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, Opt_usebackuproot,
 307         Opt_nologreplay,
 308 #ifdef CONFIG_BTRFS_DEBUG
 309         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 310 #endif
 311         Opt_err,
 312 };
 313
 314 static const match_table_t tokens = {
 315         {Opt_degraded, "degraded"},
 316         {Opt_subvol, "subvol=%s"},
 317         {Opt_subvolid, "subvolid=%s"},
 318         {Opt_device, "device=%s"},
 319         {Opt_nodatasum, "nodatasum"},
 320         {Opt_datasum, "datasum"},
 321         {Opt_nodatacow, "nodatacow"},
 322         {Opt_datacow, "datacow"},
 323         {Opt_nobarrier, "nobarrier"},
 324         {Opt_barrier, "barrier"},
 325         {Opt_max_inline, "max_inline=%s"},
 326         {Opt_alloc_start, "alloc_start=%s"},
 327         {Opt_thread_pool, "thread_pool=%d"},
 328         {Opt_compress, "compress"},
 329         {Opt_compress_type, "compress=%s"},
 330         {Opt_compress_force, "compress-force"},
 331         {Opt_compress_force_type, "compress-force=%s"},
 332         {Opt_ssd, "ssd"},
 333         {Opt_ssd_spread, "ssd_spread"},
 334         {Opt_nossd, "nossd"},
 335         {Opt_acl, "acl"},
 336         {Opt_noacl, "noacl"},
 337         {Opt_notreelog, "notreelog"},
 338         {Opt_treelog, "treelog"},
 339         {Opt_nologreplay, "nologreplay"},
 340         {Opt_flushoncommit, "flushoncommit"},
 341         {Opt_noflushoncommit, "noflushoncommit"},
 342         {Opt_ratio, "metadata_ratio=%d"},
 343         {Opt_discard, "discard"},
 344         {Opt_nodiscard, "nodiscard"},
 345         {Opt_space_cache, "space_cache"},
 346         {Opt_space_cache_version, "space_cache=%s"},
 347         {Opt_clear_cache, "clear_cache"},
 348         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 349         {Opt_enospc_debug, "enospc_debug"},
 350         {Opt_noenospc_debug, "noenospc_debug"},
 351         {Opt_subvolrootid, "subvolrootid=%d"},
 352         {Opt_defrag, "autodefrag"},
 353         {Opt_nodefrag, "noautodefrag"},
 354         {Opt_inode_cache, "inode_cache"},
 355         {Opt_noinode_cache, "noinode_cache"},
 356         {Opt_no_space_cache, "nospace_cache"},
 357         {Opt_recovery, "recovery"}, /* deprecated */
 358         {Opt_usebackuproot, "usebackuproot"},
 359         {Opt_skip_balance, "skip_balance"},
 360         {Opt_check_integrity, "check_int"},
 361         {Opt_check_integrity_including_extent_data, "check_int_data"},
 362         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 363         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 364         {Opt_fatal_errors, "fatal_errors=%s"},
 365         {Opt_commit_interval, "commit=%d"},
 366 #ifdef CONFIG_BTRFS_DEBUG
 367         {Opt_fragment_data, "fragment=data"},
 368         {Opt_fragment_metadata, "fragment=metadata"},
 369         {Opt_fragment_all, "fragment=all"},
 370 #endif
 371         {Opt_err, NULL},
 372 };
 373
 374 /*
 375  * Regular mount options parser.  Everything that is needed only when
 376  * reading in a new superblock is parsed here.
 377  * XXX JDM: This needs to be cleaned up for remount.
 378  */
 379 int btrfs_parse_options(struct btrfs_root *root, char *options,
 380                         unsigned long new_flags)
 381 {
 382         struct btrfs_fs_info *info = root->fs_info;
 383         substring_t args[MAX_OPT_ARGS];
 384         char *p, *num, *orig = NULL;
 385         u64 cache_gen;
 386         int intarg;
 387         int ret = 0;
 388         char *compress_type;
 389         bool compress_force = false;
 390         enum btrfs_compression_type saved_compress_type;
 391         bool saved_compress_force;
 392         int no_compress = 0;
 393
 394         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 395         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE))
 396                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 397         else if (cache_gen)
 398                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 399
 400         /*
 401          * Even the options are empty, we still need to do extra check
 402          * against new flags
 403          */
 404         if (!options)
 405                 goto check;
 406
 407         /*
 408          * strsep changes the string, duplicate it because parse_options
 409          * gets called twice
 410          */
 411         options = kstrdup(options, GFP_NOFS);
 412         if (!options)
 413                 return -ENOMEM;
 414
 415         orig = options;
 416
 417         while ((p = strsep(&options, ",")) != NULL) {
 418                 int token;
 419                 if (!*p)
 420                         continue;
 421
 422                 token = match_token(p, tokens, args);
 423                 switch (token) {
 424                 case Opt_degraded:
 425                         btrfs_info(root->fs_info, "allowing degraded mounts");
 426                         btrfs_set_opt(info->mount_opt, DEGRADED);
 427                         break;
 428                 case Opt_subvol:
 429                 case Opt_subvolid:
 430                 case Opt_subvolrootid:
 431                 case Opt_device:
 432                         /*
 433                          * These are parsed by btrfs_parse_early_options
 434                          * and can be happily ignored here.
 435                          */
 436                         break;
 437                 case Opt_nodatasum:
 438                         btrfs_set_and_info(root, NODATASUM,
 439                                            "setting nodatasum");
 440                         break;
 441                 case Opt_datasum:
 442                         if (btrfs_test_opt(root, NODATASUM)) {
 443                                 if (btrfs_test_opt(root, NODATACOW))
 444                                         btrfs_info(root->fs_info, "setting datasum, datacow enabled");
 445                                 else
 446                                         btrfs_info(root->fs_info, "setting datasum");
 447                         }
 448                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 449                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 450                         break;
 451                 case Opt_nodatacow:
 452                         if (!btrfs_test_opt(root, NODATACOW)) {
 453                                 if (!btrfs_test_opt(root, COMPRESS) ||
 454                                     !btrfs_test_opt(root, FORCE_COMPRESS)) {
 455                                         btrfs_info(root->fs_info,
 456                                                    "setting nodatacow, compression disabled");
 457                                 } else {
 458                                         btrfs_info(root->fs_info, "setting nodatacow");
 459                                 }
 460                         }
 461                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 462                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 463                         btrfs_set_opt(info->mount_opt, NODATACOW);
 464                         btrfs_set_opt(info->mount_opt, NODATASUM);
 465                         break;
 466                 case Opt_datacow:
 467                         btrfs_clear_and_info(root, NODATACOW,
 468                                              "setting datacow");
 469                         break;
 470                 case Opt_compress_force:
 471                 case Opt_compress_force_type:
 472                         compress_force = true;
 473                         /* Fallthrough */
 474                 case Opt_compress:
 475                 case Opt_compress_type:
 476                         saved_compress_type = btrfs_test_opt(root, COMPRESS) ?
 477                                 info->compress_type : BTRFS_COMPRESS_NONE;
 478                         saved_compress_force =
 479                                 btrfs_test_opt(root, FORCE_COMPRESS);
 480                         if (token == Opt_compress ||
 481                             token == Opt_compress_force ||
 482                             strcmp(args[0].from, "zlib") == 0) {
 483                                 compress_type = "zlib";
 484                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 485                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 486                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 487                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 488                                 no_compress = 0;
 489                         } else if (strcmp(args[0].from, "lzo") == 0) {
 490                                 compress_type = "lzo";
 491                                 info->compress_type = BTRFS_COMPRESS_LZO;
 492                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 493                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 494                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 495                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 496                                 no_compress = 0;
 497                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 498                                 compress_type = "no";
 499                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 500                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 501                                 compress_force = false;
 502                                 no_compress++;
 503                         } else {
 504                                 ret = -EINVAL;
 505                                 goto out;
 506                         }
 507
 508                         if (compress_force) {
 509                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 510                         } else {
 511                                 /*
 512                                  * If we remount from compress-force=xxx to
 513                                  * compress=xxx, we need clear FORCE_COMPRESS
 514                                  * flag, otherwise, there is no way for users
 515                                  * to disable forcible compression separately.
