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 <linux/ratelimit.h>
  44 #include "compat.h"
  45 #include "delayed-inode.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "ioctl.h"
  51 #include "print-tree.h"
  52 #include "xattr.h"
  53 #include "volumes.h"
  54 #include "version.h"
  55 #include "export.h"
  56 #include "compression.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/btrfs.h>
  60
  61 static const struct super_operations btrfs_super_ops;
  62 static struct file_system_type btrfs_fs_type;
  63
  64 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
  65                                       char nbuf[16])
  66 {
  67         char *errstr = NULL;
  68
  69         switch (errno) {
  70         case -EIO:
  71                 errstr = "IO failure";
  72                 break;
  73         case -ENOMEM:
  74                 errstr = "Out of memory";
  75                 break;
  76         case -EROFS:
  77                 errstr = "Readonly filesystem";
  78                 break;
  79         case -EEXIST:
  80                 errstr = "Object already exists";
  81                 break;
  82         default:
  83                 if (nbuf) {
  84                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  85                                 errstr = nbuf;
  86                 }
  87                 break;
  88         }
  89
  90         return errstr;
  91 }
  92
  93 static void __save_error_info(struct btrfs_fs_info *fs_info)
  94 {
  95         /*
  96          * today we only save the error info into ram.  Long term we'll
  97          * also send it down to the disk
  98          */
  99         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
 100 }
 101
 102 /* NOTE:
 103  *      We move write_super stuff at umount in order to avoid deadlock
 104  *      for umount hold all lock.
 105  */
 106 static void save_error_info(struct btrfs_fs_info *fs_info)
 107 {
 108         __save_error_info(fs_info);
 109 }
 110
 111 /* btrfs handle error by forcing the filesystem readonly */
 112 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 113 {
 114         struct super_block *sb = fs_info->sb;
 115
 116         if (sb->s_flags & MS_RDONLY)
 117                 return;
 118
 119         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 120                 sb->s_flags |= MS_RDONLY;
 121                 printk(KERN_INFO "btrfs is forced readonly\n");
 122                 __btrfs_scrub_cancel(fs_info);
 123 //              WARN_ON(1);
 124         }
 125 }
 126
 127 /*
 128  * __btrfs_std_error decodes expected errors from the caller and
 129  * invokes the approciate error response.
 130  */
 131 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 132                        unsigned int line, int errno, const char *fmt, ...)
 133 {
 134         struct super_block *sb = fs_info->sb;
 135         char nbuf[16];
 136         const char *errstr;
 137         va_list args;
 138         va_start(args, fmt);
 139
 140         /*
 141          * Special case: if the error is EROFS, and we're already
 142          * under MS_RDONLY, then it is safe here.
 143          */
 144         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 145                 return;
 146
 147         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 148         if (fmt) {
 149                 struct va_format vaf = {
 150                         .fmt = fmt,
 151                         .va = &args,
 152                 };
 153
 154                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s (%pV)\n",
 155                         sb->s_id, function, line, errstr, &vaf);
 156         } else {
 157                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 158                         sb->s_id, function, line, errstr);
 159         }
 160
 161         /* Don't go through full error handling during mount */
 162         if (sb->s_flags & MS_BORN) {
 163                 save_error_info(fs_info);
 164                 btrfs_handle_error(fs_info);
 165         }
 166         va_end(args);
 167 }
 168
 169 const char *logtypes[] = {
 170         "emergency",
 171         "alert",
 172         "critical",
 173         "error",
 174         "warning",
 175         "notice",
 176         "info",
 177         "debug",
 178 };
 179
 180 void btrfs_printk(struct btrfs_fs_info *fs_info, const char *fmt, ...)
 181 {
 182         struct super_block *sb = fs_info->sb;
 183         char lvl[4];
 184         struct va_format vaf;
 185         va_list args;
 186         const char *type = logtypes[4];
 187
 188         va_start(args, fmt);
 189
 190         if (fmt[0] == '<' && isdigit(fmt[1]) && fmt[2] == '>') {
 191                 strncpy(lvl, fmt, 3);
 192                 fmt += 3;
 193                 type = logtypes[fmt[1] - '0'];
 194         } else
 195                 *lvl = '\0';
 196
 197         vaf.fmt = fmt;
 198         vaf.va = &args;
 199         printk("%sBTRFS %s (device %s): %pV", lvl, type, sb->s_id, &vaf);
 200 }
 201
 202 /*
 203  * We only mark the transaction aborted and then set the file system read-only.
 204  * This will prevent new transactions from starting or trying to join this
 205  * one.
 206  *
 207  * This means that error recovery at the call site is limited to freeing
 208  * any local memory allocations and passing the error code up without
 209  * further cleanup. The transaction should complete as it normally would
 210  * in the call path but will return -EIO.
 211  *
 212  * We'll complete the cleanup in btrfs_end_transaction and
 213  * btrfs_commit_transaction.
 214  */
 215 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 216                                struct btrfs_root *root, const char *function,
 217                                unsigned int line, int errno)
 218 {
 219         WARN_ONCE(1, KERN_DEBUG "btrfs: Transaction aborted");
 220         trans->aborted = errno;
 221         /* Nothing used. The other threads that have joined this
 222          * transaction may be able to continue. */
 223         if (!trans->blocks_used) {
 224                 btrfs_printk(root->fs_info, "Aborting unused transaction.\n");
 225                 return;
 226         }
 227         trans->transaction->aborted = errno;
 228         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 229 }
 230 /*
 231  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 232  * issues an alert, and either panics or BUGs, depending on mount options.
 233  */
 234 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 235                    unsigned int line, int errno, const char *fmt, ...)
 236 {
 237         char nbuf[16];
 238         char *s_id = "<unknown>";
 239         const char *errstr;
 240         struct va_format vaf = { .fmt = fmt };
 241         va_list args;
 242
 243         if (fs_info)
 244                 s_id = fs_info->sb->s_id;
 245
 246         va_start(args, fmt);
 247         vaf.va = &args;
 248
 249         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 250         if (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR)
 251                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
 252                         s_id, function, line, &vaf, errstr);
 253
 254         printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
 255                s_id, function, line, &vaf, errstr);
 256         va_end(args);
 257         /* Caller calls BUG() */
 258 }
 259
 260 static void btrfs_put_super(struct super_block *sb)
 261 {
 262         (void)close_ctree(btrfs_sb(sb)->tree_root);
 263         /* FIXME: need to fix VFS to return error? */
 264         /* AV: return it _where_?  ->put_super() can be triggered by any number
 265          * of async events, up to and including delivery of SIGKILL to the
 266          * last process that kept it busy.  Or segfault in the aforementioned
 267          * process...  Whom would you report that to?
