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 <linux/btrfs.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 "print-tree.h"
  51 #include "hash.h"
  52 #include "props.h"
  53 #include "xattr.h"
  54 #include "volumes.h"
  55 #include "export.h"
  56 #include "compression.h"
  57 #include "rcu-string.h"
  58 #include "dev-replace.h"
  59 #include "free-space-cache.h"
  60 #include "backref.h"
  61 #include "tests/btrfs-tests.h"
  62
  63 #include "qgroup.h"
  64 #define CREATE_TRACE_POINTS
  65 #include <trace/events/btrfs.h>
  66
  67 static const struct super_operations btrfs_super_ops;
  68 static struct file_system_type btrfs_fs_type;
  69
  70 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  71
  72 const char *btrfs_decode_error(int errno)
  73 {
  74         char *errstr = "unknown";
  75
  76         switch (errno) {
  77         case -EIO:
  78                 errstr = "IO failure";
  79                 break;
  80         case -ENOMEM:
  81                 errstr = "Out of memory";
  82                 break;
  83         case -EROFS:
  84                 errstr = "Readonly filesystem";
  85                 break;
  86         case -EEXIST:
  87                 errstr = "Object already exists";
  88                 break;
  89         case -ENOSPC:
  90                 errstr = "No space left";
  91                 break;
  92         case -ENOENT:
  93                 errstr = "No such entry";
  94                 break;
  95         }
  96
  97         return errstr;
  98 }
  99
 100 static void save_error_info(struct btrfs_fs_info *fs_info)
 101 {
 102         /*
 103          * today we only save the error info into ram.  Long term we'll
 104          * also send it down to the disk
 105          */
 106         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 107 }
 108
 109 /* btrfs handle error by forcing the filesystem readonly */
 110 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 111 {
 112         struct super_block *sb = fs_info->sb;
 113
 114         if (sb->s_flags & MS_RDONLY)
 115                 return;
 116
 117         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 118                 sb->s_flags |= MS_RDONLY;
 119                 btrfs_info(fs_info, "forced readonly");
 120                 /*
 121                  * Note that a running device replace operation is not
 122                  * canceled here although there is no way to update
 123                  * the progress. It would add the risk of a deadlock,
 124                  * therefore the canceling is ommited. The only penalty
 125                  * is that some I/O remains active until the procedure
 126                  * completes. The next time when the filesystem is
 127                  * mounted writeable again, the device replace
 128                  * operation continues.
 129                  */
 130         }
 131 }
 132
 133 /*
 134  * __btrfs_std_error decodes expected errors from the caller and
 135  * invokes the approciate error response.
 136  */
 137 __cold
 138 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 139                        unsigned int line, int errno, const char *fmt, ...)
 140 {
 141         struct super_block *sb = fs_info->sb;
 142 #ifdef CONFIG_PRINTK
 143         const char *errstr;
 144 #endif
 145
 146         /*
 147          * Special case: if the error is EROFS, and we're already
 148          * under MS_RDONLY, then it is safe here.
 149          */
 150         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 151                 return;
 152
 153 #ifdef CONFIG_PRINTK
 154         errstr = btrfs_decode_error(errno);
 155         if (fmt) {
 156                 struct va_format vaf;
 157                 va_list args;
 158
 159                 va_start(args, fmt);
 160                 vaf.fmt = fmt;
 161                 vaf.va = &args;
 162
 163                 printk(KERN_CRIT
 164                         "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 165                         sb->s_id, function, line, errno, errstr, &vaf);
 166                 va_end(args);
 167         } else {
 168                 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 169                         sb->s_id, function, line, errno, errstr);
 170         }
 171 #endif
 172
 173         /* Don't go through full error handling during mount */
 174         save_error_info(fs_info);
 175         if (sb->s_flags & MS_BORN)
 176                 btrfs_handle_error(fs_info);
 177 }
 178
 179 #ifdef CONFIG_PRINTK
 180 static const char * const logtypes[] = {
 181         "emergency",
 182         "alert",
 183         "critical",
 184         "error",
 185         "warning",
 186         "notice",
 187         "info",
 188         "debug",
 189 };
 190
 191 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 192 {
 193         struct super_block *sb = fs_info->sb;
 194         char lvl[4];
 195         struct va_format vaf;
 196         va_list args;
 197         const char *type = logtypes[4];
 198         int kern_level;
 199
 200         va_start(args, fmt);
 201
 202         kern_level = printk_get_level(fmt);
 203         if (kern_level) {
 204                 size_t size = printk_skip_level(fmt) - fmt;
 205                 memcpy(lvl, fmt,  size);
 206                 lvl[size] = '\0';
 207                 fmt += size;
 208                 type = logtypes[kern_level - '0'];
 209         } else
 210                 *lvl = '\0';
 211
 212         vaf.fmt = fmt;
 213         vaf.va = &args;
 214
 215         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 216
 217         va_end(args);
 218 }
 219 #endif
 220
 221 /*
 222  * We only mark the transaction aborted and then set the file system read-only.
 223  * This will prevent new transactions from starting or trying to join this
 224  * one.
 225  *
 226  * This means that error recovery at the call site is limited to freeing
 227  * any local memory allocations and passing the error code up without
 228  * further cleanup. The transaction should complete as it normally would
 229  * in the call path but will return -EIO.
 230  *
 231  * We'll complete the cleanup in btrfs_end_transaction and
 232  * btrfs_commit_transaction.
 233  */
 234 __cold
 235 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 236                                struct btrfs_root *root, const char *function,
 237                                unsigned int line, int errno)
 238 {
 239         trans->aborted = errno;
 240         /* Nothing used. The other threads that have joined this
 241          * transaction may be able to continue. */
 242         if (!trans->blocks_used && list_empty(&trans->new_bgs)) {
 243                 const char *errstr;
 244
 245                 errstr = btrfs_decode_error(errno);
 246                 btrfs_warn(root->fs_info,
 247                            "%s:%d: Aborting unused transaction(%s).",
 248                            function, line, errstr);
 249                 return;
 250         }
 251         ACCESS_ONCE(trans->transaction->aborted) = errno;
 252         /* Wake up anybody who may be waiting on this transaction */
 253         wake_up(&root->fs_info->transaction_wait);
 254         wake_up(&root->fs_info->transaction_blocked_wait);
 255         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 256 }
 257 /*
 258  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 259  * issues an alert, and either panics or BUGs, depending on mount options.
 260  */
 261 __cold
 262 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 263                    unsigned int line, int errno, const char *fmt, ...)
 264 {
 265         char *s_id = "<unknown>";
 266         const char *errstr;
 267         struct va_format vaf = { .fmt = fmt };
 268         va_list args;
 269
 270         if (fs_info)
 271                 s_id = fs_info->sb->s_id;
 272
 273         va_start(args, fmt);
 274         vaf.va = &args;
 275
 276         errstr = btrfs_decode_error(errno);
 277         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 278                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 279                         s_id, function, line, &vaf, errno, errstr);
 280
 281         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 282                    function, line, &vaf, errno, errstr);
 283         va_end(args);
 284         /* Caller calls BUG() */
 285 }
 286
 287 static void btrfs_put_super(struct super_block *sb)
 288 {
 289         close_ctree(btrfs_sb(sb)->tree_root);
 290 }
 291
 292 enum {
 293         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 294         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 295         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 296         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 297         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 298         Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
 299         Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
 300         Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
 301         Opt_skip_balance, Opt_check_integrity,
 302         Opt_check_integrity_including_extent_data,
 303         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 304         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
 305         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
 306         Opt_datasum, Opt_treelog, Opt_noinode_cache,
 307 #ifdef CONFIG_BTRFS_DEBUG
 308         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 309 #endif
 310         Opt_err,
 311 };
 312
 313 static const match_table_t tokens = {
 314         {Opt_degraded, "degraded"},
 315         {Opt_subvol, "subvol=%s"},
 316         {Opt_subvolid, "subvolid=%s"},
 317         {Opt_device, "device=%s"},
 318         {Opt_nodatasum, "nodatasum"},
 319         {Opt_datasum, "datasum"},
 320         {Opt_nodatacow, "nodatacow"},
 321         {Opt_datacow, "datacow"},
 322         {Opt_nobarrier, "nobarrier"},
 323         {Opt_barrier, "barrier"},
 324         {Opt_max_inline, "max_inline=%s"},
 325         {Opt_alloc_start, "alloc_start=%s"},
 326         {Opt_thread_pool, "thread_pool=%d"},
 327         {Opt_compress, "compress"},
 328         {Opt_compress_type, "compress=%s"},
 329         {Opt_compress_force, "compress-force"},
 330         {Opt_compress_force_type, "compress-force=%s"},
 331         {Opt_ssd, "ssd"},
 332         {Opt_ssd_spread, "ssd_spread"},
 333         {Opt_nossd, "nossd"},
 334         {Opt_acl, "acl"},
 335         {Opt_noacl, "noacl"},
 336         {Opt_notreelog, "notreelog"},
 337         {Opt_treelog, "treelog"},
 338         {Opt_flushoncommit, "flushoncommit"},
 339         {Opt_noflushoncommit, "noflushoncommit"},
 340         {Opt_ratio, "metadata_ratio=%d"},
 341         {Opt_discard, "discard"},
 342         {Opt_nodiscard, "nodiscard"},
 343         {Opt_space_cache, "space_cache"},
 344         {Opt_space_cache_version, "space_cache=%s"},
 345         {Opt_clear_cache, "clear_cache"},
 346         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 347         {Opt_enospc_debug, "enospc_debug"},
 348         {Opt_noenospc_debug, "noenospc_debug"},
 349         {Opt_subvolrootid, "subvolrootid=%d"},
 350         {Opt_defrag, "autodefrag"},
 351         {Opt_nodefrag, "noautodefrag"},
 352         {Opt_inode_cache, "inode_cache"},
 353         {Opt_noinode_cache, "noinode_cache"},
 354         {Opt_no_space_cache, "nospace_cache"},
 355         {Opt_recovery, "recovery"},
 356         {Opt_skip_balance, "skip_balance"},
 357         {Opt_check_integrity, "check_int"},
 358         {Opt_check_integrity_including_extent_data, "check_int_data"},
 359         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 360         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 361         {Opt_fatal_errors, "fatal_errors=%s"},
 362         {Opt_commit_interval, "commit=%d"},
 363 #ifdef CONFIG_BTRFS_DEBUG
 364         {Opt_fragment_data, "fragment=data"},
 365         {Opt_fragment_metadata, "fragment=metadata"},
 366         {Opt_fragment_all, "fragment=all"},
 367 #endif
 368         {Opt_err, NULL},
 369 };
 370
 371 /*
 372  * Regular mount options parser.  Everything that is needed only when
 373  * reading in a new superblock is parsed here.