 516                                  */
 517                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 518                         }
 519                         if ((btrfs_test_opt(root, COMPRESS) &&
 520                              (info->compress_type != saved_compress_type ||
 521                               compress_force != saved_compress_force)) ||
 522                             (!btrfs_test_opt(root, COMPRESS) &&
 523                              no_compress == 1)) {
 524                                 btrfs_info(root->fs_info,
 525                                            "%s %s compression",
 526                                            (compress_force) ? "force" : "use",
 527                                            compress_type);
 528                         }
 529                         compress_force = false;
 530                         break;
 531                 case Opt_ssd:
 532                         btrfs_set_and_info(root, SSD,
 533                                            "use ssd allocation scheme");
 534                         break;
 535                 case Opt_ssd_spread:
 536                         btrfs_set_and_info(root, SSD_SPREAD,
 537                                            "use spread ssd allocation scheme");
 538                         btrfs_set_opt(info->mount_opt, SSD);
 539                         break;
 540                 case Opt_nossd:
 541                         btrfs_set_and_info(root, NOSSD,
 542                                              "not using ssd allocation scheme");
 543                         btrfs_clear_opt(info->mount_opt, SSD);
 544                         break;
 545                 case Opt_barrier:
 546                         btrfs_clear_and_info(root, NOBARRIER,
 547                                              "turning on barriers");
 548                         break;
 549                 case Opt_nobarrier:
 550                         btrfs_set_and_info(root, NOBARRIER,
 551                                            "turning off barriers");
 552                         break;
 553                 case Opt_thread_pool:
 554                         ret = match_int(&args[0], &intarg);
 555                         if (ret) {
 556                                 goto out;
 557                         } else if (intarg > 0) {
 558                                 info->thread_pool_size = intarg;
 559                         } else {
 560                                 ret = -EINVAL;
 561                                 goto out;
 562                         }
 563                         break;
 564                 case Opt_max_inline:
 565                         num = match_strdup(&args[0]);
 566                         if (num) {
 567                                 info->max_inline = memparse(num, NULL);
 568                                 kfree(num);
 569
 570                                 if (info->max_inline) {
 571                                         info->max_inline = min_t(u64,
 572                                                 info->max_inline,
 573                                                 root->sectorsize);
 574                                 }
 575                                 btrfs_info(root->fs_info, "max_inline at %llu",
 576                                         info->max_inline);
 577                         } else {
 578                                 ret = -ENOMEM;
 579                                 goto out;
 580                         }
 581                         break;
 582                 case Opt_alloc_start:
 583                         num = match_strdup(&args[0]);
 584                         if (num) {
 585                                 mutex_lock(&info->chunk_mutex);
 586                                 info->alloc_start = memparse(num, NULL);
 587                                 mutex_unlock(&info->chunk_mutex);
 588                                 kfree(num);
 589                                 btrfs_info(root->fs_info, "allocations start at %llu",
 590                                         info->alloc_start);
 591                         } else {
 592                                 ret = -ENOMEM;
 593                                 goto out;
 594                         }
 595                         break;
 596                 case Opt_acl:
 597 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 598                         root->fs_info->sb->s_flags |= MS_POSIXACL;
 599                         break;
 600 #else
 601                         btrfs_err(root->fs_info,
 602                                 "support for ACL not compiled in!");
 603                         ret = -EINVAL;
 604                         goto out;
 605 #endif
 606                 case Opt_noacl:
 607                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 608                         break;
 609                 case Opt_notreelog:
 610                         btrfs_set_and_info(root, NOTREELOG,
 611                                            "disabling tree log");
 612                         break;
 613                 case Opt_treelog:
 614                         btrfs_clear_and_info(root, NOTREELOG,
 615                                              "enabling tree log");
 616                         break;
 617                 case Opt_nologreplay:
 618                         btrfs_set_and_info(root, NOLOGREPLAY,
 619                                            "disabling log replay at mount time");
 620                         break;
 621                 case Opt_flushoncommit:
 622                         btrfs_set_and_info(root, FLUSHONCOMMIT,
 623                                            "turning on flush-on-commit");
 624                         break;
 625                 case Opt_noflushoncommit:
 626                         btrfs_clear_and_info(root, FLUSHONCOMMIT,
 627                                              "turning off flush-on-commit");
 628                         break;
 629                 case Opt_ratio:
 630                         ret = match_int(&args[0], &intarg);
 631                         if (ret) {
 632                                 goto out;
 633                         } else if (intarg >= 0) {
 634                                 info->metadata_ratio = intarg;
 635                                 btrfs_info(root->fs_info, "metadata ratio %d",
 636                                        info->metadata_ratio);
 637                         } else {
 638                                 ret = -EINVAL;
 639                                 goto out;
 640                         }
 641                         break;
 642                 case Opt_discard:
 643                         btrfs_set_and_info(root, DISCARD,
 644                                            "turning on discard");
 645                         break;
 646                 case Opt_nodiscard:
 647                         btrfs_clear_and_info(root, DISCARD,
 648                                              "turning off discard");
 649                         break;
 650                 case Opt_space_cache:
 651                 case Opt_space_cache_version:
 652                         if (token == Opt_space_cache ||
 653                             strcmp(args[0].from, "v1") == 0) {
 654                                 btrfs_clear_opt(root->fs_info->mount_opt,
 655                                                 FREE_SPACE_TREE);
 656                                 btrfs_set_and_info(root, SPACE_CACHE,
 657                                                    "enabling disk space caching");
 658                         } else if (strcmp(args[0].from, "v2") == 0) {
 659                                 btrfs_clear_opt(root->fs_info->mount_opt,
 660                                                 SPACE_CACHE);
 661                                 btrfs_set_and_info(root, FREE_SPACE_TREE,
 662                                                    "enabling free space tree");
 663                         } else {
 664                                 ret = -EINVAL;
 665                                 goto out;
 666                         }
 667                         break;
 668                 case Opt_rescan_uuid_tree:
 669                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 670                         break;
 671                 case Opt_no_space_cache:
 672                         if (btrfs_test_opt(root, SPACE_CACHE)) {
 673                                 btrfs_clear_and_info(root, SPACE_CACHE,
 674                                                      "disabling disk space caching");
 675                         }
 676                         if (btrfs_test_opt(root, FREE_SPACE_TREE)) {
 677                                 btrfs_clear_and_info(root, FREE_SPACE_TREE,
 678                                                      "disabling free space tree");
 679                         }
 680                         break;
 681                 case Opt_inode_cache:
 682                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 683                                            "enabling inode map caching");
 684                         break;
 685                 case Opt_noinode_cache:
 686                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 687                                              "disabling inode map caching");
 688                         break;
 689                 case Opt_clear_cache:
 690                         btrfs_set_and_info(root, CLEAR_CACHE,
 691                                            "force clearing of disk cache");
 692                         break;
 693                 case Opt_user_subvol_rm_allowed:
 694                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 695                         break;
 696                 case Opt_enospc_debug:
 697                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 698                         break;
 699                 case Opt_noenospc_debug:
 700                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 701                         break;
 702                 case Opt_defrag:
 703                         btrfs_set_and_info(root, AUTO_DEFRAG,
 704                                            "enabling auto defrag");
 705                         break;
 706                 case Opt_nodefrag:
 707                         btrfs_clear_and_info(root, AUTO_DEFRAG,
 708                                              "disabling auto defrag");
 709                         break;
 710                 case Opt_recovery:
 711                         btrfs_warn(root->fs_info,
 712                                    "'recovery' is deprecated, use 'usebackuproot' instead");
 713                 case Opt_usebackuproot:
 714                         btrfs_info(root->fs_info,
 715                                    "trying to use backup root at mount time");
 716                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 717                         break;
 718                 case Opt_skip_balance:
 719                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 720                         break;
 721 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 722                 case Opt_check_integrity_including_extent_data:
 723                         btrfs_info(root->fs_info,
 724                                    "enabling check integrity including extent data");
 725                         btrfs_set_opt(info->mount_opt,
 726                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 727                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 728                         break;
 729                 case Opt_check_integrity:
 730                         btrfs_info(root->fs_info, "enabling check integrity");
 731                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 732                         break;
 733                 case Opt_check_integrity_print_mask:
 734                         ret = match_int(&args[0], &intarg);
 735                         if (ret) {
 736                                 goto out;
 737                         } else if (intarg >= 0) {
 738                                 info->check_integrity_print_mask = intarg;
 739                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
 740                                        info->check_integrity_print_mask);
 741                         } else {
 742                                 ret = -EINVAL;
 743                                 goto out;
 744                         }
 745                         break;
 746 #else
 747                 case Opt_check_integrity_including_extent_data:
 748                 case Opt_check_integrity:
 749                 case Opt_check_integrity_print_mask:
 750                         btrfs_err(root->fs_info,
 751                                 "support for check_integrity* not compiled in!");
 752                         ret = -EINVAL;
 753                         goto out;
 754 #endif
 755                 case Opt_fatal_errors:
 756                         if (strcmp(args[0].from, "panic") == 0)
 757                                 btrfs_set_opt(info->mount_opt,
 758                                               PANIC_ON_FATAL_ERROR);
 759                         else if (strcmp(args[0].from, "bug") == 0)
 760                                 btrfs_clear_opt(info->mount_opt,
 761                                               PANIC_ON_FATAL_ERROR);
 762                         else {
 763                                 ret = -EINVAL;
 764                                 goto out;
 765                         }
 766                         break;
 767                 case Opt_commit_interval:
 768                         intarg = 0;
 769                         ret = match_int(&args[0], &intarg);
 770                         if (ret < 0) {
 771                                 btrfs_err(root->fs_info, "invalid commit interval");
 772                                 ret = -EINVAL;
 773                                 goto out;
 774                         }
 775                         if (intarg > 0) {
 776                                 if (intarg > 300) {
 777                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
 778                                                         intarg);
 779                                 }
 780                                 info->commit_interval = intarg;
 781                         } else {
 782                                 btrfs_info(root->fs_info, "using default commit interval %ds",
 783                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
 784                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 785                         }
 786                         break;
 787 #ifdef CONFIG_BTRFS_DEBUG
 788                 case Opt_fragment_all:
 789                         btrfs_info(root->fs_info, "fragmenting all space");
 790                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 791                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 792                         break;
 793                 case Opt_fragment_metadata:
 794                         btrfs_info(root->fs_info, "fragmenting metadata");
 795                         btrfs_set_opt(info->mount_opt,
 796                                       FRAGMENT_METADATA);
 797                         break;
 798                 case Opt_fragment_data:
 799                         btrfs_info(root->fs_info, "fragmenting data");
 800                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 801                         break;
 802 #endif
 803                 case Opt_err:
 804                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
 805                         ret = -EINVAL;
 806                         goto out;
 807                 default:
 808                         break;
 809                 }
 810         }
 811 check:
 812         /*
 813          * Extra check for current option against current flag
 814          */
 815         if (btrfs_test_opt(root, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
 816                 btrfs_err(root->fs_info,
 817                           "nologreplay must be used with ro mount option");
 818                 ret = -EINVAL;
 819         }
 820 out:
 821         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE) &&
 822             !btrfs_test_opt(root, FREE_SPACE_TREE) &&
 823             !btrfs_test_opt(root, CLEAR_CACHE)) {
 824                 btrfs_err(root->fs_info, "cannot disable free space tree");
 825                 ret = -EINVAL;
 826
 827         }
 828         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 829                 btrfs_info(root->fs_info, "disk space caching is enabled");
 830         if (!ret && btrfs_test_opt(root, FREE_SPACE_TREE))
 831                 btrfs_info(root->fs_info, "using free space tree");
 832         kfree(orig);
 833         return ret;
 834 }
 835
 836 /*
 837  * Parse mount options that are required early in the mount process.
