fs/btrfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/blkdev.h>
   7 #include <linux/module.h>
   8 #include <linux/fs.h>
   9 #include <linux/pagemap.h>
  10 #include <linux/highmem.h>
  11 #include <linux/time.h>
  12 #include <linux/init.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/writeback.h>
  18 #include <linux/statfs.h>
  19 #include <linux/compat.h>
  20 #include <linux/parser.h>
  21 #include <linux/ctype.h>
  22 #include <linux/namei.h>
  23 #include <linux/miscdevice.h>
  24 #include <linux/magic.h>
  25 #include <linux/slab.h>
  26 #include <linux/cleancache.h>
  27 #include <linux/ratelimit.h>
  28 #include <linux/crc32c.h>
  29 #include <linux/btrfs.h>
  30 #include "delayed-inode.h"
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "print-tree.h"
  36 #include "props.h"
  37 #include "xattr.h"
  38 #include "volumes.h"
  39 #include "export.h"
  40 #include "compression.h"
  41 #include "rcu-string.h"
  42 #include "dev-replace.h"
  43 #include "free-space-cache.h"
  44 #include "backref.h"
  45 #include "space-info.h"
  46 #include "tests/btrfs-tests.h"
  47
  48 #include "qgroup.h"
  49 #define CREATE_TRACE_POINTS
  50 #include <trace/events/btrfs.h>
  51
  52 static const struct super_operations btrfs_super_ops;
  53
  54 /*
  55  * Types for mounting the default subvolume and a subvolume explicitly
  56  * requested by subvol=/path. That way the callchain is straightforward and we
  57  * don't have to play tricks with the mount options and recursive calls to
  58  * btrfs_mount.
  59  *
  60  * The new btrfs_root_fs_type also servers as a tag for the bdev_holder.
  61  */
  62 static struct file_system_type btrfs_fs_type;
  63 static struct file_system_type btrfs_root_fs_type;
  64
  65 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  66
  67 const char *btrfs_decode_error(int errno)
  68 {
  69         char *errstr = "unknown";
  70
  71         switch (errno) {
  72         case -EIO:
  73                 errstr = "IO failure";
  74                 break;
  75         case -ENOMEM:
  76                 errstr = "Out of memory";
  77                 break;
  78         case -EROFS:
  79                 errstr = "Readonly filesystem";
  80                 break;
  81         case -EEXIST:
  82                 errstr = "Object already exists";
  83                 break;
  84         case -ENOSPC:
  85                 errstr = "No space left";
  86                 break;
  87         case -ENOENT:
  88                 errstr = "No such entry";
  89                 break;
  90         }
  91
  92         return errstr;
  93 }
  94
  95 /*
  96  * __btrfs_handle_fs_error decodes expected errors from the caller and
  97  * invokes the appropriate error response.
  98  */
  99 __cold
 100 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
 101                        unsigned int line, int errno, const char *fmt, ...)
 102 {
 103         struct super_block *sb = fs_info->sb;
 104 #ifdef CONFIG_PRINTK
 105         const char *errstr;
 106 #endif
 107
 108         /*
 109          * Special case: if the error is EROFS, and we're already
 110          * under SB_RDONLY, then it is safe here.
 111          */
 112         if (errno == -EROFS && sb_rdonly(sb))
 113                 return;
 114
 115 #ifdef CONFIG_PRINTK
 116         errstr = btrfs_decode_error(errno);
 117         if (fmt) {
 118                 struct va_format vaf;
 119                 va_list args;
 120
 121                 va_start(args, fmt);
 122                 vaf.fmt = fmt;
 123                 vaf.va = &args;
 124
 125                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 126                         sb->s_id, function, line, errno, errstr, &vaf);
 127                 va_end(args);
 128         } else {
 129                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 130                         sb->s_id, function, line, errno, errstr);
 131         }
 132 #endif
 133
 134         /*
 135          * Today we only save the error info to memory.  Long term we'll
 136          * also send it down to the disk
 137          */
 138         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 139
 140         /* Don't go through full error handling during mount */
 141         if (!(sb->s_flags & SB_BORN))
 142                 return;
 143
 144         if (sb_rdonly(sb))
 145                 return;
 146
 147         /* btrfs handle error by forcing the filesystem readonly */
 148         sb->s_flags |= SB_RDONLY;
 149         btrfs_info(fs_info, "forced readonly");
 150         /*
 151          * Note that a running device replace operation is not canceled here
 152          * although there is no way to update the progress. It would add the
 153          * risk of a deadlock, therefore the canceling is omitted. The only
 154          * penalty is that some I/O remains active until the procedure
 155          * completes. The next time when the filesystem is mounted writable
 156          * again, the device replace operation continues.
 157          */
 158 }
 159
 160 #ifdef CONFIG_PRINTK
 161 static const char * const logtypes[] = {
 162         "emergency",
 163         "alert",
 164         "critical",
 165         "error",
 166         "warning",
 167         "notice",
 168         "info",
 169         "debug",
 170 };
 171
 172
 173 /*
 174  * Use one ratelimit state per log level so that a flood of less important
 175  * messages doesn't cause more important ones to be dropped.
