fs/btrfs/super.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/time.h>
  26 #include <linux/init.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mount.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/parser.h>
  37 #include <linux/ctype.h>
  38 #include <linux/namei.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/magic.h>
  41 #include <linux/slab.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/ratelimit.h>
  44 #include <linux/btrfs.h>
  45 #include "delayed-inode.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "hash.h"
  52 #include "props.h"
  53 #include "xattr.h"
  54 #include "volumes.h"
  55 #include "export.h"
  56 #include "compression.h"
  57 #include "rcu-string.h"
  58 #include "dev-replace.h"
  59 #include "free-space-cache.h"
  60 #include "backref.h"
  61 #include "tests/btrfs-tests.h"
  62
  63 #include "qgroup.h"
  64 #define CREATE_TRACE_POINTS
  65 #include <trace/events/btrfs.h>
  66
  67 static const struct super_operations btrfs_super_ops;
  68 static struct file_system_type btrfs_fs_type;
  69
  70 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  71
  72 const char *btrfs_decode_error(int errno)
  73 {
  74         char *errstr = "unknown";
  75
  76         switch (errno) {
  77         case -EIO:
  78                 errstr = "IO failure";
  79                 break;
  80         case -ENOMEM:
  81                 errstr = "Out of memory";
  82                 break;
  83         case -EROFS:
  84                 errstr = "Readonly filesystem";
  85                 break;
  86         case -EEXIST:
  87                 errstr = "Object already exists";
  88                 break;
  89         case -ENOSPC:
  90                 errstr = "No space left";
  91                 break;
  92         case -ENOENT:
  93                 errstr = "No such entry";
  94                 break;
  95         }
  96
  97         return errstr;
  98 }
  99
 100 static void save_error_info(struct btrfs_fs_info *fs_info)
 101 {
 102         /*
 103          * today we only save the error info into ram.  Long term we'll
 104          * also send it down to the disk
 105          */
 106         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 107 }
 108
 109 /* btrfs handle error by forcing the filesystem readonly */
 110 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 111 {
 112         struct super_block *sb = fs_info->sb;
 113
 114         if (sb->s_flags & MS_RDONLY)
 115                 return;
 116
 117         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 118                 sb->s_flags |= MS_RDONLY;
 119                 btrfs_info(fs_info, "forced readonly");
 120                 /*
 121                  * Note that a running device replace operation is not
 122                  * canceled here although there is no way to update
 123                  * the progress. It would add the risk of a deadlock,
 124                  * therefore the canceling is ommited. The only penalty
 125                  * is that some I/O remains active until the procedure
 126                  * completes. The next time when the filesystem is
 127                  * mounted writeable again, the device replace
 128                  * operation continues.
 129                  */
 130         }
 131 }
 132
 133 /*
 134  * __btrfs_std_error decodes expected errors from the caller and
 135  * invokes the approciate error response.
 136  */
 137 __cold
 138 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 139                        unsigned int line, int errno, const char *fmt, ...)
 140 {
 141         struct super_block *sb = fs_info->sb;
 142 #ifdef CONFIG_PRINTK
 143         const char *errstr;
 144 #endif
 145
 146         /*
 147          * Special case: if the error is EROFS, and we're already
 148          * under MS_RDONLY, then it is safe here.
 149          */
 150         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 151                 return;
 152
 153 #ifdef CONFIG_PRINTK
 154         errstr = btrfs_decode_error(errno);
 155         if (fmt) {
 156                 struct va_format vaf;
 157                 va_list args;
 158
 159                 va_start(args, fmt);
 160                 vaf.fmt = fmt;
 161                 vaf.va = &args;
 162
 163                 printk(KERN_CRIT
 164                         "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 165                         sb->s_id, function, line, errno, errstr, &vaf);
 166                 va_end(args);
 167         } else {
 168                 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 169                         sb->s_id, function, line, errno, errstr);
 170         }
 171 #endif
 172
 173         /* Don't go through full error handling during mount */
 174         save_error_info(fs_info);
 175         if (sb->s_flags & MS_BORN)
 176                 btrfs_handle_error(fs_info);
 177 }
 178
 179 #ifdef CONFIG_PRINTK
 180 static const char * const logtypes[] = {
 181         "emergency",
 182         "alert",
 183         "critical",
 184         "error",
 185         "warning",
 186         "notice",
 187         "info",
 188         "debug",
 189 };
 190
 191 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 192 {
 193         struct super_block *sb = fs_info->sb;
 194         char lvl[4];
 195         struct va_format vaf;
 196         va_list args;
 197         const char *type = logtypes[4];
 198         int kern_level;
 199
 200         va_start(args, fmt);
 201
 202         kern_level = printk_get_level(fmt);
 203         if (kern_level) {
 204                 size_t size = printk_skip_level(fmt) - fmt;
 205                 memcpy(lvl, fmt,  size);
 206                 lvl[size] = '\0';
 207                 fmt += size;
 208                 type = logtypes[kern_level - '0'];
 209         } else
 210                 *lvl = '\0';
 211
 212         vaf.fmt = fmt;
 213         vaf.va = &args;
 214
 215         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 216
 217         va_end(args);
 218 }
 219 #endif
 220
 221 /*
 222  * We only mark the transaction aborted and then set the file system read-only.
 223  * This will prevent new transactions from starting or trying to join this
 224  * one.
 225  *
 226  * This means that error recovery at the call site is limited to freeing
 227  * any local memory allocations and passing the error code up without
 228  * further cleanup. The transaction should complete as it normally would
 229  * in the call path but will return -EIO.
 230  *
 231  * We'll complete the cleanup in btrfs_end_transaction and
 232  * btrfs_commit_transaction.
 233  */
 234 __cold
 235 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 236                                struct btrfs_root *root, const char *function,
 237                                unsigned int line, int errno)
 238 {
 239         trans->aborted = errno;
 240         /* Nothing used. The other threads that have joined this
 241          * transaction may be able to continue. */
 242         if (!trans->blocks_used && list_empty(&trans->new_bgs)) {
 243                 const char *errstr;
 244
 245                 errstr = btrfs_decode_error(errno);
 246                 btrfs_warn(root->fs_info,
 247                            "%s:%d: Aborting unused transaction(%s).",
 248                            function, line, errstr);
 249                 return;
 250         }
 251         ACCESS_ONCE(trans->transaction->aborted) = errno;
 252         /* Wake up anybody who may be waiting on this transaction */
 253         wake_up(&root->fs_info->transaction_wait);
 254         wake_up(&root->fs_info->transaction_blocked_wait);
 255         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 256 }
 257 /*
 258  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 259  * issues an alert, and either panics or BUGs, depending on mount options.
 260  */
 261 __cold
 262 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 263                    unsigned int line, int errno, const char *fmt, ...)
 264 {
 265         char *s_id = "<unknown>";
 266         const char *errstr;
 267         struct va_format vaf = { .fmt = fmt };
 268         va_list args;
 269
 270         if (fs_info)
 271                 s_id = fs_info->sb->s_id;
 272
 273         va_start(args, fmt);
 274         vaf.va = &args;
 275
 276         errstr = btrfs_decode_error(errno);
 277         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 278                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 279                         s_id, function, line, &vaf, errno, errstr);
 280
 281         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 282                    function, line, &vaf, errno, errstr);
 283         va_end(args);
 284         /* Caller calls BUG() */
 285 }
 286
 287 static void btrfs_put_super(struct super_block *sb)
 288 {
 289         close_ctree(btrfs_sb(sb)->tree_root);
 290 }
 291
 292 enum {
 293         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 294         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 295         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 296         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 297         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 298         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 299         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 300         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 301         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 302         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 303         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
 304         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
 305         Opt_datasum, Opt_treelog, Opt_noinode_cache,
 306 #ifdef CONFIG_BTRFS_DEBUG
 307         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 308 #endif
 309         Opt_err,
 310 };
 311
 312 static match_table_t tokens = {
 313         {Opt_degraded, "degraded"},
 314         {Opt_subvol, "subvol=%s"},
 315         {Opt_subvolid, "subvolid=%s"},
 316         {Opt_device, "device=%s"},
 317         {Opt_nodatasum, "nodatasum"},
 318         {Opt_datasum, "datasum"},
 319         {Opt_nodatacow, "nodatacow"},
 320         {Opt_datacow, "datacow"},
 321         {Opt_nobarrier, "nobarrier"},
 322         {Opt_barrier, "barrier"},
 323         {Opt_max_inline, "max_inline=%s"},
 324         {Opt_alloc_start, "alloc_start=%s"},
 325         {Opt_thread_pool, "thread_pool=%d"},
 326         {Opt_compress, "compress"},
 327         {Opt_compress_type, "compress=%s"},
 328         {Opt_compress_force, "compress-force"},
 329         {Opt_compress_force_type, "compress-force=%s"},
 330         {Opt_ssd, "ssd"},
 331         {Opt_ssd_spread, "ssd_spread"},
 332         {Opt_nossd, "nossd"},
 333         {Opt_acl, "acl"},
 334         {Opt_noacl, "noacl"},
 335         {Opt_notreelog, "notreelog"},
 336         {Opt_treelog, "treelog"},
 337         {Opt_flushoncommit, "flushoncommit"},
 338         {Opt_noflushoncommit, "noflushoncommit"},
 339         {Opt_ratio, "metadata_ratio=%d"},
 340         {Opt_discard, "discard"},
 341         {Opt_nodiscard, "nodiscard"},
 342         {Opt_space_cache, "space_cache"},
 343         {Opt_clear_cache, "clear_cache"},
 344         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 345         {Opt_enospc_debug, "enospc_debug"},
 346         {Opt_noenospc_debug, "noenospc_debug"},
 347         {Opt_subvolrootid, "subvolrootid=%d"},
 348         {Opt_defrag, "autodefrag"},
 349         {Opt_nodefrag, "noautodefrag"},
 350         {Opt_inode_cache, "inode_cache"},
 351         {Opt_noinode_cache, "noinode_cache"},
 352         {Opt_no_space_cache, "nospace_cache"},
 353         {Opt_recovery, "recovery"},
 354         {Opt_skip_balance, "skip_balance"},
 355         {Opt_check_integrity, "check_int"},
 356         {Opt_check_integrity_including_extent_data, "check_int_data"},
 357         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 358         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 359         {Opt_fatal_errors, "fatal_errors=%s"},
 360         {Opt_commit_interval, "commit=%d"},
 361 #ifdef CONFIG_BTRFS_DEBUG
 362         {Opt_fragment_data, "fragment=data"},
 363         {Opt_fragment_metadata, "fragment=metadata"},
 364         {Opt_fragment_all, "fragment=all"},
 365 #endif
 366         {Opt_err, NULL},
 367 };
 368
 369 /*
 370  * Regular mount options parser.  Everything that is needed only when
 371  * reading in a new superblock is parsed here.
