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 "delayed-inode.h"
  44 #include "ctree.h"
  45 #include "disk-io.h"
  46 #include "transaction.h"
  47 #include "btrfs_inode.h"
  48 #include "ioctl.h"
  49 #include "print-tree.h"
  50 #include "xattr.h"
  51 #include "volumes.h"
  52 #include "version.h"
  53 #include "export.h"
  54 #include "compression.h"
  55
  56 #define CREATE_TRACE_POINTS
  57 #include <trace/events/btrfs.h>
  58
  59 static const struct super_operations btrfs_super_ops;
  60
  61 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
  62                                       char nbuf[16])
  63 {
  64         char *errstr = NULL;
  65
  66         switch (errno) {
  67         case -EIO:
  68                 errstr = "IO failure";
  69                 break;
  70         case -ENOMEM:
  71                 errstr = "Out of memory";
  72                 break;
  73         case -EROFS:
  74                 errstr = "Readonly filesystem";
  75                 break;
  76         default:
  77                 if (nbuf) {
  78                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  79                                 errstr = nbuf;
  80                 }
  81                 break;
  82         }
  83
  84         return errstr;
  85 }
  86
  87 static void __save_error_info(struct btrfs_fs_info *fs_info)
  88 {
  89         /*
  90          * today we only save the error info into ram.  Long term we'll
  91          * also send it down to the disk
  92          */
  93         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
  94 }
  95
  96 /* NOTE:
  97  *      We move write_super stuff at umount in order to avoid deadlock
  98  *      for umount hold all lock.
  99  */
 100 static void save_error_info(struct btrfs_fs_info *fs_info)
 101 {
 102         __save_error_info(fs_info);
 103 }
 104
 105 /* btrfs handle error by forcing the filesystem readonly */
 106 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 107 {
 108         struct super_block *sb = fs_info->sb;
 109
 110         if (sb->s_flags & MS_RDONLY)
 111                 return;
 112
 113         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 114                 sb->s_flags |= MS_RDONLY;
 115                 printk(KERN_INFO "btrfs is forced readonly\n");
 116         }
 117 }
 118
 119 /*
 120  * __btrfs_std_error decodes expected errors from the caller and
 121  * invokes the approciate error response.
 122  */
 123 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 124                      unsigned int line, int errno)
 125 {
 126         struct super_block *sb = fs_info->sb;
 127         char nbuf[16];
 128         const char *errstr;
 129
 130         /*
 131          * Special case: if the error is EROFS, and we're already
 132          * under MS_RDONLY, then it is safe here.
 133          */
 134         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 135                 return;
 136
 137         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 138         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 139                 sb->s_id, function, line, errstr);
 140         save_error_info(fs_info);
 141
 142         btrfs_handle_error(fs_info);
 143 }
 144
 145 static void btrfs_put_super(struct super_block *sb)
 146 {
 147         struct btrfs_root *root = btrfs_sb(sb);
 148         int ret;
 149
 150         ret = close_ctree(root);
 151         sb->s_fs_info = NULL;
 152
 153         (void)ret; /* FIXME: need to fix VFS to return error? */
 154 }
 155
 156 enum {
 157         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 158         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 159         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 160         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 161         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 162         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 163         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag,
 164         Opt_inode_cache, Opt_err,
 165 };
 166
 167 static match_table_t tokens = {
 168         {Opt_degraded, "degraded"},
 169         {Opt_subvol, "subvol=%s"},
 170         {Opt_subvolid, "subvolid=%d"},
 171         {Opt_device, "device=%s"},
 172         {Opt_nodatasum, "nodatasum"},
 173         {Opt_nodatacow, "nodatacow"},
 174         {Opt_nobarrier, "nobarrier"},
 175         {Opt_max_inline, "max_inline=%s"},
 176         {Opt_alloc_start, "alloc_start=%s"},
 177         {Opt_thread_pool, "thread_pool=%d"},
 178         {Opt_compress, "compress"},
 179         {Opt_compress_type, "compress=%s"},
 180         {Opt_compress_force, "compress-force"},
 181         {Opt_compress_force_type, "compress-force=%s"},
 182         {Opt_ssd, "ssd"},
 183         {Opt_ssd_spread, "ssd_spread"},
 184         {Opt_nossd, "nossd"},
 185         {Opt_noacl, "noacl"},
 186         {Opt_notreelog, "notreelog"},
 187         {Opt_flushoncommit, "flushoncommit"},
 188         {Opt_ratio, "metadata_ratio=%d"},
 189         {Opt_discard, "discard"},
 190         {Opt_space_cache, "space_cache"},
 191         {Opt_clear_cache, "clear_cache"},
 192         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 193         {Opt_enospc_debug, "enospc_debug"},
 194         {Opt_subvolrootid, "subvolrootid=%d"},
 195         {Opt_defrag, "autodefrag"},
 196         {Opt_inode_cache, "inode_cache"},
 197         {Opt_err, NULL},
 198 };
 199
 200 /*
 201  * Regular mount options parser.  Everything that is needed only when
 202  * reading in a new superblock is parsed here.
