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