fs/btrfs/super.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/time.h>
  26 #include <linux/init.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mount.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/parser.h>
  37 #include <linux/ctype.h>
  38 #include <linux/namei.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/magic.h>
  41 #include <linux/slab.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "xattr.h"
  50 #include "volumes.h"
  51 #include "version.h"
  52 #include "export.h"
  53 #include "compression.h"
  54
  55 static const struct super_operations btrfs_super_ops;
  56
  57 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
  58                                       char nbuf[16])
  59 {
  60         char *errstr = NULL;
  61
  62         switch (errno) {
  63         case -EIO:
  64                 errstr = "IO failure";
  65                 break;
  66         case -ENOMEM:
  67                 errstr = "Out of memory";
  68                 break;
  69         case -EROFS:
  70                 errstr = "Readonly filesystem";
  71                 break;
  72         default:
  73                 if (nbuf) {
  74                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  75                                 errstr = nbuf;
  76                 }
  77                 break;
  78         }
  79
  80         return errstr;
  81 }
  82
  83 static void __save_error_info(struct btrfs_fs_info *fs_info)
  84 {
  85         /*
  86          * today we only save the error info into ram.  Long term we'll
  87          * also send it down to the disk
  88          */
  89         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
  90 }
  91
  92 /* NOTE:
  93  *      We move write_super stuff at umount in order to avoid deadlock
  94  *      for umount hold all lock.
  95  */
  96 static void save_error_info(struct btrfs_fs_info *fs_info)
  97 {
  98         __save_error_info(fs_info);
  99 }
 100
 101 /* btrfs handle error by forcing the filesystem readonly */
 102 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 103 {
 104         struct super_block *sb = fs_info->sb;
 105
 106         if (sb->s_flags & MS_RDONLY)
 107                 return;
 108
 109         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 110                 sb->s_flags |= MS_RDONLY;
 111                 printk(KERN_INFO "btrfs is forced readonly\n");
 112         }
 113 }
 114
 115 /*
 116  * __btrfs_std_error decodes expected errors from the caller and
 117  * invokes the approciate error response.
 118  */
 119 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 120                      unsigned int line, int errno)
 121 {
 122         struct super_block *sb = fs_info->sb;
 123         char nbuf[16];
 124         const char *errstr;
 125
 126         /*
 127          * Special case: if the error is EROFS, and we're already
 128          * under MS_RDONLY, then it is safe here.
 129          */
 130         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 131                 return;
 132
 133         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 134         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 135                 sb->s_id, function, line, errstr);
 136         save_error_info(fs_info);
 137
 138         btrfs_handle_error(fs_info);
 139 }
 140
 141 static void btrfs_put_super(struct super_block *sb)
 142 {
 143         struct btrfs_root *root = btrfs_sb(sb);
 144         int ret;
 145
 146         ret = close_ctree(root);
 147         sb->s_fs_info = NULL;
 148
 149         (void)ret; /* FIXME: need to fix VFS to return error? */
 150 }
 151
 152 enum {
 153         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 154         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 155         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 156         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 157         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 158         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 159         Opt_enospc_debug, Opt_err,
 160 };
 161
 162 static match_table_t tokens = {
 163         {Opt_degraded, "degraded"},
 164         {Opt_subvol, "subvol=%s"},
 165         {Opt_subvolid, "subvolid=%d"},
 166         {Opt_device, "device=%s"},
 167         {Opt_nodatasum, "nodatasum"},
 168         {Opt_nodatacow, "nodatacow"},
 169         {Opt_nobarrier, "nobarrier"},
 170         {Opt_max_inline, "max_inline=%s"},
 171         {Opt_alloc_start, "alloc_start=%s"},
 172         {Opt_thread_pool, "thread_pool=%d"},
 173         {Opt_compress, "compress"},
 174         {Opt_compress_type, "compress=%s"},
 175         {Opt_compress_force, "compress-force"},
 176         {Opt_compress_force_type, "compress-force=%s"},
 177         {Opt_ssd, "ssd"},
 178         {Opt_ssd_spread, "ssd_spread"},
 179         {Opt_nossd, "nossd"},
 180         {Opt_noacl, "noacl"},
 181         {Opt_notreelog, "notreelog"},
 182         {Opt_flushoncommit, "flushoncommit"},
 183         {Opt_ratio, "metadata_ratio=%d"},
 184         {Opt_discard, "discard"},
 185         {Opt_space_cache, "space_cache"},
 186         {Opt_clear_cache, "clear_cache"},
 187         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 188         {Opt_enospc_debug, "enospc_debug"},
 189         {Opt_err, NULL},
 190 };
 191
 192 /*
 193  * Regular mount options parser.  Everything that is needed only when
 194  * reading in a new superblock is parsed here.