 176  */
 177 static struct ratelimit_state printk_limits[] = {
 178         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
 179         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
 180         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
 181         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
 182         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
 183         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
 184         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
 185         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
 186 };
 187
 188 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 189 {
 190         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
 191         struct va_format vaf;
 192         va_list args;
 193         int kern_level;
 194         const char *type = logtypes[4];
 195         struct ratelimit_state *ratelimit = &printk_limits[4];
 196
 197         va_start(args, fmt);
 198
 199         while ((kern_level = printk_get_level(fmt)) != 0) {
 200                 size_t size = printk_skip_level(fmt) - fmt;
 201
 202                 if (kern_level >= '0' && kern_level <= '7') {
 203                         memcpy(lvl, fmt,  size);
 204                         lvl[size] = '\0';
 205                         type = logtypes[kern_level - '0'];
 206                         ratelimit = &printk_limits[kern_level - '0'];
 207                 }
 208                 fmt += size;
 209         }
 210
 211         vaf.fmt = fmt;
 212         vaf.va = &args;
 213
 214         if (__ratelimit(ratelimit))
 215                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type,
 216                         fs_info ? fs_info->sb->s_id : "<unknown>", &vaf);
 217
 218         va_end(args);
 219 }
 220 #endif
 221
 222 /*
 223  * We only mark the transaction aborted and then set the file system read-only.
 224  * This will prevent new transactions from starting or trying to join this
 225  * one.
 226  *
 227  * This means that error recovery at the call site is limited to freeing
 228  * any local memory allocations and passing the error code up without
 229  * further cleanup. The transaction should complete as it normally would
 230  * in the call path but will return -EIO.
 231  *
 232  * We'll complete the cleanup in btrfs_end_transaction and
 233  * btrfs_commit_transaction.
 234  */
 235 __cold
 236 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 237                                const char *function,
 238                                unsigned int line, int errno)
 239 {
 240         struct btrfs_fs_info *fs_info = trans->fs_info;
 241
 242         trans->aborted = errno;
 243         /* Nothing used. The other threads that have joined this
 244          * transaction may be able to continue. */
 245         if (!trans->dirty && list_empty(&trans->new_bgs)) {
 246                 const char *errstr;
 247
 248                 errstr = btrfs_decode_error(errno);
 249                 btrfs_warn(fs_info,
 250                            "%s:%d: Aborting unused transaction(%s).",
 251                            function, line, errstr);
 252                 return;
 253         }
 254         WRITE_ONCE(trans->transaction->aborted, errno);
 255         /* Wake up anybody who may be waiting on this transaction */
 256         wake_up(&fs_info->transaction_wait);
 257         wake_up(&fs_info->transaction_blocked_wait);
 258         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
 259 }
 260 /*
 261  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 262  * issues an alert, and either panics or BUGs, depending on mount options.
 263  */
 264 __cold
 265 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 266                    unsigned int line, int errno, const char *fmt, ...)
 267 {
 268         char *s_id = "<unknown>";
 269         const char *errstr;
 270         struct va_format vaf = { .fmt = fmt };
 271         va_list args;
 272
 273         if (fs_info)
 274                 s_id = fs_info->sb->s_id;
 275
 276         va_start(args, fmt);
 277         vaf.va = &args;
 278
 279         errstr = btrfs_decode_error(errno);
 280         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
 281                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 282                         s_id, function, line, &vaf, errno, errstr);
 283
 284         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 285                    function, line, &vaf, errno, errstr);
 286         va_end(args);
 287         /* Caller calls BUG() */
 288 }
 289
 290 static void btrfs_put_super(struct super_block *sb)
 291 {
 292         close_ctree(btrfs_sb(sb));
 293 }
 294
 295 enum {
 296         Opt_acl, Opt_noacl,
 297         Opt_clear_cache,
 298         Opt_commit_interval,
 299         Opt_compress,