 372  * XXX JDM: This needs to be cleaned up for remount.
 373  */
 374 int btrfs_parse_options(struct btrfs_root *root, char *options)
 375 {
 376         struct btrfs_fs_info *info = root->fs_info;
 377         substring_t args[MAX_OPT_ARGS];
 378         char *p, *num, *orig = NULL;
 379         u64 cache_gen;
 380         int intarg;
 381         int ret = 0;
 382         char *compress_type;
 383         bool compress_force = false;
 384
 385         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 386         if (cache_gen)
 387                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 388
 389         if (!options)
 390                 goto out;
 391
 392         /*
 393          * strsep changes the string, duplicate it because parse_options
 394          * gets called twice
 395          */
 396         options = kstrdup(options, GFP_NOFS);
 397         if (!options)
 398                 return -ENOMEM;
 399
 400         orig = options;
 401
 402         while ((p = strsep(&options, ",")) != NULL) {
 403                 int token;
 404                 if (!*p)
 405                         continue;
 406
 407                 token = match_token(p, tokens, args);
 408                 switch (token) {
 409                 case Opt_degraded:
 410                         btrfs_info(root->fs_info, "allowing degraded mounts");
 411                         btrfs_set_opt(info->mount_opt, DEGRADED);
 412                         break;
 413                 case Opt_subvol:
 414                 case Opt_subvolid:
 415                 case Opt_subvolrootid:
 416                 case Opt_device:
 417                         /*
 418                          * These are parsed by btrfs_parse_early_options
 419                          * and can be happily ignored here.
 420                          */
 421                         break;
 422                 case Opt_nodatasum:
 423                         btrfs_set_and_info(root, NODATASUM,
 424                                            "setting nodatasum");
 425                         break;
 426                 case Opt_datasum:
 427                         if (btrfs_test_opt(root, NODATASUM)) {
 428                                 if (btrfs_test_opt(root, NODATACOW))
 429                                         btrfs_info(root->fs_info, "setting datasum, datacow enabled");
 430                                 else
 431                                         btrfs_info(root->fs_info, "setting datasum");
 432                         }
 433                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 434                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 435                         break;
 436                 case Opt_nodatacow:
 437                         if (!btrfs_test_opt(root, NODATACOW)) {
 438                                 if (!btrfs_test_opt(root, COMPRESS) ||
 439                                     !btrfs_test_opt(root, FORCE_COMPRESS)) {
 440                                         btrfs_info(root->fs_info,
 441                                                    "setting nodatacow, compression disabled");
 442                                 } else {
 443                                         btrfs_info(root->fs_info, "setting nodatacow");
 444                                 }
 445                         }
 446                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 447                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 448                         btrfs_set_opt(info->mount_opt, NODATACOW);
 449                         btrfs_set_opt(info->mount_opt, NODATASUM);
 450                         break;
 451                 case Opt_datacow:
 452                         btrfs_clear_and_info(root, NODATACOW,
 453                                              "setting datacow");
 454                         break;
 455                 case Opt_compress_force:
 456                 case Opt_compress_force_type:
 457                         compress_force = true;
 458                         /* Fallthrough */
 459                 case Opt_compress:
 460                 case Opt_compress_type:
 461                         if (token == Opt_compress ||
 462                             token == Opt_compress_force ||
 463                             strcmp(args[0].from, "zlib") == 0) {
 464                                 compress_type = "zlib";
 465                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 466                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 467                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 468                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 469                         } else if (strcmp(args[0].from, "lzo") == 0) {
 470                                 compress_type = "lzo";
 471                                 info->compress_type = BTRFS_COMPRESS_LZO;
 472                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 473                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 474                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 475                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 476                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 477                                 compress_type = "no";
 478                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 479                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 480                                 compress_force = false;
 481                         } else {
 482                                 ret = -EINVAL;
 483                                 goto out;
 484                         }
 485
 486                         if (compress_force) {
 487                                 btrfs_set_and_info(root, FORCE_COMPRESS,
 488                                                    "force %s compression",
 489                                                    compress_type);
 490                         } else {
 491                                 if (!btrfs_test_opt(root, COMPRESS))
 492                                         btrfs_info(root->fs_info,
 493                                                    "btrfs: use %s compression",
 494                                                    compress_type);
 495                                 /*
 496                                  * If we remount from compress-force=xxx to
 497                                  * compress=xxx, we need clear FORCE_COMPRESS
 498                                  * flag, otherwise, there is no way for users
 499                                  * to disable forcible compression separately.
 500                                  */
 501                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 502                         }
 503                         break;
 504                 case Opt_ssd:
 505                         btrfs_set_and_info(root, SSD,
 506                                            "use ssd allocation scheme");
 507                         break;
 508                 case Opt_ssd_spread:
 509                         btrfs_set_and_info(root, SSD_SPREAD,
 510                                            "use spread ssd allocation scheme");
 511                         btrfs_set_opt(info->mount_opt, SSD);
 512                         break;
 513                 case Opt_nossd:
 514                         btrfs_set_and_info(root, NOSSD,
 515                                              "not using ssd allocation scheme");
 516                         btrfs_clear_opt(info->mount_opt, SSD);
 517                         break;
 518                 case Opt_barrier:
 519                         btrfs_clear_and_info(root, NOBARRIER,
 520                                              "turning on barriers");
 521                         break;
 522                 case Opt_nobarrier:
 523                         btrfs_set_and_info(root, NOBARRIER,
 524                                            "turning off barriers");
 525                         break;
 526                 case Opt_thread_pool:
 527                         ret = match_int(&args[0], &intarg);
 528                         if (ret) {
 529                                 goto out;
 530                         } else if (intarg > 0) {
 531                                 info->thread_pool_size = intarg;
 532                         } else {
 533                                 ret = -EINVAL;
 534                                 goto out;
 535                         }
 536                         break;
 537                 case Opt_max_inline:
 538                         num = match_strdup(&args[0]);
 539                         if (num) {
 540                                 info->max_inline = memparse(num, NULL);
 541                                 kfree(num);
 542
 543                                 if (info->max_inline) {
 544                                         info->max_inline = min_t(u64,
 545                                                 info->max_inline,
 546                                                 root->sectorsize);
 547                                 }
 548                                 btrfs_info(root->fs_info, "max_inline at %llu",
 549                                         info->max_inline);
 550                         } else {
 551                                 ret = -ENOMEM;
 552                                 goto out;
 553                         }
 554                         break;
 555                 case Opt_alloc_start:
 556                         num = match_strdup(&args[0]);
 557                         if (num) {
 558                                 mutex_lock(&info->chunk_mutex);
 559                                 info->alloc_start = memparse(num, NULL);
 560                                 mutex_unlock(&info->chunk_mutex);
 561                                 kfree(num);
 562                                 btrfs_info(root->fs_info, "allocations start at %llu",
 563                                         info->alloc_start);
 564                         } else {
 565                                 ret = -ENOMEM;
 566                                 goto out;
 567                         }
 568                         break;
 569                 case Opt_acl:
 570 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 571                         root->fs_info->sb->s_flags |= MS_POSIXACL;
 572                         break;
 573 #else
 574                         btrfs_err(root->fs_info,
 575                                 "support for ACL not compiled in!");
 576                         ret = -EINVAL;
 577                         goto out;
 578 #endif
 579                 case Opt_noacl:
 580                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 581                         break;
 582                 case Opt_notreelog:
 583                         btrfs_set_and_info(root, NOTREELOG,
 584                                            "disabling tree log");
 585                         break;
 586                 case Opt_treelog:
 587                         btrfs_clear_and_info(root, NOTREELOG,
 588                                              "enabling tree log");
 589                         break;
 590                 case Opt_flushoncommit:
 591                         btrfs_set_and_info(root, FLUSHONCOMMIT,
 592                                            "turning on flush-on-commit");
 593                         break;
 594                 case Opt_noflushoncommit:
 595                         btrfs_clear_and_info(root, FLUSHONCOMMIT,
 596                                              "turning off flush-on-commit");
 597                         break;
 598                 case Opt_ratio:
 599                         ret = match_int(&args[0], &intarg);
 600                         if (ret) {
 601                                 goto out;
 602                         } else if (intarg >= 0) {
 603                                 info->metadata_ratio = intarg;
 604                                 btrfs_info(root->fs_info, "metadata ratio %d",
 605                                        info->metadata_ratio);
 606                         } else {
 607                                 ret = -EINVAL;
 608                                 goto out;
 609                         }
 610                         break;