 203  */
 204 int btrfs_parse_options(struct btrfs_root *root, char *options)
 205 {
 206         struct btrfs_fs_info *info = root->fs_info;
 207         substring_t args[MAX_OPT_ARGS];
 208         char *p, *num, *orig;
 209         int intarg;
 210         int ret = 0;
 211         char *compress_type;
 212         bool compress_force = false;
 213
 214         if (!options)
 215                 return 0;
 216
 217         /*
 218          * strsep changes the string, duplicate it because parse_options
 219          * gets called twice
 220          */
 221         options = kstrdup(options, GFP_NOFS);
 222         if (!options)
 223                 return -ENOMEM;
 224
 225         orig = options;
 226
 227         while ((p = strsep(&options, ",")) != NULL) {
 228                 int token;
 229                 if (!*p)
 230                         continue;
 231
 232                 token = match_token(p, tokens, args);
 233                 switch (token) {
 234                 case Opt_degraded:
 235                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 236                         btrfs_set_opt(info->mount_opt, DEGRADED);
 237                         break;
 238                 case Opt_subvol:
 239                 case Opt_subvolid:
 240                 case Opt_subvolrootid:
 241                 case Opt_device:
 242                         /*
 243                          * These are parsed by btrfs_parse_early_options
 244                          * and can be happily ignored here.
 245                          */
 246                         break;
 247                 case Opt_nodatasum:
 248                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 249                         btrfs_set_opt(info->mount_opt, NODATASUM);
 250                         break;
 251                 case Opt_nodatacow:
 252                         printk(KERN_INFO "btrfs: setting nodatacow\n");
 253                         btrfs_set_opt(info->mount_opt, NODATACOW);
 254                         btrfs_set_opt(info->mount_opt, NODATASUM);
 255                         break;
 256                 case Opt_compress_force:
 257                 case Opt_compress_force_type:
 258                         compress_force = true;
 259                 case Opt_compress:
 260                 case Opt_compress_type:
 261                         if (token == Opt_compress ||
 262                             token == Opt_compress_force ||
 263                             strcmp(args[0].from, "zlib") == 0) {
 264                                 compress_type = "zlib";
 265                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 266                         } else if (strcmp(args[0].from, "lzo") == 0) {
 267                                 compress_type = "lzo";
 268                                 info->compress_type = BTRFS_COMPRESS_LZO;
 269                         } else {
 270                                 ret = -EINVAL;
 271                                 goto out;
 272                         }
 273
 274                         btrfs_set_opt(info->mount_opt, COMPRESS);
 275                         if (compress_force) {
 276                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 277                                 pr_info("btrfs: force %s compression\n",
 278                                         compress_type);
 279                         } else
 280                                 pr_info("btrfs: use %s compression\n",
 281                                         compress_type);
 282                         break;
 283                 case Opt_ssd:
 284                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 285                         btrfs_set_opt(info->mount_opt, SSD);
 286                         break;
 287                 case Opt_ssd_spread:
 288                         printk(KERN_INFO "btrfs: use spread ssd "
 289                                "allocation scheme\n");
 290                         btrfs_set_opt(info->mount_opt, SSD);
 291                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 292                         break;
 293                 case Opt_nossd:
 294                         printk(KERN_INFO "btrfs: not using ssd allocation "
 295                                "scheme\n");
 296                         btrfs_set_opt(info->mount_opt, NOSSD);
 297                         btrfs_clear_opt(info->mount_opt, SSD);
 298                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 299                         break;
 300                 case Opt_nobarrier:
 301                         printk(KERN_INFO "btrfs: turning off barriers\n");
 302                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 303                         break;
 304                 case Opt_thread_pool:
 305                         intarg = 0;
 306                         match_int(&args[0], &intarg);
 307                         if (intarg) {
 308                                 info->thread_pool_size = intarg;
 309                                 printk(KERN_INFO "btrfs: thread pool %d\n",
 310                                        info->thread_pool_size);
 311                         }
 312                         break;
 313                 case Opt_max_inline:
 314                         num = match_strdup(&args[0]);
 315                         if (num) {
 316                                 info->max_inline = memparse(num, NULL);
 317                                 kfree(num);
 318
 319                                 if (info->max_inline) {
 320                                         info->max_inline = max_t(u64,
 321                                                 info->max_inline,
 322                                                 root->sectorsize);
 323                                 }
 324                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 325                                         (unsigned long long)info->max_inline);
 326                         }
 327                         break;
 328                 case Opt_alloc_start:
 329                         num = match_strdup(&args[0]);
 330                         if (num) {
 331                                 info->alloc_start = memparse(num, NULL);
 332                                 kfree(num);
 333                                 printk(KERN_INFO
 334                                         "btrfs: allocations start at %llu\n",
 335                                         (unsigned long long)info->alloc_start);
 336                         }
 337                         break;
 338                 case Opt_noacl:
 339                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 340                         break;
 341                 case Opt_notreelog:
 342                         printk(KERN_INFO "btrfs: disabling tree log\n");
 343                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 344                         break;
 345                 case Opt_flushoncommit:
 346                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 347                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 348                         break;
 349                 case Opt_ratio:
 350                         intarg = 0;
 351                         match_int(&args[0], &intarg);
 352                         if (intarg) {
 353                                 info->metadata_ratio = intarg;
 354                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 355                                        info->metadata_ratio);
 356                         }
 357                         break;
 358                 case Opt_discard:
 359                         btrfs_set_opt(info->mount_opt, DISCARD);
 360                         break;
 361                 case Opt_space_cache:
 362                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
 363                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 364                         break;
 365                 case Opt_inode_cache:
 366                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 367                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 368                         break;
 369                 case Opt_clear_cache:
 370                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 371                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 372                         break;
 373                 case Opt_user_subvol_rm_allowed:
 374                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 375                         break;
 376                 case Opt_enospc_debug:
 377                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 378                         break;
 379                 case Opt_defrag:
 380                         printk(KERN_INFO "btrfs: enabling auto defrag");
 381                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 382                         break;
 383                 case Opt_err:
 384                         printk(KERN_INFO "btrfs: unrecognized mount option "
 385                                "'%s'\n", p);
 386                         ret = -EINVAL;
 387                         goto out;
 388                 default:
 389                         break;
 390                 }
 391         }
 392 out:
 393         kfree(orig);
 394         return ret;
 395 }
 396
 397 /*
 398  * Parse mount options that are required early in the mount process.