 300         Opt_compress_force,
 301         Opt_compress_force_type,
 302         Opt_compress_type,
 303         Opt_degraded,
 304         Opt_device,
 305         Opt_fatal_errors,
 306         Opt_flushoncommit, Opt_noflushoncommit,
 307         Opt_inode_cache, Opt_noinode_cache,
 308         Opt_max_inline,
 309         Opt_barrier, Opt_nobarrier,
 310         Opt_datacow, Opt_nodatacow,
 311         Opt_datasum, Opt_nodatasum,
 312         Opt_defrag, Opt_nodefrag,
 313         Opt_discard, Opt_nodiscard,
 314         Opt_nologreplay,
 315         Opt_norecovery,
 316         Opt_ratio,
 317         Opt_rescan_uuid_tree,
 318         Opt_skip_balance,
 319         Opt_space_cache, Opt_no_space_cache,
 320         Opt_space_cache_version,
 321         Opt_ssd, Opt_nossd,
 322         Opt_ssd_spread, Opt_nossd_spread,
 323         Opt_subvol,
 324         Opt_subvol_empty,
 325         Opt_subvolid,
 326         Opt_thread_pool,
 327         Opt_treelog, Opt_notreelog,
 328         Opt_usebackuproot,
 329         Opt_user_subvol_rm_allowed,
 330
 331         /* Deprecated options */
 332         Opt_alloc_start,
 333         Opt_recovery,
 334         Opt_subvolrootid,
 335
 336         /* Debugging options */
 337         Opt_check_integrity,
 338         Opt_check_integrity_including_extent_data,
 339         Opt_check_integrity_print_mask,
 340         Opt_enospc_debug, Opt_noenospc_debug,
 341 #ifdef CONFIG_BTRFS_DEBUG
 342         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 343 #endif
 344 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 345         Opt_ref_verify,
 346 #endif
 347         Opt_err,
 348 };
 349
 350 static const match_table_t tokens = {
 351         {Opt_acl, "acl"},
 352         {Opt_noacl, "noacl"},
 353         {Opt_clear_cache, "clear_cache"},
 354         {Opt_commit_interval, "commit=%u"},
 355         {Opt_compress, "compress"},
 356         {Opt_compress_type, "compress=%s"},
 357         {Opt_compress_force, "compress-force"},
 358         {Opt_compress_force_type, "compress-force=%s"},
 359         {Opt_degraded, "degraded"},
 360         {Opt_device, "device=%s"},
 361         {Opt_fatal_errors, "fatal_errors=%s"},
 362         {Opt_flushoncommit, "flushoncommit"},
 363         {Opt_noflushoncommit, "noflushoncommit"},
 364         {Opt_inode_cache, "inode_cache"},
 365         {Opt_noinode_cache, "noinode_cache"},
 366         {Opt_max_inline, "max_inline=%s"},
 367         {Opt_barrier, "barrier"},
 368         {Opt_nobarrier, "nobarrier"},
 369         {Opt_datacow, "datacow"},
 370         {Opt_nodatacow, "nodatacow"},
 371         {Opt_datasum, "datasum"},
 372         {Opt_nodatasum, "nodatasum"},
 373         {Opt_defrag, "autodefrag"},
 374         {Opt_nodefrag, "noautodefrag"},
 375         {Opt_discard, "discard"},
 376         {Opt_nodiscard, "nodiscard"},
 377         {Opt_nologreplay, "nologreplay"},
 378         {Opt_norecovery, "norecovery"},
 379         {Opt_ratio, "metadata_ratio=%u"},
 380         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 381         {Opt_skip_balance, "skip_balance"},
 382         {Opt_space_cache, "space_cache"},
 383         {Opt_no_space_cache, "nospace_cache"},
 384         {Opt_space_cache_version, "space_cache=%s"},
 385         {Opt_ssd, "ssd"},
 386         {Opt_nossd, "nossd"},
 387         {Opt_ssd_spread, "ssd_spread"},
 388         {Opt_nossd_spread, "nossd_spread"},
 389         {Opt_subvol, "subvol=%s"},
 390         {Opt_subvol_empty, "subvol="},
 391         {Opt_subvolid, "subvolid=%s"},
 392         {Opt_thread_pool, "thread_pool=%u"},
 393         {Opt_treelog, "treelog"},
 394         {Opt_notreelog, "notreelog"},
 395         {Opt_usebackuproot, "usebackuproot"},
 396         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 397
 398         /* Deprecated options */
 399         {Opt_alloc_start, "alloc_start=%s"},
 400         {Opt_recovery, "recovery"},
 401         {Opt_subvolrootid, "subvolrootid=%d"},
 402
 403         /* Debugging options */
 404         {Opt_check_integrity, "check_int"},
 405         {Opt_check_integrity_including_extent_data, "check_int_data"},
 406         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
 407         {Opt_enospc_debug, "enospc_debug"},
 408         {Opt_noenospc_debug, "noenospc_debug"},
 409 #ifdef CONFIG_BTRFS_DEBUG
 410         {Opt_fragment_data, "fragment=data"},
 411         {Opt_fragment_metadata, "fragment=metadata"},
 412         {Opt_fragment_all, "fragment=all"},
 413 #endif
 414 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 415         {Opt_ref_verify, "ref_verify"},
 416 #endif
 417         {Opt_err, NULL},
 418 };
 419
 420 /*
 421  * Regular mount options parser.  Everything that is needed only when
 422  * reading in a new superblock is parsed here.
 423  * XXX JDM: This needs to be cleaned up for remount.