 611                 case Opt_discard:
 612                         btrfs_set_and_info(root, DISCARD,
 613                                            "turning on discard");
 614                         break;
 615                 case Opt_nodiscard:
 616                         btrfs_clear_and_info(root, DISCARD,
 617                                              "turning off discard");
 618                         break;
 619                 case Opt_space_cache:
 620                         btrfs_set_and_info(root, SPACE_CACHE,
 621                                            "enabling disk space caching");
 622                         break;
 623                 case Opt_rescan_uuid_tree:
 624                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 625                         break;
 626                 case Opt_no_space_cache:
 627                         btrfs_clear_and_info(root, SPACE_CACHE,
 628                                              "disabling disk space caching");
 629                         break;
 630                 case Opt_inode_cache:
 631                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 632                                            "enabling inode map caching");
 633                         break;
 634                 case Opt_noinode_cache:
 635                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 636                                              "disabling inode map caching");
 637                         break;
 638                 case Opt_clear_cache:
 639                         btrfs_set_and_info(root, CLEAR_CACHE,
 640                                            "force clearing of disk cache");
 641                         break;
 642                 case Opt_user_subvol_rm_allowed:
 643                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 644                         break;
 645                 case Opt_enospc_debug:
 646                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 647                         break;
 648                 case Opt_noenospc_debug:
 649                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 650                         break;
 651                 case Opt_defrag:
 652                         btrfs_set_and_info(root, AUTO_DEFRAG,
 653                                            "enabling auto defrag");
 654                         break;
 655                 case Opt_nodefrag:
 656                         btrfs_clear_and_info(root, AUTO_DEFRAG,
 657                                              "disabling auto defrag");
 658                         break;
 659                 case Opt_recovery:
 660                         btrfs_info(root->fs_info, "enabling auto recovery");
 661                         btrfs_set_opt(info->mount_opt, RECOVERY);
 662                         break;
 663                 case Opt_skip_balance:
 664                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 665                         break;
 666 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 667                 case Opt_check_integrity_including_extent_data:
 668                         btrfs_info(root->fs_info,
 669                                    "enabling check integrity including extent data");
 670                         btrfs_set_opt(info->mount_opt,
 671                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 672                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 673                         break;
 674                 case Opt_check_integrity:
 675                         btrfs_info(root->fs_info, "enabling check integrity");
 676                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 677                         break;
 678                 case Opt_check_integrity_print_mask:
 679                         ret = match_int(&args[0], &intarg);
 680                         if (ret) {
 681                                 goto out;
 682                         } else if (intarg >= 0) {
 683                                 info->check_integrity_print_mask = intarg;
 684                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
 685                                        info->check_integrity_print_mask);
 686                         } else {
 687                                 ret = -EINVAL;
 688                                 goto out;
 689                         }
 690                         break;
 691 #else
 692                 case Opt_check_integrity_including_extent_data:
 693                 case Opt_check_integrity:
 694                 case Opt_check_integrity_print_mask:
 695                         btrfs_err(root->fs_info,
 696                                 "support for check_integrity* not compiled in!");
 697                         ret = -EINVAL;
 698                         goto out;
 699 #endif
 700                 case Opt_fatal_errors:
 701                         if (strcmp(args[0].from, "panic") == 0)
 702                                 btrfs_set_opt(info->mount_opt,
 703                                               PANIC_ON_FATAL_ERROR);
 704                         else if (strcmp(args[0].from, "bug") == 0)
 705                                 btrfs_clear_opt(info->mount_opt,
 706                                               PANIC_ON_FATAL_ERROR);
 707                         else {
 708                                 ret = -EINVAL;
 709                                 goto out;
 710                         }
 711                         break;
 712                 case Opt_commit_interval:
 713                         intarg = 0;
 714                         ret = match_int(&args[0], &intarg);
 715                         if (ret < 0) {
 716                                 btrfs_err(root->fs_info, "invalid commit interval");
 717                                 ret = -EINVAL;
 718                                 goto out;
 719                         }
 720                         if (intarg > 0) {
 721                                 if (intarg > 300) {
 722                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
 723                                                         intarg);
 724                                 }
 725                                 info->commit_interval = intarg;
 726                         } else {
 727                                 btrfs_info(root->fs_info, "using default commit interval %ds",
 728                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
 729                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 730                         }
 731                         break;
 732 #ifdef CONFIG_BTRFS_DEBUG
 733                 case Opt_fragment_all:
 734                         btrfs_info(root->fs_info, "fragmenting all space");
 735                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 736                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 737                         break;
 738                 case Opt_fragment_metadata:
 739                         btrfs_info(root->fs_info, "fragmenting metadata");
 740                         btrfs_set_opt(info->mount_opt,
 741                                       FRAGMENT_METADATA);
 742                         break;
 743                 case Opt_fragment_data:
 744                         btrfs_info(root->fs_info, "fragmenting data");
 745                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 746                         break;
 747 #endif
 748                 case Opt_err:
 749                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
 750                         ret = -EINVAL;
 751                         goto out;
 752                 default:
 753                         break;
 754                 }
 755         }
 756 out:
 757         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 758                 btrfs_info(root->fs_info, "disk space caching is enabled");
 759         kfree(orig);
 760         return ret;
 761 }
 762
 763 /*
 764  * Parse mount options that are required early in the mount process.
 765  *
 766  * All other options will be parsed on much later in the mount process and
 767  * only when we need to allocate a new super block.
 768  */
 769 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 770                 void *holder, char **subvol_name, u64 *subvol_objectid,
 771                 struct btrfs_fs_devices **fs_devices)
 772 {
 773         substring_t args[MAX_OPT_ARGS];
 774         char *device_name, *opts, *orig, *p;
 775         char *num = NULL;
 776         int error = 0;
 777
 778         if (!options)
 779                 return 0;
 780
 781         /*
 782          * strsep changes the string, duplicate it because parse_options
 783          * gets called twice
 784          */
 785         opts = kstrdup(options, GFP_KERNEL);
 786         if (!opts)
 787                 return -ENOMEM;
 788         orig = opts;
 789
 790         while ((p = strsep(&opts, ",")) != NULL) {
 791                 int token;
 792                 if (!*p)
 793                         continue;
 794
 795                 token = match_token(p, tokens, args);
 796                 switch (token) {
 797                 case Opt_subvol:
 798                         kfree(*subvol_name);
 799                         *subvol_name = match_strdup(&args[0]);
 800                         if (!*subvol_name) {
 801                                 error = -ENOMEM;
 802                                 goto out;
 803                         }
 804                         break;
 805                 case Opt_subvolid:
 806                         num = match_strdup(&args[0]);
 807                         if (num) {
 808                                 *subvol_objectid = memparse(num, NULL);
 809                                 kfree(num);
 810                                 /* we want the original fs_tree */
 811                                 if (!*subvol_objectid)
 812                                         *subvol_objectid =
 813                                                 BTRFS_FS_TREE_OBJECTID;
 814                         } else {
 815                                 error = -EINVAL;
 816                                 goto out;
 817                         }
 818                         break;
 819                 case Opt_subvolrootid:
 820                         printk(KERN_WARNING
 821                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
 822                                 "no effect\n");
 823                         break;
 824                 case Opt_device:
 825                         device_name = match_strdup(&args[0]);
 826                         if (!device_name) {
 827                                 error = -ENOMEM;
 828                                 goto out;
 829                         }
 830                         error = btrfs_scan_one_device(device_name,
 831                                         flags, holder, fs_devices);
 832                         kfree(device_name);
 833                         if (error)
 834                                 goto out;
 835                         break;
 836                 default:
 837                         break;
 838                 }
 839         }
 840
 841 out:
 842         kfree(orig);
 843         return error;
 844 }
 845
 846 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 847                                            u64 subvol_objectid)
 848 {
 849         struct btrfs_root *root = fs_info->tree_root;
 850         struct btrfs_root *fs_root;
 851         struct btrfs_root_ref *root_ref;
 852         struct btrfs_inode_ref *inode_ref;
 853         struct btrfs_key key;
 854         struct btrfs_path *path = NULL;
 855         char *name = NULL, *ptr;
 856         u64 dirid;
 857         int len;
 858         int ret;
 859
 860         path = btrfs_alloc_path();
 861         if (!path) {
 862                 ret = -ENOMEM;
 863                 goto err;
 864         }
 865         path->leave_spinning = 1;
 866
 867         name = kmalloc(PATH_MAX, GFP_NOFS);
 868         if (!name) {
 869                 ret = -ENOMEM;
 870                 goto err;
 871         }
 872         ptr = name + PATH_MAX - 1;
 873         ptr[0] = '\0';
 874
 875         /*
 876          * Walk up the subvolume trees in the tree of tree roots by root
 877          * backrefs until we hit the top-level subvolume.