 195  */
 196 int btrfs_parse_options(struct btrfs_root *root, char *options)
 197 {
 198         struct btrfs_fs_info *info = root->fs_info;
 199         substring_t args[MAX_OPT_ARGS];
 200         char *p, *num, *orig;
 201         int intarg;
 202         int ret = 0;
 203         char *compress_type;
 204         bool compress_force = false;
 205
 206         if (!options)
 207                 return 0;
 208
 209         /*
 210          * strsep changes the string, duplicate it because parse_options
 211          * gets called twice
 212          */
 213         options = kstrdup(options, GFP_NOFS);
 214         if (!options)
 215                 return -ENOMEM;
 216
 217         orig = options;
 218
 219         while ((p = strsep(&options, ",")) != NULL) {
 220                 int token;
 221                 if (!*p)
 222                         continue;
 223
 224                 token = match_token(p, tokens, args);
 225                 switch (token) {
 226                 case Opt_degraded:
 227                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 228                         btrfs_set_opt(info->mount_opt, DEGRADED);
 229                         break;
 230                 case Opt_subvol:
 231                 case Opt_subvolid:
 232                 case Opt_device:
 233                         /*
 234                          * These are parsed by btrfs_parse_early_options
 235                          * and can be happily ignored here.
 236                          */
 237                         break;
 238                 case Opt_nodatasum:
 239                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 240                         btrfs_set_opt(info->mount_opt, NODATASUM);
 241                         break;
 242                 case Opt_nodatacow:
 243                         printk(KERN_INFO "btrfs: setting nodatacow\n");
 244                         btrfs_set_opt(info->mount_opt, NODATACOW);
 245                         btrfs_set_opt(info->mount_opt, NODATASUM);
 246                         break;
 247                 case Opt_compress_force:
 248                 case Opt_compress_force_type:
 249                         compress_force = true;
 250                 case Opt_compress:
 251                 case Opt_compress_type:
 252                         if (token == Opt_compress ||
 253                             token == Opt_compress_force ||
 254                             strcmp(args[0].from, "zlib") == 0) {
 255                                 compress_type = "zlib";
 256                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 257                         } else if (strcmp(args[0].from, "lzo") == 0) {
 258                                 compress_type = "lzo";
 259                                 info->compress_type = BTRFS_COMPRESS_LZO;
 260                         } else {
 261                                 ret = -EINVAL;
 262                                 goto out;
 263                         }
 264
 265                         btrfs_set_opt(info->mount_opt, COMPRESS);
 266                         if (compress_force) {
 267                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 268                                 pr_info("btrfs: force %s compression\n",
 269                                         compress_type);
 270                         } else
 271                                 pr_info("btrfs: use %s compression\n",
 272                                         compress_type);
 273                         break;
 274                 case Opt_ssd:
 275                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 276                         btrfs_set_opt(info->mount_opt, SSD);
 277                         break;
 278                 case Opt_ssd_spread:
 279                         printk(KERN_INFO "btrfs: use spread ssd "
 280                                "allocation scheme\n");
 281                         btrfs_set_opt(info->mount_opt, SSD);
 282                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 283                         break;
 284                 case Opt_nossd:
 285                         printk(KERN_INFO "btrfs: not using ssd allocation "
 286                                "scheme\n");
 287                         btrfs_set_opt(info->mount_opt, NOSSD);
 288                         btrfs_clear_opt(info->mount_opt, SSD);
 289                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 290                         break;
 291                 case Opt_nobarrier:
 292                         printk(KERN_INFO "btrfs: turning off barriers\n");
 293                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 294                         break;
 295                 case Opt_thread_pool:
 296                         intarg = 0;
 297                         match_int(&args[0], &intarg);
 298                         if (intarg) {
 299                                 info->thread_pool_size = intarg;
 300                                 printk(KERN_INFO "btrfs: thread pool %d\n",
 301                                        info->thread_pool_size);
 302                         }
 303                         break;
 304                 case Opt_max_inline:
 305                         num = match_strdup(&args[0]);
 306                         if (num) {
 307                                 info->max_inline = memparse(num, NULL);
 308                                 kfree(num);
 309
 310                                 if (info->max_inline) {
 311                                         info->max_inline = max_t(u64,
 312                                                 info->max_inline,
 313                                                 root->sectorsize);
 314                                 }
 315                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 316                                         (unsigned long long)info->max_inline);
 317                         }
 318                         break;
 319                 case Opt_alloc_start:
 320                         num = match_strdup(&args[0]);
 321                         if (num) {
 322                                 info->alloc_start = memparse(num, NULL);
 323                                 kfree(num);
 324                                 printk(KERN_INFO
 325                                         "btrfs: allocations start at %llu\n",
 326                                         (unsigned long long)info->alloc_start);
 327                         }
 328                         break;
 329                 case Opt_noacl:
 330                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 331                         break;
 332                 case Opt_notreelog:
 333                         printk(KERN_INFO "btrfs: disabling tree log\n");
 334                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 335                         break;
 336                 case Opt_flushoncommit:
 337                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 338                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 339                         break;
 340                 case Opt_ratio:
 341                         intarg = 0;
 342                         match_int(&args[0], &intarg);
 343                         if (intarg) {
 344                                 info->metadata_ratio = intarg;
 345                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 346                                        info->metadata_ratio);
 347                         }
 348                         break;
 349                 case Opt_discard:
 350                         btrfs_set_opt(info->mount_opt, DISCARD);
 351                         break;
 352                 case Opt_space_cache:
 353                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
 354                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 355                         break;
 356                 case Opt_clear_cache:
 357                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 358                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 359                         break;
 360                 case Opt_user_subvol_rm_allowed:
 361                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 362                         break;
 363                 case Opt_enospc_debug:
 364                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 365                         break;
 366                 case Opt_err:
 367                         printk(KERN_INFO "btrfs: unrecognized mount option "
 368                                "'%s'\n", p);
 369                         ret = -EINVAL;
 370                         goto out;
 371                 default:
 372                         break;
 373                 }
 374         }
 375 out:
 376         kfree(orig);
 377         return ret;
 378 }
 379
 380 /*
 381  * Parse mount options that are required early in the mount process.