 838  *
 839  * All other options will be parsed on much later in the mount process and
 840  * only when we need to allocate a new super block.
 841  */
 842 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 843                 void *holder, char **subvol_name, u64 *subvol_objectid,
 844                 struct btrfs_fs_devices **fs_devices)
 845 {
 846         substring_t args[MAX_OPT_ARGS];
 847         char *device_name, *opts, *orig, *p;
 848         char *num = NULL;
 849         int error = 0;
 850
 851         if (!options)
 852                 return 0;
 853
 854         /*
 855          * strsep changes the string, duplicate it because parse_options
 856          * gets called twice
 857          */
 858         opts = kstrdup(options, GFP_KERNEL);
 859         if (!opts)
 860                 return -ENOMEM;
 861         orig = opts;
 862
 863         while ((p = strsep(&opts, ",")) != NULL) {
 864                 int token;
 865                 if (!*p)
 866                         continue;
 867
 868                 token = match_token(p, tokens, args);
 869                 switch (token) {
 870                 case Opt_subvol:
 871                         kfree(*subvol_name);
 872                         *subvol_name = match_strdup(&args[0]);
 873                         if (!*subvol_name) {
 874                                 error = -ENOMEM;
 875                                 goto out;
 876                         }
 877                         break;
 878                 case Opt_subvolid:
 879                         num = match_strdup(&args[0]);
 880                         if (num) {
 881                                 *subvol_objectid = memparse(num, NULL);
 882                                 kfree(num);
 883                                 /* we want the original fs_tree */
 884                                 if (!*subvol_objectid)
 885                                         *subvol_objectid =
 886                                                 BTRFS_FS_TREE_OBJECTID;
 887                         } else {
 888                                 error = -EINVAL;
 889                                 goto out;
 890                         }
 891                         break;
 892                 case Opt_subvolrootid:
 893                         printk(KERN_WARNING
 894                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
 895                                 "no effect\n");
 896                         break;
 897                 case Opt_device:
 898                         device_name = match_strdup(&args[0]);
 899                         if (!device_name) {
 900                                 error = -ENOMEM;
 901                                 goto out;
 902                         }
 903                         error = btrfs_scan_one_device(device_name,
 904                                         flags, holder, fs_devices);
 905                         kfree(device_name);
 906                         if (error)
 907                                 goto out;
 908                         break;
 909                 default:
 910                         break;
 911                 }
 912         }
 913
 914 out:
 915         kfree(orig);
 916         return error;
 917 }
 918
 919 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 920                                            u64 subvol_objectid)
 921 {
 922         struct btrfs_root *root = fs_info->tree_root;
 923         struct btrfs_root *fs_root;
 924         struct btrfs_root_ref *root_ref;
 925         struct btrfs_inode_ref *inode_ref;
 926         struct btrfs_key key;
 927         struct btrfs_path *path = NULL;
 928         char *name = NULL, *ptr;
 929         u64 dirid;
 930         int len;
 931         int ret;
 932
 933         path = btrfs_alloc_path();
 934         if (!path) {
 935                 ret = -ENOMEM;
 936                 goto err;
 937         }
 938         path->leave_spinning = 1;
 939
 940         name = kmalloc(PATH_MAX, GFP_NOFS);
 941         if (!name) {
 942                 ret = -ENOMEM;
 943                 goto err;
 944         }
 945         ptr = name + PATH_MAX - 1;
 946         ptr[0] = '\0';
 947
 948         /*
 949          * Walk up the subvolume trees in the tree of tree roots by root
 950          * backrefs until we hit the top-level subvolume.
 951          */
 952         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 953                 key.objectid = subvol_objectid;
 954                 key.type = BTRFS_ROOT_BACKREF_KEY;
 955                 key.offset = (u64)-1;
 956
 957                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 958                 if (ret < 0) {
 959                         goto err;
 960                 } else if (ret > 0) {
 961                         ret = btrfs_previous_item(root, path, subvol_objectid,
 962                                                   BTRFS_ROOT_BACKREF_KEY);
 963                         if (ret < 0) {
 964                                 goto err;
 965                         } else if (ret > 0) {
 966                                 ret = -ENOENT;
 967                                 goto err;
 968                         }
 969                 }
 970
 971                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 972                 subvol_objectid = key.offset;
 973
 974                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 975                                           struct btrfs_root_ref);
 976                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 977                 ptr -= len + 1;
 978                 if (ptr < name) {
 979                         ret = -ENAMETOOLONG;
 980                         goto err;
 981                 }
 982                 read_extent_buffer(path->nodes[0], ptr + 1,
 983                                    (unsigned long)(root_ref + 1), len);
 984                 ptr[0] = '/';
 985                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
 986                 btrfs_release_path(path);
 987
 988                 key.objectid = subvol_objectid;
 989                 key.type = BTRFS_ROOT_ITEM_KEY;
 990                 key.offset = (u64)-1;
 991                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
 992                 if (IS_ERR(fs_root)) {
 993                         ret = PTR_ERR(fs_root);
 994                         goto err;
 995                 }
 996
 997                 /*
 998                  * Walk up the filesystem tree by inode refs until we hit the
 999                  * root directory.
1000                  */
1001                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1002                         key.objectid = dirid;
1003                         key.type = BTRFS_INODE_REF_KEY;
1004                         key.offset = (u64)-1;
1005
1006                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1007                         if (ret < 0) {
1008                                 goto err;
1009                         } else if (ret > 0) {
1010                                 ret = btrfs_previous_item(fs_root, path, dirid,
1011                                                           BTRFS_INODE_REF_KEY);
1012                                 if (ret < 0) {
1013                                         goto err;
1014                                 } else if (ret > 0) {
1015                                         ret = -ENOENT;
1016                                         goto err;
1017                                 }
1018                         }
1019
1020                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1021                         dirid = key.offset;
1022
1023                         inode_ref = btrfs_item_ptr(path->nodes[0],
1024                                                    path->slots[0],
1025                                                    struct btrfs_inode_ref);
1026                         len = btrfs_inode_ref_name_len(path->nodes[0],
1027                                                        inode_ref);
1028                         ptr -= len + 1;
1029                         if (ptr < name) {
1030                                 ret = -ENAMETOOLONG;
1031                                 goto err;
1032                         }
1033                         read_extent_buffer(path->nodes[0], ptr + 1,
1034                                            (unsigned long)(inode_ref + 1), len);
1035                         ptr[0] = '/';
1036                         btrfs_release_path(path);
1037                 }
1038         }
1039
1040         btrfs_free_path(path);
1041         if (ptr == name + PATH_MAX - 1) {
1042                 name[0] = '/';
1043                 name[1] = '\0';
1044         } else {
1045                 memmove(name, ptr, name + PATH_MAX - ptr);
1046         }
1047         return name;
1048
1049 err:
1050         btrfs_free_path(path);
1051         kfree(name);
1052         return ERR_PTR(ret);
1053 }
1054
1055 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1056 {
1057         struct btrfs_root *root = fs_info->tree_root;
1058         struct btrfs_dir_item *di;
1059         struct btrfs_path *path;
1060         struct btrfs_key location;
1061         u64 dir_id;
1062
1063         path = btrfs_alloc_path();
1064         if (!path)
1065                 return -ENOMEM;
1066         path->leave_spinning = 1;
1067
1068         /*
1069          * Find the "default" dir item which points to the root item that we
1070          * will mount by default if we haven't been given a specific subvolume
1071          * to mount.