 424  */
 425 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 426                         unsigned long new_flags)
 427 {
 428         substring_t args[MAX_OPT_ARGS];
 429         char *p, *num;
 430         u64 cache_gen;
 431         int intarg;
 432         int ret = 0;
 433         char *compress_type;
 434         bool compress_force = false;
 435         enum btrfs_compression_type saved_compress_type;
 436         bool saved_compress_force;
 437         int no_compress = 0;
 438
 439         cache_gen = btrfs_super_cache_generation(info->super_copy);
 440         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 441                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 442         else if (cache_gen)
 443                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 444
 445         /*
 446          * Even the options are empty, we still need to do extra check
 447          * against new flags
 448          */
 449         if (!options)
 450                 goto check;
 451
 452         while ((p = strsep(&options, ",")) != NULL) {
 453                 int token;
 454                 if (!*p)
 455                         continue;
 456
 457                 token = match_token(p, tokens, args);
 458                 switch (token) {
 459                 case Opt_degraded:
 460                         btrfs_info(info, "allowing degraded mounts");
 461                         btrfs_set_opt(info->mount_opt, DEGRADED);
 462                         break;
 463                 case Opt_subvol:
 464                 case Opt_subvol_empty:
 465                 case Opt_subvolid:
 466                 case Opt_subvolrootid:
 467                 case Opt_device:
 468                         /*
 469                          * These are parsed by btrfs_parse_subvol_options or
 470                          * btrfs_parse_device_options and can be ignored here.
 471                          */
 472                         break;
 473                 case Opt_nodatasum:
 474                         btrfs_set_and_info(info, NODATASUM,
 475                                            "setting nodatasum");
 476                         break;
 477                 case Opt_datasum:
 478                         if (btrfs_test_opt(info, NODATASUM)) {
 479                                 if (btrfs_test_opt(info, NODATACOW))
 480                                         btrfs_info(info,
 481                                                    "setting datasum, datacow enabled");
 482                                 else
 483                                         btrfs_info(info, "setting datasum");
 484                         }
 485                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 486                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 487                         break;
 488                 case Opt_nodatacow:
 489                         if (!btrfs_test_opt(info, NODATACOW)) {
 490                                 if (!btrfs_test_opt(info, COMPRESS) ||
 491                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 492                                         btrfs_info(info,
 493                                                    "setting nodatacow, compression disabled");
 494                                 } else {
 495                                         btrfs_info(info, "setting nodatacow");
 496                                 }
 497                         }
 498                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 499                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 500                         btrfs_set_opt(info->mount_opt, NODATACOW);
 501                         btrfs_set_opt(info->mount_opt, NODATASUM);
 502                         break;
 503                 case Opt_datacow:
 504                         btrfs_clear_and_info(info, NODATACOW,
 505                                              "setting datacow");
 506                         break;
 507                 case Opt_compress_force:
 508                 case Opt_compress_force_type:
 509                         compress_force = true;
 510                         /* Fallthrough */
 511                 case Opt_compress:
 512                 case Opt_compress_type:
 513                         saved_compress_type = btrfs_test_opt(info,
 514                                                              COMPRESS) ?
 515                                 info->compress_type : BTRFS_COMPRESS_NONE;
 516                         saved_compress_force =
 517                                 btrfs_test_opt(info, FORCE_COMPRESS);
 518                         if (token == Opt_compress ||
 519                             token == Opt_compress_force ||
 520                             strncmp(args[0].from, "zlib", 4) == 0) {
 521                                 compress_type = "zlib";
 522
 523                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 524                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 525                                 /*
 526                                  * args[0] contains uninitialized data since
 527                                  * for these tokens we don't expect any
 528                                  * parameter.
 529                                  */
 530                                 if (token != Opt_compress &&
 531                                     token != Opt_compress_force)
 532                                         info->compress_level =
 533                                           btrfs_compress_str2level(
 534                                                         BTRFS_COMPRESS_ZLIB,
 535                                                         args[0].from + 4);
 536                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 537                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 538                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 539                                 no_compress = 0;
 540                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
 541                                 compress_type = "lzo";
 542                                 info->compress_type = BTRFS_COMPRESS_LZO;
 543                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 544                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 545                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 546                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 547                                 no_compress = 0;
 548                         } else if (strncmp(args[0].from, "zstd", 4) == 0) {
 549                                 compress_type = "zstd";
 550                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
 551                                 info->compress_level =
 552                                         btrfs_compress_str2level(
 553                                                          BTRFS_COMPRESS_ZSTD,
 554                                                          args[0].from + 4);
 555                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 556                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 557                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 558                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
 559                                 no_compress = 0;
 560                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 561                                 compress_type = "no";
 562                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 563                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 564                                 compress_force = false;
 565                                 no_compress++;
 566                         } else {
 567                                 ret = -EINVAL;
 568                                 goto out;
 569                         }
 570
 571                         if (compress_force) {
 572                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 573                         } else {
 574                                 /*
 575                                  * If we remount from compress-force=xxx to
 576                                  * compress=xxx, we need clear FORCE_COMPRESS
 577                                  * flag, otherwise, there is no way for users
 578                                  * to disable forcible compression separately.