 382  *
 383  * All other options will be parsed on much later in the mount process and
 384  * only when we need to allocate a new super block.
 385  */
 386 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 387                 void *holder, char **subvol_name, u64 *subvol_objectid,
 388                 struct btrfs_fs_devices **fs_devices)
 389 {
 390         substring_t args[MAX_OPT_ARGS];
 391         char *opts, *orig, *p;
 392         int error = 0;
 393         int intarg;
 394
 395         if (!options)
 396                 goto out;
 397
 398         /*
 399          * strsep changes the string, duplicate it because parse_options
 400          * gets called twice
 401          */
 402         opts = kstrdup(options, GFP_KERNEL);
 403         if (!opts)
 404                 return -ENOMEM;
 405         orig = opts;
 406
 407         while ((p = strsep(&opts, ",")) != NULL) {
 408                 int token;
 409                 if (!*p)
 410                         continue;
 411
 412                 token = match_token(p, tokens, args);
 413                 switch (token) {
 414                 case Opt_subvol:
 415                         *subvol_name = match_strdup(&args[0]);
 416                         break;
 417                 case Opt_subvolid:
 418                         intarg = 0;
 419                         error = match_int(&args[0], &intarg);
 420                         if (!error) {
 421                                 /* we want the original fs_tree */
 422                                 if (!intarg)
 423                                         *subvol_objectid =
 424                                                 BTRFS_FS_TREE_OBJECTID;
 425                                 else
 426                                         *subvol_objectid = intarg;
 427                         }
 428                         break;
 429                 case Opt_device:
 430                         error = btrfs_scan_one_device(match_strdup(&args[0]),
 431                                         flags, holder, fs_devices);
 432                         if (error)
 433                                 goto out_free_opts;
 434                         break;
 435                 default:
 436                         break;
 437                 }
 438         }
 439
 440  out_free_opts:
 441         kfree(orig);
 442  out:
 443         /*
 444          * If no subvolume name is specified we use the default one.  Allocate
 445          * a copy of the string "." here so that code later in the
 446          * mount path doesn't care if it's the default volume or another one.
 447          */
 448         if (!*subvol_name) {
 449                 *subvol_name = kstrdup(".", GFP_KERNEL);
 450                 if (!*subvol_name)
 451                         return -ENOMEM;
 452         }
 453         return error;
 454 }
 455
 456 static struct dentry *get_default_root(struct super_block *sb,
 457                                        u64 subvol_objectid)
 458 {
 459         struct btrfs_root *root = sb->s_fs_info;
 460         struct btrfs_root *new_root;
 461         struct btrfs_dir_item *di;
 462         struct btrfs_path *path;
 463         struct btrfs_key location;
 464         struct inode *inode;
 465         struct dentry *dentry;
 466         u64 dir_id;
 467         int new = 0;
 468
 469         /*
 470          * We have a specific subvol we want to mount, just setup location and
 471          * go look up the root.
 472          */
 473         if (subvol_objectid) {
 474                 location.objectid = subvol_objectid;
 475                 location.type = BTRFS_ROOT_ITEM_KEY;
 476                 location.offset = (u64)-1;
 477                 goto find_root;
 478         }
 479
 480         path = btrfs_alloc_path();
 481         if (!path)
 482                 return ERR_PTR(-ENOMEM);
 483         path->leave_spinning = 1;
 484
 485         /*
 486          * Find the "default" dir item which points to the root item that we
 487          * will mount by default if we haven't been given a specific subvolume
 488          * to mount.