 268          */
 269 }
 270
 271 enum {
 272         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 273         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 274         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 275         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 276         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 277         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 278         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 279         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 280         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 281         Opt_check_integrity_print_mask, Opt_fatal_errors,
 282         Opt_err,
 283 };
 284
 285 static match_table_t tokens = {
 286         {Opt_degraded, "degraded"},
 287         {Opt_subvol, "subvol=%s"},
 288         {Opt_subvolid, "subvolid=%d"},
 289         {Opt_device, "device=%s"},
 290         {Opt_nodatasum, "nodatasum"},
 291         {Opt_nodatacow, "nodatacow"},
 292         {Opt_nobarrier, "nobarrier"},
 293         {Opt_max_inline, "max_inline=%s"},
 294         {Opt_alloc_start, "alloc_start=%s"},
 295         {Opt_thread_pool, "thread_pool=%d"},
 296         {Opt_compress, "compress"},
 297         {Opt_compress_type, "compress=%s"},
 298         {Opt_compress_force, "compress-force"},
 299         {Opt_compress_force_type, "compress-force=%s"},
 300         {Opt_ssd, "ssd"},
 301         {Opt_ssd_spread, "ssd_spread"},
 302         {Opt_nossd, "nossd"},
 303         {Opt_noacl, "noacl"},
 304         {Opt_notreelog, "notreelog"},
 305         {Opt_flushoncommit, "flushoncommit"},
 306         {Opt_ratio, "metadata_ratio=%d"},
 307         {Opt_discard, "discard"},
 308         {Opt_space_cache, "space_cache"},
 309         {Opt_clear_cache, "clear_cache"},
 310         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 311         {Opt_enospc_debug, "enospc_debug"},
 312         {Opt_subvolrootid, "subvolrootid=%d"},
 313         {Opt_defrag, "autodefrag"},
 314         {Opt_inode_cache, "inode_cache"},
 315         {Opt_no_space_cache, "nospace_cache"},
 316         {Opt_recovery, "recovery"},
 317         {Opt_skip_balance, "skip_balance"},
 318         {Opt_check_integrity, "check_int"},
 319         {Opt_check_integrity_including_extent_data, "check_int_data"},
 320         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 321         {Opt_fatal_errors, "fatal_errors=%s"},
 322         {Opt_err, NULL},
 323 };
 324
 325 /*
 326  * Regular mount options parser.  Everything that is needed only when
 327  * reading in a new superblock is parsed here.
 328  * XXX JDM: This needs to be cleaned up for remount.
 329  */
 330 int btrfs_parse_options(struct btrfs_root *root, char *options)
 331 {
 332         struct btrfs_fs_info *info = root->fs_info;
 333         substring_t args[MAX_OPT_ARGS];
 334         char *p, *num, *orig = NULL;
 335         u64 cache_gen;
 336         int intarg;
 337         int ret = 0;
 338         char *compress_type;
 339         bool compress_force = false;
 340
 341         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 342         if (cache_gen)
 343                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 344
 345         if (!options)
 346                 goto out;
 347
 348         /*
 349          * strsep changes the string, duplicate it because parse_options
 350          * gets called twice
 351          */
 352         options = kstrdup(options, GFP_NOFS);
 353         if (!options)
 354                 return -ENOMEM;
 355
 356         orig = options;
 357
 358         while ((p = strsep(&options, ",")) != NULL) {
 359                 int token;
 360                 if (!*p)
 361                         continue;
 362
 363                 token = match_token(p, tokens, args);
 364                 switch (token) {
 365                 case Opt_degraded:
 366                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 367                         btrfs_set_opt(info->mount_opt, DEGRADED);
 368                         break;
 369                 case Opt_subvol:
 370                 case Opt_subvolid:
 371                 case Opt_subvolrootid:
 372                 case Opt_device:
 373                         /*
 374                          * These are parsed by btrfs_parse_early_options
 375                          * and can be happily ignored here.
 376                          */
 377                         break;
 378                 case Opt_nodatasum:
 379                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 380                         btrfs_set_opt(info->mount_opt, NODATASUM);
 381                         break;
 382                 case Opt_nodatacow:
 383                         printk(KERN_INFO "btrfs: setting nodatacow\n");
 384                         btrfs_set_opt(info->mount_opt, NODATACOW);
 385                         btrfs_set_opt(info->mount_opt, NODATASUM);
 386                         break;
 387                 case Opt_compress_force:
 388                 case Opt_compress_force_type:
 389                         compress_force = true;
 390                 case Opt_compress:
 391                 case Opt_compress_type:
 392                         if (token == Opt_compress ||
 393                             token == Opt_compress_force ||
 394                             strcmp(args[0].from, "zlib") == 0) {
 395                                 compress_type = "zlib";
 396                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 397                         } else if (strcmp(args[0].from, "lzo") == 0) {
 398                                 compress_type = "lzo";
 399                                 info->compress_type = BTRFS_COMPRESS_LZO;
 400                         } else {
 401                                 ret = -EINVAL;
 402                                 goto out;
 403                         }
 404
 405                         btrfs_set_opt(info->mount_opt, COMPRESS);
 406                         if (compress_force) {
 407                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 408                                 pr_info("btrfs: force %s compression\n",
 409                                         compress_type);
 410                         } else
 411                                 pr_info("btrfs: use %s compression\n",
 412                                         compress_type);
 413                         break;
 414                 case Opt_ssd:
 415                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 416                         btrfs_set_opt(info->mount_opt, SSD);
 417                         break;
 418                 case Opt_ssd_spread:
 419                         printk(KERN_INFO "btrfs: use spread ssd "
 420                                "allocation scheme\n");
 421                         btrfs_set_opt(info->mount_opt, SSD);
 422                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 423                         break;
 424                 case Opt_nossd:
 425                         printk(KERN_INFO "btrfs: not using ssd allocation "
 426                                "scheme\n");
 427                         btrfs_set_opt(info->mount_opt, NOSSD);
 428                         btrfs_clear_opt(info->mount_opt, SSD);
 429                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 430                         break;
 431                 case Opt_nobarrier:
 432                         printk(KERN_INFO "btrfs: turning off barriers\n");
 433                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 434                         break;
 435                 case Opt_thread_pool:
 436                         intarg = 0;
 437                         match_int(&args[0], &intarg);
