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