 878          */
 879         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 880                 key.objectid = subvol_objectid;
 881                 key.type = BTRFS_ROOT_BACKREF_KEY;
 882                 key.offset = (u64)-1;
 883
 884                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 885                 if (ret < 0) {
 886                         goto err;
 887                 } else if (ret > 0) {
 888                         ret = btrfs_previous_item(root, path, subvol_objectid,
 889                                                   BTRFS_ROOT_BACKREF_KEY);
 890                         if (ret < 0) {
 891                                 goto err;
 892                         } else if (ret > 0) {
 893                                 ret = -ENOENT;
 894                                 goto err;
 895                         }
 896                 }
 897
 898                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 899                 subvol_objectid = key.offset;
 900
 901                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 902                                           struct btrfs_root_ref);
 903                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 904                 ptr -= len + 1;
 905                 if (ptr < name) {
 906                         ret = -ENAMETOOLONG;
 907                         goto err;
 908                 }
 909                 read_extent_buffer(path->nodes[0], ptr + 1,
 910                                    (unsigned long)(root_ref + 1), len);
 911                 ptr[0] = '/';
 912                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
 913                 btrfs_release_path(path);
 914
 915                 key.objectid = subvol_objectid;
 916                 key.type = BTRFS_ROOT_ITEM_KEY;
 917                 key.offset = (u64)-1;
 918                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
 919                 if (IS_ERR(fs_root)) {
 920                         ret = PTR_ERR(fs_root);
 921                         goto err;
 922                 }
 923
 924                 /*
 925                  * Walk up the filesystem tree by inode refs until we hit the
 926                  * root directory.
 927                  */
 928                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
 929                         key.objectid = dirid;
 930                         key.type = BTRFS_INODE_REF_KEY;
 931                         key.offset = (u64)-1;
 932
 933                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
 934                         if (ret < 0) {
 935                                 goto err;
 936                         } else if (ret > 0) {
 937                                 ret = btrfs_previous_item(fs_root, path, dirid,
 938                                                           BTRFS_INODE_REF_KEY);
 939                                 if (ret < 0) {
 940                                         goto err;
 941                                 } else if (ret > 0) {
 942                                         ret = -ENOENT;
 943                                         goto err;
 944                                 }
 945                         }
 946
 947                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 948                         dirid = key.offset;
 949
 950                         inode_ref = btrfs_item_ptr(path->nodes[0],
 951                                                    path->slots[0],
 952                                                    struct btrfs_inode_ref);
 953                         len = btrfs_inode_ref_name_len(path->nodes[0],
 954                                                        inode_ref);
 955                         ptr -= len + 1;
 956                         if (ptr < name) {
 957                                 ret = -ENAMETOOLONG;
 958                                 goto err;
 959                         }
 960                         read_extent_buffer(path->nodes[0], ptr + 1,
 961                                            (unsigned long)(inode_ref + 1), len);
 962                         ptr[0] = '/';
 963                         btrfs_release_path(path);
 964                 }
 965         }
 966
 967         btrfs_free_path(path);
 968         if (ptr == name + PATH_MAX - 1) {
 969                 name[0] = '/';
 970                 name[1] = '\0';
 971         } else {
 972                 memmove(name, ptr, name + PATH_MAX - ptr);
 973         }
 974         return name;
 975
 976 err:
 977         btrfs_free_path(path);
 978         kfree(name);
 979         return ERR_PTR(ret);
 980 }
 981
 982 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
 983 {
 984         struct btrfs_root *root = fs_info->tree_root;
 985         struct btrfs_dir_item *di;
 986         struct btrfs_path *path;
 987         struct btrfs_key location;
 988         u64 dir_id;
 989
 990         path = btrfs_alloc_path();
 991         if (!path)
 992                 return -ENOMEM;
 993         path->leave_spinning = 1;
 994
 995         /*
 996          * Find the "default" dir item which points to the root item that we
 997          * will mount by default if we haven't been given a specific subvolume
 998          * to mount.
 999          */
1000         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1001         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1002         if (IS_ERR(di)) {
1003                 btrfs_free_path(path);
1004                 return PTR_ERR(di);
1005         }
1006         if (!di) {
1007                 /*
1008                  * Ok the default dir item isn't there.  This is weird since
1009                  * it's always been there, but don't freak out, just try and
1010                  * mount the top-level subvolume.
1011                  */
1012                 btrfs_free_path(path);
1013                 *objectid = BTRFS_FS_TREE_OBJECTID;
1014                 return 0;
1015         }
1016
1017         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1018         btrfs_free_path(path);
1019         *objectid = location.objectid;
1020         return 0;
1021 }
1022
1023 static int btrfs_fill_super(struct super_block *sb,
1024                             struct btrfs_fs_devices *fs_devices,
1025                             void *data, int silent)
1026 {
1027         struct inode *inode;
1028         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1029         struct btrfs_key key;
1030         int err;
1031
1032         sb->s_maxbytes = MAX_LFS_FILESIZE;
1033         sb->s_magic = BTRFS_SUPER_MAGIC;
1034         sb->s_op = &btrfs_super_ops;
1035         sb->s_d_op = &btrfs_dentry_operations;
1036         sb->s_export_op = &btrfs_export_ops;
1037         sb->s_xattr = btrfs_xattr_handlers;
1038         sb->s_time_gran = 1;
1039 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1040         sb->s_flags |= MS_POSIXACL;
1041 #endif
1042         sb->s_flags |= MS_I_VERSION;
1043         sb->s_iflags |= SB_I_CGROUPWB;
1044         err = open_ctree(sb, fs_devices, (char *)data);
1045         if (err) {
1046                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
1047                 return err;
1048         }
1049
1050         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1051         key.type = BTRFS_INODE_ITEM_KEY;
1052         key.offset = 0;
1053         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1054         if (IS_ERR(inode)) {
1055                 err = PTR_ERR(inode);
1056                 goto fail_close;
1057         }
1058
1059         sb->s_root = d_make_root(inode);
1060         if (!sb->s_root) {
1061                 err = -ENOMEM;
1062                 goto fail_close;
1063         }
1064
1065         save_mount_options(sb, data);
1066         cleancache_init_fs(sb);
1067         sb->s_flags |= MS_ACTIVE;
1068         return 0;
1069
1070 fail_close:
1071         close_ctree(fs_info->tree_root);
1072         return err;
1073 }
1074
1075 int btrfs_sync_fs(struct super_block *sb, int wait)
1076 {
1077         struct btrfs_trans_handle *trans;
1078         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1079         struct btrfs_root *root = fs_info->tree_root;
1080
1081         trace_btrfs_sync_fs(wait);
1082
1083         if (!wait) {
1084                 filemap_flush(fs_info->btree_inode->i_mapping);
1085                 return 0;
1086         }
1087
1088         btrfs_wait_ordered_roots(fs_info, -1);
1089
1090         trans = btrfs_attach_transaction_barrier(root);
1091         if (IS_ERR(trans)) {
1092                 /* no transaction, don't bother */
1093                 if (PTR_ERR(trans) == -ENOENT) {
1094                         /*
1095                          * Exit unless we have some pending changes
1096                          * that need to go through commit
1097                          */
1098                         if (fs_info->pending_changes == 0)
1099                                 return 0;
1100                         /*
1101                          * A non-blocking test if the fs is frozen. We must not
1102                          * start a new transaction here otherwise a deadlock
1103                          * happens. The pending operations are delayed to the
1104                          * next commit after thawing.