 489          */
 490         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
 491         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 492         if (IS_ERR(di))
 493                 return ERR_CAST(di);
 494         if (!di) {
 495                 /*
 496                  * Ok the default dir item isn't there.  This is weird since
 497                  * it's always been there, but don't freak out, just try and
 498                  * mount to root most subvolume.
 499                  */
 500                 btrfs_free_path(path);
 501                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 502                 new_root = root->fs_info->fs_root;
 503                 goto setup_root;
 504         }
 505
 506         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 507         btrfs_free_path(path);
 508
 509 find_root:
 510         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
 511         if (IS_ERR(new_root))
 512                 return ERR_CAST(new_root);
 513
 514         if (btrfs_root_refs(&new_root->root_item) == 0)
 515                 return ERR_PTR(-ENOENT);
 516
 517         dir_id = btrfs_root_dirid(&new_root->root_item);
 518 setup_root:
 519         location.objectid = dir_id;
 520         location.type = BTRFS_INODE_ITEM_KEY;
 521         location.offset = 0;
 522
 523         inode = btrfs_iget(sb, &location, new_root, &new);
 524         if (IS_ERR(inode))
 525                 return ERR_CAST(inode);
 526
 527         /*
 528          * If we're just mounting the root most subvol put the inode and return
 529          * a reference to the dentry.  We will have already gotten a reference
 530          * to the inode in btrfs_fill_super so we're good to go.
 531          */
 532         if (!new && sb->s_root->d_inode == inode) {
 533                 iput(inode);
 534                 return dget(sb->s_root);
 535         }
 536
 537         if (new) {
 538                 const struct qstr name = { .name = "/", .len = 1 };
 539
 540                 /*
 541                  * New inode, we need to make the dentry a sibling of s_root so
 542                  * everything gets cleaned up properly on unmount.
 543                  */
 544                 dentry = d_alloc(sb->s_root, &name);
 545                 if (!dentry) {
 546                         iput(inode);
 547                         return ERR_PTR(-ENOMEM);
 548                 }
 549                 d_splice_alias(inode, dentry);
 550         } else {
 551                 /*
 552                  * We found the inode in cache, just find a dentry for it and
 553                  * put the reference to the inode we just got.
 554                  */
 555                 dentry = d_find_alias(inode);
 556                 iput(inode);
 557         }
 558
 559         return dentry;
 560 }
 561
 562 static int btrfs_fill_super(struct super_block *sb,
 563                             struct btrfs_fs_devices *fs_devices,
 564                             void *data, int silent)
 565 {
 566         struct inode *inode;
 567         struct dentry *root_dentry;
 568         struct btrfs_root *tree_root;
 569         struct btrfs_key key;
 570         int err;
 571
 572         sb->s_maxbytes = MAX_LFS_FILESIZE;
 573         sb->s_magic = BTRFS_SUPER_MAGIC;
 574         sb->s_op = &btrfs_super_ops;
 575         sb->s_export_op = &btrfs_export_ops;
 576         sb->s_xattr = btrfs_xattr_handlers;
 577         sb->s_time_gran = 1;
 578 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 579         sb->s_flags |= MS_POSIXACL;
 580 #endif
 581
 582         tree_root = open_ctree(sb, fs_devices, (char *)data);
 583
 584         if (IS_ERR(tree_root)) {
 585                 printk("btrfs: open_ctree failed\n");
 586                 return PTR_ERR(tree_root);
 587         }
 588         sb->s_fs_info = tree_root;
 589
 590         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 591         key.type = BTRFS_INODE_ITEM_KEY;
 592         key.offset = 0;
 593         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
 594         if (IS_ERR(inode)) {
 595                 err = PTR_ERR(inode);
 596                 goto fail_close;
 597         }
 598
 599         root_dentry = d_alloc_root(inode);
 600         if (!root_dentry) {
 601                 iput(inode);
 602                 err = -ENOMEM;
 603                 goto fail_close;
 604         }
 605
 606         sb->s_root = root_dentry;
 607
 608         save_mount_options(sb, data);
 609         return 0;
 610
 611 fail_close:
 612         close_ctree(tree_root);
 613         return err;
 614 }
 615
 616 int btrfs_sync_fs(struct super_block *sb, int wait)
 617 {
 618         struct btrfs_trans_handle *trans;
 619         struct btrfs_root *root = btrfs_sb(sb);
 620         int ret;
 621
 622         if (!wait) {
 623                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 624                 return 0;
 625         }