 438                         if (intarg) {
 439                                 info->thread_pool_size = intarg;
 440                                 printk(KERN_INFO "btrfs: thread pool %d\n",
 441                                        info->thread_pool_size);
 442                         }
 443                         break;
 444                 case Opt_max_inline:
 445                         num = match_strdup(&args[0]);
 446                         if (num) {
 447                                 info->max_inline = memparse(num, NULL);
 448                                 kfree(num);
 449
 450                                 if (info->max_inline) {
 451                                         info->max_inline = max_t(u64,
 452                                                 info->max_inline,
 453                                                 root->sectorsize);
 454                                 }
 455                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 456                                         (unsigned long long)info->max_inline);
 457                         }
 458                         break;
 459                 case Opt_alloc_start:
 460                         num = match_strdup(&args[0]);
 461                         if (num) {
 462                                 info->alloc_start = memparse(num, NULL);
 463                                 kfree(num);
 464                                 printk(KERN_INFO
 465                                         "btrfs: allocations start at %llu\n",
 466                                         (unsigned long long)info->alloc_start);
 467                         }
 468                         break;
 469                 case Opt_noacl:
 470                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 471                         break;
 472                 case Opt_notreelog:
 473                         printk(KERN_INFO "btrfs: disabling tree log\n");
 474                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 475                         break;
 476                 case Opt_flushoncommit:
 477                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 478                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 479                         break;
 480                 case Opt_ratio:
 481                         intarg = 0;
 482                         match_int(&args[0], &intarg);
 483                         if (intarg) {
 484                                 info->metadata_ratio = intarg;
 485                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 486                                        info->metadata_ratio);
 487                         }
 488                         break;
 489                 case Opt_discard:
 490                         btrfs_set_opt(info->mount_opt, DISCARD);
 491                         break;
 492                 case Opt_space_cache:
 493                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 494                         break;
 495                 case Opt_no_space_cache:
 496                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 497                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 498                         break;
 499                 case Opt_inode_cache:
 500                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 501                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 502                         break;
 503                 case Opt_clear_cache:
 504                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 505                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 506                         break;
 507                 case Opt_user_subvol_rm_allowed:
 508                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 509                         break;
 510                 case Opt_enospc_debug:
 511                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 512                         break;
 513                 case Opt_defrag:
 514                         printk(KERN_INFO "btrfs: enabling auto defrag");
 515                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 516                         break;
 517                 case Opt_recovery:
 518                         printk(KERN_INFO "btrfs: enabling auto recovery");
 519                         btrfs_set_opt(info->mount_opt, RECOVERY);
 520                         break;
 521                 case Opt_skip_balance:
 522                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 523                         break;
 524 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 525                 case Opt_check_integrity_including_extent_data:
 526                         printk(KERN_INFO "btrfs: enabling check integrity"
 527                                " including extent data\n");
 528                         btrfs_set_opt(info->mount_opt,
 529                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 530                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 531                         break;
 532                 case Opt_check_integrity:
 533                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 534                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 535                         break;
 536                 case Opt_check_integrity_print_mask:
 537                         intarg = 0;
 538                         match_int(&args[0], &intarg);
 539                         if (intarg) {
 540                                 info->check_integrity_print_mask = intarg;
 541                                 printk(KERN_INFO "btrfs:"
 542                                        " check_integrity_print_mask 0x%x\n",
 543                                        info->check_integrity_print_mask);
 544                         }
 545                         break;
 546 #else
 547                 case Opt_check_integrity_including_extent_data:
 548                 case Opt_check_integrity:
 549                 case Opt_check_integrity_print_mask:
 550                         printk(KERN_ERR "btrfs: support for check_integrity*"
 551                                " not compiled in!\n");
 552                         ret = -EINVAL;
 553                         goto out;
 554 #endif
 555                 case Opt_fatal_errors:
 556                         if (strcmp(args[0].from, "panic") == 0)
 557                                 btrfs_set_opt(info->mount_opt,
 558                                               PANIC_ON_FATAL_ERROR);
 559                         else if (strcmp(args[0].from, "bug") == 0)
 560                                 btrfs_clear_opt(info->mount_opt,
 561                                               PANIC_ON_FATAL_ERROR);
 562                         else {
 563                                 ret = -EINVAL;
 564                                 goto out;
 565                         }
 566                         break;
 567                 case Opt_err:
 568                         printk(KERN_INFO "btrfs: unrecognized mount option "
 569                                "'%s'\n", p);
 570                         ret = -EINVAL;
 571                         goto out;
 572                 default:
 573                         break;
 574                 }
 575         }
 576 out:
 577         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 578                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 579         kfree(orig);
 580         return ret;
 581 }
 582
 583 /*
 584  * Parse mount options that are required early in the mount process.
 585  *
 586  * All other options will be parsed on much later in the mount process and
 587  * only when we need to allocate a new super block.