 399  *
 400  * All other options will be parsed on much later in the mount process and
 401  * only when we need to allocate a new super block.
 402  */
 403 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 404                 void *holder, char **subvol_name, u64 *subvol_objectid,
 405                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
 406 {
 407         substring_t args[MAX_OPT_ARGS];
 408         char *opts, *orig, *p;
 409         int error = 0;
 410         int intarg;
 411
 412         if (!options)
 413                 goto out;
 414
 415         /*
 416          * strsep changes the string, duplicate it because parse_options
 417          * gets called twice
 418          */
 419         opts = kstrdup(options, GFP_KERNEL);
 420         if (!opts)
 421                 return -ENOMEM;
 422         orig = opts;
 423
 424         while ((p = strsep(&opts, ",")) != NULL) {
 425                 int token;
 426                 if (!*p)
 427                         continue;
 428
 429                 token = match_token(p, tokens, args);
 430                 switch (token) {
 431                 case Opt_subvol:
 432                         *subvol_name = match_strdup(&args[0]);
 433                         break;
 434                 case Opt_subvolid:
 435                         intarg = 0;
 436                         error = match_int(&args[0], &intarg);
 437                         if (!error) {
 438                                 /* we want the original fs_tree */
 439                                 if (!intarg)
 440                                         *subvol_objectid =
 441                                                 BTRFS_FS_TREE_OBJECTID;
 442                                 else
 443                                         *subvol_objectid = intarg;
 444                         }
 445                         break;
 446                 case Opt_subvolrootid:
 447                         intarg = 0;
 448                         error = match_int(&args[0], &intarg);
 449                         if (!error) {
 450                                 /* we want the original fs_tree */
 451                                 if (!intarg)
 452                                         *subvol_rootid =
 453                                                 BTRFS_FS_TREE_OBJECTID;
 454                                 else
 455                                         *subvol_rootid = intarg;
 456                         }
 457                         break;
 458                 case Opt_device:
 459                         error = btrfs_scan_one_device(match_strdup(&args[0]),
 460                                         flags, holder, fs_devices);
 461                         if (error)
 462                                 goto out_free_opts;
 463                         break;
 464                 default:
 465                         break;
 466                 }
 467         }
 468
 469  out_free_opts:
 470         kfree(orig);
 471  out:
 472         /*
 473          * If no subvolume name is specified we use the default one.  Allocate
 474          * a copy of the string "." here so that code later in the
 475          * mount path doesn't care if it's the default volume or another one.
 476          */
 477         if (!*subvol_name) {
 478                 *subvol_name = kstrdup(".", GFP_KERNEL);
 479                 if (!*subvol_name)
 480                         return -ENOMEM;
 481         }
 482         return error;
 483 }
 484
 485 static struct dentry *get_default_root(struct super_block *sb,
 486                                        u64 subvol_objectid)
 487 {
 488         struct btrfs_root *root = sb->s_fs_info;
 489         struct btrfs_root *new_root;
 490         struct btrfs_dir_item *di;
 491         struct btrfs_path *path;
 492         struct btrfs_key location;
 493         struct inode *inode;
 494         struct dentry *dentry;
 495         u64 dir_id;
 496         int new = 0;
 497
 498         /*
 499          * We have a specific subvol we want to mount, just setup location and
 500          * go look up the root.
 501          */
 502         if (subvol_objectid) {
 503                 location.objectid = subvol_objectid;
 504                 location.type = BTRFS_ROOT_ITEM_KEY;
 505                 location.offset = (u64)-1;
 506                 goto find_root;
 507         }
 508
 509         path = btrfs_alloc_path();
 510         if (!path)
 511                 return ERR_PTR(-ENOMEM);
 512         path->leave_spinning = 1;
 513
 514         /*
 515          * Find the "default" dir item which points to the root item that we
 516          * will mount by default if we haven't been given a specific subvolume
 517          * to mount.