 626
 627         btrfs_start_delalloc_inodes(root, 0);
 628         btrfs_wait_ordered_extents(root, 0, 0);
 629
 630         trans = btrfs_start_transaction(root, 0);
 631         if (IS_ERR(trans))
 632                 return PTR_ERR(trans);
 633         ret = btrfs_commit_transaction(trans, root);
 634         return ret;
 635 }
 636
 637 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 638 {
 639         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
 640         struct btrfs_fs_info *info = root->fs_info;
 641
 642         if (btrfs_test_opt(root, DEGRADED))
 643                 seq_puts(seq, ",degraded");
 644         if (btrfs_test_opt(root, NODATASUM))
 645                 seq_puts(seq, ",nodatasum");
 646         if (btrfs_test_opt(root, NODATACOW))
 647                 seq_puts(seq, ",nodatacow");
 648         if (btrfs_test_opt(root, NOBARRIER))
 649                 seq_puts(seq, ",nobarrier");
 650         if (info->max_inline != 8192 * 1024)
 651                 seq_printf(seq, ",max_inline=%llu",
 652                            (unsigned long long)info->max_inline);
 653         if (info->alloc_start != 0)
 654                 seq_printf(seq, ",alloc_start=%llu",
 655                            (unsigned long long)info->alloc_start);
 656         if (info->thread_pool_size !=  min_t(unsigned long,
 657                                              num_online_cpus() + 2, 8))
 658                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 659         if (btrfs_test_opt(root, COMPRESS))
 660                 seq_puts(seq, ",compress");
 661         if (btrfs_test_opt(root, NOSSD))
 662                 seq_puts(seq, ",nossd");
 663         if (btrfs_test_opt(root, SSD_SPREAD))
 664                 seq_puts(seq, ",ssd_spread");
 665         else if (btrfs_test_opt(root, SSD))
 666                 seq_puts(seq, ",ssd");
 667         if (btrfs_test_opt(root, NOTREELOG))
 668                 seq_puts(seq, ",notreelog");
 669         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 670                 seq_puts(seq, ",flushoncommit");
 671         if (btrfs_test_opt(root, DISCARD))
 672                 seq_puts(seq, ",discard");
 673         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 674                 seq_puts(seq, ",noacl");
 675         return 0;
 676 }
 677
 678 static int btrfs_test_super(struct super_block *s, void *data)
 679 {
 680         struct btrfs_root *test_root = data;
 681         struct btrfs_root *root = btrfs_sb(s);
 682
 683         /*
 684          * If this super block is going away, return false as it
 685          * can't match as an existing super block.
 686          */
 687         if (!atomic_read(&s->s_active))
 688                 return 0;
 689         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
 690 }
 691
 692 static int btrfs_set_super(struct super_block *s, void *data)
 693 {
 694         s->s_fs_info = data;
 695
 696         return set_anon_super(s, data);
 697 }
 698
 699
 700 /*
 701  * Find a superblock for the given device / mount point.
 702  *
 703  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
 704  *        for multiple device setup.  Make sure to keep it in sync.
 705  */
 706 static int btrfs_get_sb(struct file_system_type *fs_type, int flags,
 707                 const char *dev_name, void *data, struct vfsmount *mnt)
 708 {
 709         struct block_device *bdev = NULL;
 710         struct super_block *s;
 711         struct dentry *root;
 712         struct btrfs_fs_devices *fs_devices = NULL;
 713         struct btrfs_root *tree_root = NULL;
 714         struct btrfs_fs_info *fs_info = NULL;
 715         fmode_t mode = FMODE_READ;
 716         char *subvol_name = NULL;
 717         u64 subvol_objectid = 0;
 718         int error = 0;
 719
 720         if (!(flags & MS_RDONLY))
 721                 mode |= FMODE_WRITE;
 722
 723         error = btrfs_parse_early_options(data, mode, fs_type,
 724                                           &subvol_name, &subvol_objectid,
 725                                           &fs_devices);
 726         if (error)
 727                 return error;
 728
 729         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
 730         if (error)
 731                 goto error_free_subvol_name;
 732
 733         error = btrfs_open_devices(fs_devices, mode, fs_type);
 734         if (error)
 735                 goto error_free_subvol_name;
 736
 737         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
 738                 error = -EACCES;
 739                 goto error_close_devices;
 740         }
 741
 742         /*
 743          * Setup a dummy root and fs_info for test/set super.  This is because
 744          * we don't actually fill this stuff out until open_ctree, but we need
 745          * it for searching for existing supers, so this lets us do that and
 746          * then open_ctree will properly initialize everything later.