 588  */
 589 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 590                 void *holder, char **subvol_name, u64 *subvol_objectid,
 591                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
 592 {
 593         substring_t args[MAX_OPT_ARGS];
 594         char *device_name, *opts, *orig, *p;
 595         int error = 0;
 596         int intarg;
 597
 598         if (!options)
 599                 return 0;
 600
 601         /*
 602          * strsep changes the string, duplicate it because parse_options
 603          * gets called twice
 604          */
 605         opts = kstrdup(options, GFP_KERNEL);
 606         if (!opts)
 607                 return -ENOMEM;
 608         orig = opts;
 609
 610         while ((p = strsep(&opts, ",")) != NULL) {
 611                 int token;
 612                 if (!*p)
 613                         continue;
 614
 615                 token = match_token(p, tokens, args);
 616                 switch (token) {
 617                 case Opt_subvol:
 618                         kfree(*subvol_name);
 619                         *subvol_name = match_strdup(&args[0]);
 620                         break;
 621                 case Opt_subvolid:
 622                         intarg = 0;
 623                         error = match_int(&args[0], &intarg);
 624                         if (!error) {
 625                                 /* we want the original fs_tree */
 626                                 if (!intarg)
 627                                         *subvol_objectid =
 628                                                 BTRFS_FS_TREE_OBJECTID;
 629                                 else
 630                                         *subvol_objectid = intarg;
 631                         }
 632                         break;
 633                 case Opt_subvolrootid:
 634                         intarg = 0;
 635                         error = match_int(&args[0], &intarg);
 636                         if (!error) {
 637                                 /* we want the original fs_tree */
 638                                 if (!intarg)
 639                                         *subvol_rootid =
 640                                                 BTRFS_FS_TREE_OBJECTID;
 641                                 else
 642                                         *subvol_rootid = intarg;
 643                         }
 644                         break;
 645                 case Opt_device:
 646                         device_name = match_strdup(&args[0]);
 647                         if (!device_name) {
 648                                 error = -ENOMEM;
 649                                 goto out;
 650                         }
 651                         error = btrfs_scan_one_device(device_name,
 652                                         flags, holder, fs_devices);
 653                         kfree(device_name);
 654                         if (error)
 655                                 goto out;
 656                         break;
 657                 default:
 658                         break;
 659                 }
 660         }
 661
 662 out:
 663         kfree(orig);
 664         return error;
 665 }
 666
 667 static struct dentry *get_default_root(struct super_block *sb,
 668                                        u64 subvol_objectid)
 669 {
 670         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 671         struct btrfs_root *root = fs_info->tree_root;
 672         struct btrfs_root *new_root;
 673         struct btrfs_dir_item *di;
 674         struct btrfs_path *path;
 675         struct btrfs_key location;
 676         struct inode *inode;
 677         u64 dir_id;
 678         int new = 0;
 679
 680         /*
 681          * We have a specific subvol we want to mount, just setup location and
 682          * go look up the root.
 683          */
 684         if (subvol_objectid) {
 685                 location.objectid = subvol_objectid;
 686                 location.type = BTRFS_ROOT_ITEM_KEY;
 687                 location.offset = (u64)-1;
 688                 goto find_root;
 689         }
 690
 691         path = btrfs_alloc_path();
 692         if (!path)
 693                 return ERR_PTR(-ENOMEM);
 694         path->leave_spinning = 1;
 695
 696         /*
 697          * Find the "default" dir item which points to the root item that we
 698          * will mount by default if we haven't been given a specific subvolume
 699          * to mount.
 700          */
 701         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 702         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 703         if (IS_ERR(di)) {
 704                 btrfs_free_path(path);
 705                 return ERR_CAST(di);
 706         }
 707         if (!di) {
 708                 /*
 709                  * Ok the default dir item isn't there.  This is weird since
 710                  * it's always been there, but don't freak out, just try and
 711                  * mount to root most subvolume.
 712                  */
 713                 btrfs_free_path(path);
 714                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 715                 new_root = fs_info->fs_root;
 716                 goto setup_root;
 717         }
 718
 719         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 720         btrfs_free_path(path);
 721
 722 find_root:
 723         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 724         if (IS_ERR(new_root))
 725                 return ERR_CAST(new_root);
 726
 727         if (btrfs_root_refs(&new_root->root_item) == 0)
 728                 return ERR_PTR(-ENOENT);
 729
 730         dir_id = btrfs_root_dirid(&new_root->root_item);
 731 setup_root:
 732         location.objectid = dir_id;
 733         location.type = BTRFS_INODE_ITEM_KEY;
 734         location.offset = 0;
 735
 736         inode = btrfs_iget(sb, &location, new_root, &new);
 737         if (IS_ERR(inode))
 738                 return ERR_CAST(inode);
 739
 740         /*
 741          * If we're just mounting the root most subvol put the inode and return
 742          * a reference to the dentry.  We will have already gotten a reference
 743          * to the inode in btrfs_fill_super so we're good to go.
 744          */
 745         if (!new && sb->s_root->d_inode == inode) {
 746                 iput(inode);
 747                 return dget(sb->s_root);
 748         }
 749
 750         return d_obtain_alias(inode);
 751 }
 752
 753 static int btrfs_fill_super(struct super_block *sb,
 754                             struct btrfs_fs_devices *fs_devices,
 755                             void *data, int silent)
 756 {
 757         struct inode *inode;
 758         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 759         struct btrfs_key key;
 760         int err;
 761
 762         sb->s_maxbytes = MAX_LFS_FILESIZE;
 763         sb->s_magic = BTRFS_SUPER_MAGIC;
 764         sb->s_op = &btrfs_super_ops;
 765         sb->s_d_op = &btrfs_dentry_operations;
 766         sb->s_export_op = &btrfs_export_ops;
 767         sb->s_xattr = btrfs_xattr_handlers;
 768         sb->s_time_gran = 1;
 769 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 770         sb->s_flags |= MS_POSIXACL;
 771 #endif
 772
 773         err = open_ctree(sb, fs_devices, (char *)data);
 774         if (err) {
 775                 printk("btrfs: open_ctree failed\n");
 776                 return err;
