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