 374  * XXX JDM: This needs to be cleaned up for remount.
 375  */
 376 int btrfs_parse_options(struct btrfs_root *root, char *options)
 377 {
 378         struct btrfs_fs_info *info = root->fs_info;
 379         substring_t args[MAX_OPT_ARGS];
 380         char *p, *num, *orig = NULL;
 381         u64 cache_gen;
 382         int intarg;
 383         int ret = 0;
 384         char *compress_type;
 385         bool compress_force = false;
 386
 387         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 388         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE))
 389                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 390         else if (cache_gen)
 391                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 392
 393         if (!options)
 394                 goto out;
 395
 396         /*
 397          * strsep changes the string, duplicate it because parse_options
 398          * gets called twice
 399          */
 400         options = kstrdup(options, GFP_NOFS);
 401         if (!options)
 402                 return -ENOMEM;
 403
 404         orig = options;
 405
 406         while ((p = strsep(&options, ",")) != NULL) {
 407                 int token;
 408                 if (!*p)
 409                         continue;
 410
 411                 token = match_token(p, tokens, args);
 412                 switch (token) {
 413                 case Opt_degraded:
 414                         btrfs_info(root->fs_info, "allowing degraded mounts");
 415                         btrfs_set_opt(info->mount_opt, DEGRADED);
 416                         break;
 417                 case Opt_subvol:
 418                 case Opt_subvolid:
 419                 case Opt_subvolrootid:
 420                 case Opt_device:
 421                         /*
 422                          * These are parsed by btrfs_parse_early_options
 423                          * and can be happily ignored here.
 424                          */
 425                         break;
 426                 case Opt_nodatasum:
 427                         btrfs_set_and_info(root, NODATASUM,
 428                                            "setting nodatasum");
 429                         break;
 430                 case Opt_datasum:
 431                         if (btrfs_test_opt(root, NODATASUM)) {
 432                                 if (btrfs_test_opt(root, NODATACOW))
 433                                         btrfs_info(root->fs_info, "setting datasum, datacow enabled");
 434                                 else
 435                                         btrfs_info(root->fs_info, "setting datasum");
 436                         }
 437                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 438                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 439                         break;
 440                 case Opt_nodatacow:
 441                         if (!btrfs_test_opt(root, NODATACOW)) {
 442                                 if (!btrfs_test_opt(root, COMPRESS) ||
 443                                     !btrfs_test_opt(root, FORCE_COMPRESS)) {
 444                                         btrfs_info(root->fs_info,
 445                                                    "setting nodatacow, compression disabled");
 446                                 } else {
 447                                         btrfs_info(root->fs_info, "setting nodatacow");
 448                                 }
 449                         }
 450                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 451                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 452                         btrfs_set_opt(info->mount_opt, NODATACOW);
 453                         btrfs_set_opt(info->mount_opt, NODATASUM);
 454                         break;
 455                 case Opt_datacow:
 456                         btrfs_clear_and_info(root, NODATACOW,
 457                                              "setting datacow");
 458                         break;
 459                 case Opt_compress_force:
 460                 case Opt_compress_force_type:
 461                         compress_force = true;
 462                         /* Fallthrough */
 463                 case Opt_compress:
 464                 case Opt_compress_type:
 465                         if (token == Opt_compress ||
 466                             token == Opt_compress_force ||
 467                             strcmp(args[0].from, "zlib") == 0) {
 468                                 compress_type = "zlib";
 469                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 470                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 471                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 472                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 473                         } else if (strcmp(args[0].from, "lzo") == 0) {
 474                                 compress_type = "lzo";
 475                                 info->compress_type = BTRFS_COMPRESS_LZO;
 476                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 477                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 478                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 479                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 480                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 481                                 compress_type = "no";
 482                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 483                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 484                                 compress_force = false;
 485                         } else {
 486                                 ret = -EINVAL;
 487                                 goto out;
 488                         }
 489
 490                         if (compress_force) {
 491                                 btrfs_set_and_info(root, FORCE_COMPRESS,
 492                                                    "force %s compression",
 493                                                    compress_type);
 494                         } else {
 495                                 if (!btrfs_test_opt(root, COMPRESS))
 496                                         btrfs_info(root->fs_info,
 497                                                    "btrfs: use %s compression",
 498                                                    compress_type);
 499                                 /*
 500                                  * If we remount from compress-force=xxx to
 501                                  * compress=xxx, we need clear FORCE_COMPRESS
 502                                  * flag, otherwise, there is no way for users
 503                                  * to disable forcible compression separately.
 504                                  */
 505                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 506                         }
 507                         break;
 508                 case Opt_ssd:
 509                         btrfs_set_and_info(root, SSD,
 510                                            "use ssd allocation scheme");
 511                         break;
 512                 case Opt_ssd_spread:
 513                         btrfs_set_and_info(root, SSD_SPREAD,
 514                                            "use spread ssd allocation scheme");
 515                         btrfs_set_opt(info->mount_opt, SSD);
 516                         break;
 517                 case Opt_nossd:
 518                         btrfs_set_and_info(root, NOSSD,
 519                                              "not using ssd allocation scheme");
 520                         btrfs_clear_opt(info->mount_opt, SSD);
 521                         break;
 522                 case Opt_barrier:
 523                         btrfs_clear_and_info(root, NOBARRIER,
 524                                              "turning on barriers");
 525                         break;
 526                 case Opt_nobarrier:
 527                         btrfs_set_and_info(root, NOBARRIER,
 528                                            "turning off barriers");
 529                         break;
 530                 case Opt_thread_pool:
 531                         ret = match_int(&args[0], &intarg);
 532                         if (ret) {
 533                                 goto out;
 534                         } else if (intarg > 0) {
 535                                 info->thread_pool_size = intarg;
 536                         } else {
 537                                 ret = -EINVAL;
 538                                 goto out;
 539                         }
 540                         break;
 541                 case Opt_max_inline:
 542                         num = match_strdup(&args[0]);
 543                         if (num) {
 544                                 info->max_inline = memparse(num, NULL);
 545                                 kfree(num);
 546
 547                                 if (info->max_inline) {
 548                                         info->max_inline = min_t(u64,
 549                                                 info->max_inline,
 550                                                 root->sectorsize);
 551                                 }
 552                                 btrfs_info(root->fs_info, "max_inline at %llu",
 553                                         info->max_inline);
 554                         } else {
 555                                 ret = -ENOMEM;
 556                                 goto out;
 557                         }
 558                         break;
 559                 case Opt_alloc_start:
 560                         num = match_strdup(&args[0]);
 561                         if (num) {
 562                                 mutex_lock(&info->chunk_mutex);
 563                                 info->alloc_start = memparse(num, NULL);
 564                                 mutex_unlock(&info->chunk_mutex);
 565                                 kfree(num);
 566                                 btrfs_info(root->fs_info, "allocations start at %llu",
 567                                         info->alloc_start);
 568                         } else {
 569                                 ret = -ENOMEM;
 570                                 goto out;
 571                         }
 572                         break;
 573                 case Opt_acl:
 574 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 575                         root->fs_info->sb->s_flags |= MS_POSIXACL;
 576                         break;
 577 #else
 578                         btrfs_err(root->fs_info,
 579                                 "support for ACL not compiled in!");
 580                         ret = -EINVAL;
 581                         goto out;
 582 #endif
 583                 case Opt_noacl:
 584                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 585                         break;
 586                 case Opt_notreelog:
 587                         btrfs_set_and_info(root, NOTREELOG,
 588                                            "disabling tree log");
 589                         break;
 590                 case Opt_treelog:
 591                         btrfs_clear_and_info(root, NOTREELOG,
 592                                              "enabling tree log");
 593                         break;
 594                 case Opt_flushoncommit:
 595                         btrfs_set_and_info(root, FLUSHONCOMMIT,
 596                                            "turning on flush-on-commit");
 597                         break;
 598                 case Opt_noflushoncommit:
 599                         btrfs_clear_and_info(root, FLUSHONCOMMIT,
 600                                              "turning off flush-on-commit");
 601                         break;
 602                 case Opt_ratio:
 603                         ret = match_int(&args[0], &intarg);
 604                         if (ret) {
 605                                 goto out;
 606                         } else if (intarg >= 0) {
 607                                 info->metadata_ratio = intarg;
 608                                 btrfs_info(root->fs_info, "metadata ratio %d",
 609                                        info->metadata_ratio);
 610                         } else {
 611                                 ret = -EINVAL;
 612                                 goto out;
 613                         }
 614                         break;
 615                 case Opt_discard:
 616                         btrfs_set_and_info(root, DISCARD,
 617                                            "turning on discard");
 618                         break;
 619                 case Opt_nodiscard:
 620                         btrfs_clear_and_info(root, DISCARD,
 621                                              "turning off discard");
 622                         break;
 623                 case Opt_space_cache:
 624                 case Opt_space_cache_version:
 625                         if (token == Opt_space_cache ||
 626                             strcmp(args[0].from, "v1") == 0) {
 627                                 btrfs_clear_opt(root->fs_info->mount_opt,
 628                                                 FREE_SPACE_TREE);