 777         }
 778
 779         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 780         key.type = BTRFS_INODE_ITEM_KEY;
 781         key.offset = 0;
 782         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 783         if (IS_ERR(inode)) {
 784                 err = PTR_ERR(inode);
 785                 goto fail_close;
 786         }
 787
 788         sb->s_root = d_make_root(inode);
 789         if (!sb->s_root) {
 790                 err = -ENOMEM;
 791                 goto fail_close;
 792         }
 793
 794         save_mount_options(sb, data);
 795         cleancache_init_fs(sb);
 796         sb->s_flags |= MS_ACTIVE;
 797         return 0;
 798
 799 fail_close:
 800         close_ctree(fs_info->tree_root);
 801         return err;
 802 }
 803
 804 int btrfs_sync_fs(struct super_block *sb, int wait)
 805 {
 806         struct btrfs_trans_handle *trans;
 807         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 808         struct btrfs_root *root = fs_info->tree_root;
 809         int ret;
 810
 811         trace_btrfs_sync_fs(wait);
 812
 813         if (!wait) {
 814                 filemap_flush(fs_info->btree_inode->i_mapping);
 815                 return 0;
 816         }
 817
 818         btrfs_start_delalloc_inodes(root, 0);
 819         btrfs_wait_ordered_extents(root, 0, 0);
 820
 821         trans = btrfs_start_transaction(root, 0);
 822         if (IS_ERR(trans))
 823                 return PTR_ERR(trans);
 824         ret = btrfs_commit_transaction(trans, root);
 825         return ret;
 826 }
 827
 828 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 829 {
 830         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 831         struct btrfs_root *root = info->tree_root;
 832         char *compress_type;
 833
 834         if (btrfs_test_opt(root, DEGRADED))
 835                 seq_puts(seq, ",degraded");
 836         if (btrfs_test_opt(root, NODATASUM))
 837                 seq_puts(seq, ",nodatasum");
 838         if (btrfs_test_opt(root, NODATACOW))
 839                 seq_puts(seq, ",nodatacow");
 840         if (btrfs_test_opt(root, NOBARRIER))
 841                 seq_puts(seq, ",nobarrier");
 842         if (info->max_inline != 8192 * 1024)
 843                 seq_printf(seq, ",max_inline=%llu",
 844                            (unsigned long long)info->max_inline);
 845         if (info->alloc_start != 0)
 846                 seq_printf(seq, ",alloc_start=%llu",
 847                            (unsigned long long)info->alloc_start);
 848         if (info->thread_pool_size !=  min_t(unsigned long,
 849                                              num_online_cpus() + 2, 8))
 850                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 851         if (btrfs_test_opt(root, COMPRESS)) {
 852                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 853                         compress_type = "zlib";
 854                 else
 855                         compress_type = "lzo";
 856                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 857                         seq_printf(seq, ",compress-force=%s", compress_type);
 858                 else
 859                         seq_printf(seq, ",compress=%s", compress_type);
 860         }
 861         if (btrfs_test_opt(root, NOSSD))
 862                 seq_puts(seq, ",nossd");
 863         if (btrfs_test_opt(root, SSD_SPREAD))
 864                 seq_puts(seq, ",ssd_spread");
 865         else if (btrfs_test_opt(root, SSD))
 866                 seq_puts(seq, ",ssd");
 867         if (btrfs_test_opt(root, NOTREELOG))
 868                 seq_puts(seq, ",notreelog");
 869         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 870                 seq_puts(seq, ",flushoncommit");
 871         if (btrfs_test_opt(root, DISCARD))
 872                 seq_puts(seq, ",discard");
 873         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 874                 seq_puts(seq, ",noacl");
 875         if (btrfs_test_opt(root, SPACE_CACHE))
 876                 seq_puts(seq, ",space_cache");
 877         else
 878                 seq_puts(seq, ",nospace_cache");
 879         if (btrfs_test_opt(root, CLEAR_CACHE))
 880                 seq_puts(seq, ",clear_cache");
 881         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 882                 seq_puts(seq, ",user_subvol_rm_allowed");
 883         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 884                 seq_puts(seq, ",enospc_debug");
 885         if (btrfs_test_opt(root, AUTO_DEFRAG))
 886                 seq_puts(seq, ",autodefrag");
 887         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 888                 seq_puts(seq, ",inode_cache");
 889         if (btrfs_test_opt(root, SKIP_BALANCE))
 890                 seq_puts(seq, ",skip_balance");
 891         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 892                 seq_puts(seq, ",fatal_errors=panic");
 893         return 0;
 894 }
 895
 896 static int btrfs_test_super(struct super_block *s, void *data)
 897 {
 898         struct btrfs_fs_info *p = data;
 899         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 900
 901         return fs_info->fs_devices == p->fs_devices;
 902 }
 903
 904 static int btrfs_set_super(struct super_block *s, void *data)
 905 {
 906         int err = set_anon_super(s, data);
 907         if (!err)
 908                 s->s_fs_info = data;
 909         return err;
 910 }
 911
 912 /*
 913  * subvolumes are identified by ino 256
 914  */
 915 static inline int is_subvolume_inode(struct inode *inode)
 916 {
 917         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 918                 return 1;
 919         return 0;
 920 }
 921
 922 /*
 923  * This will strip out the subvol=%s argument for an argument string and add
 924  * subvolid=0 to make sure we get the actual tree root for path walking to the
 925  * subvol we want.
 926  */
 927 static char *setup_root_args(char *args)
 928 {
 929         unsigned copied = 0;
 930         unsigned len = strlen(args) + 2;
 931         char *pos;
 932         char *ret;
 933
 934         /*
 935          * We need the same args as before, but minus
 936          *
 937          * subvol=a
 938          *
 939          * and add
 940          *
 941          * subvolid=0
 942          *
 943          * which is a difference of 2 characters, so we allocate strlen(args) +
 944          * 2 characters.
 945          */
 946         ret = kzalloc(len * sizeof(char), GFP_NOFS);
 947         if (!ret)
 948                 return NULL;
 949         pos = strstr(args, "subvol=");
 950
 951         /* This shouldn't happen, but just in case.. */
 952         if (!pos) {
 953                 kfree(ret);
 954                 return NULL;
 955         }
 956
 957         /*
 958          * The subvol=<> arg is not at the front of the string, copy everybody
 959          * up to that into ret.
 960          */
 961         if (pos != args) {
 962                 *pos = '\0';
 963                 strcpy(ret, args);
 964                 copied += strlen(args);
 965                 pos++;
 966         }
 967
 968         strncpy(ret + copied, "subvolid=0", len - copied);
 969
 970         /* Length of subvolid=0 */
 971         copied += 10;
 972
 973         /*
 974          * If there is no , after the subvol= option then we know there's no
 975          * other options and we can just return.