 747          */
 748         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
 749         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
 750         if (!fs_info || !tree_root) {
 751                 error = -ENOMEM;
 752                 goto error_close_devices;
 753         }
 754         fs_info->tree_root = tree_root;
 755         fs_info->fs_devices = fs_devices;
 756         tree_root->fs_info = fs_info;
 757
 758         bdev = fs_devices->latest_bdev;
 759         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
 760         if (IS_ERR(s))
 761                 goto error_s;
 762
 763         if (s->s_root) {
 764                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 765                         deactivate_locked_super(s);
 766                         error = -EBUSY;
 767                         goto error_close_devices;
 768                 }
 769
 770                 btrfs_close_devices(fs_devices);
 771                 kfree(fs_info);
 772                 kfree(tree_root);
 773         } else {
 774                 char b[BDEVNAME_SIZE];
 775
 776                 s->s_flags = flags;
 777                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 778                 error = btrfs_fill_super(s, fs_devices, data,
 779                                          flags & MS_SILENT ? 1 : 0);
 780                 if (error) {
 781                         deactivate_locked_super(s);
 782                         goto error_free_subvol_name;
 783                 }
 784
 785                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
 786                 s->s_flags |= MS_ACTIVE;
 787         }
 788
 789         root = get_default_root(s, subvol_objectid);
 790         if (IS_ERR(root)) {
 791                 error = PTR_ERR(root);
 792                 deactivate_locked_super(s);
 793                 goto error_free_subvol_name;
 794         }
 795         /* if they gave us a subvolume name bind mount into that */
 796         if (strcmp(subvol_name, ".")) {
 797                 struct dentry *new_root;
 798                 mutex_lock(&root->d_inode->i_mutex);
 799                 new_root = lookup_one_len(subvol_name, root,
 800                                       strlen(subvol_name));
 801                 mutex_unlock(&root->d_inode->i_mutex);
 802
 803                 if (IS_ERR(new_root)) {
 804                         dput(root);
 805                         deactivate_locked_super(s);
 806                         error = PTR_ERR(new_root);
 807                         goto error_free_subvol_name;
 808                 }
 809                 if (!new_root->d_inode) {
 810                         dput(root);
 811                         dput(new_root);
 812                         deactivate_locked_super(s);
 813                         error = -ENXIO;
 814                         goto error_free_subvol_name;
 815                 }
 816                 dput(root);
 817                 root = new_root;
 818         }
 819
 820         mnt->mnt_sb = s;
 821         mnt->mnt_root = root;
 822
 823         kfree(subvol_name);
 824         return 0;
 825
 826 error_s:
 827         error = PTR_ERR(s);
 828 error_close_devices:
 829         btrfs_close_devices(fs_devices);
 830         kfree(fs_info);
 831         kfree(tree_root);
 832 error_free_subvol_name:
 833         kfree(subvol_name);
 834         return error;
 835 }
 836
 837 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 838 {
 839         struct btrfs_root *root = btrfs_sb(sb);
 840         int ret;
 841
 842         ret = btrfs_parse_options(root, data);
 843         if (ret)
 844                 return -EINVAL;
 845
 846         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 847                 return 0;
 848
 849         if (*flags & MS_RDONLY) {
 850                 sb->s_flags |= MS_RDONLY;
 851
 852                 ret =  btrfs_commit_super(root);
 853                 WARN_ON(ret);
 854         } else {
 855                 if (root->fs_info->fs_devices->rw_devices == 0)
 856                         return -EACCES;
 857
 858                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
 859                         return -EINVAL;
 860
 861                 ret = btrfs_cleanup_fs_roots(root->fs_info);
 862                 WARN_ON(ret);
 863
 864                 /* recover relocation */
 865                 ret = btrfs_recover_relocation(root);
 866                 WARN_ON(ret);
 867
 868                 sb->s_flags &= ~MS_RDONLY;
 869         }
 870
 871         return 0;
 872 }
 873
 874 /*
 875  * The helper to calc the free space on the devices that can be used to store
 876  * file data.
 877  */
 878 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
 879 {
 880         struct btrfs_fs_info *fs_info = root->fs_info;
 881         struct btrfs_device_info *devices_info;
 882         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 883         struct btrfs_device *device;
 884         u64 skip_space;
 885         u64 type;
 886         u64 avail_space;
 887         u64 used_space;
 888         u64 min_stripe_size;
 889         int min_stripes = 1;
 890         int i = 0, nr_devices;
 891         int ret;
 892
 893         nr_devices = fs_info->fs_devices->rw_devices;
 894         BUG_ON(!nr_devices);
 895
 896         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
 897                                GFP_NOFS);
 898         if (!devices_info)
 899                 return -ENOMEM;
 900
 901         /* calc min stripe number for data space alloction */
 902         type = btrfs_get_alloc_profile(root, 1);
 903         if (type & BTRFS_BLOCK_GROUP_RAID0)
 904                 min_stripes = 2;
 905         else if (type & BTRFS_BLOCK_GROUP_RAID1)
 906                 min_stripes = 2;
 907         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 908                 min_stripes = 4;
 909
 910         if (type & BTRFS_BLOCK_GROUP_DUP)
 911                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
 912         else
 913                 min_stripe_size = BTRFS_STRIPE_LEN;
 914
 915         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
 916                 if (!device->in_fs_metadata)
 917                         continue;
 918
 919                 avail_space = device->total_bytes - device->bytes_used;
 920
 921                 /* align with stripe_len */
 922                 do_div(avail_space, BTRFS_STRIPE_LEN);
 923                 avail_space *= BTRFS_STRIPE_LEN;
 924
 925                 /*
 926                  * In order to avoid overwritting the superblock on the drive,
 927                  * btrfs starts at an offset of at least 1MB when doing chunk
 928                  * allocation.