1105                          */
1106                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1107                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1108                         else
1109                                 return 0;
1110                         trans = btrfs_start_transaction(root, 0);
1111                 }
1112                 if (IS_ERR(trans))
1113                         return PTR_ERR(trans);
1114         }
1115         return btrfs_commit_transaction(trans, root);
1116 }
1117
1118 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1119 {
1120         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1121         struct btrfs_root *root = info->tree_root;
1122         char *compress_type;
1123
1124         if (btrfs_test_opt(root, DEGRADED))
1125                 seq_puts(seq, ",degraded");
1126         if (btrfs_test_opt(root, NODATASUM))
1127                 seq_puts(seq, ",nodatasum");
1128         if (btrfs_test_opt(root, NODATACOW))
1129                 seq_puts(seq, ",nodatacow");
1130         if (btrfs_test_opt(root, NOBARRIER))
1131                 seq_puts(seq, ",nobarrier");
1132         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1133                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1134         if (info->alloc_start != 0)
1135                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1136         if (info->thread_pool_size !=  min_t(unsigned long,
1137                                              num_online_cpus() + 2, 8))
1138                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1139         if (btrfs_test_opt(root, COMPRESS)) {
1140                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1141                         compress_type = "zlib";
1142                 else
1143                         compress_type = "lzo";
1144                 if (btrfs_test_opt(root, FORCE_COMPRESS))
1145                         seq_printf(seq, ",compress-force=%s", compress_type);
1146                 else
1147                         seq_printf(seq, ",compress=%s", compress_type);
1148         }
1149         if (btrfs_test_opt(root, NOSSD))
1150                 seq_puts(seq, ",nossd");
1151         if (btrfs_test_opt(root, SSD_SPREAD))
1152                 seq_puts(seq, ",ssd_spread");
1153         else if (btrfs_test_opt(root, SSD))
1154                 seq_puts(seq, ",ssd");
1155         if (btrfs_test_opt(root, NOTREELOG))
1156                 seq_puts(seq, ",notreelog");
1157         if (btrfs_test_opt(root, FLUSHONCOMMIT))
1158                 seq_puts(seq, ",flushoncommit");
1159         if (btrfs_test_opt(root, DISCARD))
1160                 seq_puts(seq, ",discard");
1161         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1162                 seq_puts(seq, ",noacl");
1163         if (btrfs_test_opt(root, SPACE_CACHE))
1164                 seq_puts(seq, ",space_cache");
1165         else
1166                 seq_puts(seq, ",nospace_cache");
1167         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1168                 seq_puts(seq, ",rescan_uuid_tree");
1169         if (btrfs_test_opt(root, CLEAR_CACHE))
1170                 seq_puts(seq, ",clear_cache");
1171         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1172                 seq_puts(seq, ",user_subvol_rm_allowed");
1173         if (btrfs_test_opt(root, ENOSPC_DEBUG))
1174                 seq_puts(seq, ",enospc_debug");
1175         if (btrfs_test_opt(root, AUTO_DEFRAG))
1176                 seq_puts(seq, ",autodefrag");
1177         if (btrfs_test_opt(root, INODE_MAP_CACHE))
1178                 seq_puts(seq, ",inode_cache");
1179         if (btrfs_test_opt(root, SKIP_BALANCE))
1180                 seq_puts(seq, ",skip_balance");
1181         if (btrfs_test_opt(root, RECOVERY))
1182                 seq_puts(seq, ",recovery");
1183 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1184         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1185                 seq_puts(seq, ",check_int_data");
1186         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1187                 seq_puts(seq, ",check_int");
1188         if (info->check_integrity_print_mask)
1189                 seq_printf(seq, ",check_int_print_mask=%d",
1190                                 info->check_integrity_print_mask);
1191 #endif
1192         if (info->metadata_ratio)
1193                 seq_printf(seq, ",metadata_ratio=%d",
1194                                 info->metadata_ratio);
1195         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1196                 seq_puts(seq, ",fatal_errors=panic");
1197         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1198                 seq_printf(seq, ",commit=%d", info->commit_interval);
1199 #ifdef CONFIG_BTRFS_DEBUG
1200         if (btrfs_test_opt(root, FRAGMENT_DATA))
1201                 seq_puts(seq, ",fragment=data");
1202         if (btrfs_test_opt(root, FRAGMENT_METADATA))
1203                 seq_puts(seq, ",fragment=metadata");
1204 #endif
1205         seq_printf(seq, ",subvolid=%llu",
1206                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1207         seq_puts(seq, ",subvol=");
1208         seq_dentry(seq, dentry, " \t\n\\");
1209         return 0;
1210 }
1211
1212 static int btrfs_test_super(struct super_block *s, void *data)
1213 {
1214         struct btrfs_fs_info *p = data;
1215         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1216
1217         return fs_info->fs_devices == p->fs_devices;
1218 }
1219
1220 static int btrfs_set_super(struct super_block *s, void *data)
1221 {
1222         int err = set_anon_super(s, data);
1223         if (!err)
1224                 s->s_fs_info = data;
1225         return err;
1226 }
1227
1228 /*
1229  * subvolumes are identified by ino 256
1230  */
1231 static inline int is_subvolume_inode(struct inode *inode)
1232 {
1233         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1234                 return 1;
1235         return 0;
1236 }
1237
1238 /*
1239  * This will add subvolid=0 to the argument string while removing any subvol=
1240  * and subvolid= arguments to make sure we get the top-level root for path
1241  * walking to the subvol we want.
1242  */
1243 static char *setup_root_args(char *args)
1244 {
1245         char *buf, *dst, *sep;
1246
1247         if (!args)
1248                 return kstrdup("subvolid=0", GFP_NOFS);
1249
1250         /* The worst case is that we add ",subvolid=0" to the end. */
1251         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1252         if (!buf)
1253                 return NULL;
1254
1255         while (1) {
1256                 sep = strchrnul(args, ',');
1257                 if (!strstarts(args, "subvol=") &&
1258                     !strstarts(args, "subvolid=")) {
1259                         memcpy(dst, args, sep - args);
1260                         dst += sep - args;
1261                         *dst++ = ',';
1262                 }
1263                 if (*sep)
1264                         args = sep + 1;
1265                 else
1266                         break;
1267         }
1268         strcpy(dst, "subvolid=0");
1269
1270         return buf;
1271 }
1272
1273 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1274                                    int flags, const char *device_name,
1275                                    char *data)
1276 {
1277         struct dentry *root;
1278         struct vfsmount *mnt = NULL;
1279         char *newargs;
1280         int ret;
1281
1282         newargs = setup_root_args(data);
1283         if (!newargs) {
1284                 root = ERR_PTR(-ENOMEM);
1285                 goto out;
1286         }
1287
1288         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1289         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1290                 if (flags & MS_RDONLY) {
1291                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1292                                              device_name, newargs);
1293                 } else {
1294                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1295                                              device_name, newargs);
1296                         if (IS_ERR(mnt)) {
1297                                 root = ERR_CAST(mnt);
1298                                 mnt = NULL;
1299                                 goto out;
1300                         }
1301
1302                         down_write(&mnt->mnt_sb->s_umount);
1303                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1304                         up_write(&mnt->mnt_sb->s_umount);
1305                         if (ret < 0) {
1306                                 root = ERR_PTR(ret);
1307                                 goto out;
1308                         }
1309                 }
1310         }
1311         if (IS_ERR(mnt)) {
1312                 root = ERR_CAST(mnt);
1313                 mnt = NULL;
1314                 goto out;
1315         }
1316
1317         if (!subvol_name) {
1318                 if (!subvol_objectid) {
1319                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1320                                                           &subvol_objectid);
1321                         if (ret) {
1322                                 root = ERR_PTR(ret);
1323                                 goto out;
1324                         }
1325                 }
1326                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1327                                                             subvol_objectid);
1328                 if (IS_ERR(subvol_name)) {
1329                         root = ERR_CAST(subvol_name);
1330                         subvol_name = NULL;
1331                         goto out;
1332                 }
1333
1334         }
1335
1336         root = mount_subtree(mnt, subvol_name);
1337         /* mount_subtree() drops our reference on the vfsmount. */
1338         mnt = NULL;
1339
1340         if (!IS_ERR(root)) {
1341                 struct super_block *s = root->d_sb;
1342                 struct inode *root_inode = d_inode(root);
1343                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1344
1345                 ret = 0;
1346                 if (!is_subvolume_inode(root_inode)) {
1347                         pr_err("BTRFS: '%s' is not a valid subvolume\n",
1348                                subvol_name);
1349                         ret = -EINVAL;
1350                 }
1351                 if (subvol_objectid && root_objectid != subvol_objectid) {
1352                         /*
1353                          * This will also catch a race condition where a
1354                          * subvolume which was passed by ID is renamed and
1355                          * another subvolume is renamed over the old location.
1356                          */
1357                         pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1358                                subvol_name, subvol_objectid);
1359                         ret = -EINVAL;
1360                 }
1361                 if (ret) {
1362                         dput(root);
1363                         root = ERR_PTR(ret);
1364                         deactivate_locked_super(s);
1365                 }
1366         }
1367
1368 out:
1369         mntput(mnt);
1370         kfree(newargs);
1371         kfree(subvol_name);
1372         return root;
1373 }
1374
1375 static int parse_security_options(char *orig_opts,
1376                                   struct security_mnt_opts *sec_opts)
1377 {
1378         char *secdata = NULL;
1379         int ret = 0;
1380
1381         secdata = alloc_secdata();
1382         if (!secdata)
1383                 return -ENOMEM;
1384         ret = security_sb_copy_data(orig_opts, secdata);
1385         if (ret) {
1386                 free_secdata(secdata);
1387                 return ret;
1388         }
1389         ret = security_sb_parse_opts_str(secdata, sec_opts);
1390         free_secdata(secdata);
1391         return ret;
1392 }
1393
1394 static int setup_security_options(struct btrfs_fs_info *fs_info,
1395                                   struct super_block *sb,
1396                                   struct security_mnt_opts *sec_opts)
1397 {
1398         int ret = 0;
1399
1400         /*
1401          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1402          * is valid.
1403          */
1404         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1405         if (ret)
1406                 return ret;
1407
1408 #ifdef CONFIG_SECURITY
1409         if (!fs_info->security_opts.num_mnt_opts) {
1410                 /* first time security setup, copy sec_opts to fs_info */
1411                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1412         } else {
1413                 /*
1414                  * Since SELinux(the only one supports security_mnt_opts) does
1415                  * NOT support changing context during remount/mount same sb,
1416                  * This must be the same or part of the same security options,
1417                  * just free it.
1418                  */
1419                 security_free_mnt_opts(sec_opts);
1420         }
1421 #endif
1422         return ret;
1423 }
1424
1425 /*
1426  * Find a superblock for the given device / mount point.
1427  *
1428  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1429  *        for multiple device setup.  Make sure to keep it in sync.
1430  */
1431 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1432                 const char *device_name, void *data)
1433 {
1434         struct block_device *bdev = NULL;
1435         struct super_block *s;
1436         struct btrfs_fs_devices *fs_devices = NULL;
1437         struct btrfs_fs_info *fs_info = NULL;
1438         struct security_mnt_opts new_sec_opts;
1439         fmode_t mode = FMODE_READ;
1440         char *subvol_name = NULL;
1441         u64 subvol_objectid = 0;
1442         int error = 0;
1443
1444         if (!(flags & MS_RDONLY))
1445                 mode |= FMODE_WRITE;
1446
1447         error = btrfs_parse_early_options(data, mode, fs_type,
1448                                           &subvol_name, &subvol_objectid,
1449                                           &fs_devices);
1450         if (error) {
1451                 kfree(subvol_name);
1452                 return ERR_PTR(error);
1453         }
1454
1455         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1456                 /* mount_subvol() will free subvol_name. */
1457                 return mount_subvol(subvol_name, subvol_objectid, flags,
1458                                     device_name, data);
1459         }
1460
1461         security_init_mnt_opts(&new_sec_opts);
1462         if (data) {
1463                 error = parse_security_options(data, &new_sec_opts);
1464                 if (error)
1465                         return ERR_PTR(error);
1466         }
1467
1468         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1469         if (error)
1470                 goto error_sec_opts;
1471
1472         /*
1473          * Setup a dummy root and fs_info for test/set super.  This is because
1474          * we don't actually fill this stuff out until open_ctree, but we need
1475          * it for searching for existing supers, so this lets us do that and
1476          * then open_ctree will properly initialize everything later.