 629                                 btrfs_set_and_info(root, SPACE_CACHE,
 630                                                    "enabling disk space caching");
 631                         } else if (strcmp(args[0].from, "v2") == 0) {
 632                                 btrfs_clear_opt(root->fs_info->mount_opt,
 633                                                 SPACE_CACHE);
 634                                 btrfs_set_and_info(root, FREE_SPACE_TREE,
 635                                                    "enabling free space tree");
 636                         } else {
 637                                 ret = -EINVAL;
 638                                 goto out;
 639                         }
 640                         break;
 641                 case Opt_rescan_uuid_tree:
 642                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 643                         break;
 644                 case Opt_no_space_cache:
 645                         if (btrfs_test_opt(root, SPACE_CACHE)) {
 646                                 btrfs_clear_and_info(root, SPACE_CACHE,
 647                                                      "disabling disk space caching");
 648                         }
 649                         if (btrfs_test_opt(root, FREE_SPACE_TREE)) {
 650                                 btrfs_clear_and_info(root, FREE_SPACE_TREE,
 651                                                      "disabling free space tree");
 652                         }
 653                         break;
 654                 case Opt_inode_cache:
 655                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 656                                            "enabling inode map caching");
 657                         break;
 658                 case Opt_noinode_cache:
 659                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 660                                              "disabling inode map caching");
 661                         break;
 662                 case Opt_clear_cache:
 663                         btrfs_set_and_info(root, CLEAR_CACHE,
 664                                            "force clearing of disk cache");
 665                         break;
 666                 case Opt_user_subvol_rm_allowed:
 667                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 668                         break;
 669                 case Opt_enospc_debug:
 670                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 671                         break;
 672                 case Opt_noenospc_debug:
 673                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 674                         break;
 675                 case Opt_defrag:
 676                         btrfs_set_and_info(root, AUTO_DEFRAG,
 677                                            "enabling auto defrag");
 678                         break;
 679                 case Opt_nodefrag:
 680                         btrfs_clear_and_info(root, AUTO_DEFRAG,
 681                                              "disabling auto defrag");
 682                         break;
 683                 case Opt_recovery:
 684                         btrfs_info(root->fs_info, "enabling auto recovery");
 685                         btrfs_set_opt(info->mount_opt, RECOVERY);
 686                         break;
 687                 case Opt_skip_balance:
 688                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 689                         break;
 690 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 691                 case Opt_check_integrity_including_extent_data:
 692                         btrfs_info(root->fs_info,
 693                                    "enabling check integrity including extent data");
 694                         btrfs_set_opt(info->mount_opt,
 695                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 696                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 697                         break;
 698                 case Opt_check_integrity:
 699                         btrfs_info(root->fs_info, "enabling check integrity");
 700                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 701                         break;
 702                 case Opt_check_integrity_print_mask:
 703                         ret = match_int(&args[0], &intarg);
 704                         if (ret) {
 705                                 goto out;
 706                         } else if (intarg >= 0) {
 707                                 info->check_integrity_print_mask = intarg;
 708                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
 709                                        info->check_integrity_print_mask);
 710                         } else {
 711                                 ret = -EINVAL;
 712                                 goto out;
 713                         }
 714                         break;
 715 #else
 716                 case Opt_check_integrity_including_extent_data:
 717                 case Opt_check_integrity:
 718                 case Opt_check_integrity_print_mask:
 719                         btrfs_err(root->fs_info,
 720                                 "support for check_integrity* not compiled in!");
 721                         ret = -EINVAL;
 722                         goto out;
 723 #endif
 724                 case Opt_fatal_errors:
 725                         if (strcmp(args[0].from, "panic") == 0)
 726                                 btrfs_set_opt(info->mount_opt,
 727                                               PANIC_ON_FATAL_ERROR);
 728                         else if (strcmp(args[0].from, "bug") == 0)
 729                                 btrfs_clear_opt(info->mount_opt,
 730                                               PANIC_ON_FATAL_ERROR);
 731                         else {
 732                                 ret = -EINVAL;
 733                                 goto out;
 734                         }
 735                         break;
 736                 case Opt_commit_interval:
 737                         intarg = 0;
 738                         ret = match_int(&args[0], &intarg);
 739                         if (ret < 0) {
 740                                 btrfs_err(root->fs_info, "invalid commit interval");
 741                                 ret = -EINVAL;
 742                                 goto out;
 743                         }
 744                         if (intarg > 0) {
 745                                 if (intarg > 300) {
 746                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
 747                                                         intarg);
 748                                 }
 749                                 info->commit_interval = intarg;
 750                         } else {
 751                                 btrfs_info(root->fs_info, "using default commit interval %ds",
 752                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
 753                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 754                         }
 755                         break;
 756 #ifdef CONFIG_BTRFS_DEBUG
 757                 case Opt_fragment_all:
 758                         btrfs_info(root->fs_info, "fragmenting all space");
 759                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 760                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 761                         break;
 762                 case Opt_fragment_metadata:
 763                         btrfs_info(root->fs_info, "fragmenting metadata");
 764                         btrfs_set_opt(info->mount_opt,
 765                                       FRAGMENT_METADATA);
 766                         break;
 767                 case Opt_fragment_data:
 768                         btrfs_info(root->fs_info, "fragmenting data");
 769                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 770                         break;
 771 #endif
 772                 case Opt_err:
 773                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
 774                         ret = -EINVAL;
 775                         goto out;
 776                 default:
 777                         break;
 778                 }
 779         }
 780 out:
 781         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE) &&
 782             !btrfs_test_opt(root, FREE_SPACE_TREE) &&
 783             !btrfs_test_opt(root, CLEAR_CACHE)) {
 784                 btrfs_err(root->fs_info, "cannot disable free space tree");
 785                 ret = -EINVAL;
 786
 787         }
 788         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 789                 btrfs_info(root->fs_info, "disk space caching is enabled");
 790         if (!ret && btrfs_test_opt(root, FREE_SPACE_TREE))
 791                 btrfs_info(root->fs_info, "using free space tree");
 792         kfree(orig);
 793         return ret;
 794 }
 795
 796 /*
 797  * Parse mount options that are required early in the mount process.
 798  *
 799  * All other options will be parsed on much later in the mount process and
 800  * only when we need to allocate a new super block.
 801  */
 802 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 803                 void *holder, char **subvol_name, u64 *subvol_objectid,
 804                 struct btrfs_fs_devices **fs_devices)
 805 {
 806         substring_t args[MAX_OPT_ARGS];
 807         char *device_name, *opts, *orig, *p;
 808         char *num = NULL;
 809         int error = 0;
 810
 811         if (!options)
 812                 return 0;
 813
 814         /*
 815          * strsep changes the string, duplicate it because parse_options
 816          * gets called twice
 817          */
 818         opts = kstrdup(options, GFP_KERNEL);
 819         if (!opts)
 820                 return -ENOMEM;
 821         orig = opts;
 822
 823         while ((p = strsep(&opts, ",")) != NULL) {
 824                 int token;
 825                 if (!*p)
 826                         continue;
 827
 828                 token = match_token(p, tokens, args);
 829                 switch (token) {
 830                 case Opt_subvol:
 831                         kfree(*subvol_name);
 832                         *subvol_name = match_strdup(&args[0]);
 833                         if (!*subvol_name) {
 834                                 error = -ENOMEM;
 835                                 goto out;
 836                         }
 837                         break;
 838                 case Opt_subvolid:
 839                         num = match_strdup(&args[0]);
 840                         if (num) {
 841                                 *subvol_objectid = memparse(num, NULL);
 842                                 kfree(num);
 843                                 /* we want the original fs_tree */
 844                                 if (!*subvol_objectid)
 845                                         *subvol_objectid =
 846                                                 BTRFS_FS_TREE_OBJECTID;
 847                         } else {
 848                                 error = -EINVAL;
 849                                 goto out;
 850                         }
 851                         break;
 852                 case Opt_subvolrootid:
 853                         printk(KERN_WARNING
 854                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
 855                                 "no effect\n");
 856                         break;
 857                 case Opt_device:
 858                         device_name = match_strdup(&args[0]);
 859                         if (!device_name) {
 860                                 error = -ENOMEM;
 861                                 goto out;
 862                         }
 863                         error = btrfs_scan_one_device(device_name,
 864                                         flags, holder, fs_devices);
 865                         kfree(device_name);
 866                         if (error)
 867                                 goto out;
 868                         break;
 869                 default:
 870                         break;
 871                 }
 872         }
 873
 874 out:
 875         kfree(orig);
 876         return error;
 877 }
 878
 879 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 880                                            u64 subvol_objectid)
 881 {
 882         struct btrfs_root *root = fs_info->tree_root;
 883         struct btrfs_root *fs_root;
 884         struct btrfs_root_ref *root_ref;
 885         struct btrfs_inode_ref *inode_ref;
 886         struct btrfs_key key;
 887         struct btrfs_path *path = NULL;
 888         char *name = NULL, *ptr;
 889         u64 dirid;
 890         int len;
 891         int ret;
 892
 893         path = btrfs_alloc_path();
 894         if (!path) {
 895                 ret = -ENOMEM;
 896                 goto err;
 897         }
 898         path->leave_spinning = 1;
 899
 900         name = kmalloc(PATH_MAX, GFP_NOFS);
 901         if (!name) {
 902                 ret = -ENOMEM;
 903                 goto err;
 904         }
 905         ptr = name + PATH_MAX - 1;
 906         ptr[0] = '\0';
 907
 908         /*
 909          * Walk up the subvolume trees in the tree of tree roots by root
 910          * backrefs until we hit the top-level subvolume.