 976          */
 977         pos = strchr(pos, ',');
 978         if (!pos)
 979                 return ret;
 980
 981         /* Copy the rest of the arguments into our buffer */
 982         strncpy(ret + copied, pos, len - copied);
 983         copied += strlen(pos);
 984
 985         return ret;
 986 }
 987
 988 static struct dentry *mount_subvol(const char *subvol_name, int flags,
 989                                    const char *device_name, char *data)
 990 {
 991         struct dentry *root;
 992         struct vfsmount *mnt;
 993         char *newargs;
 994
 995         newargs = setup_root_args(data);
 996         if (!newargs)
 997                 return ERR_PTR(-ENOMEM);
 998         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
 999                              newargs);
1000         kfree(newargs);
1001         if (IS_ERR(mnt))
1002                 return ERR_CAST(mnt);
1003
1004         root = mount_subtree(mnt, subvol_name);
1005
1006         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1007                 struct super_block *s = root->d_sb;
1008                 dput(root);
1009                 root = ERR_PTR(-EINVAL);
1010                 deactivate_locked_super(s);
1011                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1012                                 subvol_name);
1013         }
1014
1015         return root;
1016 }
1017
1018 /*
1019  * Find a superblock for the given device / mount point.
1020  *
1021  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1022  *        for multiple device setup.  Make sure to keep it in sync.
1023  */
1024 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1025                 const char *device_name, void *data)
1026 {
1027         struct block_device *bdev = NULL;
1028         struct super_block *s;
1029         struct dentry *root;
1030         struct btrfs_fs_devices *fs_devices = NULL;
1031         struct btrfs_fs_info *fs_info = NULL;
1032         fmode_t mode = FMODE_READ;
1033         char *subvol_name = NULL;
1034         u64 subvol_objectid = 0;
1035         u64 subvol_rootid = 0;
1036         int error = 0;
1037
1038         if (!(flags & MS_RDONLY))
1039                 mode |= FMODE_WRITE;
1040
1041         error = btrfs_parse_early_options(data, mode, fs_type,
1042                                           &subvol_name, &subvol_objectid,
1043                                           &subvol_rootid, &fs_devices);
1044         if (error) {
1045                 kfree(subvol_name);
1046                 return ERR_PTR(error);
1047         }
1048
1049         if (subvol_name) {
1050                 root = mount_subvol(subvol_name, flags, device_name, data);
1051                 kfree(subvol_name);
1052                 return root;
1053         }
1054
1055         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1056         if (error)
1057                 return ERR_PTR(error);
1058
1059         /*
1060          * Setup a dummy root and fs_info for test/set super.  This is because
1061          * we don't actually fill this stuff out until open_ctree, but we need
1062          * it for searching for existing supers, so this lets us do that and
1063          * then open_ctree will properly initialize everything later.
1064          */
1065         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1066         if (!fs_info)
1067                 return ERR_PTR(-ENOMEM);
1068
1069         fs_info->fs_devices = fs_devices;
1070
1071         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1072         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1073         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1074                 error = -ENOMEM;
1075                 goto error_fs_info;
1076         }
1077
1078         error = btrfs_open_devices(fs_devices, mode, fs_type);
1079         if (error)
1080                 goto error_fs_info;
1081
1082         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1083                 error = -EACCES;
1084                 goto error_close_devices;
1085         }
1086
1087         bdev = fs_devices->latest_bdev;
1088         s = sget(fs_type, btrfs_test_super, btrfs_set_super, fs_info);
1089         if (IS_ERR(s)) {
1090                 error = PTR_ERR(s);
1091                 goto error_close_devices;
1092         }
1093
1094         if (s->s_root) {
1095                 btrfs_close_devices(fs_devices);
1096                 free_fs_info(fs_info);
1097                 if ((flags ^ s->s_flags) & MS_RDONLY)
1098                         error = -EBUSY;
1099         } else {
1100                 char b[BDEVNAME_SIZE];
1101
1102                 s->s_flags = flags | MS_NOSEC;
1103                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1104                 btrfs_sb(s)->bdev_holder = fs_type;
1105                 error = btrfs_fill_super(s, fs_devices, data,
1106                                          flags & MS_SILENT ? 1 : 0);
1107         }
1108
1109         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1110         if (IS_ERR(root))
1111                 deactivate_locked_super(s);
1112
1113         return root;
1114
1115 error_close_devices:
1116         btrfs_close_devices(fs_devices);
1117 error_fs_info:
1118         free_fs_info(fs_info);
1119         return ERR_PTR(error);
1120 }
1121
1122 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1123 {
1124         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1125         struct btrfs_root *root = fs_info->tree_root;
1126         unsigned old_flags = sb->s_flags;
1127         unsigned long old_opts = fs_info->mount_opt;
1128         unsigned long old_compress_type = fs_info->compress_type;
1129         u64 old_max_inline = fs_info->max_inline;
1130         u64 old_alloc_start = fs_info->alloc_start;
1131         int old_thread_pool_size = fs_info->thread_pool_size;
1132         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1133         int ret;
1134
1135         ret = btrfs_parse_options(root, data);
1136         if (ret) {
1137                 ret = -EINVAL;
1138                 goto restore;
1139         }
1140
1141         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1142                 return 0;
1143
1144         if (*flags & MS_RDONLY) {
1145                 sb->s_flags |= MS_RDONLY;
1146
1147                 ret = btrfs_commit_super(root);
1148                 if (ret)
1149                         goto restore;
1150         } else {
1151                 if (fs_info->fs_devices->rw_devices == 0)
1152                         ret = -EACCES;
1153                         goto restore;
1154
1155                 if (btrfs_super_log_root(fs_info->super_copy) != 0)
1156                         ret = -EINVAL;
1157                         goto restore;
1158
1159                 ret = btrfs_cleanup_fs_roots(fs_info);
1160                 if (ret)
1161                         goto restore;
1162
1163                 /* recover relocation */
1164                 ret = btrfs_recover_relocation(root);
1165                 if (ret)
1166                         goto restore;
1167
1168                 sb->s_flags &= ~MS_RDONLY;
1169         }
1170
1171         return 0;
1172
1173 restore:
1174         /* We've hit an error - don't reset MS_RDONLY */
1175         if (sb->s_flags & MS_RDONLY)
1176                 old_flags |= MS_RDONLY;
1177         sb->s_flags = old_flags;
1178         fs_info->mount_opt = old_opts;
1179         fs_info->compress_type = old_compress_type;
1180         fs_info->max_inline = old_max_inline;
1181         fs_info->alloc_start = old_alloc_start;
1182         fs_info->thread_pool_size = old_thread_pool_size;
1183         fs_info->metadata_ratio = old_metadata_ratio;
1184         return ret;
1185 }
1186
1187 /* Used to sort the devices by max_avail(descending sort) */
1188 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1189                                        const void *dev_info2)
1190 {
1191         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1192             ((struct btrfs_device_info *)dev_info2)->max_avail)
1193                 return -1;
1194         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1195                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1196                 return 1;
1197         else
1198         return 0;
1199 }
1200
1201 /*
1202  * sort the devices by max_avail, in which max free extent size of each device
1203  * is stored.(Descending Sort)
1204  */
1205 static inline void btrfs_descending_sort_devices(
1206                                         struct btrfs_device_info *devices,
1207                                         size_t nr_devices)
1208 {
1209         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1210              btrfs_cmp_device_free_bytes, NULL);
1211 }
1212
1213 /*
1214  * The helper to calc the free space on the devices that can be used to store
1215  * file data.