1477          */
1478         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1479         if (!fs_info) {
1480                 error = -ENOMEM;
1481                 goto error_sec_opts;
1482         }
1483
1484         fs_info->fs_devices = fs_devices;
1485
1486         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1487         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1488         security_init_mnt_opts(&fs_info->security_opts);
1489         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1490                 error = -ENOMEM;
1491                 goto error_fs_info;
1492         }
1493
1494         error = btrfs_open_devices(fs_devices, mode, fs_type);
1495         if (error)
1496                 goto error_fs_info;
1497
1498         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1499                 error = -EACCES;
1500                 goto error_close_devices;
1501         }
1502
1503         bdev = fs_devices->latest_bdev;
1504         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1505                  fs_info);
1506         if (IS_ERR(s)) {
1507                 error = PTR_ERR(s);
1508                 goto error_close_devices;
1509         }
1510
1511         if (s->s_root) {
1512                 btrfs_close_devices(fs_devices);
1513                 free_fs_info(fs_info);
1514                 if ((flags ^ s->s_flags) & MS_RDONLY)
1515                         error = -EBUSY;
1516         } else {
1517                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1518                 btrfs_sb(s)->bdev_holder = fs_type;
1519                 error = btrfs_fill_super(s, fs_devices, data,
1520                                          flags & MS_SILENT ? 1 : 0);
1521         }
1522         if (error) {
1523                 deactivate_locked_super(s);
1524                 goto error_sec_opts;
1525         }
1526
1527         fs_info = btrfs_sb(s);
1528         error = setup_security_options(fs_info, s, &new_sec_opts);
1529         if (error) {
1530                 deactivate_locked_super(s);
1531                 goto error_sec_opts;
1532         }
1533
1534         return dget(s->s_root);
1535
1536 error_close_devices:
1537         btrfs_close_devices(fs_devices);
1538 error_fs_info:
1539         free_fs_info(fs_info);
1540 error_sec_opts:
1541         security_free_mnt_opts(&new_sec_opts);
1542         return ERR_PTR(error);
1543 }
1544
1545 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1546                                      int new_pool_size, int old_pool_size)
1547 {
1548         if (new_pool_size == old_pool_size)
1549                 return;
1550
1551         fs_info->thread_pool_size = new_pool_size;
1552
1553         btrfs_info(fs_info, "resize thread pool %d -> %d",
1554                old_pool_size, new_pool_size);
1555
1556         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1557         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1558         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1559         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1560         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1561         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1562         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1563                                 new_pool_size);
1564         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1565         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1566         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1567         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1568         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1569                                 new_pool_size);
1570 }
1571
1572 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1573 {
1574         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1575 }
1576
1577 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1578                                        unsigned long old_opts, int flags)
1579 {
1580         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1581             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1582              (flags & MS_RDONLY))) {
1583                 /* wait for any defraggers to finish */
1584                 wait_event(fs_info->transaction_wait,
1585                            (atomic_read(&fs_info->defrag_running) == 0));
1586                 if (flags & MS_RDONLY)
1587                         sync_filesystem(fs_info->sb);
1588         }
1589 }
1590
1591 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1592                                          unsigned long old_opts)
1593 {
1594         /*
1595          * We need cleanup all defragable inodes if the autodefragment is
1596          * close or the fs is R/O.
1597          */
1598         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1599             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1600              (fs_info->sb->s_flags & MS_RDONLY))) {
1601                 btrfs_cleanup_defrag_inodes(fs_info);
1602         }
1603
1604         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1605 }
1606
1607 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1608 {
1609         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1610         struct btrfs_root *root = fs_info->tree_root;
1611         unsigned old_flags = sb->s_flags;
1612         unsigned long old_opts = fs_info->mount_opt;
1613         unsigned long old_compress_type = fs_info->compress_type;
1614         u64 old_max_inline = fs_info->max_inline;
1615         u64 old_alloc_start = fs_info->alloc_start;
1616         int old_thread_pool_size = fs_info->thread_pool_size;
1617         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1618         int ret;
1619
1620         sync_filesystem(sb);
1621         btrfs_remount_prepare(fs_info);
1622
1623         if (data) {
1624                 struct security_mnt_opts new_sec_opts;
1625
1626                 security_init_mnt_opts(&new_sec_opts);
1627                 ret = parse_security_options(data, &new_sec_opts);
1628                 if (ret)
1629                         goto restore;
1630                 ret = setup_security_options(fs_info, sb,
1631                                              &new_sec_opts);
1632                 if (ret) {
1633                         security_free_mnt_opts(&new_sec_opts);
1634                         goto restore;
1635                 }
1636         }
1637
1638         ret = btrfs_parse_options(root, data);
1639         if (ret) {
1640                 ret = -EINVAL;
1641                 goto restore;
1642         }
1643
1644         btrfs_remount_begin(fs_info, old_opts, *flags);
1645         btrfs_resize_thread_pool(fs_info,
1646                 fs_info->thread_pool_size, old_thread_pool_size);
1647
1648         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1649                 goto out;
1650
1651         if (*flags & MS_RDONLY) {
1652                 /*
1653                  * this also happens on 'umount -rf' or on shutdown, when
1654                  * the filesystem is busy.
1655                  */
1656                 cancel_work_sync(&fs_info->async_reclaim_work);
1657
1658                 /* wait for the uuid_scan task to finish */
1659                 down(&fs_info->uuid_tree_rescan_sem);
1660                 /* avoid complains from lockdep et al. */
1661                 up(&fs_info->uuid_tree_rescan_sem);
1662
1663                 sb->s_flags |= MS_RDONLY;
1664
1665                 /*
1666                  * Setting MS_RDONLY will put the cleaner thread to
1667                  * sleep at the next loop if it's already active.
1668                  * If it's already asleep, we'll leave unused block
1669                  * groups on disk until we're mounted read-write again
1670                  * unless we clean them up here.
1671                  */
1672                 btrfs_delete_unused_bgs(fs_info);
1673
1674                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1675                 btrfs_scrub_cancel(fs_info);
1676                 btrfs_pause_balance(fs_info);
1677
1678                 ret = btrfs_commit_super(root);
1679                 if (ret)
1680                         goto restore;
1681         } else {
1682                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1683                         btrfs_err(fs_info,
1684                                 "Remounting read-write after error is not allowed");
1685                         ret = -EINVAL;
1686                         goto restore;
1687                 }
1688                 if (fs_info->fs_devices->rw_devices == 0) {
1689                         ret = -EACCES;
1690                         goto restore;
1691                 }
1692
1693                 if (fs_info->fs_devices->missing_devices >
1694                      fs_info->num_tolerated_disk_barrier_failures &&
1695                     !(*flags & MS_RDONLY)) {
1696                         btrfs_warn(fs_info,
1697                                 "too many missing devices, writeable remount is not allowed");
1698                         ret = -EACCES;
1699                         goto restore;
1700                 }
1701
1702                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1703                         ret = -EINVAL;
1704                         goto restore;
1705                 }
1706
1707                 ret = btrfs_cleanup_fs_roots(fs_info);
1708                 if (ret)
1709                         goto restore;
1710
1711                 /* recover relocation */
1712                 mutex_lock(&fs_info->cleaner_mutex);
1713                 ret = btrfs_recover_relocation(root);
1714                 mutex_unlock(&fs_info->cleaner_mutex);
1715                 if (ret)
1716                         goto restore;
1717
1718                 ret = btrfs_resume_balance_async(fs_info);
1719                 if (ret)
1720                         goto restore;
1721
1722                 ret = btrfs_resume_dev_replace_async(fs_info);
1723                 if (ret) {
1724                         btrfs_warn(fs_info, "failed to resume dev_replace");
1725                         goto restore;
1726                 }
1727
1728                 if (!fs_info->uuid_root) {
1729                         btrfs_info(fs_info, "creating UUID tree");
1730                         ret = btrfs_create_uuid_tree(fs_info);
1731                         if (ret) {
1732                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1733                                 goto restore;
1734                         }
1735                 }
1736                 sb->s_flags &= ~MS_RDONLY;
1737         }
1738 out:
1739         wake_up_process(fs_info->transaction_kthread);
1740         btrfs_remount_cleanup(fs_info, old_opts);
1741         return 0;
1742
1743 restore:
1744         /* We've hit an error - don't reset MS_RDONLY */
1745         if (sb->s_flags & MS_RDONLY)
1746                 old_flags |= MS_RDONLY;
1747         sb->s_flags = old_flags;
1748         fs_info->mount_opt = old_opts;
1749         fs_info->compress_type = old_compress_type;
1750         fs_info->max_inline = old_max_inline;
1751         mutex_lock(&fs_info->chunk_mutex);
1752         fs_info->alloc_start = old_alloc_start;
1753         mutex_unlock(&fs_info->chunk_mutex);
1754         btrfs_resize_thread_pool(fs_info,
1755                 old_thread_pool_size, fs_info->thread_pool_size);
1756         fs_info->metadata_ratio = old_metadata_ratio;
1757         btrfs_remount_cleanup(fs_info, old_opts);
1758         return ret;
1759 }
1760
1761 /* Used to sort the devices by max_avail(descending sort) */
1762 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1763                                        const void *dev_info2)
1764 {
1765         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1766             ((struct btrfs_device_info *)dev_info2)->max_avail)
1767                 return -1;
1768         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1769                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1770                 return 1;
1771         else
1772         return 0;
1773 }
1774
1775 /*
1776  * sort the devices by max_avail, in which max free extent size of each device
1777  * is stored.(Descending Sort)
1778  */
1779 static inline void btrfs_descending_sort_devices(
1780                                         struct btrfs_device_info *devices,
1781                                         size_t nr_devices)
1782 {
1783         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1784              btrfs_cmp_device_free_bytes, NULL);
1785 }
1786
1787 /*
1788  * The helper to calc the free space on the devices that can be used to store
1789  * file data.