 911          */
 912         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 913                 key.objectid = subvol_objectid;
 914                 key.type = BTRFS_ROOT_BACKREF_KEY;
 915                 key.offset = (u64)-1;
 916
 917                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 918                 if (ret < 0) {
 919                         goto err;
 920                 } else if (ret > 0) {
 921                         ret = btrfs_previous_item(root, path, subvol_objectid,
 922                                                   BTRFS_ROOT_BACKREF_KEY);
 923                         if (ret < 0) {
 924                                 goto err;
 925                         } else if (ret > 0) {
 926                                 ret = -ENOENT;
 927                                 goto err;
 928                         }
 929                 }
 930
 931                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 932                 subvol_objectid = key.offset;
 933
 934                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 935                                           struct btrfs_root_ref);
 936                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 937                 ptr -= len + 1;
 938                 if (ptr < name) {
 939                         ret = -ENAMETOOLONG;
 940                         goto err;
 941                 }
 942                 read_extent_buffer(path->nodes[0], ptr + 1,
 943                                    (unsigned long)(root_ref + 1), len);
 944                 ptr[0] = '/';
 945                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
 946                 btrfs_release_path(path);
 947
 948                 key.objectid = subvol_objectid;
 949                 key.type = BTRFS_ROOT_ITEM_KEY;
 950                 key.offset = (u64)-1;
 951                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
 952                 if (IS_ERR(fs_root)) {
 953                         ret = PTR_ERR(fs_root);
 954                         goto err;
 955                 }
 956
 957                 /*
 958                  * Walk up the filesystem tree by inode refs until we hit the
 959                  * root directory.
 960                  */
 961                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
 962                         key.objectid = dirid;
 963                         key.type = BTRFS_INODE_REF_KEY;
 964                         key.offset = (u64)-1;
 965
 966                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
 967                         if (ret < 0) {
 968                                 goto err;
 969                         } else if (ret > 0) {
 970                                 ret = btrfs_previous_item(fs_root, path, dirid,
 971                                                           BTRFS_INODE_REF_KEY);
 972                                 if (ret < 0) {
 973                                         goto err;
 974                                 } else if (ret > 0) {
 975                                         ret = -ENOENT;
 976                                         goto err;
 977                                 }
 978                         }
 979
 980                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 981                         dirid = key.offset;
 982
 983                         inode_ref = btrfs_item_ptr(path->nodes[0],
 984                                                    path->slots[0],
 985                                                    struct btrfs_inode_ref);
 986                         len = btrfs_inode_ref_name_len(path->nodes[0],
 987                                                        inode_ref);
 988                         ptr -= len + 1;
 989                         if (ptr < name) {
 990                                 ret = -ENAMETOOLONG;
 991                                 goto err;
 992                         }
 993                         read_extent_buffer(path->nodes[0], ptr + 1,
 994                                            (unsigned long)(inode_ref + 1), len);
 995                         ptr[0] = '/';
 996                         btrfs_release_path(path);
 997                 }
 998         }
 999
1000         btrfs_free_path(path);
1001         if (ptr == name + PATH_MAX - 1) {
1002                 name[0] = '/';
1003                 name[1] = '\0';
1004         } else {
1005                 memmove(name, ptr, name + PATH_MAX - ptr);
1006         }
1007         return name;
1008
1009 err:
1010         btrfs_free_path(path);
1011         kfree(name);
1012         return ERR_PTR(ret);
1013 }
1014
1015 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1016 {
1017         struct btrfs_root *root = fs_info->tree_root;
1018         struct btrfs_dir_item *di;
1019         struct btrfs_path *path;
1020         struct btrfs_key location;
1021         u64 dir_id;
1022
1023         path = btrfs_alloc_path();
1024         if (!path)
1025                 return -ENOMEM;
1026         path->leave_spinning = 1;
1027
1028         /*
1029          * Find the "default" dir item which points to the root item that we
1030          * will mount by default if we haven't been given a specific subvolume
1031          * to mount.
1032          */
1033         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1034         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1035         if (IS_ERR(di)) {
1036                 btrfs_free_path(path);
1037                 return PTR_ERR(di);
1038         }
1039         if (!di) {
1040                 /*
1041                  * Ok the default dir item isn't there.  This is weird since
1042                  * it's always been there, but don't freak out, just try and
1043                  * mount the top-level subvolume.
1044                  */
1045                 btrfs_free_path(path);
1046                 *objectid = BTRFS_FS_TREE_OBJECTID;
1047                 return 0;
1048         }
1049
1050         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1051         btrfs_free_path(path);
1052         *objectid = location.objectid;
1053         return 0;
1054 }
1055
1056 static int btrfs_fill_super(struct super_block *sb,
1057                             struct btrfs_fs_devices *fs_devices,
1058                             void *data, int silent)
1059 {
1060         struct inode *inode;
1061         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1062         struct btrfs_key key;
1063         int err;
1064
1065         sb->s_maxbytes = MAX_LFS_FILESIZE;
1066         sb->s_magic = BTRFS_SUPER_MAGIC;
1067         sb->s_op = &btrfs_super_ops;
1068         sb->s_d_op = &btrfs_dentry_operations;
1069         sb->s_export_op = &btrfs_export_ops;
1070         sb->s_xattr = btrfs_xattr_handlers;
1071         sb->s_time_gran = 1;
1072 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1073         sb->s_flags |= MS_POSIXACL;
1074 #endif
1075         sb->s_flags |= MS_I_VERSION;
1076         sb->s_iflags |= SB_I_CGROUPWB;
1077         err = open_ctree(sb, fs_devices, (char *)data);
1078         if (err) {
1079                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
1080                 return err;
1081         }
1082
1083         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1084         key.type = BTRFS_INODE_ITEM_KEY;
1085         key.offset = 0;
1086         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1087         if (IS_ERR(inode)) {
1088                 err = PTR_ERR(inode);
1089                 goto fail_close;
1090         }
1091
1092         sb->s_root = d_make_root(inode);
1093         if (!sb->s_root) {
1094                 err = -ENOMEM;
1095                 goto fail_close;
1096         }
1097
1098         save_mount_options(sb, data);
1099         cleancache_init_fs(sb);
1100         sb->s_flags |= MS_ACTIVE;
1101         return 0;
1102
1103 fail_close:
1104         close_ctree(fs_info->tree_root);
1105         return err;
1106 }
1107
1108 int btrfs_sync_fs(struct super_block *sb, int wait)
1109 {
1110         struct btrfs_trans_handle *trans;
1111         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1112         struct btrfs_root *root = fs_info->tree_root;
1113
1114         trace_btrfs_sync_fs(wait);
1115
1116         if (!wait) {
1117                 filemap_flush(fs_info->btree_inode->i_mapping);
1118                 return 0;
1119         }
1120
1121         btrfs_wait_ordered_roots(fs_info, -1);
1122
1123         trans = btrfs_attach_transaction_barrier(root);
1124         if (IS_ERR(trans)) {
1125                 /* no transaction, don't bother */
1126                 if (PTR_ERR(trans) == -ENOENT) {
1127                         /*
1128                          * Exit unless we have some pending changes
1129                          * that need to go through commit
1130                          */
1131                         if (fs_info->pending_changes == 0)
1132                                 return 0;
1133                         /*
1134                          * A non-blocking test if the fs is frozen. We must not
1135                          * start a new transaction here otherwise a deadlock
1136                          * happens. The pending operations are delayed to the
1137                          * next commit after thawing.