 929                  */
 930                 skip_space = 1024 * 1024;
 931
 932                 /* user can set the offset in fs_info->alloc_start. */
 933                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
 934                     device->total_bytes)
 935                         skip_space = max(fs_info->alloc_start, skip_space);
 936
 937                 /*
 938                  * btrfs can not use the free space in [0, skip_space - 1],
 939                  * we must subtract it from the total. In order to implement
 940                  * it, we account the used space in this range first.
 941                  */
 942                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
 943                                                      &used_space);
 944                 if (ret) {
 945                         kfree(devices_info);
 946                         return ret;
 947                 }
 948
 949                 /* calc the free space in [0, skip_space - 1] */
 950                 skip_space -= used_space;
 951
 952                 /*
 953                  * we can use the free space in [0, skip_space - 1], subtract
 954                  * it from the total.
 955                  */
 956                 if (avail_space && avail_space >= skip_space)
 957                         avail_space -= skip_space;
 958                 else
 959                         avail_space = 0;
 960
 961                 if (avail_space < min_stripe_size)
 962                         continue;
 963
 964                 devices_info[i].dev = device;
 965                 devices_info[i].max_avail = avail_space;
 966
 967                 i++;
 968         }
 969
 970         nr_devices = i;
 971
 972         btrfs_descending_sort_devices(devices_info, nr_devices);
 973
 974         i = nr_devices - 1;
 975         avail_space = 0;
 976         while (nr_devices >= min_stripes) {
 977                 if (devices_info[i].max_avail >= min_stripe_size) {
 978                         int j;
 979                         u64 alloc_size;
 980
 981                         avail_space += devices_info[i].max_avail * min_stripes;
 982                         alloc_size = devices_info[i].max_avail;
 983                         for (j = i + 1 - min_stripes; j <= i; j++)
 984                                 devices_info[j].max_avail -= alloc_size;
 985                 }
 986                 i--;
 987                 nr_devices--;
 988         }
 989
 990         kfree(devices_info);
 991         *free_bytes = avail_space;
 992         return 0;
 993 }
 994
 995 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 996 {
 997         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
 998         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
 999         struct list_head *head = &root->fs_info->space_info;
1000         struct btrfs_space_info *found;
1001         u64 total_used = 0;
1002         u64 total_free_data = 0;
1003         int bits = dentry->d_sb->s_blocksize_bits;
1004         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1005         int ret;
1006
1007         /* holding chunk_muext to avoid allocating new chunks */
1008         mutex_lock(&root->fs_info->chunk_mutex);
1009         rcu_read_lock();
1010         list_for_each_entry_rcu(found, head, list) {
1011                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1012                         total_free_data += found->disk_total - found->disk_used;
1013                         total_free_data -=
1014                                 btrfs_account_ro_block_groups_free_space(found);
1015                 }
1016
1017                 total_used += found->disk_used;
1018         }
1019         rcu_read_unlock();
1020
1021         buf->f_namelen = BTRFS_NAME_LEN;
1022         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1023         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1024         buf->f_bsize = dentry->d_sb->s_blocksize;
1025         buf->f_type = BTRFS_SUPER_MAGIC;
1026         buf->f_bavail = total_free_data;
1027         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1028         if (ret) {
1029                 mutex_unlock(&root->fs_info->chunk_mutex);
1030                 return ret;
1031         }
1032         buf->f_bavail += total_free_data;
1033         buf->f_bavail = buf->f_bavail >> bits;
1034         mutex_unlock(&root->fs_info->chunk_mutex);
1035
1036         /* We treat it as constant endianness (it doesn't matter _which_)
1037            because we want the fsid to come out the same whether mounted
1038            on a big-endian or little-endian host */
1039         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1040         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1041         /* Mask in the root object ID too, to disambiguate subvols */
1042         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1043         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1044
1045         return 0;
