name:
 823         kfree(subvol_name);
 824         return error;
 825 }
 826
 827 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 828 {
 829         struct btrfs_root *root = btrfs_sb(sb);
 830         int ret;
 831
 832         ret = btrfs_parse_options(root, data);
 833         if (ret)
 834                 return -EINVAL;
 835
 836         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 837                 return 0;
 838
 839         if (*flags & MS_RDONLY) {
 840                 sb->s_flags |= MS_RDONLY;
 841
 842                 ret =  btrfs_commit_super(root);
 843                 WARN_ON(ret);
 844         } else {
 845                 if (root->fs_info->fs_devices->rw_devices == 0)
 846                         return -EACCES;
 847
 848                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
 849                         return -EINVAL;
 850
 851                 ret = btrfs_cleanup_fs_roots(root->fs_info);
 852                 WARN_ON(ret);
 853
 854                 /* recover relocation */
 855                 ret = btrfs_recover_relocation(root);
 856                 WARN_ON(ret);
 857
 858                 sb->s_flags &= ~MS_RDONLY;
 859         }
 860
 861         return 0;
 862 }
 863
 864 /*
 865  * The helper to calc the free space on the devices that can be used to store
 866  * file data.
 867  */
 868 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
 869 {
 870         struct btrfs_fs_info *fs_info = root->fs_info;
 871         struct btrfs_device_info *devices_info;
 872         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 873         struct btrfs_device *device;
 874         u64 skip_space;
 875         u64 type;
 876         u64 avail_space;
 877         u64 used_space;
 878         u64 min_stripe_size;
 879         int min_stripes = 1;
 880         int i = 0, nr_devices;
 881         int ret;
 882
 883         nr_devices = fs_info->fs_devices->rw_devices;
 884         BUG_ON(!nr_devices);
 885
 886         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
 887                                GFP_NOFS);
 888         if (!devices_info)
 889                 return -ENOMEM;
 890
 891         /* calc min stripe number for data space alloction */
 892         type = btrfs_get_alloc_profile(root, 1);
 893         if (type & BTRFS_BLOCK_GROUP_RAID0)
 894                 min_stripes = 2;
 895         else if (type & BTRFS_BLOCK_GROUP_RAID1)
 896                 min_stripes = 2;
 897         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 898                 min_stripes = 4;
 899
 900         if (type & BTRFS_BLOCK_GROUP_DUP)
 901                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
 902         else
 903                 min_stripe_size = BTRFS_STRIPE_LEN;
 904
 905         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
 906                 if (!device->in_fs_metadata)
 907                         continue;
 908
 909                 avail_space = device->total_bytes - device->bytes_used;
 910
 911                 /* align with stripe_len */
 912                 do_div(avail_space, BTRFS_STRIPE_LEN);
 913                 avail_space *= BTRFS_STRIPE_LEN;
 914
 915                 /*
 916                  * In order to avoid overwritting the superblock on the drive,
 917                  * btrfs starts at an offset of at least 1MB when doing chunk
 918                  * allocation.
 919                  */
 920                 skip_space = 1024 * 1024;
 921
 922                 /* user can set the offset in fs_info->alloc_start. */
 923                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
 924                     device->total_bytes)
 925                         skip_space = max(fs_info->alloc_start, skip_space);
 926
 927                 /*
 928                  * btrfs can not use the free space in [0, skip_space - 1],
 929                  * we must subtract it from the total. In order to implement
 930                  * it, we account the used space in this range first.
 931                  */
 932                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
 933                                                      &used_space);
 934                 if (ret) {
 935                         kfree(devices_info);
 936                         return ret;
 937                 }
 938
 939                 /* calc the free space in [0, skip_space - 1] */
 940                 skip_space -= used_space;
 941
 942                 /*
 943                  * we can use the free space in [0, skip_space - 1], subtract
 944                  * it from the total.