1138                          */
1139                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1140                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1141                         else
1142                                 return 0;
1143                         trans = btrfs_start_transaction(root, 0);
1144                 }
1145                 if (IS_ERR(trans))
1146                         return PTR_ERR(trans);
1147         }
1148         return btrfs_commit_transaction(trans, root);
1149 }
1150
1151 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1152 {
1153         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1154         struct btrfs_root *root = info->tree_root;
1155         char *compress_type;
1156
1157         if (btrfs_test_opt(root, DEGRADED))
1158                 seq_puts(seq, ",degraded");
1159         if (btrfs_test_opt(root, NODATASUM))
1160                 seq_puts(seq, ",nodatasum");
1161         if (btrfs_test_opt(root, NODATACOW))
1162                 seq_puts(seq, ",nodatacow");
1163         if (btrfs_test_opt(root, NOBARRIER))
1164                 seq_puts(seq, ",nobarrier");
1165         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1166                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1167         if (info->alloc_start != 0)
1168                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1169         if (info->thread_pool_size !=  min_t(unsigned long,
1170                                              num_online_cpus() + 2, 8))
1171                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1172         if (btrfs_test_opt(root, COMPRESS)) {
1173                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1174                         compress_type = "zlib";
1175                 else
1176                         compress_type = "lzo";
1177                 if (btrfs_test_opt(root, FORCE_COMPRESS))
1178                         seq_printf(seq, ",compress-force=%s", compress_type);
1179                 else
1180                         seq_printf(seq, ",compress=%s", compress_type);
1181         }
1182         if (btrfs_test_opt(root, NOSSD))
1183                 seq_puts(seq, ",nossd");
1184         if (btrfs_test_opt(root, SSD_SPREAD))
1185                 seq_puts(seq, ",ssd_spread");
1186         else if (btrfs_test_opt(root, SSD))
1187                 seq_puts(seq, ",ssd");
1188         if (btrfs_test_opt(root, NOTREELOG))
1189                 seq_puts(seq, ",notreelog");
1190         if (btrfs_test_opt(root, FLUSHONCOMMIT))
1191                 seq_puts(seq, ",flushoncommit");
1192         if (btrfs_test_opt(root, DISCARD))
1193                 seq_puts(seq, ",discard");
1194         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1195                 seq_puts(seq, ",noacl");
1196         if (btrfs_test_opt(root, SPACE_CACHE))
1197                 seq_puts(seq, ",space_cache");
1198         else if (btrfs_test_opt(root, FREE_SPACE_TREE))
1199                 seq_puts(seq, ",space_cache=v2");
1200         else
1201                 seq_puts(seq, ",nospace_cache");
1202         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1203                 seq_puts(seq, ",rescan_uuid_tree");
1204         if (btrfs_test_opt(root, CLEAR_CACHE))
1205                 seq_puts(seq, ",clear_cache");
1206         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1207                 seq_puts(seq, ",user_subvol_rm_allowed");
1208         if (btrfs_test_opt(root, ENOSPC_DEBUG))
1209                 seq_puts(seq, ",enospc_debug");
1210         if (btrfs_test_opt(root, AUTO_DEFRAG))
1211                 seq_puts(seq, ",autodefrag");
1212         if (btrfs_test_opt(root, INODE_MAP_CACHE))
1213                 seq_puts(seq, ",inode_cache");
1214         if (btrfs_test_opt(root, SKIP_BALANCE))
1215                 seq_puts(seq, ",skip_balance");
1216         if (btrfs_test_opt(root, RECOVERY))
1217                 seq_puts(seq, ",recovery");
1218 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1219         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1220                 seq_puts(seq, ",check_int_data");
1221         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1222                 seq_puts(seq, ",check_int");
1223         if (info->check_integrity_print_mask)
1224                 seq_printf(seq, ",check_int_print_mask=%d",
1225                                 info->check_integrity_print_mask);
1226 #endif
1227         if (info->metadata_ratio)
1228                 seq_printf(seq, ",metadata_ratio=%d",
1229                                 info->metadata_ratio);
1230         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1231                 seq_puts(seq, ",fatal_errors=panic");
1232         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1233                 seq_printf(seq, ",commit=%d", info->commit_interval);
1234 #ifdef CONFIG_BTRFS_DEBUG
1235         if (btrfs_test_opt(root, FRAGMENT_DATA))
1236                 seq_puts(seq, ",fragment=data");
1237         if (btrfs_test_opt(root, FRAGMENT_METADATA))
1238                 seq_puts(seq, ",fragment=metadata");
1239 #endif
1240         seq_printf(seq, ",subvolid=%llu",
1241                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1242         seq_puts(seq, ",subvol=");
1243         seq_dentry(seq, dentry, " \t\n\\");
1244         return 0;
1245 }
1246
1247 static int btrfs_test_super(struct super_block *s, void *data)
1248 {
1249         struct btrfs_fs_info *p = data;
1250         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1251
1252         return fs_info->fs_devices == p->fs_devices;
1253 }
1254
1255 static int btrfs_set_super(struct super_block *s, void *data)
1256 {
1257         int err = set_anon_super(s, data);
1258         if (!err)
1259                 s->s_fs_info = data;
1260         return err;
1261 }
1262
1263 /*
1264  * subvolumes are identified by ino 256
1265  */
1266 static inline int is_subvolume_inode(struct inode *inode)
1267 {
1268         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1269                 return 1;
1270         return 0;
1271 }
1272
1273 /*
1274  * This will add subvolid=0 to the argument string while removing any subvol=
1275  * and subvolid= arguments to make sure we get the top-level root for path
1276  * walking to the subvol we want.
1277  */
1278 static char *setup_root_args(char *args)
1279 {
1280         char *buf, *dst, *sep;
1281
1282         if (!args)
1283                 return kstrdup("subvolid=0", GFP_NOFS);
1284
1285         /* The worst case is that we add ",subvolid=0" to the end. */
1286         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1287         if (!buf)
1288                 return NULL;
1289
1290         while (1) {
1291                 sep = strchrnul(args, ',');
1292                 if (!strstarts(args, "subvol=") &&
1293                     !strstarts(args, "subvolid=")) {
1294                         memcpy(dst, args, sep - args);
1295                         dst += sep - args;
1296                         *dst++ = ',';
1297                 }
1298                 if (*sep)
1299                         args = sep + 1;
1300                 else
1301                         break;
1302         }
1303         strcpy(dst, "subvolid=0");
1304
1305         return buf;
1306 }
1307
1308 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1309                                    int flags, const char *device_name,
1310                                    char *data)
1311 {
1312         struct dentry *root;
1313         struct vfsmount *mnt = NULL;
1314         char *newargs;
1315         int ret;
1316
1317         newargs = setup_root_args(data);
1318         if (!newargs) {
1319                 root = ERR_PTR(-ENOMEM);
1320                 goto out;
1321         }
1322
1323         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1324         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1325                 if (flags & MS_RDONLY) {
1326                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1327                                              device_name, newargs);
1328                 } else {
1329                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1330                                              device_name, newargs);
1331                         if (IS_ERR(mnt)) {
1332                                 root = ERR_CAST(mnt);
1333                                 mnt = NULL;
1334                                 goto out;
1335                         }
1336
1337                         down_write(&mnt->mnt_sb->s_umount);
1338                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1339                         up_write(&mnt->mnt_sb->s_umount);
1340                         if (ret < 0) {
1341                                 root = ERR_PTR(ret);
1342                                 goto out;
1343                         }
1344                 }
1345         }
1346         if (IS_ERR(mnt)) {
1347                 root = ERR_CAST(mnt);
1348                 mnt = NULL;
1349                 goto out;
1350         }
1351
1352         if (!subvol_name) {
1353                 if (!subvol_objectid) {
1354                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1355                                                           &subvol_objectid);
1356                         if (ret) {
1357                                 root = ERR_PTR(ret);
1358                                 goto out;
1359                         }
1360                 }
1361                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1362                                                             subvol_objectid);
1363                 if (IS_ERR(subvol_name)) {
1364                         root = ERR_CAST(subvol_name);
1365                         subvol_name = NULL;
1366                         goto out;
1367                 }
1368
1369         }
1370
1371         root = mount_subtree(mnt, subvol_name);
1372         /* mount_subtree() drops our reference on the vfsmount. */
1373         mnt = NULL;
1374
1375         if (!IS_ERR(root)) {
1376                 struct super_block *s = root->d_sb;
1377                 struct inode *root_inode = d_inode(root);
1378                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1379
1380                 ret = 0;
1381                 if (!is_subvolume_inode(root_inode)) {
1382                         pr_err("BTRFS: '%s' is not a valid subvolume\n",
1383                                subvol_name);
1384                         ret = -EINVAL;
1385                 }
1386                 if (subvol_objectid && root_objectid != subvol_objectid) {
1387                         /*
1388                          * This will also catch a race condition where a
1389                          * subvolume which was passed by ID is renamed and
1390                          * another subvolume is renamed over the old location.
1391                          */
1392                         pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1393                                subvol_name, subvol_objectid);
1394                         ret = -EINVAL;
1395                 }
1396                 if (ret) {
1397                         dput(root);
1398                         root = ERR_PTR(ret);
1399                         deactivate_locked_super(s);
1400                 }
1401         }
1402
1403 out:
1404         mntput(mnt);
1405         kfree(newargs);
1406         kfree(subvol_name);
1407         return root;
1408 }
1409
1410 static int parse_security_options(char *orig_opts,
1411                                   struct security_mnt_opts *sec_opts)
1412 {
1413         char *secdata = NULL;
1414         int ret = 0;
1415
1416         secdata = alloc_secdata();
1417         if (!secdata)
1418                 return -ENOMEM;
1419         ret = security_sb_copy_data(orig_opts, secdata);
1420         if (ret) {
1421                 free_secdata(secdata);
1422                 return ret;
1423         }
1424         ret = security_sb_parse_opts_str(secdata, sec_opts);
1425         free_secdata(secdata);
1426         return ret;
1427 }
1428
1429 static int setup_security_options(struct btrfs_fs_info *fs_info,
1430                                   struct super_block *sb,
1431                                   struct security_mnt_opts *sec_opts)
1432 {
1433         int ret = 0;
1434
1435         /*
1436          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1437          * is valid.
1438          */
1439         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1440         if (ret)
1441                 return ret;
1442
1443 #ifdef CONFIG_SECURITY
1444         if (!fs_info->security_opts.num_mnt_opts) {
1445                 /* first time security setup, copy sec_opts to fs_info */
1446                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1447         } else {
1448                 /*
1449                  * Since SELinux(the only one supports security_mnt_opts) does
1450                  * NOT support changing context during remount/mount same sb,
1451                  * This must be the same or part of the same security options,
1452                  * just free it.
1453                  */
1454                 security_free_mnt_opts(sec_opts);
1455         }
1456 #endif
1457         return ret;
1458 }
1459
1460 /*
1461  * Find a superblock for the given device / mount point.
1462  *
1463  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1464  *        for multiple device setup.  Make sure to keep it in sync.
1465  */
1466 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1467                 const char *device_name, void *data)
1468 {
1469         struct block_device *bdev = NULL;
1470         struct super_block *s;
1471         struct btrfs_fs_devices *fs_devices = NULL;
1472         struct btrfs_fs_info *fs_info = NULL;
1473         struct security_mnt_opts new_sec_opts;
1474         fmode_t mode = FMODE_READ;
1475         char *subvol_name = NULL;
1476         u64 subvol_objectid = 0;
1477         int error = 0;
1478
1479         if (!(flags & MS_RDONLY))
1480                 mode |= FMODE_WRITE;
1481
1482         error = btrfs_parse_early_options(data, mode, fs_type,
1483                                           &subvol_name, &subvol_objectid,
1484                                           &fs_devices);
1485         if (error) {
1486                 kfree(subvol_name);
1487                 return ERR_PTR(error);
1488         }
1489
1490         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1491                 /* mount_subvol() will free subvol_name. */
1492                 return mount_subvol(subvol_name, subvol_objectid, flags,
1493                                     device_name, data);
1494         }
1495
1496         security_init_mnt_opts(&new_sec_opts);
1497         if (data) {
1498                 error = parse_security_options(data, &new_sec_opts);
1499                 if (error)
1500                         return ERR_PTR(error);
1501         }
1502
1503         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1504         if (error)
1505                 goto error_sec_opts;
1506
1507         /*
1508          * Setup a dummy root and fs_info for test/set super.  This is because
1509          * we don't actually fill this stuff out until open_ctree, but we need
1510          * it for searching for existing supers, so this lets us do that and
1511          * then open_ctree will properly initialize everything later.