1046 }
1047
1048 static struct file_system_type btrfs_fs_type = {
1049         .owner          = THIS_MODULE,
1050         .name           = "btrfs",
1051         .get_sb         = btrfs_get_sb,
1052         .kill_sb        = kill_anon_super,
1053         .fs_flags       = FS_REQUIRES_DEV,
1054 };
1055
1056 /*
1057  * used by btrfsctl to scan devices when no FS is mounted
1058  */
1059 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1060                                 unsigned long arg)
1061 {
1062         struct btrfs_ioctl_vol_args *vol;
1063         struct btrfs_fs_devices *fs_devices;
1064         int ret = -ENOTTY;
1065
1066         if (!capable(CAP_SYS_ADMIN))
1067                 return -EPERM;
1068
1069         vol = memdup_user((void __user *)arg, sizeof(*vol));
1070         if (IS_ERR(vol))
1071                 return PTR_ERR(vol);
1072
1073         switch (cmd) {
1074         case BTRFS_IOC_SCAN_DEV:
1075                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1076                                             &btrfs_fs_type, &fs_devices);
1077                 break;
1078         }
1079
1080         kfree(vol);
1081         return ret;
1082 }
1083
1084 static int btrfs_freeze(struct super_block *sb)
1085 {
1086         struct btrfs_root *root = btrfs_sb(sb);
1087         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1088         mutex_lock(&root->fs_info->cleaner_mutex);
1089         return 0;
1090 }
1091
1092 static int btrfs_unfreeze(struct super_block *sb)
1093 {
1094         struct btrfs_root *root = btrfs_sb(sb);
1095         mutex_unlock(&root->fs_info->cleaner_mutex);
1096         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1097         return 0;
1098 }
1099
1100 static const struct super_operations btrfs_super_ops = {
1101         .drop_inode     = btrfs_drop_inode,
1102         .evict_inode    = btrfs_evict_inode,
1103         .put_super      = btrfs_put_super,
1104         .sync_fs        = btrfs_sync_fs,
1105         .show_options   = btrfs_show_options,
1106         .write_inode    = btrfs_write_inode,
1107         .dirty_inode    = btrfs_dirty_inode,
1108         .alloc_inode    = btrfs_alloc_inode,
1109         .destroy_inode  = btrfs_destroy_inode,
1110         .statfs         = btrfs_statfs,
1111         .remount_fs     = btrfs_remount,
1112         .freeze_fs      = btrfs_freeze,
1113         .unfreeze_fs    = btrfs_unfreeze,
1114 };
1115
1116 static const struct file_operations btrfs_ctl_fops = {
1117         .unlocked_ioctl  = btrfs_control_ioctl,
1118         .compat_ioctl = btrfs_control_ioctl,
1119         .owner   = THIS_MODULE,
1120 };
1121
1122 static struct miscdevice btrfs_misc = {
1123         .minor          = BTRFS_MINOR,
1124         .name           = "btrfs-control",
1125         .fops           = &btrfs_ctl_fops
1126 };
1127
1128 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1129 MODULE_ALIAS("devname:btrfs-control");
1130
1131 static int btrfs_interface_init(void)
1132 {
1133         return misc_register(&btrfs_misc);
1134 }
1135
1136 static void btrfs_interface_exit(void)
1137 {
1138         if (misc_deregister(&btrfs_misc) < 0)
1139                 printk(KERN_INFO "misc_deregister failed for control device");
1140 }
1141
1142 static int __init init_btrfs_fs(void)
1143 {
1144         int err;
1145
1146         err = btrfs_init_sysfs();
1147         if (err)
1148                 return err;
1149
1150         err = btrfs_init_compress();
1151         if (err)
1152                 goto free_sysfs;
1153
1154         err = btrfs_init_cachep();
1155         if (err)
1156                 goto free_compress;
1157
1158         err = extent_io_init();
1159         if (err)
1160                 goto free_cachep;
1161
1162         err = extent_map_init();
1163         if (err)
1164                 goto free_extent_io;
1165
1166         err = btrfs_interface_init();
1167         if (err)
1168                 goto free_extent_map;
1169
1170         err = register_filesystem(&btrfs_fs_type);
1171         if (err)
1172                 goto unregister_ioctl;
1173
1174         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1175         return 0;
1176
1177 unregister_ioctl:
1178         btrfs_interface_exit();
1179 free_extent_map:
1180         extent_map_exit();
1181 free_extent_io:
1182         extent_io_exit();
1183 free_cachep:
1184         btrfs_destroy_cachep();
1185 free_compress:
1186         btrfs_exit_compress();
1187 free_sysfs:
1188         btrfs_exit_sysfs();
1189         return err;
1190 }
1191
1192 static void __exit exit_btrfs_fs(void)
1193 {
1194         btrfs_destroy_cachep();
1195         extent_map_exit();
1196         extent_io_exit();
1197         btrfs_interface_exit();
1198         unregister_filesystem(&btrfs_fs_type);
1199         btrfs_exit_sysfs();
1200         btrfs_cleanup_fs_uuids();
1201         btrfs_exit_compress();
1202 }
1203
1204 module_init(init_btrfs_fs)
1205 module_exit(exit_btrfs_fs)
1206
1207 MODULE_LICENSE("GPL");