 518          */
 519         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
 520         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 521         if (IS_ERR(di)) {
 522                 btrfs_free_path(path);
 523                 return ERR_CAST(di);
 524         }
 525         if (!di) {
 526                 /*
 527                  * Ok the default dir item isn't there.  This is weird since
 528                  * it's always been there, but don't freak out, just try and
 529                  * mount to root most subvolume.
 530                  */
 531                 btrfs_free_path(path);
 532                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 533                 new_root = root->fs_info->fs_root;
 534                 goto setup_root;
 535         }
 536
 537         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 538         btrfs_free_path(path);
 539
 540 find_root:
 541         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
 542         if (IS_ERR(new_root))
 543                 return ERR_CAST(new_root);
 544
 545         if (btrfs_root_refs(&new_root->root_item) == 0)
 546                 return ERR_PTR(-ENOENT);
 547
 548         dir_id = btrfs_root_dirid(&new_root->root_item);
 549 setup_root:
 550         location.objectid = dir_id;
 551         location.type = BTRFS_INODE_ITEM_KEY;
 552         location.offset = 0;
 553
 554         inode = btrfs_iget(sb, &location, new_root, &new);
 555         if (IS_ERR(inode))
 556                 return ERR_CAST(inode);
 557
 558         /*
 559          * If we're just mounting the root most subvol put the inode and return
 560          * a reference to the dentry.  We will have already gotten a reference
 561          * to the inode in btrfs_fill_super so we're good to go.
 562          */
 563         if (!new && sb->s_root->d_inode == inode) {
 564                 iput(inode);
 565                 return dget(sb->s_root);
 566         }
 567
 568         if (new) {
 569                 const struct qstr name = { .name = "/", .len = 1 };
 570
 571                 /*
 572                  * New inode, we need to make the dentry a sibling of s_root so
 573                  * everything gets cleaned up properly on unmount.
 574                  */
 575                 dentry = d_alloc(sb->s_root, &name);
 576                 if (!dentry) {
 577                         iput(inode);
 578                         return ERR_PTR(-ENOMEM);
 579                 }
 580                 d_splice_alias(inode, dentry);
 581         } else {
 582                 /*
 583                  * We found the inode in cache, just find a dentry for it and
 584                  * put the reference to the inode we just got.
 585                  */
 586                 dentry = d_find_alias(inode);
 587                 iput(inode);
 588         }
 589
 590         return dentry;
 591 }
 592
 593 static int btrfs_fill_super(struct super_block *sb,
 594                             struct btrfs_fs_devices *fs_devices,
 595                             void *data, int silent)
 596 {
 597         struct inode *inode;
 598         struct dentry *root_dentry;
 599         struct btrfs_root *tree_root;
 600         struct btrfs_key key;
 601         int err;
 602
 603         sb->s_maxbytes = MAX_LFS_FILESIZE;
 604         sb->s_magic = BTRFS_SUPER_MAGIC;
 605         sb->s_op = &btrfs_super_ops;
 606         sb->s_d_op = &btrfs_dentry_operations;
 607         sb->s_export_op = &btrfs_export_ops;
 608         sb->s_xattr = btrfs_xattr_handlers;
 609         sb->s_time_gran = 1;
 610 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 611         sb->s_flags |= MS_POSIXACL;
 612 #endif
 613
 614         tree_root = open_ctree(sb, fs_devices, (char *)data);
 615
 616         if (IS_ERR(tree_root)) {
 617                 printk("btrfs: open_ctree failed\n");
 618                 return PTR_ERR(tree_root);
 619         }
 620         sb->s_fs_info = tree_root;
 621
 622         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 623         key.type = BTRFS_INODE_ITEM_KEY;
 624         key.offset = 0;
 625         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
 626         if (IS_ERR(inode)) {
 627                 err = PTR_ERR(inode);
 628                 goto fail_close;
 629         }
 630
 631         root_dentry = d_alloc_root(inode);
 632         if (!root_dentry) {
 633                 iput(inode);
 634                 err = -ENOMEM;
 635                 goto fail_close;
 636         }
 637
 638         sb->s_root = root_dentry;
 639
 640         save_mount_options(sb, data);
 641         return 0;
 642
 643 fail_close:
 644         close_ctree(tree_root);
 645         return err;
 646 }
 647
 648 int btrfs_sync_fs(struct super_block *sb, int wait)
 649 {
 650         struct btrfs_trans_handle *trans;
 651         struct btrfs_root *root = btrfs_sb(sb);
 652         int ret;
 653
 654         trace_btrfs_sync_fs(wait);
 655
 656         if (!wait) {
 657                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 658                 return 0;
 659         }
 660
 661         btrfs_start_delalloc_inodes(root, 0);
 662         btrfs_wait_ordered_extents(root, 0, 0);
 663
 664         trans = btrfs_start_transaction(root, 0);
 665         if (IS_ERR(trans))
 666                 return PTR_ERR(trans);
 667         ret = btrfs_commit_transaction(trans, root);
 668         return ret;
 669 }