1512          */
1513         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1514         if (!fs_info) {
1515                 error = -ENOMEM;
1516                 goto error_sec_opts;
1517         }
1518
1519         fs_info->fs_devices = fs_devices;
1520
1521         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1522         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1523         security_init_mnt_opts(&fs_info->security_opts);
1524         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1525                 error = -ENOMEM;
1526                 goto error_fs_info;
1527         }
1528
1529         error = btrfs_open_devices(fs_devices, mode, fs_type);
1530         if (error)
1531                 goto error_fs_info;
1532
1533         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1534                 error = -EACCES;
1535                 goto error_close_devices;
1536         }
1537
1538         bdev = fs_devices->latest_bdev;
1539         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1540                  fs_info);
1541         if (IS_ERR(s)) {
1542                 error = PTR_ERR(s);
1543                 goto error_close_devices;
1544         }
1545
1546         if (s->s_root) {
1547                 btrfs_close_devices(fs_devices);
1548                 free_fs_info(fs_info);
1549                 if ((flags ^ s->s_flags) & MS_RDONLY)
1550                         error = -EBUSY;
1551         } else {
1552                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1553                 btrfs_sb(s)->bdev_holder = fs_type;
1554                 error = btrfs_fill_super(s, fs_devices, data,
1555                                          flags & MS_SILENT ? 1 : 0);
1556         }
1557         if (error) {
1558                 deactivate_locked_super(s);
1559                 goto error_sec_opts;
1560         }
1561
1562         fs_info = btrfs_sb(s);
1563         error = setup_security_options(fs_info, s, &new_sec_opts);
1564         if (error) {
1565                 deactivate_locked_super(s);
1566                 goto error_sec_opts;
1567         }
1568
1569         return dget(s->s_root);
1570
1571 error_close_devices:
1572         btrfs_close_devices(fs_devices);
1573 error_fs_info:
1574         free_fs_info(fs_info);
1575 error_sec_opts:
1576         security_free_mnt_opts(&new_sec_opts);
1577         return ERR_PTR(error);
1578 }
1579
1580 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1581                                      int new_pool_size, int old_pool_size)
1582 {
1583         if (new_pool_size == old_pool_size)
1584                 return;
1585
1586         fs_info->thread_pool_size = new_pool_size;
1587
1588         btrfs_info(fs_info, "resize thread pool %d -> %d",
1589                old_pool_size, new_pool_size);
1590
1591         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1592         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1593         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1594         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1595         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1596         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1597         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1598                                 new_pool_size);
1599         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1600         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1601         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1602         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1603         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1604                                 new_pool_size);
1605 }
1606
1607 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1608 {
1609         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1610 }
1611
1612 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1613                                        unsigned long old_opts, int flags)
1614 {
1615         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1616             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1617              (flags & MS_RDONLY))) {
1618                 /* wait for any defraggers to finish */
1619                 wait_event(fs_info->transaction_wait,
1620                            (atomic_read(&fs_info->defrag_running) == 0));
1621                 if (flags & MS_RDONLY)
1622                         sync_filesystem(fs_info->sb);
1623         }
1624 }
1625
1626 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1627                                          unsigned long old_opts)
1628 {
1629         /*
1630          * We need cleanup all defragable inodes if the autodefragment is
1631          * close or the fs is R/O.
1632          */
1633         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1634             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1635              (fs_info->sb->s_flags & MS_RDONLY))) {
1636                 btrfs_cleanup_defrag_inodes(fs_info);
1637         }
1638
1639         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1640 }
1641
1642 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1643 {
1644         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1645         struct btrfs_root *root = fs_info->tree_root;
1646         unsigned old_flags = sb->s_flags;
1647         unsigned long old_opts = fs_info->mount_opt;
1648         unsigned long old_compress_type = fs_info->compress_type;
1649         u64 old_max_inline = fs_info->max_inline;
1650         u64 old_alloc_start = fs_info->alloc_start;
1651         int old_thread_pool_size = fs_info->thread_pool_size;
1652         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1653         int ret;
1654
1655         sync_filesystem(sb);
1656         btrfs_remount_prepare(fs_info);
1657
1658         if (data) {
1659                 struct security_mnt_opts new_sec_opts;
1660
1661                 security_init_mnt_opts(&new_sec_opts);
1662                 ret = parse_security_options(data, &new_sec_opts);
1663                 if (ret)
1664                         goto restore;
1665                 ret = setup_security_options(fs_info, sb,
1666                                              &new_sec_opts);
1667                 if (ret) {
1668                         security_free_mnt_opts(&new_sec_opts);
1669                         goto restore;
1670                 }
1671         }
1672
1673         ret = btrfs_parse_options(root, data);
1674         if (ret) {
1675                 ret = -EINVAL;
1676                 goto restore;
1677         }
1678
1679         btrfs_remount_begin(fs_info, old_opts, *flags);
1680         btrfs_resize_thread_pool(fs_info,
1681                 fs_info->thread_pool_size, old_thread_pool_size);
1682
1683         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1684                 goto out;
1685
1686         if (*flags & MS_RDONLY) {
1687                 /*
1688                  * this also happens on 'umount -rf' or on shutdown, when
1689                  * the filesystem is busy.
1690                  */
1691                 cancel_work_sync(&fs_info->async_reclaim_work);
1692
1693                 /* wait for the uuid_scan task to finish */
1694                 down(&fs_info->uuid_tree_rescan_sem);
1695                 /* avoid complains from lockdep et al. */
1696                 up(&fs_info->uuid_tree_rescan_sem);
1697
1698                 sb->s_flags |= MS_RDONLY;
1699
1700                 /*
1701                  * Setting MS_RDONLY will put the cleaner thread to
1702                  * sleep at the next loop if it's already active.
1703                  * If it's already asleep, we'll leave unused block
1704                  * groups on disk until we're mounted read-write again
1705                  * unless we clean them up here.
1706                  */
1707                 btrfs_delete_unused_bgs(fs_info);
1708
1709                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1710                 btrfs_scrub_cancel(fs_info);
1711                 btrfs_pause_balance(fs_info);
1712
1713                 ret = btrfs_commit_super(root);
1714                 if (ret)
1715                         goto restore;
1716         } else {
1717                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1718                         btrfs_err(fs_info,
1719                                 "Remounting read-write after error is not allowed");
1720                         ret = -EINVAL;
1721                         goto restore;
1722                 }
1723                 if (fs_info->fs_devices->rw_devices == 0) {
1724                         ret = -EACCES;
1725                         goto restore;
1726                 }
1727
1728                 if (fs_info->fs_devices->missing_devices >
1729                      fs_info->num_tolerated_disk_barrier_failures &&
1730                     !(*flags & MS_RDONLY)) {
1731                         btrfs_warn(fs_info,
1732                                 "too many missing devices, writeable remount is not allowed");
1733                         ret = -EACCES;
1734                         goto restore;
1735                 }
1736
1737                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1738                         ret = -EINVAL;
1739                         goto restore;
1740                 }
1741
1742                 ret = btrfs_cleanup_fs_roots(fs_info);
1743                 if (ret)
1744                         goto restore;
1745
1746                 /* recover relocation */
1747                 mutex_lock(&fs_info->cleaner_mutex);
1748                 ret = btrfs_recover_relocation(root);
1749                 mutex_unlock(&fs_info->cleaner_mutex);
1750                 if (ret)
1751                         goto restore;
1752
1753                 ret = btrfs_resume_balance_async(fs_info);
1754                 if (ret)
1755                         goto restore;
1756
1757                 ret = btrfs_resume_dev_replace_async(fs_info);
1758                 if (ret) {
1759                         btrfs_warn(fs_info, "failed to resume dev_replace");
1760                         goto restore;
1761                 }
1762
1763                 if (!fs_info->uuid_root) {
1764                         btrfs_info(fs_info, "creating UUID tree");
1765                         ret = btrfs_create_uuid_tree(fs_info);
1766                         if (ret) {
1767                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1768                                 goto restore;
1769                         }
1770                 }
1771                 sb->s_flags &= ~MS_RDONLY;
1772         }
1773 out:
1774         wake_up_process(fs_info->transaction_kthread);
1775         btrfs_remount_cleanup(fs_info, old_opts);
1776         return 0;
1777
1778 restore:
1779         /* We've hit an error - don't reset MS_RDONLY */
1780         if (sb->s_flags & MS_RDONLY)
1781                 old_flags |= MS_RDONLY;
1782         sb->s_flags = old_flags;
1783         fs_info->mount_opt = old_opts;
1784         fs_info->compress_type = old_compress_type;
1785         fs_info->max_inline = old_max_inline;
1786         mutex_lock(&fs_info->chunk_mutex);
1787         fs_info->alloc_start = old_alloc_start;
1788         mutex_unlock(&fs_info->chunk_mutex);
1789         btrfs_resize_thread_pool(fs_info,
1790                 old_thread_pool_size, fs_info->thread_pool_size);
1791         fs_info->metadata_ratio = old_metadata_ratio;
1792         btrfs_remount_cleanup(fs_info, old_opts);
1793         return ret;
1794 }
1795
1796 /* Used to sort the devices by max_avail(descending sort) */
1797 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1798                                        const void *dev_info2)
1799 {
1800         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1801             ((struct btrfs_device_info *)dev_info2)->max_avail)
1802                 return -1;
1803         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1804                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1805                 return 1;
1806         else
1807         return 0;
1808 }
1809
1810 /*
1811  * sort the devices by max_avail, in which max free extent size of each device
1812  * is stored.(Descending Sort)
1813  */
1814 static inline void btrfs_descending_sort_devices(
1815                                         struct btrfs_device_info *devices,
1816                                         size_t nr_devices)
1817 {
1818         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1819              btrfs_cmp_device_free_bytes, NULL);
1820 }
1821
1822 /*
1823  * The helper to calc the free space on the devices that can be used to store
1824  * file data.