1072          */
1073         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1074         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1075         if (IS_ERR(di)) {
1076                 btrfs_free_path(path);
1077                 return PTR_ERR(di);
1078         }
1079         if (!di) {
1080                 /*
1081                  * Ok the default dir item isn't there.  This is weird since
1082                  * it's always been there, but don't freak out, just try and
1083                  * mount the top-level subvolume.
1084                  */
1085                 btrfs_free_path(path);
1086                 *objectid = BTRFS_FS_TREE_OBJECTID;
1087                 return 0;
1088         }
1089
1090         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1091         btrfs_free_path(path);
1092         *objectid = location.objectid;
1093         return 0;
1094 }
1095
1096 static int btrfs_fill_super(struct super_block *sb,
1097                             struct btrfs_fs_devices *fs_devices,
1098                             void *data, int silent)
1099 {
1100         struct inode *inode;
1101         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1102         struct btrfs_key key;
1103         int err;
1104
1105         sb->s_maxbytes = MAX_LFS_FILESIZE;
1106         sb->s_magic = BTRFS_SUPER_MAGIC;
1107         sb->s_op = &btrfs_super_ops;
1108         sb->s_d_op = &btrfs_dentry_operations;
1109         sb->s_export_op = &btrfs_export_ops;
1110         sb->s_xattr = btrfs_xattr_handlers;
1111         sb->s_time_gran = 1;
1112 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1113         sb->s_flags |= MS_POSIXACL;
1114 #endif
1115         sb->s_flags |= MS_I_VERSION;
1116         sb->s_iflags |= SB_I_CGROUPWB;
1117         err = open_ctree(sb, fs_devices, (char *)data);
1118         if (err) {
1119                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
1120                 return err;
1121         }
1122
1123         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1124         key.type = BTRFS_INODE_ITEM_KEY;
1125         key.offset = 0;
1126         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1127         if (IS_ERR(inode)) {
1128                 err = PTR_ERR(inode);
1129                 goto fail_close;
1130         }
1131
1132         sb->s_root = d_make_root(inode);
1133         if (!sb->s_root) {
1134                 err = -ENOMEM;
1135                 goto fail_close;
1136         }
1137
1138         save_mount_options(sb, data);
1139         cleancache_init_fs(sb);
1140         sb->s_flags |= MS_ACTIVE;
1141         return 0;
1142
1143 fail_close:
1144         close_ctree(fs_info->tree_root);
1145         return err;
1146 }
1147
1148 int btrfs_sync_fs(struct super_block *sb, int wait)
1149 {
1150         struct btrfs_trans_handle *trans;
1151         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1152         struct btrfs_root *root = fs_info->tree_root;
1153
1154         trace_btrfs_sync_fs(wait);
1155
1156         if (!wait) {
1157                 filemap_flush(fs_info->btree_inode->i_mapping);
1158                 return 0;
1159         }
1160
1161         btrfs_wait_ordered_roots(fs_info, -1);
1162
1163         trans = btrfs_attach_transaction_barrier(root);
1164         if (IS_ERR(trans)) {
1165                 /* no transaction, don't bother */
1166                 if (PTR_ERR(trans) == -ENOENT) {
1167                         /*
1168                          * Exit unless we have some pending changes
1169                          * that need to go through commit
1170                          */
1171                         if (fs_info->pending_changes == 0)
1172                                 return 0;
1173                         /*
1174                          * A non-blocking test if the fs is frozen. We must not
1175                          * start a new transaction here otherwise a deadlock
1176                          * happens. The pending operations are delayed to the
1177                          * next commit after thawing.
1178                          */
1179                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1180                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1181                         else
1182                                 return 0;
1183                         trans = btrfs_start_transaction(root, 0);
1184                 }
1185                 if (IS_ERR(trans))
1186                         return PTR_ERR(trans);
1187         }
1188         return btrfs_commit_transaction(trans, root);
1189 }
1190
1191 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1192 {
1193         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1194         struct btrfs_root *root = info->tree_root;
1195         char *compress_type;
1196
1197         if (btrfs_test_opt(root, DEGRADED))
1198                 seq_puts(seq, ",degraded");
1199         if (btrfs_test_opt(root, NODATASUM))
1200                 seq_puts(seq, ",nodatasum");
1201         if (btrfs_test_opt(root, NODATACOW))
1202                 seq_puts(seq, ",nodatacow");
1203         if (btrfs_test_opt(root, NOBARRIER))
1204                 seq_puts(seq, ",nobarrier");
1205         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1206                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1207         if (info->alloc_start != 0)
1208                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1209         if (info->thread_pool_size !=  min_t(unsigned long,
1210                                              num_online_cpus() + 2, 8))
1211                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1212         if (btrfs_test_opt(root, COMPRESS)) {
1213                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1214                         compress_type = "zlib";
1215                 else
1216                         compress_type = "lzo";
1217                 if (btrfs_test_opt(root, FORCE_COMPRESS))
1218                         seq_printf(seq, ",compress-force=%s", compress_type);
1219                 else
1220                         seq_printf(seq, ",compress=%s", compress_type);
1221         }
1222         if (btrfs_test_opt(root, NOSSD))
1223                 seq_puts(seq, ",nossd");
1224         if (btrfs_test_opt(root, SSD_SPREAD))
1225                 seq_puts(seq, ",ssd_spread");
1226         else if (btrfs_test_opt(root, SSD))
1227                 seq_puts(seq, ",ssd");
1228         if (btrfs_test_opt(root, NOTREELOG))
1229                 seq_puts(seq, ",notreelog");
1230         if (btrfs_test_opt(root, NOLOGREPLAY))
1231                 seq_puts(seq, ",nologreplay");
1232         if (btrfs_test_opt(root, FLUSHONCOMMIT))
1233                 seq_puts(seq, ",flushoncommit");
1234         if (btrfs_test_opt(root, DISCARD))
1235                 seq_puts(seq, ",discard");
1236         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1237                 seq_puts(seq, ",noacl");
1238         if (btrfs_test_opt(root, SPACE_CACHE))
1239                 seq_puts(seq, ",space_cache");
1240         else if (btrfs_test_opt(root, FREE_SPACE_TREE))
1241                 seq_puts(seq, ",space_cache=v2");
1242         else
1243                 seq_puts(seq, ",nospace_cache");
1244         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1245                 seq_puts(seq, ",rescan_uuid_tree");
1246         if (btrfs_test_opt(root, CLEAR_CACHE))
1247                 seq_puts(seq, ",clear_cache");
1248         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1249                 seq_puts(seq, ",user_subvol_rm_allowed");
1250         if (btrfs_test_opt(root, ENOSPC_DEBUG))
1251                 seq_puts(seq, ",enospc_debug");
1252         if (btrfs_test_opt(root, AUTO_DEFRAG))
1253                 seq_puts(seq, ",autodefrag");
1254         if (btrfs_test_opt(root, INODE_MAP_CACHE))
1255                 seq_puts(seq, ",inode_cache");
1256         if (btrfs_test_opt(root, SKIP_BALANCE))
1257                 seq_puts(seq, ",skip_balance");
1258 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1259         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1260                 seq_puts(seq, ",check_int_data");
1261         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1262                 seq_puts(seq, ",check_int");
1263         if (info->check_integrity_print_mask)
1264                 seq_printf(seq, ",check_int_print_mask=%d",
1265                                 info->check_integrity_print_mask);
1266 #endif
1267         if (info->metadata_ratio)
1268                 seq_printf(seq, ",metadata_ratio=%d",
1269                                 info->metadata_ratio);
1270         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1271                 seq_puts(seq, ",fatal_errors=panic");
1272         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1273                 seq_printf(seq, ",commit=%d", info->commit_interval);
1274 #ifdef CONFIG_BTRFS_DEBUG
1275         if (btrfs_test_opt(root, FRAGMENT_DATA))
1276                 seq_puts(seq, ",fragment=data");
1277         if (btrfs_test_opt(root, FRAGMENT_METADATA))
1278                 seq_puts(seq, ",fragment=metadata");
1279 #endif
1280         seq_printf(seq, ",subvolid=%llu",
1281                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1282         seq_puts(seq, ",subvol=");
1283         seq_dentry(seq, dentry, " \t\n\\");
1284         return 0;
1285 }
1286
1287 static int btrfs_test_super(struct super_block *s, void *data)
1288 {
1289         struct btrfs_fs_info *p = data;
1290         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1291
1292         return fs_info->fs_devices == p->fs_devices;
1293 }
1294
1295 static int btrfs_set_super(struct super_block *s, void *data)
1296 {
1297         int err = set_anon_super(s, data);
1298         if (!err)
1299                 s->s_fs_info = data;
1300         return err;
1301 }
1302
1303 /*
1304  * subvolumes are identified by ino 256
1305  */
1306 static inline int is_subvolume_inode(struct inode *inode)
1307 {
1308         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1309                 return 1;
1310         return 0;
1311 }
1312
1313 /*
1314  * This will add subvolid=0 to the argument string while removing any subvol=
1315  * and subvolid= arguments to make sure we get the top-level root for path
1316  * walking to the subvol we want.