 670
 671 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 672 {
 673         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
 674         struct btrfs_fs_info *info = root->fs_info;
 675         char *compress_type;
 676
 677         if (btrfs_test_opt(root, DEGRADED))
 678                 seq_puts(seq, ",degraded");
 679         if (btrfs_test_opt(root, NODATASUM))
 680                 seq_puts(seq, ",nodatasum");
 681         if (btrfs_test_opt(root, NODATACOW))
 682                 seq_puts(seq, ",nodatacow");
 683         if (btrfs_test_opt(root, NOBARRIER))
 684                 seq_puts(seq, ",nobarrier");
 685         if (info->max_inline != 8192 * 1024)
 686                 seq_printf(seq, ",max_inline=%llu",
 687                            (unsigned long long)info->max_inline);
 688         if (info->alloc_start != 0)
 689                 seq_printf(seq, ",alloc_start=%llu",
 690                            (unsigned long long)info->alloc_start);
 691         if (info->thread_pool_size !=  min_t(unsigned long,
 692                                              num_online_cpus() + 2, 8))
 693                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 694         if (btrfs_test_opt(root, COMPRESS)) {
 695                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 696                         compress_type = "zlib";
 697                 else
 698                         compress_type = "lzo";
 699                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 700                         seq_printf(seq, ",compress-force=%s", compress_type);
 701                 else
 702                         seq_printf(seq, ",compress=%s", compress_type);
 703         }
 704         if (btrfs_test_opt(root, NOSSD))
 705                 seq_puts(seq, ",nossd");
 706         if (btrfs_test_opt(root, SSD_SPREAD))
 707                 seq_puts(seq, ",ssd_spread");
 708         else if (btrfs_test_opt(root, SSD))
 709                 seq_puts(seq, ",ssd");
 710         if (btrfs_test_opt(root, NOTREELOG))
 711                 seq_puts(seq, ",notreelog");
 712         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 713                 seq_puts(seq, ",flushoncommit");
 714         if (btrfs_test_opt(root, DISCARD))
 715                 seq_puts(seq, ",discard");
 716         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 717                 seq_puts(seq, ",noacl");
 718         if (btrfs_test_opt(root, SPACE_CACHE))
 719                 seq_puts(seq, ",space_cache");
 720         if (btrfs_test_opt(root, CLEAR_CACHE))
 721                 seq_puts(seq, ",clear_cache");
 722         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 723                 seq_puts(seq, ",user_subvol_rm_allowed");
 724         return 0;
 725 }
 726
 727 static int btrfs_test_super(struct super_block *s, void *data)
 728 {
 729         struct btrfs_root *test_root = data;
 730         struct btrfs_root *root = btrfs_sb(s);
 731
 732         /*
 733          * If this super block is going away, return false as it
 734          * can't match as an existing super block.
 735          */
 736         if (!atomic_read(&s->s_active))
 737                 return 0;
 738         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
 739 }
 740
 741 static int btrfs_set_super(struct super_block *s, void *data)
 742 {
 743         s->s_fs_info = data;
 744
 745         return set_anon_super(s, data);
 746 }
 747
 748
 749 /*
 750  * Find a superblock for the given device / mount point.
 751  *
 752  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
 753  *        for multiple device setup.  Make sure to keep it in sync.
 754  */
 755 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
 756                 const char *device_name, void *data)
 757 {
 758         struct block_device *bdev = NULL;
 759         struct super_block *s;
 760         struct dentry *root;
 761         struct btrfs_fs_devices *fs_devices = NULL;
 762         struct btrfs_root *tree_root = NULL;
 763         struct btrfs_fs_info *fs_info = NULL;
 764         fmode_t mode = FMODE_READ;
 765         char *subvol_name = NULL;
 766         u64 subvol_objectid = 0;
 767         u64 subvol_rootid = 0;
 768         int error = 0;
 769
 770         if (!(flags & MS_RDONLY))
 771                 mode |= FMODE_WRITE;
 772
 773         error = btrfs_parse_early_options(data, mode, fs_type,
 774                                           &subvol_name, &subvol_objectid,
 775                                           &subvol_rootid, &fs_devices);
 776         if (error)
 777                 return ERR_PTR(error);
 778
 779         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
 780         if (error)
 781                 goto error_free_subvol_name;
 782
 783         error = btrfs_open_devices(fs_devices, mode, fs_type);
 784         if (error)
 785                 goto error_free_subvol_name;
 786
 787         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
 788                 error = -EACCES;
 789                 goto error_close_devices;
 790         }
 791
 792         /*
 793          * Setup a dummy root and fs_info for test/set super.  This is because
 794          * we don't actually fill this stuff out until open_ctree, but we need
 795          * it for searching for existing supers, so this lets us do that and
 796          * then open_ctree will properly initialize everything later.