 945                  */
 946                 if (avail_space && avail_space >= skip_space)
 947                         avail_space -= skip_space;
 948                 else
 949                         avail_space = 0;
 950
 951                 if (avail_space < min_stripe_size)
 952                         continue;
 953
 954                 devices_info[i].dev = device;
 955                 devices_info[i].max_avail = avail_space;
 956
 957                 i++;
 958         }
 959
 960         nr_devices = i;
 961
 962         btrfs_descending_sort_devices(devices_info, nr_devices);
 963
 964         i = nr_devices - 1;
 965         avail_space = 0;
 966         while (nr_devices >= min_stripes) {
 967                 if (devices_info[i].max_avail >= min_stripe_size) {
 968                         int j;
 969                         u64 alloc_size;
 970
 971                         avail_space += devices_info[i].max_avail * min_stripes;
 972                         alloc_size = devices_info[i].max_avail;
 973                         for (j = i + 1 - min_stripes; j <= i; j++)
 974                                 devices_info[j].max_avail -= alloc_size;
 975                 }
 976                 i--;
 977                 nr_devices--;
 978         }
 979
 980         kfree(devices_info);
 981         *free_bytes = avail_space;
 982         return 0;
 983 }
 984
 985 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 986 {
 987         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
 988         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
 989         struct list_head *head = &root->fs_info->space_info;
 990         struct btrfs_space_info *found;
 991         u64 total_used = 0;
 992         u64 total_free_data = 0;
 993         int bits = dentry->d_sb->s_blocksize_bits;
 994         __be32 *fsid = (__be32 *)root->fs_info->fsid;
 995         int ret;
 996
 997         /* holding chunk_muext to avoid allocating new chunks */
 998         mutex_lock(&root->fs_info->chunk_mutex);
 999         rcu_read_lock();
1000         list_for_each_entry_rcu(found, head, list) {
1001                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1002                         total_free_data += found->disk_total - found->disk_used;
1003                         total_free_data -=
1004                                 btrfs_account_ro_block_groups_free_space(found);
1005                 }
1006
1007                 total_used += found->disk_used;
1008         }
1009         rcu_read_unlock();
1010
1011         buf->f_namelen = BTRFS_NAME_LEN;
1012         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1013         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1014         buf->f_bsize = dentry->d_sb->s_blocksize;
1015         buf->f_type = BTRFS_SUPER_MAGIC;
1016         buf->f_bavail = total_free_data;
1017         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1018         if (ret) {
1019                 mutex_unlock(&root->fs_info->chunk_mutex);
1020                 return ret;
1021         }
1022         buf->f_bavail += total_free_data;
1023         buf->f_bavail = buf->f_bavail >> bits;
1024         mutex_unlock(&root->fs_info->chunk_mutex);
1025
1026         /* We treat it as constant endianness (it doesn't matter _which_)
1027            because we want the fsid to come out the same whether mounted
1028            on a big-endian or little-endian host */
1029         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1030         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1031         /* Mask in the root object ID too, to disambiguate subvols */
1032         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1033         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1034
1035         return 0;
1036 }
1037
1038 static struct file_system_type btrfs_fs_type = {
1039         .owner          = THIS_MODULE,
1040         .name           = "btrfs",
1041         .get_sb         = btrfs_get_sb,
1042         .kill_sb        = kill_anon_super,
1043         .fs_flags       = FS_REQUIRES_DEV,
1044 };
1045
1046 /*
1047  * used by btrfsctl to scan devices when no FS is mounted
1048  */
1049 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1050                                 unsigned long arg)
1051 {
1052         struct btrfs_ioctl_vol_args *vol;
1053         struct btrfs_fs_devices *fs_devices;