1825  */
1826 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1827 {
1828         struct btrfs_fs_info *fs_info = root->fs_info;
1829         struct btrfs_device_info *devices_info;
1830         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1831         struct btrfs_device *device;
1832         u64 skip_space;
1833         u64 type;
1834         u64 avail_space;
1835         u64 used_space;
1836         u64 min_stripe_size;
1837         int min_stripes = 1, num_stripes = 1;
1838         int i = 0, nr_devices;
1839         int ret;
1840
1841         /*
1842          * We aren't under the device list lock, so this is racey-ish, but good
1843          * enough for our purposes.
1844          */
1845         nr_devices = fs_info->fs_devices->open_devices;
1846         if (!nr_devices) {
1847                 smp_mb();
1848                 nr_devices = fs_info->fs_devices->open_devices;
1849                 ASSERT(nr_devices);
1850                 if (!nr_devices) {
1851                         *free_bytes = 0;
1852                         return 0;
1853                 }
1854         }
1855
1856         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1857                                GFP_NOFS);
1858         if (!devices_info)
1859                 return -ENOMEM;
1860
1861         /* calc min stripe number for data space alloction */
1862         type = btrfs_get_alloc_profile(root, 1);
1863         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1864                 min_stripes = 2;
1865                 num_stripes = nr_devices;
1866         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1867                 min_stripes = 2;
1868                 num_stripes = 2;
1869         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1870                 min_stripes = 4;
1871                 num_stripes = 4;
1872         }
1873
1874         if (type & BTRFS_BLOCK_GROUP_DUP)
1875                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1876         else
1877                 min_stripe_size = BTRFS_STRIPE_LEN;
1878
1879         if (fs_info->alloc_start)
1880                 mutex_lock(&fs_devices->device_list_mutex);
1881         rcu_read_lock();
1882         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1883                 if (!device->in_fs_metadata || !device->bdev ||
1884                     device->is_tgtdev_for_dev_replace)
1885                         continue;
1886
1887                 if (i >= nr_devices)
1888                         break;
1889
1890                 avail_space = device->total_bytes - device->bytes_used;
1891
1892                 /* align with stripe_len */
1893                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1894                 avail_space *= BTRFS_STRIPE_LEN;
1895
1896                 /*
1897                  * In order to avoid overwritting the superblock on the drive,
1898                  * btrfs starts at an offset of at least 1MB when doing chunk
1899                  * allocation.
1900                  */
1901                 skip_space = SZ_1M;
1902
1903                 /* user can set the offset in fs_info->alloc_start. */
1904                 if (fs_info->alloc_start &&
1905                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1906                     device->total_bytes) {
1907                         rcu_read_unlock();
1908                         skip_space = max(fs_info->alloc_start, skip_space);
1909
1910                         /*
1911                          * btrfs can not use the free space in
1912                          * [0, skip_space - 1], we must subtract it from the
1913                          * total. In order to implement it, we account the used
1914                          * space in this range first.
1915                          */
1916                         ret = btrfs_account_dev_extents_size(device, 0,
1917                                                              skip_space - 1,
1918                                                              &used_space);
1919                         if (ret) {
1920                                 kfree(devices_info);
1921                                 mutex_unlock(&fs_devices->device_list_mutex);
1922                                 return ret;
1923                         }
1924
1925                         rcu_read_lock();
1926
1927                         /* calc the free space in [0, skip_space - 1] */
1928                         skip_space -= used_space;
1929                 }
1930
1931                 /*
1932                  * we can use the free space in [0, skip_space - 1], subtract
1933                  * it from the total.
1934                  */
1935                 if (avail_space && avail_space >= skip_space)
1936                         avail_space -= skip_space;
1937                 else
1938                         avail_space = 0;
1939
1940                 if (avail_space < min_stripe_size)
1941                         continue;
1942
1943                 devices_info[i].dev = device;
1944                 devices_info[i].max_avail = avail_space;
1945
1946                 i++;
1947         }
1948         rcu_read_unlock();
1949         if (fs_info->alloc_start)
1950                 mutex_unlock(&fs_devices->device_list_mutex);
1951
1952         nr_devices = i;
1953
1954         btrfs_descending_sort_devices(devices_info, nr_devices);
1955
1956         i = nr_devices - 1;
1957         avail_space = 0;
1958         while (nr_devices >= min_stripes) {
1959                 if (num_stripes > nr_devices)
1960                         num_stripes = nr_devices;
1961
1962                 if (devices_info[i].max_avail >= min_stripe_size) {
1963                         int j;
1964                         u64 alloc_size;
1965
1966                         avail_space += devices_info[i].max_avail * num_stripes;
1967                         alloc_size = devices_info[i].max_avail;
1968                         for (j = i + 1 - num_stripes; j <= i; j++)
1969                                 devices_info[j].max_avail -= alloc_size;
1970                 }
1971                 i--;
1972                 nr_devices--;
1973         }
1974
1975         kfree(devices_info);
1976         *free_bytes = avail_space;
1977         return 0;
1978 }
1979
1980 /*
1981  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
1982  *
1983  * If there's a redundant raid level at DATA block groups, use the respective
1984  * multiplier to scale the sizes.
1985  *
1986  * Unused device space usage is based on simulating the chunk allocator
1987  * algorithm that respects the device sizes, order of allocations and the
1988  * 'alloc_start' value, this is a close approximation of the actual use but
1989  * there are other factors that may change the result (like a new metadata
1990  * chunk).
1991  *
1992  * If metadata is exhausted, f_bavail will be 0.
1993  *
1994  * FIXME: not accurate for mixed block groups, total and free/used are ok,
1995  * available appears slightly larger.
1996  */
1997 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1998 {
1999         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2000         struct btrfs_super_block *disk_super = fs_info->super_copy;
2001         struct list_head *head = &fs_info->space_info;
2002         struct btrfs_space_info *found;
2003         u64 total_used = 0;
2004         u64 total_free_data = 0;
2005         u64 total_free_meta = 0;
2006         int bits = dentry->d_sb->s_blocksize_bits;
2007         __be32 *fsid = (__be32 *)fs_info->fsid;
2008         unsigned factor = 1;
2009         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2010         int ret;
2011         u64 thresh = 0;
2012
2013         /*
2014          * holding chunk_muext to avoid allocating new chunks, holding
2015          * device_list_mutex to avoid the device being removed
2016          */
2017         rcu_read_lock();
2018         list_for_each_entry_rcu(found, head, list) {
2019                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2020                         int i;
2021
2022                         total_free_data += found->disk_total - found->disk_used;
2023                         total_free_data -=
2024                                 btrfs_account_ro_block_groups_free_space(found);
2025
2026                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2027                                 if (!list_empty(&found->block_groups[i])) {
2028                                         switch (i) {
2029                                         case BTRFS_RAID_DUP:
2030                                         case BTRFS_RAID_RAID1:
2031                                         case BTRFS_RAID_RAID10:
2032                                                 factor = 2;
2033                                         }
2034                                 }
2035                         }
2036                 }
2037                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2038                         total_free_meta += found->disk_total - found->disk_used;
2039
2040                 total_used += found->disk_used;
2041         }
2042
2043         rcu_read_unlock();
2044
2045         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2046         buf->f_blocks >>= bits;
2047         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2048
2049         /* Account global block reserve as used, it's in logical size already */
2050         spin_lock(&block_rsv->lock);
2051         buf->f_bfree -= block_rsv->size >> bits;
2052         spin_unlock(&block_rsv->lock);
2053
2054         buf->f_bavail = div_u64(total_free_data, factor);
2055         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2056         if (ret)
2057                 return ret;
2058         buf->f_bavail += div_u64(total_free_data, factor);
2059         buf->f_bavail = buf->f_bavail >> bits;
2060
2061         /*
2062          * We calculate the remaining metadata space minus global reserve. If
2063          * this is (supposedly) smaller than zero, there's no space. But this
2064          * does not hold in practice, the exhausted state happens where's still
2065          * some positive delta. So we apply some guesswork and compare the
2066          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2067          *
2068          * We probably cannot calculate the exact threshold value because this
2069          * depends on the internal reservations requested by various
2070          * operations, so some operations that consume a few metadata will
2071          * succeed even if the Avail is zero. But this is better than the other
2072          * way around.
2073          */
2074         thresh = 4 * 1024 * 1024;
2075
2076         if (total_free_meta - thresh < block_rsv->size)
2077                 buf->f_bavail = 0;
2078
2079         buf->f_type = BTRFS_SUPER_MAGIC;
2080         buf->f_bsize = dentry->d_sb->s_blocksize;
2081         buf->f_namelen = BTRFS_NAME_LEN;
2082
2083         /* We treat it as constant endianness (it doesn't matter _which_)
2084            because we want the fsid to come out the same whether mounted
2085            on a big-endian or little-endian host */
2086         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2087         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2088         /* Mask in the root object ID too, to disambiguate subvols */
2089         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2090         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2091
2092         return 0;
2093 }
2094
2095 static void btrfs_kill_super(struct super_block *sb)
2096 {
2097         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2098         kill_anon_super(sb);
2099         free_fs_info(fs_info);
2100 }
2101
2102 static struct file_system_type btrfs_fs_type = {
2103         .owner          = THIS_MODULE,
2104         .name           = "btrfs",
2105         .mount          = btrfs_mount,
2106         .kill_sb        = btrfs_kill_super,
2107         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2108 };
2109 MODULE_ALIAS_FS("btrfs");
2110
2111 static int btrfs_control_open(struct inode *inode, struct file *file)
2112 {
2113         /*
2114          * The control file's private_data is used to hold the
2115          * transaction when it is started and is used to keep
2116          * track of whether a transaction is already in progress.