 797          */
 798         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
 799         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
 800         if (!fs_info || !tree_root) {
 801                 error = -ENOMEM;
 802                 goto error_close_devices;
 803         }
 804         fs_info->tree_root = tree_root;
 805         fs_info->fs_devices = fs_devices;
 806         tree_root->fs_info = fs_info;
 807
 808         bdev = fs_devices->latest_bdev;
 809         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
 810         if (IS_ERR(s))
 811                 goto error_s;
 812
 813         if (s->s_root) {
 814                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 815                         deactivate_locked_super(s);
 816                         error = -EBUSY;
 817                         goto error_close_devices;
 818                 }
 819
 820                 btrfs_close_devices(fs_devices);
 821                 kfree(fs_info);
 822                 kfree(tree_root);
 823         } else {
 824                 char b[BDEVNAME_SIZE];
 825
 826                 s->s_flags = flags;
 827                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 828                 error = btrfs_fill_super(s, fs_devices, data,
 829                                          flags & MS_SILENT ? 1 : 0);
 830                 if (error) {
 831                         deactivate_locked_super(s);
 832                         goto error_free_subvol_name;
 833                 }
 834
 835                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
 836                 s->s_flags |= MS_ACTIVE;
 837         }
 838
 839         /* if they gave us a subvolume name bind mount into that */
 840         if (strcmp(subvol_name, ".")) {
 841                 struct dentry *new_root;
 842
 843                 root = get_default_root(s, subvol_rootid);
 844                 if (IS_ERR(root)) {
 845                         error = PTR_ERR(root);
 846                         deactivate_locked_super(s);
 847                         goto error_free_subvol_name;
 848                 }
 849
 850                 mutex_lock(&root->d_inode->i_mutex);
 851                 new_root = lookup_one_len(subvol_name, root,
 852                                       strlen(subvol_name));
 853                 mutex_unlock(&root->d_inode->i_mutex);
 854
 855                 if (IS_ERR(new_root)) {
 856                         dput(root);
 857                         deactivate_locked_super(s);
 858                         error = PTR_ERR(new_root);
 859                         goto error_free_subvol_name;
 860                 }
 861                 if (!new_root->d_inode) {
 862                         dput(root);
 863                         dput(new_root);
 864                         deactivate_locked_super(s);
 865                         error = -ENXIO;
 866                         goto error_free_subvol_name;
 867                 }
 868                 dput(root);
 869                 root = new_root;
 870         } else {
 871                 root = get_default_root(s, subvol_objectid);
 872                 if (IS_ERR(root)) {
 873                         error = PTR_ERR(root);
 874                         deactivate_locked_super(s);
 875                         goto error_free_subvol_name;
 876                 }
 877         }
 878
 879         kfree(subvol_name);
 880         return root;
 881
 882 error_s:
 883         error = PTR_ERR(s);
 884 error_close_devices:
 885         btrfs_close_devices(fs_devices);
 886         kfree(fs_info);
 887         kfree(tree_root);
 888 error_free_subvol_name:
 889         kfree(subvol_name);
 890         return ERR_PTR(error);
 891 }
 892
 893 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 894 {
 895         struct btrfs_root *root = btrfs_sb(sb);
 896         int ret;
 897
 898         ret = btrfs_parse_options(root, data);
 899         if (ret)
 900                 return -EINVAL;
 901
 902         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 903                 return 0;
 904
 905         if (*flags & MS_RDONLY) {
 906                 sb->s_flags |= MS_RDONLY;
 907
 908                 ret =  btrfs_commit_super(root);
 909                 WARN_ON(ret);
 910         } else {
 911                 if (root->fs_info->fs_devices->rw_devices == 0)
 912                         return -EACCES;
 913
 914                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
 915                         return -EINVAL;
 916
 917                 ret = btrfs_cleanup_fs_roots(root->fs_info);
 918                 WARN_ON(ret);
 919
 920                 /* recover relocation */
 921                 ret = btrfs_recover_relocation(root);
 922                 WARN_ON(ret);
 923
 924                 sb->s_flags &= ~MS_RDONLY;
 925         }
 926
 927         return 0;
 928 }
 929
 930 /* Used to sort the devices by max_avail(descending sort) */
 931 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
 932                                        const void *dev_info2)
 933 {
 934         if (((struct btrfs_device_info *)dev_info1)->max_avail >
 935             ((struct btrfs_device_info *)dev_info2)->max_avail)
 936                 return -1;
 937         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
 938                  ((struct btrfs_device_info *)dev_info2)->max_avail)
 939                 return 1;
 940         else
 941         return 0;
 942 }
 943
 944 /*
 945  * sort the devices by max_avail, in which max free extent size of each device
 946  * is stored.(Descending Sort)
 947  */
 948 static inline void btrfs_descending_sort_devices(
 949                                         struct btrfs_device_info *devices,
 950                                         size_t nr_devices)
 951 {
 952         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
 953              btrfs_cmp_device_free_bytes, NULL);
 954 }
 955
 956 /*
 957  * The helper to calc the free space on the devices that can be used to store
 958  * file data.