 579                                  */
 580                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 581                         }
 582                         if ((btrfs_test_opt(info, COMPRESS) &&
 583                              (info->compress_type != saved_compress_type ||
 584                               compress_force != saved_compress_force)) ||
 585                             (!btrfs_test_opt(info, COMPRESS) &&
 586                              no_compress == 1)) {
 587                                 btrfs_info(info, "%s %s compression, level %d",
 588                                            (compress_force) ? "force" : "use",
 589                                            compress_type, info->compress_level);
 590                         }
 591                         compress_force = false;
 592                         break;
 593                 case Opt_ssd:
 594                         btrfs_set_and_info(info, SSD,
 595                                            "enabling ssd optimizations");
 596                         btrfs_clear_opt(info->mount_opt, NOSSD);
 597                         break;
 598                 case Opt_ssd_spread:
 599                         btrfs_set_and_info(info, SSD,
 600                                            "enabling ssd optimizations");
 601                         btrfs_set_and_info(info, SSD_SPREAD,
 602                                            "using spread ssd allocation scheme");
 603                         btrfs_clear_opt(info->mount_opt, NOSSD);
 604                         break;
 605                 case Opt_nossd:
 606                         btrfs_set_opt(info->mount_opt, NOSSD);
 607                         btrfs_clear_and_info(info, SSD,
 608                                              "not using ssd optimizations");
 609                         /* Fallthrough */
 610                 case Opt_nossd_spread:
 611                         btrfs_clear_and_info(info, SSD_SPREAD,
 612                                              "not using spread ssd allocation scheme");
 613                         break;
 614                 case Opt_barrier:
 615                         btrfs_clear_and_info(info, NOBARRIER,
 616                                              "turning on barriers");
 617                         break;
 618                 case Opt_nobarrier:
 619                         btrfs_set_and_info(info, NOBARRIER,
 620                                            "turning off barriers");
 621                         break;
 622                 case Opt_thread_pool:
 623                         ret = match_int(&args[0], &intarg);
 624                         if (ret) {
 625                                 goto out;
 626                         } else if (intarg == 0) {
 627                                 ret = -EINVAL;
 628                                 goto out;
 629                         }
 630                         info->thread_pool_size = intarg;
 631                         break;
 632                 case Opt_max_inline:
 633                         num = match_strdup(&args[0]);
 634                         if (num) {
 635                                 info->max_inline = memparse(num, NULL);
 636                                 kfree(num);
 637
 638                                 if (info->max_inline) {
 639                                         info->max_inline = min_t(u64,
 640                                                 info->max_inline,
 641                                                 info->sectorsize);
 642                                 }
 643                                 btrfs_info(info, "max_inline at %llu",
 644                                            info->max_inline);
 645                         } else {
 646                                 ret = -ENOMEM;
 647                                 goto out;
 648                         }
 649                         break;
 650                 case Opt_alloc_start:
 651                         btrfs_info(info,
 652                                 "option alloc_start is obsolete, ignored");
 653                         break;
 654                 case Opt_acl:
 655 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 656                         info->sb->s_flags |= SB_POSIXACL;
 657                         break;
 658 #else
 659                         btrfs_err(info, "support for ACL not compiled in!");
 660                         ret = -EINVAL;
 661                         goto out;
 662 #endif
 663                 case Opt_noacl:
 664                         info->sb->s_flags &= ~SB_POSIXACL;
 665                         break;
 666                 case Opt_notreelog:
 667                         btrfs_set_and_info(info, NOTREELOG,
 668                                            "disabling tree log");
 669                         break;
 670                 case Opt_treelog:
 671                         btrfs_clear_and_info(info, NOTREELOG,
 672                                              "enabling tree log");
 673                         break;
 674                 case Opt_norecovery:
 675                 case Opt_nologreplay:
 676                         btrfs_set_and_info(info, NOLOGREPLAY,
 677                                            "disabling log replay at mount time");
 678                         break;
 679                 case Opt_flushoncommit:
 680                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 681                                            "turning on flush-on-commit");
 682                         break;
 683                 case Opt_noflushoncommit:
 684                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 685                                              "turning off flush-on-commit");
 686                         break;
 687                 case Opt_ratio:
 688                         ret = match_int(&args[0], &intarg);
 689                         if (ret)
 690                                 goto out;
 691                         info->metadata_ratio = intarg;
 692                         btrfs_info(info, "metadata ratio %u",
 693                                    info->metadata_ratio);
 694                         break;
 695                 case Opt_discard:
 696                         btrfs_set_and_info(info, DISCARD,
 697                                            "turning on discard");
 698                         break;
 699                 case Opt_nodiscard:
 700                         btrfs_clear_and_info(info, DISCARD,
 701                                              "turning off discard");
 702                         break;
 703                 case Opt_space_cache:
 704                 case Opt_space_cache_version:
 705                         if (token == Opt_space_cache ||
 706                             strcmp(args[0].from, "v1") == 0) {
 707                                 btrfs_clear_opt(info->mount_opt,
 708                                                 FREE_SPACE_TREE);
 709                                 btrfs_set_and_info(info, SPACE_CACHE,
 710                                            "enabling disk space caching");
 711                         } else if (strcmp(args[0].from, "v2") == 0) {
 712                                 btrfs_clear_opt(info->mount_opt,
 713                                                 SPACE_CACHE);
 714                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 715                                                    "enabling free space tree");
 716                         } else {
 717                                 ret = -EINVAL;
 718                                 goto out;
 719                         }
 720                         break;
 721                 case Opt_rescan_uuid_tree:
 722                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 723                         break;
 724                 case Opt_no_space_cache:
 725                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 726                                 btrfs_clear_and_info(info, SPACE_CACHE,
 727                                              "disabling disk space caching");
 728                         }