1790  */
1791 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1792 {
1793         struct btrfs_fs_info *fs_info = root->fs_info;
1794         struct btrfs_device_info *devices_info;
1795         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1796         struct btrfs_device *device;
1797         u64 skip_space;
1798         u64 type;
1799         u64 avail_space;
1800         u64 used_space;
1801         u64 min_stripe_size;
1802         int min_stripes = 1, num_stripes = 1;
1803         int i = 0, nr_devices;
1804         int ret;
1805
1806         /*
1807          * We aren't under the device list lock, so this is racey-ish, but good
1808          * enough for our purposes.
1809          */
1810         nr_devices = fs_info->fs_devices->open_devices;
1811         if (!nr_devices) {
1812                 smp_mb();
1813                 nr_devices = fs_info->fs_devices->open_devices;
1814                 ASSERT(nr_devices);
1815                 if (!nr_devices) {
1816                         *free_bytes = 0;
1817                         return 0;
1818                 }
1819         }
1820
1821         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1822                                GFP_NOFS);
1823         if (!devices_info)
1824                 return -ENOMEM;
1825
1826         /* calc min stripe number for data space alloction */
1827         type = btrfs_get_alloc_profile(root, 1);
1828         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1829                 min_stripes = 2;
1830                 num_stripes = nr_devices;
1831         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1832                 min_stripes = 2;
1833                 num_stripes = 2;
1834         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1835                 min_stripes = 4;
1836                 num_stripes = 4;
1837         }
1838
1839         if (type & BTRFS_BLOCK_GROUP_DUP)
1840                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1841         else
1842                 min_stripe_size = BTRFS_STRIPE_LEN;
1843
1844         if (fs_info->alloc_start)
1845                 mutex_lock(&fs_devices->device_list_mutex);
1846         rcu_read_lock();
1847         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1848                 if (!device->in_fs_metadata || !device->bdev ||
1849                     device->is_tgtdev_for_dev_replace)
1850                         continue;
1851
1852                 if (i >= nr_devices)
1853                         break;
1854
1855                 avail_space = device->total_bytes - device->bytes_used;
1856
1857                 /* align with stripe_len */
1858                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1859                 avail_space *= BTRFS_STRIPE_LEN;
1860
1861                 /*
1862                  * In order to avoid overwritting the superblock on the drive,
1863                  * btrfs starts at an offset of at least 1MB when doing chunk
1864                  * allocation.
1865                  */
1866                 skip_space = 1024 * 1024;
1867
1868                 /* user can set the offset in fs_info->alloc_start. */
1869                 if (fs_info->alloc_start &&
1870                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1871                     device->total_bytes) {
1872                         rcu_read_unlock();
1873                         skip_space = max(fs_info->alloc_start, skip_space);
1874
1875                         /*
1876                          * btrfs can not use the free space in
1877                          * [0, skip_space - 1], we must subtract it from the
1878                          * total. In order to implement it, we account the used
1879                          * space in this range first.
1880                          */
1881                         ret = btrfs_account_dev_extents_size(device, 0,
1882                                                              skip_space - 1,
1883                                                              &used_space);
1884                         if (ret) {
1885                                 kfree(devices_info);
1886                                 mutex_unlock(&fs_devices->device_list_mutex);
1887                                 return ret;
1888                         }
1889
1890                         rcu_read_lock();
1891
1892                         /* calc the free space in [0, skip_space - 1] */
1893                         skip_space -= used_space;
1894                 }
1895
1896                 /*
1897                  * we can use the free space in [0, skip_space - 1], subtract
1898                  * it from the total.
1899                  */
1900                 if (avail_space && avail_space >= skip_space)
1901                         avail_space -= skip_space;
1902                 else
1903                         avail_space = 0;
1904
1905                 if (avail_space < min_stripe_size)
1906                         continue;
1907
1908                 devices_info[i].dev = device;
1909                 devices_info[i].max_avail = avail_space;
1910
1911                 i++;
1912         }
1913         rcu_read_unlock();
1914         if (fs_info->alloc_start)
1915                 mutex_unlock(&fs_devices->device_list_mutex);
1916
1917         nr_devices = i;
1918
1919         btrfs_descending_sort_devices(devices_info, nr_devices);
1920
1921         i = nr_devices - 1;
1922         avail_space = 0;
1923         while (nr_devices >= min_stripes) {
1924                 if (num_stripes > nr_devices)
1925                         num_stripes = nr_devices;
1926
1927                 if (devices_info[i].max_avail >= min_stripe_size) {
1928                         int j;
1929                         u64 alloc_size;
1930
1931                         avail_space += devices_info[i].max_avail * num_stripes;
1932                         alloc_size = devices_info[i].max_avail;
1933                         for (j = i + 1 - num_stripes; j <= i; j++)
1934                                 devices_info[j].max_avail -= alloc_size;
1935                 }
1936                 i--;
1937                 nr_devices--;
1938         }
1939
1940         kfree(devices_info);
1941         *free_bytes = avail_space;
1942         return 0;
1943 }
1944
1945 /*
1946  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
1947  *
1948  * If there's a redundant raid level at DATA block groups, use the respective
1949  * multiplier to scale the sizes.
1950  *
1951  * Unused device space usage is based on simulating the chunk allocator
1952  * algorithm that respects the device sizes, order of allocations and the
1953  * 'alloc_start' value, this is a close approximation of the actual use but
1954  * there are other factors that may change the result (like a new metadata
1955  * chunk).
1956  *
1957  * FIXME: not accurate for mixed block groups, total and free/used are ok,
1958  * available appears slightly larger.
1959  */
1960 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1961 {
1962         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1963         struct btrfs_super_block *disk_super = fs_info->super_copy;
1964         struct list_head *head = &fs_info->space_info;
1965         struct btrfs_space_info *found;
1966         u64 total_used = 0;
1967         u64 total_free_data = 0;
1968         int bits = dentry->d_sb->s_blocksize_bits;
1969         __be32 *fsid = (__be32 *)fs_info->fsid;
1970         unsigned factor = 1;
1971         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
1972         int ret;
1973
1974         /*
1975          * holding chunk_muext to avoid allocating new chunks, holding
1976          * device_list_mutex to avoid the device being removed
1977          */
1978         rcu_read_lock();
1979         list_for_each_entry_rcu(found, head, list) {
1980                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1981                         int i;
1982
1983                         total_free_data += found->disk_total - found->disk_used;
1984                         total_free_data -=
1985                                 btrfs_account_ro_block_groups_free_space(found);
1986
1987                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1988                                 if (!list_empty(&found->block_groups[i])) {
1989                                         switch (i) {
1990                                         case BTRFS_RAID_DUP:
1991                                         case BTRFS_RAID_RAID1:
1992                                         case BTRFS_RAID_RAID10:
1993                                                 factor = 2;
1994                                         }
1995                                 }
1996                         }
1997                 }
1998
1999                 total_used += found->disk_used;
2000         }
2001
2002         rcu_read_unlock();
2003
2004         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2005         buf->f_blocks >>= bits;
2006         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2007
2008         /* Account global block reserve as used, it's in logical size already */
2009         spin_lock(&block_rsv->lock);
2010         buf->f_bfree -= block_rsv->size >> bits;
2011         spin_unlock(&block_rsv->lock);
2012
2013         buf->f_bavail = div_u64(total_free_data, factor);
2014         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2015         if (ret)
2016                 return ret;
2017         buf->f_bavail += div_u64(total_free_data, factor);
2018         buf->f_bavail = buf->f_bavail >> bits;
2019
2020         buf->f_type = BTRFS_SUPER_MAGIC;
2021         buf->f_bsize = dentry->d_sb->s_blocksize;
2022         buf->f_namelen = BTRFS_NAME_LEN;
2023
2024         /* We treat it as constant endianness (it doesn't matter _which_)
2025            because we want the fsid to come out the same whether mounted
2026            on a big-endian or little-endian host */
2027         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2028         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2029         /* Mask in the root object ID too, to disambiguate subvols */
2030         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2031         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2032
2033         return 0;
2034 }
2035
2036 static void btrfs_kill_super(struct super_block *sb)
2037 {
2038         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2039         kill_anon_super(sb);
2040         free_fs_info(fs_info);
2041 }
2042
2043 static struct file_system_type btrfs_fs_type = {
2044         .owner          = THIS_MODULE,
2045         .name           = "btrfs",
2046         .mount          = btrfs_mount,
2047         .kill_sb        = btrfs_kill_super,
2048         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2049 };
2050 MODULE_ALIAS_FS("btrfs");
2051
2052 static int btrfs_control_open(struct inode *inode, struct file *file)
2053 {
2054         /*
2055          * The control file's private_data is used to hold the
2056          * transaction when it is started and is used to keep
2057          * track of whether a transaction is already in progress.