1317  */
1318 static char *setup_root_args(char *args)
1319 {
1320         char *buf, *dst, *sep;
1321
1322         if (!args)
1323                 return kstrdup("subvolid=0", GFP_NOFS);
1324
1325         /* The worst case is that we add ",subvolid=0" to the end. */
1326         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1327         if (!buf)
1328                 return NULL;
1329
1330         while (1) {
1331                 sep = strchrnul(args, ',');
1332                 if (!strstarts(args, "subvol=") &&
1333                     !strstarts(args, "subvolid=")) {
1334                         memcpy(dst, args, sep - args);
1335                         dst += sep - args;
1336                         *dst++ = ',';
1337                 }
1338                 if (*sep)
1339                         args = sep + 1;
1340                 else
1341                         break;
1342         }
1343         strcpy(dst, "subvolid=0");
1344
1345         return buf;
1346 }
1347
1348 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1349                                    int flags, const char *device_name,
1350                                    char *data)
1351 {
1352         struct dentry *root;
1353         struct vfsmount *mnt = NULL;
1354         char *newargs;
1355         int ret;
1356
1357         newargs = setup_root_args(data);
1358         if (!newargs) {
1359                 root = ERR_PTR(-ENOMEM);
1360                 goto out;
1361         }
1362
1363         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1364         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1365                 if (flags & MS_RDONLY) {
1366                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1367                                              device_name, newargs);
1368                 } else {
1369                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1370                                              device_name, newargs);
1371                         if (IS_ERR(mnt)) {
1372                                 root = ERR_CAST(mnt);
1373                                 mnt = NULL;
1374                                 goto out;
1375                         }
1376
1377                         down_write(&mnt->mnt_sb->s_umount);
1378                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1379                         up_write(&mnt->mnt_sb->s_umount);
1380                         if (ret < 0) {
1381                                 root = ERR_PTR(ret);
1382                                 goto out;
1383                         }
1384                 }
1385         }
1386         if (IS_ERR(mnt)) {
1387                 root = ERR_CAST(mnt);
1388                 mnt = NULL;
1389                 goto out;
1390         }
1391
1392         if (!subvol_name) {
1393                 if (!subvol_objectid) {
1394                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1395                                                           &subvol_objectid);
1396                         if (ret) {
1397                                 root = ERR_PTR(ret);
1398                                 goto out;
1399                         }
1400                 }
1401                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1402                                                             subvol_objectid);
1403                 if (IS_ERR(subvol_name)) {
1404                         root = ERR_CAST(subvol_name);
1405                         subvol_name = NULL;
1406                         goto out;
1407                 }
1408
1409         }
1410
1411         root = mount_subtree(mnt, subvol_name);
1412         /* mount_subtree() drops our reference on the vfsmount. */
1413         mnt = NULL;
1414
1415         if (!IS_ERR(root)) {
1416                 struct super_block *s = root->d_sb;
1417                 struct inode *root_inode = d_inode(root);
1418                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1419
1420                 ret = 0;
1421                 if (!is_subvolume_inode(root_inode)) {
1422                         pr_err("BTRFS: '%s' is not a valid subvolume\n",
1423                                subvol_name);
1424                         ret = -EINVAL;
1425                 }
1426                 if (subvol_objectid && root_objectid != subvol_objectid) {
1427                         /*
1428                          * This will also catch a race condition where a
1429                          * subvolume which was passed by ID is renamed and
1430                          * another subvolume is renamed over the old location.
1431                          */
1432                         pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1433                                subvol_name, subvol_objectid);
1434                         ret = -EINVAL;
1435                 }
1436                 if (ret) {
1437                         dput(root);
1438                         root = ERR_PTR(ret);
1439                         deactivate_locked_super(s);
1440                 }
1441         }
1442
1443 out:
1444         mntput(mnt);
1445         kfree(newargs);
1446         kfree(subvol_name);
1447         return root;
1448 }
1449
1450 static int parse_security_options(char *orig_opts,
1451                                   struct security_mnt_opts *sec_opts)
1452 {
1453         char *secdata = NULL;
1454         int ret = 0;
1455
1456         secdata = alloc_secdata();
1457         if (!secdata)
1458                 return -ENOMEM;
1459         ret = security_sb_copy_data(orig_opts, secdata);
1460         if (ret) {
1461                 free_secdata(secdata);
1462                 return ret;
1463         }
1464         ret = security_sb_parse_opts_str(secdata, sec_opts);
1465         free_secdata(secdata);
1466         return ret;
1467 }
1468
1469 static int setup_security_options(struct btrfs_fs_info *fs_info,
1470                                   struct super_block *sb,
1471                                   struct security_mnt_opts *sec_opts)
1472 {
1473         int ret = 0;
1474
1475         /*
1476          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1477          * is valid.
1478          */
1479         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1480         if (ret)
1481                 return ret;
1482
1483 #ifdef CONFIG_SECURITY
1484         if (!fs_info->security_opts.num_mnt_opts) {
1485                 /* first time security setup, copy sec_opts to fs_info */
1486                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1487         } else {
1488                 /*
1489                  * Since SELinux(the only one supports security_mnt_opts) does
1490                  * NOT support changing context during remount/mount same sb,
1491                  * This must be the same or part of the same security options,
1492                  * just free it.
1493                  */
1494                 security_free_mnt_opts(sec_opts);
1495         }
1496 #endif
1497         return ret;
1498 }
1499
1500 /*
1501  * Find a superblock for the given device / mount point.
1502  *
1503  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1504  *        for multiple device setup.  Make sure to keep it in sync.
1505  */
1506 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1507                 const char *device_name, void *data)
1508 {
1509         struct block_device *bdev = NULL;
1510         struct super_block *s;
1511         struct btrfs_fs_devices *fs_devices = NULL;
1512         struct btrfs_fs_info *fs_info = NULL;
1513         struct security_mnt_opts new_sec_opts;
1514         fmode_t mode = FMODE_READ;
1515         char *subvol_name = NULL;
1516         u64 subvol_objectid = 0;
1517         int error = 0;
1518
1519         if (!(flags & MS_RDONLY))
1520                 mode |= FMODE_WRITE;
1521
1522         error = btrfs_parse_early_options(data, mode, fs_type,
1523                                           &subvol_name, &subvol_objectid,
1524                                           &fs_devices);
1525         if (error) {
1526                 kfree(subvol_name);
1527                 return ERR_PTR(error);
1528         }
1529
1530         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1531                 /* mount_subvol() will free subvol_name. */
1532                 return mount_subvol(subvol_name, subvol_objectid, flags,
1533                                     device_name, data);
1534         }
1535
1536         security_init_mnt_opts(&new_sec_opts);
1537         if (data) {
1538                 error = parse_security_options(data, &new_sec_opts);
1539                 if (error)
1540                         return ERR_PTR(error);
1541         }
1542
1543         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1544         if (error)
1545                 goto error_sec_opts;
1546
1547         /*
1548          * Setup a dummy root and fs_info for test/set super.  This is because
1549          * we don't actually fill this stuff out until open_ctree, but we need
1550          * it for searching for existing supers, so this lets us do that and
1551          * then open_ctree will properly initialize everything later.