 729                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 730                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 731                                              "disabling free space tree");
 732                         }
 733                         break;
 734                 case Opt_inode_cache:
 735                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 736                                            "enabling inode map caching");
 737                         break;
 738                 case Opt_noinode_cache:
 739                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 740                                              "disabling inode map caching");
 741                         break;
 742                 case Opt_clear_cache:
 743                         btrfs_set_and_info(info, CLEAR_CACHE,
 744                                            "force clearing of disk cache");
 745                         break;
 746                 case Opt_user_subvol_rm_allowed:
 747                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 748                         break;
 749                 case Opt_enospc_debug:
 750                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 751                         break;
 752                 case Opt_noenospc_debug:
 753                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 754                         break;
 755                 case Opt_defrag:
 756                         btrfs_set_and_info(info, AUTO_DEFRAG,
 757                                            "enabling auto defrag");
 758                         break;
 759                 case Opt_nodefrag:
 760                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 761                                              "disabling auto defrag");
 762                         break;
 763                 case Opt_recovery:
 764                         btrfs_warn(info,
 765                                    "'recovery' is deprecated, use 'usebackuproot' instead");
 766                         /* fall through */
 767                 case Opt_usebackuproot:
 768                         btrfs_info(info,
 769                                    "trying to use backup root at mount time");
 770                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 771                         break;
 772                 case Opt_skip_balance:
 773                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 774                         break;
 775 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 776                 case Opt_check_integrity_including_extent_data:
 777                         btrfs_info(info,
 778                                    "enabling check integrity including extent data");
 779                         btrfs_set_opt(info->mount_opt,
 780                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 781                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 782                         break;
 783                 case Opt_check_integrity:
 784                         btrfs_info(info, "enabling check integrity");
 785                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 786                         break;
 787                 case Opt_check_integrity_print_mask:
 788                         ret = match_int(&args[0], &intarg);
 789                         if (ret)
 790                                 goto out;
 791                         info->check_integrity_print_mask = intarg;
 792                         btrfs_info(info, "check_integrity_print_mask 0x%x",
 793                                    info->check_integrity_print_mask);
 794                         break;
 795 #else
 796                 case Opt_check_integrity_including_extent_data:
 797                 case Opt_check_integrity:
 798                 case Opt_check_integrity_print_mask:
 799                         btrfs_err(info,
 800                                   "support for check_integrity* not compiled in!");
 801                         ret = -EINVAL;
 802                         goto out;
 803 #endif
 804                 case Opt_fatal_errors:
 805                         if (strcmp(args[0].from, "panic") == 0)
 806                                 btrfs_set_opt(info->mount_opt,
 807                                               PANIC_ON_FATAL_ERROR);
 808                         else if (strcmp(args[0].from, "bug") == 0)
 809                                 btrfs_clear_opt(info->mount_opt,
 810                                               PANIC_ON_FATAL_ERROR);
 811                         else {
 812                                 ret = -EINVAL;
 813                                 goto out;
 814                         }
 815                         break;
 816                 case Opt_commit_interval:
 817                         intarg = 0;
 818                         ret = match_int(&args[0], &intarg);
 819                         if (ret)
 820                                 goto out;
 821                         if (intarg == 0) {
 822                                 btrfs_info(info,
 823                                            "using default commit interval %us",
 824                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
 825                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
 826                         } else if (intarg > 300) {
 827                                 btrfs_warn(info, "excessive commit interval %d",
 828                                            intarg);
 829                         }
 830                         info->commit_interval = intarg;
 831                         break;
 832 #ifdef CONFIG_BTRFS_DEBUG
 833                 case Opt_fragment_all:
 834                         btrfs_info(info, "fragmenting all space");
 835                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 836                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 837                         break;
 838                 case Opt_fragment_metadata:
 839                         btrfs_info(info, "fragmenting metadata");
 840                         btrfs_set_opt(info->mount_opt,
 841                                       FRAGMENT_METADATA);
 842                         break;
 843                 case Opt_fragment_data:
 844                         btrfs_info(info, "fragmenting data");
 845                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 846                         break;
 847 #endif
 848 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 849                 case Opt_ref_verify:
 850                         btrfs_info(info, "doing ref verification");
 851                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
 852                         break;
 853 #endif
 854                 case Opt_err:
 855                         btrfs_info(info, "unrecognized mount option '%s'", p);
 856                         ret = -EINVAL;
 857                         goto out;
 858                 default:
 859                         break;
 860                 }
 861         }
 862 check:
 863         /*
 864          * Extra check for current option against current flag
 865          */
 866         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & SB_RDONLY)) {
 867                 btrfs_err(info,
 868                           "nologreplay must be used with ro mount option");
 869                 ret = -EINVAL;
 870         }
 871 out:
 872         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
 873             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
 874             !btrfs_test_opt(info, CLEAR_CACHE)) {
 875                 btrfs_err(info, "cannot disable free space tree");
 876                 ret = -EINVAL;
 877
 878         }
 879         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
 880                 btrfs_info(info, "disk space caching is enabled");
 881         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
 882                 btrfs_info(info, "using free space tree");
 883         return ret;
 884 }
 885
 886 /*
 887  * Parse mount options that are required early in the mount process.
