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