1216  */
1217 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1218 {
1219         struct btrfs_fs_info *fs_info = root->fs_info;
1220         struct btrfs_device_info *devices_info;
1221         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1222         struct btrfs_device *device;
1223         u64 skip_space;
1224         u64 type;
1225         u64 avail_space;
1226         u64 used_space;
1227         u64 min_stripe_size;
1228         int min_stripes = 1, num_stripes = 1;
1229         int i = 0, nr_devices;
1230         int ret;
1231
1232         nr_devices = fs_info->fs_devices->open_devices;
1233         BUG_ON(!nr_devices);
1234
1235         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1236                                GFP_NOFS);
1237         if (!devices_info)
1238                 return -ENOMEM;
1239
1240         /* calc min stripe number for data space alloction */
1241         type = btrfs_get_alloc_profile(root, 1);
1242         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1243                 min_stripes = 2;
1244                 num_stripes = nr_devices;
1245         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1246                 min_stripes = 2;
1247                 num_stripes = 2;
1248         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1249                 min_stripes = 4;
1250                 num_stripes = 4;
1251         }
1252
1253         if (type & BTRFS_BLOCK_GROUP_DUP)
1254                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1255         else
1256                 min_stripe_size = BTRFS_STRIPE_LEN;
1257
1258         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1259                 if (!device->in_fs_metadata || !device->bdev)
1260                         continue;
1261
1262                 avail_space = device->total_bytes - device->bytes_used;
1263
1264                 /* align with stripe_len */
1265                 do_div(avail_space, BTRFS_STRIPE_LEN);
1266                 avail_space *= BTRFS_STRIPE_LEN;
1267
1268                 /*
1269                  * In order to avoid overwritting the superblock on the drive,
1270                  * btrfs starts at an offset of at least 1MB when doing chunk
1271                  * allocation.
1272                  */
1273                 skip_space = 1024 * 1024;
1274
1275                 /* user can set the offset in fs_info->alloc_start. */
1276                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1277                     device->total_bytes)
1278                         skip_space = max(fs_info->alloc_start, skip_space);
1279
1280                 /*
1281                  * btrfs can not use the free space in [0, skip_space - 1],
1282                  * we must subtract it from the total. In order to implement
1283                  * it, we account the used space in this range first.
1284                  */
1285                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1286                                                      &used_space);
1287                 if (ret) {
1288                         kfree(devices_info);
1289                         return ret;
1290                 }
1291
1292                 /* calc the free space in [0, skip_space - 1] */
1293                 skip_space -= used_space;
1294
1295                 /*
1296                  * we can use the free space in [0, skip_space - 1], subtract
1297                  * it from the total.
1298                  */
1299                 if (avail_space && avail_space >= skip_space)
1300                         avail_space -= skip_space;
1301                 else
1302                         avail_space = 0;
1303
1304                 if (avail_space < min_stripe_size)
1305                         continue;
1306
1307                 devices_info[i].dev = device;
1308                 devices_info[i].max_avail = avail_space;
1309
1310                 i++;
1311         }
1312
1313         nr_devices = i;
1314
1315         btrfs_descending_sort_devices(devices_info, nr_devices);
1316
1317         i = nr_devices - 1;
1318         avail_space = 0;
1319         while (nr_devices >= min_stripes) {
1320                 if (num_stripes > nr_devices)
1321                         num_stripes = nr_devices;
1322
1323                 if (devices_info[i].max_avail >= min_stripe_size) {
1324                         int j;
1325                         u64 alloc_size;
1326
1327                         avail_space += devices_info[i].max_avail * num_stripes;
1328                         alloc_size = devices_info[i].max_avail;
1329                         for (j = i + 1 - num_stripes; j <= i; j++)
1330                                 devices_info[j].max_avail -= alloc_size;
1331                 }
1332                 i--;
1333                 nr_devices--;
1334         }
1335
1336         kfree(devices_info);
1337         *free_bytes = avail_space;
1338         return 0;
1339 }
1340
1341 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1342 {
1343         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1344         struct btrfs_super_block *disk_super = fs_info->super_copy;
1345         struct list_head *head = &fs_info->space_info;
1346         struct btrfs_space_info *found;
1347         u64 total_used = 0;
1348         u64 total_free_data = 0;
1349         int bits = dentry->d_sb->s_blocksize_bits;
1350         __be32 *fsid = (__be32 *)fs_info->fsid;
1351         int ret;
1352
1353         /* holding chunk_muext to avoid allocating new chunks */
1354         mutex_lock(&fs_info->chunk_mutex);
1355         rcu_read_lock();
1356         list_for_each_entry_rcu(found, head, list) {
1357                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1358                         total_free_data += found->disk_total - found->disk_used;
1359                         total_free_data -=
1360                                 btrfs_account_ro_block_groups_free_space(found);
1361                 }
1362
1363                 total_used += found->disk_used;
1364         }
1365         rcu_read_unlock();
1366
1367         buf->f_namelen = BTRFS_NAME_LEN;
1368         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1369         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1370         buf->f_bsize = dentry->d_sb->s_blocksize;
1371         buf->f_type = BTRFS_SUPER_MAGIC;
1372         buf->f_bavail = total_free_data;
1373         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1374         if (ret) {
1375                 mutex_unlock(&fs_info->chunk_mutex);