 959  */
 960 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
 961 {
 962         struct btrfs_fs_info *fs_info = root->fs_info;
 963         struct btrfs_device_info *devices_info;
 964         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 965         struct btrfs_device *device;
 966         u64 skip_space;
 967         u64 type;
 968         u64 avail_space;
 969         u64 used_space;
 970         u64 min_stripe_size;
 971         int min_stripes = 1;
 972         int i = 0, nr_devices;
 973         int ret;
 974
 975         nr_devices = fs_info->fs_devices->rw_devices;
 976         BUG_ON(!nr_devices);
 977
 978         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
 979                                GFP_NOFS);
 980         if (!devices_info)
 981                 return -ENOMEM;
 982
 983         /* calc min stripe number for data space alloction */
 984         type = btrfs_get_alloc_profile(root, 1);
 985         if (type & BTRFS_BLOCK_GROUP_RAID0)
 986                 min_stripes = 2;
 987         else if (type & BTRFS_BLOCK_GROUP_RAID1)
 988                 min_stripes = 2;
 989         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 990                 min_stripes = 4;
 991
 992         if (type & BTRFS_BLOCK_GROUP_DUP)
 993                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
 994         else
 995                 min_stripe_size = BTRFS_STRIPE_LEN;
 996
 997         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
 998                 if (!device->in_fs_metadata)
 999                         continue;
1000
1001                 avail_space = device->total_bytes - device->bytes_used;
1002
1003                 /* align with stripe_len */
1004                 do_div(avail_space, BTRFS_STRIPE_LEN);
1005                 avail_space *= BTRFS_STRIPE_LEN;
1006
1007                 /*
1008                  * In order to avoid overwritting the superblock on the drive,
1009                  * btrfs starts at an offset of at least 1MB when doing chunk
1010                  * allocation.
1011                  */
1012                 skip_space = 1024 * 1024;
1013
1014                 /* user can set the offset in fs_info->alloc_start. */
1015                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1016                     device->total_bytes)
1017                         skip_space = max(fs_info->alloc_start, skip_space);
1018
1019                 /*
1020                  * btrfs can not use the free space in [0, skip_space - 1],
1021                  * we must subtract it from the total. In order to implement
1022                  * it, we account the used space in this range first.
1023                  */
1024                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1025                                                      &used_space);
1026                 if (ret) {
1027                         kfree(devices_info);
1028                         return ret;
1029                 }
1030
1031                 /* calc the free space in [0, skip_space - 1] */
1032                 skip_space -= used_space;
1033
1034                 /*
1035                  * we can use the free space in [0, skip_space - 1], subtract
1036                  * it from the total.
1037                  */
1038                 if (avail_space && avail_space >= skip_space)
1039                         avail_space -= skip_space;
1040                 else
1041                         avail_space = 0;
1042
1043                 if (avail_space < min_stripe_size)
1044                         continue;
1045
1046                 devices_info[i].dev = device;
1047                 devices_info[i].max_avail = avail_space;
1048
1049                 i++;
1050         }
1051
1052         nr_devices = i;
1053
1054         btrfs_descending_sort_devices(devices_info, nr_devices);
1055
1056         i = nr_devices - 1;
1057         avail_space = 0;
1058         while (nr_devices >= min_stripes) {
1059                 if (devices_info[i].max_avail >= min_stripe_size) {
1060                         int j;
1061                         u64 alloc_size;
1062
1063                         avail_space += devices_info[i].max_avail * min_stripes;
1064                         alloc_size = devices_info[i].max_avail;
1065                         for (j = i + 1 - min_stripes; j <= i; j++)
1066                                 devices_info[j].max_avail -= alloc_size;
1067                 }
1068                 i--;
1069                 nr_devices--;
1070         }
1071
1072         kfree(devices_info);
1073         *free_bytes = avail_space;
1074         return 0;
1075 }
1076
1077 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1078 {
1079         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
1080         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1081         struct list_head *head = &root->fs_info->space_info;
1082         struct btrfs_space_info *found;
1083         u64 total_used = 0;
1084         u64 total_free_data = 0;
1085         int bits = dentry->d_sb->s_blocksize_bits;
1086         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1087         int ret;
1088
1089         /* holding chunk_muext to avoid allocating new chunks */
1090         mutex_lock(&root->fs_info->chunk_mutex);
1091         rcu_read_lock();
1092         list_for_each_entry_rcu(found, head, list) {
1093                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1094                         total_free_data += found->disk_total - found->disk_used;
1095                         total_free_data -=
1096                                 btrfs_account_ro_block_groups_free_space(found);
1097                 }
1098
1099                 total_used += found->disk_used;
1100         }
1101         rcu_read_unlock();
1102
1103         buf->f_namelen = BTRFS_NAME_LEN;
1104         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1105         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1106         buf->f_bsize = dentry->d_sb->s_blocksize;
1107         buf->f_type = BTRFS_SUPER_MAGIC;
1108         buf->f_bavail = total_free_data;