1552          */
1553         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1554         if (!fs_info) {
1555                 error = -ENOMEM;
1556                 goto error_sec_opts;
1557         }
1558
1559         fs_info->fs_devices = fs_devices;
1560
1561         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1562         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1563         security_init_mnt_opts(&fs_info->security_opts);
1564         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1565                 error = -ENOMEM;
1566                 goto error_fs_info;
1567         }
1568
1569         error = btrfs_open_devices(fs_devices, mode, fs_type);
1570         if (error)
1571                 goto error_fs_info;
1572
1573         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1574                 error = -EACCES;
1575                 goto error_close_devices;
1576         }
1577
1578         bdev = fs_devices->latest_bdev;
1579         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1580                  fs_info);
1581         if (IS_ERR(s)) {
1582                 error = PTR_ERR(s);
1583                 goto error_close_devices;
1584         }
1585
1586         if (s->s_root) {
1587                 btrfs_close_devices(fs_devices);
1588                 free_fs_info(fs_info);
1589                 if ((flags ^ s->s_flags) & MS_RDONLY)
1590                         error = -EBUSY;
1591         } else {
1592                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1593                 btrfs_sb(s)->bdev_holder = fs_type;
1594                 error = btrfs_fill_super(s, fs_devices, data,
1595                                          flags & MS_SILENT ? 1 : 0);
1596         }
1597         if (error) {
1598                 deactivate_locked_super(s);
1599                 goto error_sec_opts;
1600         }
1601
1602         fs_info = btrfs_sb(s);
1603         error = setup_security_options(fs_info, s, &new_sec_opts);
1604         if (error) {
1605                 deactivate_locked_super(s);
1606                 goto error_sec_opts;
1607         }
1608
1609         return dget(s->s_root);
1610
1611 error_close_devices:
1612         btrfs_close_devices(fs_devices);
1613 error_fs_info:
1614         free_fs_info(fs_info);
1615 error_sec_opts:
1616         security_free_mnt_opts(&new_sec_opts);
1617         return ERR_PTR(error);
1618 }
1619
1620 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1621                                      int new_pool_size, int old_pool_size)
1622 {
1623         if (new_pool_size == old_pool_size)
1624                 return;
1625
1626         fs_info->thread_pool_size = new_pool_size;
1627
1628         btrfs_info(fs_info, "resize thread pool %d -> %d",
1629                old_pool_size, new_pool_size);
1630
1631         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1632         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1633         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1634         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1635         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1636         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1637         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1638                                 new_pool_size);
1639         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1640         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1641         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1642         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1643         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1644                                 new_pool_size);
1645 }
1646
1647 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1648 {
1649         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1650 }
1651
1652 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1653                                        unsigned long old_opts, int flags)
1654 {
1655         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1656             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1657              (flags & MS_RDONLY))) {
1658                 /* wait for any defraggers to finish */
1659                 wait_event(fs_info->transaction_wait,
1660                            (atomic_read(&fs_info->defrag_running) == 0));
1661                 if (flags & MS_RDONLY)
1662                         sync_filesystem(fs_info->sb);
1663         }
1664 }
1665
1666 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1667                                          unsigned long old_opts)
1668 {
1669         /*
1670          * We need cleanup all defragable inodes if the autodefragment is
1671          * close or the fs is R/O.
1672          */
1673         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1674             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1675              (fs_info->sb->s_flags & MS_RDONLY))) {
1676                 btrfs_cleanup_defrag_inodes(fs_info);
1677         }
1678
1679         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1680 }
1681
1682 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1683 {
1684         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1685         struct btrfs_root *root = fs_info->tree_root;
1686         unsigned old_flags = sb->s_flags;
1687         unsigned long old_opts = fs_info->mount_opt;
1688         unsigned long old_compress_type = fs_info->compress_type;
1689         u64 old_max_inline = fs_info->max_inline;
1690         u64 old_alloc_start = fs_info->alloc_start;
1691         int old_thread_pool_size = fs_info->thread_pool_size;
1692         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1693         int ret;
1694
1695         sync_filesystem(sb);
1696         btrfs_remount_prepare(fs_info);
1697
1698         if (data) {
1699                 struct security_mnt_opts new_sec_opts;
1700
1701                 security_init_mnt_opts(&new_sec_opts);
1702                 ret = parse_security_options(data, &new_sec_opts);
1703                 if (ret)
1704                         goto restore;
1705                 ret = setup_security_options(fs_info, sb,
1706                                              &new_sec_opts);
1707                 if (ret) {
1708                         security_free_mnt_opts(&new_sec_opts);
1709                         goto restore;
1710                 }
1711         }
1712
1713         ret = btrfs_parse_options(root, data, *flags);
1714         if (ret) {
1715                 ret = -EINVAL;
1716                 goto restore;
1717         }
1718
1719         btrfs_remount_begin(fs_info, old_opts, *flags);
1720         btrfs_resize_thread_pool(fs_info,
1721                 fs_info->thread_pool_size, old_thread_pool_size);
1722
1723         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1724                 goto out;
1725
1726         if (*flags & MS_RDONLY) {
1727                 /*
1728                  * this also happens on 'umount -rf' or on shutdown, when
1729                  * the filesystem is busy.
1730                  */
1731                 cancel_work_sync(&fs_info->async_reclaim_work);
1732
1733                 /* wait for the uuid_scan task to finish */
1734                 down(&fs_info->uuid_tree_rescan_sem);
1735                 /* avoid complains from lockdep et al. */
1736                 up(&fs_info->uuid_tree_rescan_sem);
1737
1738                 sb->s_flags |= MS_RDONLY;
1739
1740                 /*
1741                  * Setting MS_RDONLY will put the cleaner thread to
1742                  * sleep at the next loop if it's already active.
1743                  * If it's already asleep, we'll leave unused block
1744                  * groups on disk until we're mounted read-write again
1745                  * unless we clean them up here.
1746                  */
1747                 btrfs_delete_unused_bgs(fs_info);
1748
1749                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1750                 btrfs_scrub_cancel(fs_info);
1751                 btrfs_pause_balance(fs_info);
1752
1753                 ret = btrfs_commit_super(root);
1754                 if (ret)
1755                         goto restore;
1756         } else {
1757                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1758                         btrfs_err(fs_info,
1759                                 "Remounting read-write after error is not allowed");
1760                         ret = -EINVAL;
1761                         goto restore;
1762                 }
1763                 if (fs_info->fs_devices->rw_devices == 0) {
1764                         ret = -EACCES;
1765                         goto restore;
1766                 }
1767
1768                 if (fs_info->fs_devices->missing_devices >
1769                      fs_info->num_tolerated_disk_barrier_failures &&
1770                     !(*flags & MS_RDONLY)) {
1771                         btrfs_warn(fs_info,
1772                                 "too many missing devices, writeable remount is not allowed");
1773                         ret = -EACCES;
1774                         goto restore;
1775                 }
1776
1777                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1778                         ret = -EINVAL;
1779                         goto restore;
1780                 }
1781
1782                 ret = btrfs_cleanup_fs_roots(fs_info);
1783                 if (ret)
1784                         goto restore;
1785
1786                 /* recover relocation */
1787                 mutex_lock(&fs_info->cleaner_mutex);
1788                 ret = btrfs_recover_relocation(root);
1789                 mutex_unlock(&fs_info->cleaner_mutex);
1790                 if (ret)
1791                         goto restore;
1792
1793                 ret = btrfs_resume_balance_async(fs_info);
1794                 if (ret)
1795                         goto restore;
1796
1797                 ret = btrfs_resume_dev_replace_async(fs_info);
1798                 if (ret) {
1799                         btrfs_warn(fs_info, "failed to resume dev_replace");
1800                         goto restore;
1801                 }
1802
1803                 if (!fs_info->uuid_root) {
1804                         btrfs_info(fs_info, "creating UUID tree");
1805                         ret = btrfs_create_uuid_tree(fs_info);
1806                         if (ret) {
1807                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1808                                 goto restore;
1809                         }
1810                 }
1811                 sb->s_flags &= ~MS_RDONLY;
1812         }
1813 out:
1814         wake_up_process(fs_info->transaction_kthread);
1815         btrfs_remount_cleanup(fs_info, old_opts);
1816         return 0;
1817
1818 restore:
1819         /* We've hit an error - don't reset MS_RDONLY */
1820         if (sb->s_flags & MS_RDONLY)
1821                 old_flags |= MS_RDONLY;
1822         sb->s_flags = old_flags;
1823         fs_info->mount_opt = old_opts;
1824         fs_info->compress_type = old_compress_type;
1825         fs_info->max_inline = old_max_inline;
1826         mutex_lock(&fs_info->chunk_mutex);
1827         fs_info->alloc_start = old_alloc_start;
1828         mutex_unlock(&fs_info->chunk_mutex);
1829         btrfs_resize_thread_pool(fs_info,
1830                 old_thread_pool_size, fs_info->thread_pool_size);
1831         fs_info->metadata_ratio = old_metadata_ratio;
1832         btrfs_remount_cleanup(fs_info, old_opts);
1833         return ret;
1834 }
1835
1836 /* Used to sort the devices by max_avail(descending sort) */
1837 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1838                                        const void *dev_info2)
1839 {
1840         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1841             ((struct btrfs_device_info *)dev_info2)->max_avail)
1842                 return -1;
1843         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1844                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1845                 return 1;
1846         else
1847         return 0;
1848 }
1849
1850 /*
1851  * sort the devices by max_avail, in which max free extent size of each device
1852  * is stored.(Descending Sort)
1853  */
1854 static inline void btrfs_descending_sort_devices(
1855                                         struct btrfs_device_info *devices,
1856                                         size_t nr_devices)
1857 {
1858         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1859              btrfs_cmp_device_free_bytes, NULL);
1860 }
1861
1862 /*
1863  * The helper to calc the free space on the devices that can be used to store
1864  * file data.
1865  */
1866 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1867 {
1868         struct btrfs_fs_info *fs_info = root->fs_info;
1869         struct btrfs_device_info *devices_info;
1870         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1871         struct btrfs_device *device;
1872         u64 skip_space;
1873         u64 type;
1874         u64 avail_space;
1875         u64 used_space;
1876         u64 min_stripe_size;
1877         int min_stripes = 1, num_stripes = 1;
1878         int i = 0, nr_devices;
1879         int ret;
1880
1881         /*
1882          * We aren't under the device list lock, so this is racey-ish, but good
1883          * enough for our purposes.