 888  *
 889  * All other options will be parsed on much later in the mount process and
 890  * only when we need to allocate a new super block.
 891  */
 892 static int btrfs_parse_device_options(const char *options, fmode_t flags,
 893                                       void *holder)
 894 {
 895         substring_t args[MAX_OPT_ARGS];
 896         char *device_name, *opts, *orig, *p;
 897         struct btrfs_device *device = NULL;
 898         int error = 0;
 899
 900         lockdep_assert_held(&uuid_mutex);
 901
 902         if (!options)
 903                 return 0;
 904
 905         /*
 906          * strsep changes the string, duplicate it because btrfs_parse_options
 907          * gets called later
 908          */
 909         opts = kstrdup(options, GFP_KERNEL);
 910         if (!opts)
 911                 return -ENOMEM;
 912         orig = opts;
 913
 914         while ((p = strsep(&opts, ",")) != NULL) {
 915                 int token;
 916
 917                 if (!*p)
 918                         continue;
 919
 920                 token = match_token(p, tokens, args);
 921                 if (token == Opt_device) {
 922                         device_name = match_strdup(&args[0]);
 923                         if (!device_name) {
 924                                 error = -ENOMEM;
 925                                 goto out;
 926                         }
 927                         device = btrfs_scan_one_device(device_name, flags,
 928                                         holder);
 929                         kfree(device_name);
 930                         if (IS_ERR(device)) {
 931                                 error = PTR_ERR(device);
 932                                 goto out;
 933                         }
 934                 }
 935         }
 936
 937 out:
 938         kfree(orig);
 939         return error;
 940 }
 941
 942 /*
 943  * Parse mount options that are related to subvolume id
 944  *
 945  * The value is later passed to mount_subvol()
 946  */
 947 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
 948                 u64 *subvol_objectid)
 949 {
 950         substring_t args[MAX_OPT_ARGS];
 951         char *opts, *orig, *p;
 952         int error = 0;
 953         u64 subvolid;
 954
 955         if (!options)
 956                 return 0;
 957
 958         /*
 959          * strsep changes the string, duplicate it because
 960          * btrfs_parse_device_options gets called later
 961          */
 962         opts = kstrdup(options, GFP_KERNEL);
 963         if (!opts)
 964                 return -ENOMEM;
 965         orig = opts;
 966
 967         while ((p = strsep(&opts, ",")) != NULL) {
 968                 int token;
 969                 if (!*p)
 970                         continue;
 971
 972                 token = match_token(p, tokens, args);
 973                 switch (token) {
 974                 case Opt_subvol:
 975                         kfree(*subvol_name);
 976                         *subvol_name = match_strdup(&args[0]);
 977                         if (!*subvol_name) {
 978                                 error = -ENOMEM;
 979                                 goto out;
 980                         }
 981                         break;
 982                 case Opt_subvolid:
 983                         error = match_u64(&args[0], &subvolid);
 984                         if (error)
 985                                 goto out;
 986
 987                         /* we want the original fs_tree */
 988                         if (subvolid == 0)
 989                                 subvolid = BTRFS_FS_TREE_OBJECTID;
 990
 991                         *subvol_objectid = subvolid;
 992                         break;
 993                 case Opt_subvolrootid:
 994                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
 995                         break;
 996                 default:
 997                         break;
 998                 }
 999         }
1000
1001 out:
1002         kfree(orig);
1003         return error;
1004 }
1005
1006 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
1007                                            u64 subvol_objectid)
1008 {
1009         struct btrfs_root *root = fs_info->tree_root;
1010         struct btrfs_root *fs_root;
1011         struct btrfs_root_ref *root_ref;
1012         struct btrfs_inode_ref *inode_ref;
1013         struct btrfs_key key;
1014         struct btrfs_path *path = NULL;
1015         char *name = NULL, *ptr;
1016         u64 dirid;
1017         int len;
1018         int ret;
1019
1020         path = btrfs_alloc_path();
1021         if (!path) {
1022                 ret = -ENOMEM;
1023                 goto err;
1024         }
1025         path->leave_spinning = 1;
1026
1027         name = kmalloc(PATH_MAX, GFP_KERNEL);
1028         if (!name) {
1029                 ret = -ENOMEM;
1030                 goto err;
1031         }
1032         ptr = name + PATH_MAX - 1;
1033         ptr[0] = '\0';
1034
1035         /*
1036          * Walk up the subvolume trees in the tree of tree roots by root
1037          * backrefs until we hit the top-level subvolume.