1109         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1110         if (ret) {
1111                 mutex_unlock(&root->fs_info->chunk_mutex);
1112                 return ret;
1113         }
1114         buf->f_bavail += total_free_data;
1115         buf->f_bavail = buf->f_bavail >> bits;
1116         mutex_unlock(&root->fs_info->chunk_mutex);
1117
1118         /* We treat it as constant endianness (it doesn't matter _which_)
1119            because we want the fsid to come out the same whether mounted
1120            on a big-endian or little-endian host */
1121         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1122         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1123         /* Mask in the root object ID too, to disambiguate subvols */
1124         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1125         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1126
1127         return 0;
1128 }
1129
1130 static struct file_system_type btrfs_fs_type = {
1131         .owner          = THIS_MODULE,
1132         .name           = "btrfs",
1133         .mount          = btrfs_mount,
1134         .kill_sb        = kill_anon_super,
1135         .fs_flags       = FS_REQUIRES_DEV,
1136 };
1137
1138 /*
1139  * used by btrfsctl to scan devices when no FS is mounted
1140  */
1141 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1142                                 unsigned long arg)
1143 {
1144         struct btrfs_ioctl_vol_args *vol;
1145         struct btrfs_fs_devices *fs_devices;
1146         int ret = -ENOTTY;
1147
1148         if (!capable(CAP_SYS_ADMIN))
1149                 return -EPERM;
1150
1151         vol = memdup_user((void __user *)arg, sizeof(*vol));
1152         if (IS_ERR(vol))
1153                 return PTR_ERR(vol);
1154
1155         switch (cmd) {
1156         case BTRFS_IOC_SCAN_DEV:
1157                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1158                                             &btrfs_fs_type, &fs_devices);
1159                 break;
1160         }
1161
1162         kfree(vol);
1163         return ret;
1164 }
1165
1166 static int btrfs_freeze(struct super_block *sb)
1167 {
1168         struct btrfs_root *root = btrfs_sb(sb);
1169         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1170         mutex_lock(&root->fs_info->cleaner_mutex);
1171         return 0;
1172 }
1173
1174 static int btrfs_unfreeze(struct super_block *sb)
1175 {
1176         struct btrfs_root *root = btrfs_sb(sb);
1177         mutex_unlock(&root->fs_info->cleaner_mutex);
1178         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1179         return 0;
1180 }
1181
1182 static const struct super_operations btrfs_super_ops = {
1183         .drop_inode     = btrfs_drop_inode,
1184         .evict_inode    = btrfs_evict_inode,
1185         .put_super      = btrfs_put_super,
1186         .sync_fs        = btrfs_sync_fs,
1187         .show_options   = btrfs_show_options,
1188         .write_inode    = btrfs_write_inode,
1189         .dirty_inode    = btrfs_dirty_inode,
1190         .alloc_inode    = btrfs_alloc_inode,
1191         .destroy_inode  = btrfs_destroy_inode,
1192         .statfs         = btrfs_statfs,
1193         .remount_fs     = btrfs_remount,
1194         .freeze_fs      = btrfs_freeze,
1195         .unfreeze_fs    = btrfs_unfreeze,
1196 };
1197
1198 static const struct file_operations btrfs_ctl_fops = {
1199         .unlocked_ioctl  = btrfs_control_ioctl,
1200         .compat_ioctl = btrfs_control_ioctl,
1201         .owner   = THIS_MODULE,
1202         .llseek = noop_llseek,
1203 };
1204
1205 static struct miscdevice btrfs_misc = {
1206         .minor          = BTRFS_MINOR,
1207         .name           = "btrfs-control",
1208         .fops           = &btrfs_ctl_fops
1209 };
1210
1211 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1212 MODULE_ALIAS("devname:btrfs-control");
1213
1214 static int btrfs_interface_init(void)
1215 {
1216         return misc_register(&btrfs_misc);
1217 }
1218
1219 static void btrfs_interface_exit(void)
1220 {
1221         if (misc_deregister(&btrfs_misc) < 0)
1222                 printk(KERN_INFO "misc_deregister failed for control device");
1223 }
1224
1225 static int __init init_btrfs_fs(void)
1226 {
1227         int err;
1228
1229         err = btrfs_init_sysfs();
1230         if (err)
1231                 return err;
1232
1233         err = btrfs_init_compress();
1234         if (err)
1235                 goto free_sysfs;
1236
1237         err = btrfs_init_cachep();
1238         if (err)
1239                 goto free_compress;
1240
1241         err = extent_io_init();
1242         if (err)
1243                 goto free_cachep;
1244
1245         err = extent_map_init();
1246         if (err)
1247                 goto free_extent_io;
1248
1249         err = btrfs_delayed_inode_init();
1250         if (err)
1251                 goto free_extent_map;
1252
1253         err = btrfs_interface_init();
1254         if (err)
1255                 goto free_delayed_inode;
1256
1257         err = register_filesystem(&btrfs_fs_type);
1258         if (err)
1259                 goto unregister_ioctl;
1260
1261         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1262         return 0;
1263
1264 unregister_ioctl:
1265         btrfs_interface_exit();
1266 free_delayed_inode:
1267         btrfs_delayed_inode_exit();
1268 free_extent_map:
1269         extent_map_exit();
1270 free_extent_io:
1271         extent_io_exit();
1272 free_cachep:
1273         btrfs_destroy_cachep();
1274 free_compress:
1275         btrfs_exit_compress();
1276 free_sysfs:
1277         btrfs_exit_sysfs();
1278         return err;
1279 }
1280
1281 static void __exit exit_btrfs_fs(void)
1282 {
1283         btrfs_destroy_cachep();
1284         btrfs_delayed_inode_exit();
1285         extent_map_exit();
1286         extent_io_exit();
1287         btrfs_interface_exit();
1288         unregister_filesystem(&btrfs_fs_type);
1289         btrfs_exit_sysfs();
1290         btrfs_cleanup_fs_uuids();
1291         btrfs_exit_compress();
1292 }
1293
1294 module_init(init_btrfs_fs)
1295 module_exit(exit_btrfs_fs)
1296
1297 MODULE_LICENSE("GPL");