1884          */
1885         nr_devices = fs_info->fs_devices->open_devices;
1886         if (!nr_devices) {
1887                 smp_mb();
1888                 nr_devices = fs_info->fs_devices->open_devices;
1889                 ASSERT(nr_devices);
1890                 if (!nr_devices) {
1891                         *free_bytes = 0;
1892                         return 0;
1893                 }
1894         }
1895
1896         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1897                                GFP_NOFS);
1898         if (!devices_info)
1899                 return -ENOMEM;
1900
1901         /* calc min stripe number for data space alloction */
1902         type = btrfs_get_alloc_profile(root, 1);
1903         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1904                 min_stripes = 2;
1905                 num_stripes = nr_devices;
1906         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1907                 min_stripes = 2;
1908                 num_stripes = 2;
1909         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1910                 min_stripes = 4;
1911                 num_stripes = 4;
1912         }
1913
1914         if (type & BTRFS_BLOCK_GROUP_DUP)
1915                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1916         else
1917                 min_stripe_size = BTRFS_STRIPE_LEN;
1918
1919         if (fs_info->alloc_start)
1920                 mutex_lock(&fs_devices->device_list_mutex);
1921         rcu_read_lock();
1922         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1923                 if (!device->in_fs_metadata || !device->bdev ||
1924                     device->is_tgtdev_for_dev_replace)
1925                         continue;
1926
1927                 if (i >= nr_devices)
1928                         break;
1929
1930                 avail_space = device->total_bytes - device->bytes_used;
1931
1932                 /* align with stripe_len */
1933                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1934                 avail_space *= BTRFS_STRIPE_LEN;
1935
1936                 /*
1937                  * In order to avoid overwritting the superblock on the drive,
1938                  * btrfs starts at an offset of at least 1MB when doing chunk
1939                  * allocation.
1940                  */
1941                 skip_space = SZ_1M;
1942
1943                 /* user can set the offset in fs_info->alloc_start. */
1944                 if (fs_info->alloc_start &&
1945                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1946                     device->total_bytes) {
1947                         rcu_read_unlock();
1948                         skip_space = max(fs_info->alloc_start, skip_space);
1949
1950                         /*
1951                          * btrfs can not use the free space in
1952                          * [0, skip_space - 1], we must subtract it from the
1953                          * total. In order to implement it, we account the used
1954                          * space in this range first.
1955                          */
1956                         ret = btrfs_account_dev_extents_size(device, 0,
1957                                                              skip_space - 1,
1958                                                              &used_space);
1959                         if (ret) {
1960                                 kfree(devices_info);
1961                                 mutex_unlock(&fs_devices->device_list_mutex);
1962                                 return ret;
1963                         }
1964
1965                         rcu_read_lock();
1966
1967                         /* calc the free space in [0, skip_space - 1] */
1968                         skip_space -= used_space;
1969                 }
1970
1971                 /*
1972                  * we can use the free space in [0, skip_space - 1], subtract
1973                  * it from the total.
1974                  */
1975                 if (avail_space && avail_space >= skip_space)
1976                         avail_space -= skip_space;
1977                 else
1978                         avail_space = 0;
1979
1980                 if (avail_space < min_stripe_size)
1981                         continue;
1982
1983                 devices_info[i].dev = device;
1984                 devices_info[i].max_avail = avail_space;
1985
1986                 i++;
1987         }
1988         rcu_read_unlock();
1989         if (fs_info->alloc_start)
1990                 mutex_unlock(&fs_devices->device_list_mutex);
1991
1992         nr_devices = i;
1993
1994         btrfs_descending_sort_devices(devices_info, nr_devices);
1995
1996         i = nr_devices - 1;
1997         avail_space = 0;
1998         while (nr_devices >= min_stripes) {
1999                 if (num_stripes > nr_devices)
2000                         num_stripes = nr_devices;
2001
2002                 if (devices_info[i].max_avail >= min_stripe_size) {
2003                         int j;
2004                         u64 alloc_size;
2005
2006                         avail_space += devices_info[i].max_avail * num_stripes;
2007                         alloc_size = devices_info[i].max_avail;
2008                         for (j = i + 1 - num_stripes; j <= i; j++)
2009                                 devices_info[j].max_avail -= alloc_size;
2010                 }
2011                 i--;
2012                 nr_devices--;
2013         }
2014
2015         kfree(devices_info);
2016         *free_bytes = avail_space;
2017         return 0;
2018 }
2019
2020 /*
2021  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2022  *
2023  * If there's a redundant raid level at DATA block groups, use the respective
2024  * multiplier to scale the sizes.
2025  *
2026  * Unused device space usage is based on simulating the chunk allocator
2027  * algorithm that respects the device sizes, order of allocations and the
2028  * 'alloc_start' value, this is a close approximation of the actual use but
2029  * there are other factors that may change the result (like a new metadata
2030  * chunk).
2031  *
2032  * If metadata is exhausted, f_bavail will be 0.
2033  *
2034  * FIXME: not accurate for mixed block groups, total and free/used are ok,
2035  * available appears slightly larger.
2036  */
2037 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2038 {
2039         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2040         struct btrfs_super_block *disk_super = fs_info->super_copy;
2041         struct list_head *head = &fs_info->space_info;
2042         struct btrfs_space_info *found;
2043         u64 total_used = 0;
2044         u64 total_free_data = 0;
2045         u64 total_free_meta = 0;
2046         int bits = dentry->d_sb->s_blocksize_bits;
2047         __be32 *fsid = (__be32 *)fs_info->fsid;
2048         unsigned factor = 1;
2049         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2050         int ret;
2051         u64 thresh = 0;
2052
2053         /*
2054          * holding chunk_muext to avoid allocating new chunks, holding
2055          * device_list_mutex to avoid the device being removed
2056          */
2057         rcu_read_lock();
2058         list_for_each_entry_rcu(found, head, list) {
2059                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2060                         int i;
2061
2062                         total_free_data += found->disk_total - found->disk_used;
2063                         total_free_data -=
2064                                 btrfs_account_ro_block_groups_free_space(found);
2065
2066                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2067                                 if (!list_empty(&found->block_groups[i])) {
2068                                         switch (i) {
2069                                         case BTRFS_RAID_DUP:
2070                                         case BTRFS_RAID_RAID1:
2071                                         case BTRFS_RAID_RAID10:
2072                                                 factor = 2;
2073                                         }
2074                                 }
2075                         }
2076                 }
2077                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2078                         total_free_meta += found->disk_total - found->disk_used;
2079
2080                 total_used += found->disk_used;
2081         }
2082
2083         rcu_read_unlock();
2084
2085         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2086         buf->f_blocks >>= bits;
2087         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2088
2089         /* Account global block reserve as used, it's in logical size already */
2090         spin_lock(&block_rsv->lock);
2091         buf->f_bfree -= block_rsv->size >> bits;
2092         spin_unlock(&block_rsv->lock);
2093
2094         buf->f_bavail = div_u64(total_free_data, factor);
2095         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2096         if (ret)
2097                 return ret;
2098         buf->f_bavail += div_u64(total_free_data, factor);
2099         buf->f_bavail = buf->f_bavail >> bits;
2100
2101         /*
2102          * We calculate the remaining metadata space minus global reserve. If
2103          * this is (supposedly) smaller than zero, there's no space. But this
2104          * does not hold in practice, the exhausted state happens where's still
2105          * some positive delta. So we apply some guesswork and compare the
2106          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2107          *
2108          * We probably cannot calculate the exact threshold value because this
2109          * depends on the internal reservations requested by various
2110          * operations, so some operations that consume a few metadata will
2111          * succeed even if the Avail is zero. But this is better than the other
2112          * way around.
2113          */
2114         thresh = 4 * 1024 * 1024;
2115
2116         if (total_free_meta - thresh < block_rsv->size)
2117                 buf->f_bavail = 0;
2118
2119         buf->f_type = BTRFS_SUPER_MAGIC;
2120         buf->f_bsize = dentry->d_sb->s_blocksize;
2121         buf->f_namelen = BTRFS_NAME_LEN;
2122
2123         /* We treat it as constant endianness (it doesn't matter _which_)
2124            because we want the fsid to come out the same whether mounted
2125            on a big-endian or little-endian host */
2126         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2127         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2128         /* Mask in the root object ID too, to disambiguate subvols */
2129         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2130         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2131
2132         return 0;
2133 }
2134
2135 static void btrfs_kill_super(struct super_block *sb)
2136 {
2137         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2138         kill_anon_super(sb);
2139         free_fs_info(fs_info);
2140 }
2141
2142 static struct file_system_type btrfs_fs_type = {
2143         .owner          = THIS_MODULE,
2144         .name           = "btrfs",
2145         .mount          = btrfs_mount,
2146         .kill_sb        = btrfs_kill_super,
2147         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2148 };
2149 MODULE_ALIAS_FS("btrfs");
2150
2151 static int btrfs_control_open(struct inode *inode, struct file *file)
2152 {
2153         /*
2154          * The control file's private_data is used to hold the
2155          * transaction when it is started and is used to keep
2156          * track of whether a transaction is already in progress.