1054         int ret = -ENOTTY;
1055
1056         if (!capable(CAP_SYS_ADMIN))
1057                 return -EPERM;
1058
1059         vol = memdup_user((void __user *)arg, sizeof(*vol));
1060         if (IS_ERR(vol))
1061                 return PTR_ERR(vol);
1062
1063         switch (cmd) {
1064         case BTRFS_IOC_SCAN_DEV:
1065                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1066                                             &btrfs_fs_type, &fs_devices);
1067                 break;
1068         }
1069
1070         kfree(vol);
1071         return ret;
1072 }
1073
1074 static int btrfs_freeze(struct super_block *sb)
1075 {
1076         struct btrfs_root *root = btrfs_sb(sb);
1077         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1078         mutex_lock(&root->fs_info->cleaner_mutex);
1079         return 0;
1080 }
1081
1082 static int btrfs_unfreeze(struct super_block *sb)
1083 {
1084         struct btrfs_root *root = btrfs_sb(sb);
1085         mutex_unlock(&root->fs_info->cleaner_mutex);
1086         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1087         return 0;
1088 }
1089
1090 static const struct super_operations btrfs_super_ops = {
1091         .drop_inode     = btrfs_drop_inode,
1092         .evict_inode    = btrfs_evict_inode,
1093         .put_super      = btrfs_put_super,
1094         .sync_fs        = btrfs_sync_fs,
1095         .show_options   = btrfs_show_options,
1096         .write_inode    = btrfs_write_inode,
1097         .dirty_inode    = btrfs_dirty_inode,
1098         .alloc_inode    = btrfs_alloc_inode,
1099         .destroy_inode  = btrfs_destroy_inode,
1100         .statfs         = btrfs_statfs,
1101         .remount_fs     = btrfs_remount,
1102         .freeze_fs      = btrfs_freeze,
1103         .unfreeze_fs    = btrfs_unfreeze,
1104 };
1105
1106 static const struct file_operations btrfs_ctl_fops = {
1107         .unlocked_ioctl  = btrfs_control_ioctl,
1108         .compat_ioctl = btrfs_control_ioctl,
1109         .owner   = THIS_MODULE,
1110 };
1111
1112 static struct miscdevice btrfs_misc = {
1113         .minor          = BTRFS_MINOR,
1114         .name           = "btrfs-control",
1115         .fops           = &btrfs_ctl_fops
1116 };
1117
1118 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1119 MODULE_ALIAS("devname:btrfs-control");
1120
1121 static int btrfs_interface_init(void)
1122 {
1123         return misc_register(&btrfs_misc);
1124 }
1125
1126 static void btrfs_interface_exit(void)
1127 {
1128         if (misc_deregister(&btrfs_misc) < 0)
1129                 printk(KERN_INFO "misc_deregister failed for control device");
1130 }
1131
1132 static int __init init_btrfs_fs(void)
1133 {
1134         int err;
1135
1136         err = btrfs_init_sysfs();
1137         if (err)
1138                 return err;
1139
1140         err = btrfs_init_compress();
1141         if (err)
1142                 goto free_sysfs;
1143
1144         err = btrfs_init_cachep();
1145         if (err)
1146                 goto free_compress;
1147
1148         err = extent_io_init();
1149         if (err)
1150                 goto free_cachep;
1151
1152         err = extent_map_init();
1153         if (err)
1154                 goto free_extent_io;
1155
1156         err = btrfs_interface_init();
1157         if (err)
1158                 goto free_extent_map;
1159
1160         err = register_filesystem(&btrfs_fs_type);
1161         if (err)
1162                 goto unregister_ioctl;
1163
1164         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1165         return 0;
1166
1167 unregister_ioctl:
1168         btrfs_interface_exit();
1169 free_extent_map:
1170         extent_map_exit();
1171 free_extent_io:
1172         extent_io_exit();
1173 free_cachep:
1174         btrfs_destroy_cachep();
1175 free_compress:
1176         btrfs_exit_compress();
1177 free_sysfs:
1178         btrfs_exit_sysfs();
1179         return err;
1180 }
1181
1182 static void __exit exit_btrfs_fs(void)
1183 {
1184         btrfs_destroy_cachep();
1185         extent_map_exit();
1186         extent_io_exit();
1187         btrfs_interface_exit();
1188         unregister_filesystem(&btrfs_fs_type);
1189         btrfs_exit_sysfs();
1190         btrfs_cleanup_fs_uuids();
1191         btrfs_exit_compress();
1192 }
1193
1194 module_init(init_btrfs_fs)
1195 module_exit(exit_btrfs_fs)
1196
1197 MODULE_LICENSE("GPL");