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