1376                 return ret;
1377         }
1378         buf->f_bavail += total_free_data;
1379         buf->f_bavail = buf->f_bavail >> bits;
1380         mutex_unlock(&fs_info->chunk_mutex);
1381
1382         /* We treat it as constant endianness (it doesn't matter _which_)
1383            because we want the fsid to come out the same whether mounted
1384            on a big-endian or little-endian host */
1385         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1386         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1387         /* Mask in the root object ID too, to disambiguate subvols */
1388         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1389         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1390
1391         return 0;
1392 }
1393
1394 static void btrfs_kill_super(struct super_block *sb)
1395 {
1396         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1397         kill_anon_super(sb);
1398         free_fs_info(fs_info);
1399 }
1400
1401 static struct file_system_type btrfs_fs_type = {
1402         .owner          = THIS_MODULE,
1403         .name           = "btrfs",
1404         .mount          = btrfs_mount,
1405         .kill_sb        = btrfs_kill_super,
1406         .fs_flags       = FS_REQUIRES_DEV,
1407 };
1408
1409 /*
1410  * used by btrfsctl to scan devices when no FS is mounted
1411  */
1412 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1413                                 unsigned long arg)
1414 {
1415         struct btrfs_ioctl_vol_args *vol;
1416         struct btrfs_fs_devices *fs_devices;
1417         int ret = -ENOTTY;
1418
1419         if (!capable(CAP_SYS_ADMIN))
1420                 return -EPERM;
1421
1422         vol = memdup_user((void __user *)arg, sizeof(*vol));
1423         if (IS_ERR(vol))
1424                 return PTR_ERR(vol);
1425
1426         switch (cmd) {
1427         case BTRFS_IOC_SCAN_DEV:
1428                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1429                                             &btrfs_fs_type, &fs_devices);
1430                 break;
1431         }
1432
1433         kfree(vol);
1434         return ret;
1435 }
1436
1437 static int btrfs_freeze(struct super_block *sb)
1438 {
1439         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1440         mutex_lock(&fs_info->transaction_kthread_mutex);
1441         mutex_lock(&fs_info->cleaner_mutex);
1442         return 0;
1443 }
1444
1445 static int btrfs_unfreeze(struct super_block *sb)
1446 {
1447         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1448         mutex_unlock(&fs_info->cleaner_mutex);
1449         mutex_unlock(&fs_info->transaction_kthread_mutex);
1450         return 0;
1451 }
1452
1453 static void btrfs_fs_dirty_inode(struct inode *inode, int flags)
1454 {
1455         int ret;
1456
1457         ret = btrfs_dirty_inode(inode);
1458         if (ret)
1459                 printk_ratelimited(KERN_ERR "btrfs: fail to dirty inode %Lu "
1460                                    "error %d\n", btrfs_ino(inode), ret);
1461 }
1462
1463 static const struct super_operations btrfs_super_ops = {
1464         .drop_inode     = btrfs_drop_inode,
1465         .evict_inode    = btrfs_evict_inode,
1466         .put_super      = btrfs_put_super,
1467         .sync_fs        = btrfs_sync_fs,
1468         .show_options   = btrfs_show_options,
1469         .write_inode    = btrfs_write_inode,
1470         .dirty_inode    = btrfs_fs_dirty_inode,
1471         .alloc_inode    = btrfs_alloc_inode,
1472         .destroy_inode  = btrfs_destroy_inode,
1473         .statfs         = btrfs_statfs,
1474         .remount_fs     = btrfs_remount,
1475         .freeze_fs      = btrfs_freeze,
1476         .unfreeze_fs    = btrfs_unfreeze,
1477 };
1478
1479 static const struct file_operations btrfs_ctl_fops = {
1480         .unlocked_ioctl  = btrfs_control_ioctl,
1481         .compat_ioctl = btrfs_control_ioctl,
1482         .owner   = THIS_MODULE,
1483         .llseek = noop_llseek,
1484 };
1485
1486 static struct miscdevice btrfs_misc = {
1487         .minor          = BTRFS_MINOR,
1488         .name           = "btrfs-control",
1489         .fops           = &btrfs_ctl_fops
1490 };
1491
1492 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1493 MODULE_ALIAS("devname:btrfs-control");
1494
1495 static int btrfs_interface_init(void)
1496 {
1497         return misc_register(&btrfs_misc);
1498 }
1499
1500 static void btrfs_interface_exit(void)
1501 {
1502         if (misc_deregister(&btrfs_misc) < 0)
1503                 printk(KERN_INFO "misc_deregister failed for control device");
1504 }
1505
1506 static int __init init_btrfs_fs(void)
1507 {
1508         int err;
1509
1510         err = btrfs_init_sysfs();
1511         if (err)
1512                 return err;
1513
1514         btrfs_init_compress();
1515
1516         err = btrfs_init_cachep();
1517         if (err)
1518                 goto free_compress;
1519
1520         err = extent_io_init();
1521         if (err)
1522                 goto free_cachep;
1523
1524         err = extent_map_init();
1525         if (err)
1526                 goto free_extent_io;
1527
1528         err = btrfs_delayed_inode_init();
1529         if (err)
1530                 goto free_extent_map;
1531
1532         err = btrfs_interface_init();
1533         if (err)
1534                 goto free_delayed_inode;
1535
1536         err = register_filesystem(&btrfs_fs_type);
1537         if (err)
1538                 goto unregister_ioctl;
1539
1540         btrfs_init_lockdep();
1541
1542         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1543         return 0;
1544
1545 unregister_ioctl:
1546         btrfs_interface_exit();
1547 free_delayed_inode:
1548         btrfs_delayed_inode_exit();
1549 free_extent_map:
1550         extent_map_exit();
1551 free_extent_io:
1552         extent_io_exit();
1553 free_cachep:
1554         btrfs_destroy_cachep();
1555 free_compress:
1556         btrfs_exit_compress();
1557         btrfs_exit_sysfs();
1558         return err;
1559 }
1560
1561 static void __exit exit_btrfs_fs(void)
1562 {
1563         btrfs_destroy_cachep();
1564         btrfs_delayed_inode_exit();
1565         extent_map_exit();
1566         extent_io_exit();
1567         btrfs_interface_exit();
1568         unregister_filesystem(&btrfs_fs_type);
1569         btrfs_exit_sysfs();
1570         btrfs_cleanup_fs_uuids();
1571         btrfs_exit_compress();
1572 }
1573
1574 module_init(init_btrfs_fs)
1575 module_exit(exit_btrfs_fs)
1576
1577 MODULE_LICENSE("GPL");