2117          */
2118         file->private_data = NULL;
2119         return 0;
2120 }
2121
2122 /*
2123  * used by btrfsctl to scan devices when no FS is mounted
2124  */
2125 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2126                                 unsigned long arg)
2127 {
2128         struct btrfs_ioctl_vol_args *vol;
2129         struct btrfs_fs_devices *fs_devices;
2130         int ret = -ENOTTY;
2131
2132         if (!capable(CAP_SYS_ADMIN))
2133                 return -EPERM;
2134
2135         vol = memdup_user((void __user *)arg, sizeof(*vol));
2136         if (IS_ERR(vol))
2137                 return PTR_ERR(vol);
2138
2139         switch (cmd) {
2140         case BTRFS_IOC_SCAN_DEV:
2141                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2142                                             &btrfs_fs_type, &fs_devices);
2143                 break;
2144         case BTRFS_IOC_DEVICES_READY:
2145                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2146                                             &btrfs_fs_type, &fs_devices);
2147                 if (ret)
2148                         break;
2149                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2150                 break;
2151         }
2152
2153         kfree(vol);
2154         return ret;
2155 }
2156
2157 static int btrfs_freeze(struct super_block *sb)
2158 {
2159         struct btrfs_trans_handle *trans;
2160         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2161
2162         trans = btrfs_attach_transaction_barrier(root);
2163         if (IS_ERR(trans)) {
2164                 /* no transaction, don't bother */
2165                 if (PTR_ERR(trans) == -ENOENT)
2166                         return 0;
2167                 return PTR_ERR(trans);
2168         }
2169         return btrfs_commit_transaction(trans, root);
2170 }
2171
2172 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2173 {
2174         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2175         struct btrfs_fs_devices *cur_devices;
2176         struct btrfs_device *dev, *first_dev = NULL;
2177         struct list_head *head;
2178         struct rcu_string *name;
2179
2180         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2181         cur_devices = fs_info->fs_devices;
2182         while (cur_devices) {
2183                 head = &cur_devices->devices;
2184                 list_for_each_entry(dev, head, dev_list) {
2185                         if (dev->missing)
2186                                 continue;
2187                         if (!dev->name)
2188                                 continue;
2189                         if (!first_dev || dev->devid < first_dev->devid)
2190                                 first_dev = dev;
2191                 }
2192                 cur_devices = cur_devices->seed;
2193         }
2194
2195         if (first_dev) {
2196                 rcu_read_lock();
2197                 name = rcu_dereference(first_dev->name);
2198                 seq_escape(m, name->str, " \t\n\\");
2199                 rcu_read_unlock();
2200         } else {
2201                 WARN_ON(1);
2202         }
2203         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2204         return 0;
2205 }
2206
2207 static const struct super_operations btrfs_super_ops = {
2208         .drop_inode     = btrfs_drop_inode,
2209         .evict_inode    = btrfs_evict_inode,
2210         .put_super      = btrfs_put_super,
2211         .sync_fs        = btrfs_sync_fs,
2212         .show_options   = btrfs_show_options,
2213         .show_devname   = btrfs_show_devname,
2214         .write_inode    = btrfs_write_inode,
2215         .alloc_inode    = btrfs_alloc_inode,
2216         .destroy_inode  = btrfs_destroy_inode,
2217         .statfs         = btrfs_statfs,
2218         .remount_fs     = btrfs_remount,
2219         .freeze_fs      = btrfs_freeze,
2220 };
2221
2222 static const struct file_operations btrfs_ctl_fops = {
2223         .open = btrfs_control_open,
2224         .unlocked_ioctl  = btrfs_control_ioctl,
2225         .compat_ioctl = btrfs_control_ioctl,
2226         .owner   = THIS_MODULE,
2227         .llseek = noop_llseek,
2228 };
2229
2230 static struct miscdevice btrfs_misc = {
2231         .minor          = BTRFS_MINOR,
2232         .name           = "btrfs-control",
2233         .fops           = &btrfs_ctl_fops
2234 };
2235
2236 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2237 MODULE_ALIAS("devname:btrfs-control");
2238
2239 static int btrfs_interface_init(void)
2240 {
2241         return misc_register(&btrfs_misc);
2242 }
2243
2244 static void btrfs_interface_exit(void)
2245 {
2246         misc_deregister(&btrfs_misc);
2247 }
2248
2249 static void btrfs_print_info(void)
2250 {
2251         printk(KERN_INFO "Btrfs loaded"
2252 #ifdef CONFIG_BTRFS_DEBUG
2253                         ", debug=on"
2254 #endif
2255 #ifdef CONFIG_BTRFS_ASSERT
2256                         ", assert=on"
2257 #endif
2258 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2259                         ", integrity-checker=on"
2260 #endif
2261                         "\n");
2262 }
2263
2264 static int btrfs_run_sanity_tests(void)
2265 {
2266         int ret;
2267
2268         ret = btrfs_init_test_fs();
2269         if (ret)
2270                 return ret;
2271
2272         ret = btrfs_test_free_space_cache();
2273         if (ret)
2274                 goto out;
2275         ret = btrfs_test_extent_buffer_operations();
2276         if (ret)
2277                 goto out;
2278         ret = btrfs_test_extent_io();
2279         if (ret)
2280                 goto out;
2281         ret = btrfs_test_inodes();
2282         if (ret)
2283                 goto out;
2284         ret = btrfs_test_qgroups();
2285         if (ret)
2286                 goto out;
2287         ret = btrfs_test_free_space_tree();
2288 out:
2289         btrfs_destroy_test_fs();
2290         return ret;
2291 }
2292
2293 static int __init init_btrfs_fs(void)
2294 {
2295         int err;
2296
2297         err = btrfs_hash_init();
2298         if (err)
2299                 return err;
2300
2301         btrfs_props_init();
2302
2303         err = btrfs_init_sysfs();
2304         if (err)
2305                 goto free_hash;
2306
2307         btrfs_init_compress();
2308
2309         err = btrfs_init_cachep();
2310         if (err)
2311                 goto free_compress;
2312
2313         err = extent_io_init();
2314         if (err)
2315                 goto free_cachep;
2316
2317         err = extent_map_init();
2318         if (err)
2319                 goto free_extent_io;
2320
2321         err = ordered_data_init();
2322         if (err)
2323                 goto free_extent_map;
2324
2325         err = btrfs_delayed_inode_init();
2326         if (err)
2327                 goto free_ordered_data;
2328
2329         err = btrfs_auto_defrag_init();
2330         if (err)
2331                 goto free_delayed_inode;
2332
2333         err = btrfs_delayed_ref_init();
2334         if (err)
2335                 goto free_auto_defrag;
2336
2337         err = btrfs_prelim_ref_init();
2338         if (err)
2339                 goto free_delayed_ref;
2340
2341         err = btrfs_end_io_wq_init();
2342         if (err)
2343                 goto free_prelim_ref;
2344
2345         err = btrfs_interface_init();
2346         if (err)
2347                 goto free_end_io_wq;
2348
2349         btrfs_init_lockdep();
2350
2351         btrfs_print_info();
2352
2353         err = btrfs_run_sanity_tests();
2354         if (err)
2355                 goto unregister_ioctl;
2356
2357         err = register_filesystem(&btrfs_fs_type);
2358         if (err)
2359                 goto unregister_ioctl;
2360
2361         return 0;
2362
2363 unregister_ioctl:
2364         btrfs_interface_exit();
2365 free_end_io_wq:
2366         btrfs_end_io_wq_exit();
2367 free_prelim_ref:
2368         btrfs_prelim_ref_exit();
2369 free_delayed_ref:
2370         btrfs_delayed_ref_exit();
2371 free_auto_defrag:
2372         btrfs_auto_defrag_exit();
2373 free_delayed_inode:
2374         btrfs_delayed_inode_exit();
2375 free_ordered_data:
2376         ordered_data_exit();
2377 free_extent_map:
2378         extent_map_exit();
2379 free_extent_io:
2380         extent_io_exit();
2381 free_cachep:
2382         btrfs_destroy_cachep();
2383 free_compress:
2384         btrfs_exit_compress();
2385         btrfs_exit_sysfs();
2386 free_hash:
2387         btrfs_hash_exit();
2388         return err;
2389 }
2390
2391 static void __exit exit_btrfs_fs(void)
2392 {
2393         btrfs_destroy_cachep();
2394         btrfs_delayed_ref_exit();
2395         btrfs_auto_defrag_exit();
2396         btrfs_delayed_inode_exit();
2397         btrfs_prelim_ref_exit();
2398         ordered_data_exit();
2399         extent_map_exit();
2400         extent_io_exit();
2401         btrfs_interface_exit();
2402         btrfs_end_io_wq_exit();
2403         unregister_filesystem(&btrfs_fs_type);
2404         btrfs_exit_sysfs();
2405         btrfs_cleanup_fs_uuids();
2406         btrfs_exit_compress();
2407         btrfs_hash_exit();
2408 }
2409
2410 late_initcall(init_btrfs_fs);
2411 module_exit(exit_btrfs_fs)
2412
2413 MODULE_LICENSE("GPL");