2157          */
2158         file->private_data = NULL;
2159         return 0;
2160 }
2161
2162 /*
2163  * used by btrfsctl to scan devices when no FS is mounted
2164  */
2165 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2166                                 unsigned long arg)
2167 {
2168         struct btrfs_ioctl_vol_args *vol;
2169         struct btrfs_fs_devices *fs_devices;
2170         int ret = -ENOTTY;
2171
2172         if (!capable(CAP_SYS_ADMIN))
2173                 return -EPERM;
2174
2175         vol = memdup_user((void __user *)arg, sizeof(*vol));
2176         if (IS_ERR(vol))
2177                 return PTR_ERR(vol);
2178
2179         switch (cmd) {
2180         case BTRFS_IOC_SCAN_DEV:
2181                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2182                                             &btrfs_fs_type, &fs_devices);
2183                 break;
2184         case BTRFS_IOC_DEVICES_READY:
2185                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2186                                             &btrfs_fs_type, &fs_devices);
2187                 if (ret)
2188                         break;
2189                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2190                 break;
2191         }
2192
2193         kfree(vol);
2194         return ret;
2195 }
2196
2197 static int btrfs_freeze(struct super_block *sb)
2198 {
2199         struct btrfs_trans_handle *trans;
2200         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2201
2202         trans = btrfs_attach_transaction_barrier(root);
2203         if (IS_ERR(trans)) {
2204                 /* no transaction, don't bother */
2205                 if (PTR_ERR(trans) == -ENOENT)
2206                         return 0;
2207                 return PTR_ERR(trans);
2208         }
2209         return btrfs_commit_transaction(trans, root);
2210 }
2211
2212 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2213 {
2214         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2215         struct btrfs_fs_devices *cur_devices;
2216         struct btrfs_device *dev, *first_dev = NULL;
2217         struct list_head *head;
2218         struct rcu_string *name;
2219
2220         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2221         cur_devices = fs_info->fs_devices;
2222         while (cur_devices) {
2223                 head = &cur_devices->devices;
2224                 list_for_each_entry(dev, head, dev_list) {
2225                         if (dev->missing)
2226                                 continue;
2227                         if (!dev->name)
2228                                 continue;
2229                         if (!first_dev || dev->devid < first_dev->devid)
2230                                 first_dev = dev;
2231                 }
2232                 cur_devices = cur_devices->seed;
2233         }
2234
2235         if (first_dev) {
2236                 rcu_read_lock();
2237                 name = rcu_dereference(first_dev->name);
2238                 seq_escape(m, name->str, " \t\n\\");
2239                 rcu_read_unlock();
2240         } else {
2241                 WARN_ON(1);
2242         }
2243         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2244         return 0;
2245 }
2246
2247 static const struct super_operations btrfs_super_ops = {
2248         .drop_inode     = btrfs_drop_inode,
2249         .evict_inode    = btrfs_evict_inode,
2250         .put_super      = btrfs_put_super,
2251         .sync_fs        = btrfs_sync_fs,
2252         .show_options   = btrfs_show_options,
2253         .show_devname   = btrfs_show_devname,
2254         .write_inode    = btrfs_write_inode,
2255         .alloc_inode    = btrfs_alloc_inode,
2256         .destroy_inode  = btrfs_destroy_inode,
2257         .statfs         = btrfs_statfs,
2258         .remount_fs     = btrfs_remount,
2259         .freeze_fs      = btrfs_freeze,
2260 };
2261
2262 static const struct file_operations btrfs_ctl_fops = {
2263         .open = btrfs_control_open,
2264         .unlocked_ioctl  = btrfs_control_ioctl,
2265         .compat_ioctl = btrfs_control_ioctl,
2266         .owner   = THIS_MODULE,
2267         .llseek = noop_llseek,
2268 };
2269
2270 static struct miscdevice btrfs_misc = {
2271         .minor          = BTRFS_MINOR,
2272         .name           = "btrfs-control",
2273         .fops           = &btrfs_ctl_fops
2274 };
2275
2276 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2277 MODULE_ALIAS("devname:btrfs-control");
2278
2279 static int btrfs_interface_init(void)
2280 {
2281         return misc_register(&btrfs_misc);
2282 }
2283
2284 static void btrfs_interface_exit(void)
2285 {
2286         misc_deregister(&btrfs_misc);
2287 }
2288
2289 static void btrfs_print_info(void)
2290 {
2291         printk(KERN_INFO "Btrfs loaded"
2292 #ifdef CONFIG_BTRFS_DEBUG
2293                         ", debug=on"
2294 #endif
2295 #ifdef CONFIG_BTRFS_ASSERT
2296                         ", assert=on"
2297 #endif
2298 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2299                         ", integrity-checker=on"
2300 #endif
2301                         "\n");
2302 }
2303
2304 static int btrfs_run_sanity_tests(void)
2305 {
2306         int ret;
2307
2308         ret = btrfs_init_test_fs();
2309         if (ret)
2310                 return ret;
2311
2312         ret = btrfs_test_free_space_cache();
2313         if (ret)
2314                 goto out;
2315         ret = btrfs_test_extent_buffer_operations();
2316         if (ret)
2317                 goto out;
2318         ret = btrfs_test_extent_io();
2319         if (ret)
2320                 goto out;
2321         ret = btrfs_test_inodes();
2322         if (ret)
2323                 goto out;
2324         ret = btrfs_test_qgroups();
2325         if (ret)
2326                 goto out;
2327         ret = btrfs_test_free_space_tree();
2328 out:
2329         btrfs_destroy_test_fs();
2330         return ret;
2331 }
2332
2333 static int __init init_btrfs_fs(void)
2334 {
2335         int err;
2336
2337         err = btrfs_hash_init();
2338         if (err)
2339                 return err;
2340
2341         btrfs_props_init();
2342
2343         err = btrfs_init_sysfs();
2344         if (err)
2345                 goto free_hash;
2346
2347         btrfs_init_compress();
2348
2349         err = btrfs_init_cachep();
2350         if (err)
2351                 goto free_compress;
2352
2353         err = extent_io_init();
2354         if (err)
2355                 goto free_cachep;
2356
2357         err = extent_map_init();
2358         if (err)
2359                 goto free_extent_io;
2360
2361         err = ordered_data_init();
2362         if (err)
2363                 goto free_extent_map;
2364
2365         err = btrfs_delayed_inode_init();
2366         if (err)
2367                 goto free_ordered_data;
2368
2369         err = btrfs_auto_defrag_init();
2370         if (err)
2371                 goto free_delayed_inode;
2372
2373         err = btrfs_delayed_ref_init();
2374         if (err)
2375                 goto free_auto_defrag;
2376
2377         err = btrfs_prelim_ref_init();
2378         if (err)
2379                 goto free_delayed_ref;
2380
2381         err = btrfs_end_io_wq_init();
2382         if (err)
2383                 goto free_prelim_ref;
2384
2385         err = btrfs_interface_init();
2386         if (err)
2387                 goto free_end_io_wq;
2388
2389         btrfs_init_lockdep();
2390
2391         btrfs_print_info();
2392
2393         err = btrfs_run_sanity_tests();
2394         if (err)
2395                 goto unregister_ioctl;
2396
2397         err = register_filesystem(&btrfs_fs_type);
2398         if (err)
2399                 goto unregister_ioctl;
2400
2401         return 0;
2402
2403 unregister_ioctl:
2404         btrfs_interface_exit();
2405 free_end_io_wq:
2406         btrfs_end_io_wq_exit();
2407 free_prelim_ref:
2408         btrfs_prelim_ref_exit();
2409 free_delayed_ref:
2410         btrfs_delayed_ref_exit();
2411 free_auto_defrag:
2412         btrfs_auto_defrag_exit();
2413 free_delayed_inode:
2414         btrfs_delayed_inode_exit();
2415 free_ordered_data:
2416         ordered_data_exit();
2417 free_extent_map:
2418         extent_map_exit();
2419 free_extent_io:
2420         extent_io_exit();
2421 free_cachep:
2422         btrfs_destroy_cachep();
2423 free_compress:
2424         btrfs_exit_compress();
2425         btrfs_exit_sysfs();
2426 free_hash:
2427         btrfs_hash_exit();
2428         return err;
2429 }
2430
2431 static void __exit exit_btrfs_fs(void)
2432 {
2433         btrfs_destroy_cachep();
2434         btrfs_delayed_ref_exit();
2435         btrfs_auto_defrag_exit();
2436         btrfs_delayed_inode_exit();
2437         btrfs_prelim_ref_exit();
2438         ordered_data_exit();
2439         extent_map_exit();
2440         extent_io_exit();
2441         btrfs_interface_exit();
2442         btrfs_end_io_wq_exit();
2443         unregister_filesystem(&btrfs_fs_type);
2444         btrfs_exit_sysfs();
2445         btrfs_cleanup_fs_uuids();
2446         btrfs_exit_compress();
2447         btrfs_hash_exit();
2448 }
2449
2450 late_initcall(init_btrfs_fs);
2451 module_exit(exit_btrfs_fs)
2452
2453 MODULE_LICENSE("GPL");