2058          */
2059         file->private_data = NULL;
2060         return 0;
2061 }
2062
2063 /*
2064  * used by btrfsctl to scan devices when no FS is mounted
2065  */
2066 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2067                                 unsigned long arg)
2068 {
2069         struct btrfs_ioctl_vol_args *vol;
2070         struct btrfs_fs_devices *fs_devices;
2071         int ret = -ENOTTY;
2072
2073         if (!capable(CAP_SYS_ADMIN))
2074                 return -EPERM;
2075
2076         vol = memdup_user((void __user *)arg, sizeof(*vol));
2077         if (IS_ERR(vol))
2078                 return PTR_ERR(vol);
2079
2080         switch (cmd) {
2081         case BTRFS_IOC_SCAN_DEV:
2082                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2083                                             &btrfs_fs_type, &fs_devices);
2084                 break;
2085         case BTRFS_IOC_DEVICES_READY:
2086                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2087                                             &btrfs_fs_type, &fs_devices);
2088                 if (ret)
2089                         break;
2090                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2091                 break;
2092         }
2093
2094         kfree(vol);
2095         return ret;
2096 }
2097
2098 static int btrfs_freeze(struct super_block *sb)
2099 {
2100         struct btrfs_trans_handle *trans;
2101         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2102
2103         trans = btrfs_attach_transaction_barrier(root);
2104         if (IS_ERR(trans)) {
2105                 /* no transaction, don't bother */
2106                 if (PTR_ERR(trans) == -ENOENT)
2107                         return 0;
2108                 return PTR_ERR(trans);
2109         }
2110         return btrfs_commit_transaction(trans, root);
2111 }
2112
2113 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2114 {
2115         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2116         struct btrfs_fs_devices *cur_devices;
2117         struct btrfs_device *dev, *first_dev = NULL;
2118         struct list_head *head;
2119         struct rcu_string *name;
2120
2121         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2122         cur_devices = fs_info->fs_devices;
2123         while (cur_devices) {
2124                 head = &cur_devices->devices;
2125                 list_for_each_entry(dev, head, dev_list) {
2126                         if (dev->missing)
2127                                 continue;
2128                         if (!dev->name)
2129                                 continue;
2130                         if (!first_dev || dev->devid < first_dev->devid)
2131                                 first_dev = dev;
2132                 }
2133                 cur_devices = cur_devices->seed;
2134         }
2135
2136         if (first_dev) {
2137                 rcu_read_lock();
2138                 name = rcu_dereference(first_dev->name);
2139                 seq_escape(m, name->str, " \t\n\\");
2140                 rcu_read_unlock();
2141         } else {
2142                 WARN_ON(1);
2143         }
2144         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2145         return 0;
2146 }
2147
2148 static const struct super_operations btrfs_super_ops = {
2149         .drop_inode     = btrfs_drop_inode,
2150         .evict_inode    = btrfs_evict_inode,
2151         .put_super      = btrfs_put_super,
2152         .sync_fs        = btrfs_sync_fs,
2153         .show_options   = btrfs_show_options,
2154         .show_devname   = btrfs_show_devname,
2155         .write_inode    = btrfs_write_inode,
2156         .alloc_inode    = btrfs_alloc_inode,
2157         .destroy_inode  = btrfs_destroy_inode,
2158         .statfs         = btrfs_statfs,
2159         .remount_fs     = btrfs_remount,
2160         .freeze_fs      = btrfs_freeze,
2161 };
2162
2163 static const struct file_operations btrfs_ctl_fops = {
2164         .open = btrfs_control_open,
2165         .unlocked_ioctl  = btrfs_control_ioctl,
2166         .compat_ioctl = btrfs_control_ioctl,
2167         .owner   = THIS_MODULE,
2168         .llseek = noop_llseek,
2169 };
2170
2171 static struct miscdevice btrfs_misc = {
2172         .minor          = BTRFS_MINOR,
2173         .name           = "btrfs-control",
2174         .fops           = &btrfs_ctl_fops
2175 };
2176
2177 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2178 MODULE_ALIAS("devname:btrfs-control");
2179
2180 static int btrfs_interface_init(void)
2181 {
2182         return misc_register(&btrfs_misc);
2183 }
2184
2185 static void btrfs_interface_exit(void)
2186 {
2187         misc_deregister(&btrfs_misc);
2188 }
2189
2190 static void btrfs_print_info(void)
2191 {
2192         printk(KERN_INFO "Btrfs loaded"
2193 #ifdef CONFIG_BTRFS_DEBUG
2194                         ", debug=on"
2195 #endif
2196 #ifdef CONFIG_BTRFS_ASSERT
2197                         ", assert=on"
2198 #endif
2199 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2200                         ", integrity-checker=on"
2201 #endif
2202                         "\n");
2203 }
2204
2205 static int btrfs_run_sanity_tests(void)
2206 {
2207         int ret;
2208
2209         ret = btrfs_init_test_fs();
2210         if (ret)
2211                 return ret;
2212
2213         ret = btrfs_test_free_space_cache();
2214         if (ret)
2215                 goto out;
2216         ret = btrfs_test_extent_buffer_operations();
2217         if (ret)
2218                 goto out;
2219         ret = btrfs_test_extent_io();
2220         if (ret)
2221                 goto out;
2222         ret = btrfs_test_inodes();
2223         if (ret)
2224                 goto out;
2225         ret = btrfs_test_qgroups();
2226 out:
2227         btrfs_destroy_test_fs();
2228         return ret;
2229 }
2230
2231 static int __init init_btrfs_fs(void)
2232 {
2233         int err;
2234
2235         err = btrfs_hash_init();
2236         if (err)
2237                 return err;
2238
2239         btrfs_props_init();
2240
2241         err = btrfs_init_sysfs();
2242         if (err)
2243                 goto free_hash;
2244
2245         btrfs_init_compress();
2246
2247         err = btrfs_init_cachep();
2248         if (err)
2249                 goto free_compress;
2250
2251         err = extent_io_init();
2252         if (err)
2253                 goto free_cachep;
2254
2255         err = extent_map_init();
2256         if (err)
2257                 goto free_extent_io;
2258
2259         err = ordered_data_init();
2260         if (err)
2261                 goto free_extent_map;
2262
2263         err = btrfs_delayed_inode_init();
2264         if (err)
2265                 goto free_ordered_data;
2266
2267         err = btrfs_auto_defrag_init();
2268         if (err)
2269                 goto free_delayed_inode;
2270
2271         err = btrfs_delayed_ref_init();
2272         if (err)
2273                 goto free_auto_defrag;
2274
2275         err = btrfs_prelim_ref_init();
2276         if (err)
2277                 goto free_delayed_ref;
2278
2279         err = btrfs_end_io_wq_init();
2280         if (err)
2281                 goto free_prelim_ref;
2282
2283         err = btrfs_interface_init();
2284         if (err)
2285                 goto free_end_io_wq;
2286
2287         btrfs_init_lockdep();
2288
2289         btrfs_print_info();
2290
2291         err = btrfs_run_sanity_tests();
2292         if (err)
2293                 goto unregister_ioctl;
2294
2295         err = register_filesystem(&btrfs_fs_type);
2296         if (err)
2297                 goto unregister_ioctl;
2298
2299         return 0;
2300
2301 unregister_ioctl:
2302         btrfs_interface_exit();
2303 free_end_io_wq:
2304         btrfs_end_io_wq_exit();
2305 free_prelim_ref:
2306         btrfs_prelim_ref_exit();
2307 free_delayed_ref:
2308         btrfs_delayed_ref_exit();
2309 free_auto_defrag:
2310         btrfs_auto_defrag_exit();
2311 free_delayed_inode:
2312         btrfs_delayed_inode_exit();
2313 free_ordered_data:
2314         ordered_data_exit();
2315 free_extent_map:
2316         extent_map_exit();
2317 free_extent_io:
2318         extent_io_exit();
2319 free_cachep:
2320         btrfs_destroy_cachep();
2321 free_compress:
2322         btrfs_exit_compress();
2323         btrfs_exit_sysfs();
2324 free_hash:
2325         btrfs_hash_exit();
2326         return err;
2327 }
2328
2329 static void __exit exit_btrfs_fs(void)
2330 {
2331         btrfs_destroy_cachep();
2332         btrfs_delayed_ref_exit();
2333         btrfs_auto_defrag_exit();
2334         btrfs_delayed_inode_exit();
2335         btrfs_prelim_ref_exit();
2336         ordered_data_exit();
2337         extent_map_exit();
2338         extent_io_exit();
2339         btrfs_interface_exit();
2340         btrfs_end_io_wq_exit();
2341         unregister_filesystem(&btrfs_fs_type);
2342         btrfs_exit_sysfs();
2343         btrfs_cleanup_fs_uuids();
2344         btrfs_exit_compress();
2345         btrfs_hash_exit();
2346 }
2347
2348 late_initcall(init_btrfs_fs);
2349 module_exit(exit_btrfs_fs)
2350
2351 MODULE_LICENSE("GPL");