1038          */
1039         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1040                 key.objectid = subvol_objectid;
1041                 key.type = BTRFS_ROOT_BACKREF_KEY;
1042                 key.offset = (u64)-1;
1043
1044                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1045                 if (ret < 0) {
1046                         goto err;
1047                 } else if (ret > 0) {
1048                         ret = btrfs_previous_item(root, path, subvol_objectid,
1049                                                   BTRFS_ROOT_BACKREF_KEY);
1050                         if (ret < 0) {
1051                                 goto err;
1052                         } else if (ret > 0) {
1053                                 ret = -ENOENT;
1054                                 goto err;
1055                         }
1056                 }
1057
1058                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1059                 subvol_objectid = key.offset;
1060
1061                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1062                                           struct btrfs_root_ref);
1063                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1064                 ptr -= len + 1;
1065                 if (ptr < name) {
1066                         ret = -ENAMETOOLONG;
1067                         goto err;
1068                 }
1069                 read_extent_buffer(path->nodes[0], ptr + 1,
1070                                    (unsigned long)(root_ref + 1), len);
1071                 ptr[0] = '/';
1072                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1073                 btrfs_release_path(path);
1074
1075                 key.objectid = subvol_objectid;
1076                 key.type = BTRFS_ROOT_ITEM_KEY;
1077                 key.offset = (u64)-1;
1078                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1079                 if (IS_ERR(fs_root)) {
1080                         ret = PTR_ERR(fs_root);
1081                         goto err;
1082                 }
1083
1084                 /*
1085                  * Walk up the filesystem tree by inode refs until we hit the
1086                  * root directory.
1087                  */
1088                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1089                         key.objectid = dirid;
1090                         key.type = BTRFS_INODE_REF_KEY;
1091                         key.offset = (u64)-1;
1092
1093                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1094                         if (ret < 0) {
1095                                 goto err;
1096                         } else if (ret > 0) {
1097                                 ret = btrfs_previous_item(fs_root, path, dirid,
1098                                                           BTRFS_INODE_REF_KEY);
1099                                 if (ret < 0) {
1100                                         goto err;
1101                                 } else if (ret > 0) {
1102                                         ret = -ENOENT;
1103                                         goto err;
1104                                 }
1105                         }
1106
1107                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1108                         dirid = key.offset;
1109
1110                         inode_ref = btrfs_item_ptr(path->nodes[0],
1111                                                    path->slots[0],
1112                                                    struct btrfs_inode_ref);
1113                         len = btrfs_inode_ref_name_len(path->nodes[0],
1114                                                        inode_ref);
1115                         ptr -= len + 1;
1116                         if (ptr < name) {
1117                                 ret = -ENAMETOOLONG;
1118                                 goto err;
1119                         }
1120                         read_extent_buffer(path->nodes[0], ptr + 1,
1121                                            (unsigned long)(inode_ref + 1), len);
1122                         ptr[0] = '/';
1123                         btrfs_release_path(path);
1124                 }
1125         }
1126
1127         btrfs_free_path(path);
1128         if (ptr == name + PATH_MAX - 1) {
1129                 name[0] = '/';
1130                 name[1] = '\0';
1131         } else {
1132                 memmove(name, ptr, name + PATH_MAX - ptr);
1133         }
1134         return name;
1135
1136 err:
1137         btrfs_free_path(path);
1138         kfree(name);
1139         return ERR_PTR(ret);
1140 }
1141
1142 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1143 {
1144         struct btrfs_root *root = fs_info->tree_root;
1145         struct btrfs_dir_item *di;
1146         struct btrfs_path *path;
1147         struct btrfs_key location;
1148         u64 dir_id;
1149
1150         path = btrfs_alloc_path();
1151         if (!path)
1152                 return -ENOMEM;
1153         path->leave_spinning = 1;
1154
1155         /*
1156          * Find the "default" dir item which points to the root item that we
1157          * will mount by default if we haven't been given a specific subvolume
1158          * to mount.
1159          */
1160         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1161         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1162         if (IS_ERR(di)) {
1163                 btrfs_free_path(path);
1164                 return PTR_ERR(di);
1165         }
1166         if (!di) {
1167                 /*
1168                  * Ok the default dir item isn't there.  This is weird since
1169                  * it's always been there, but don't freak out, just try and
1170                  * mount the top-level subvolume.
1171                  */
1172                 btrfs_free_path(path);
1173                 *objectid = BTRFS_FS_TREE_OBJECTID;
1174                 return 0;
1175         }
1176
1177         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1178         btrfs_free_path(path);
1179         *objectid = location.objectid;
1180         return 0;
1181 }
1182
1183 static int btrfs_fill_super(struct super_block *sb,
1184                             struct btrfs_fs_devices *fs_devices,
1185                             void *data)
1186 {
1187         struct inode *inode;
1188         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1189         struct btrfs_key key;
1190         int err;
1191
1192         sb->s_maxbytes = MAX_LFS_FILESIZE;
1193         sb->s_magic = BTRFS_SUPER_MAGIC;
1194         sb->s_op = &btrfs_super_ops;
1195         sb->s_d_op = &btrfs_dentry_operations;
1196         sb->s_export_op = &btrfs_export_ops;
1197         sb->s_xattr = btrfs_xattr_handlers;
1198         sb->s_time_gran = 1;
1199 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1200         sb->s_flags |= SB_POSIXACL;
1201 #endif
1202         sb->s_flags |= SB_I_VERSION;
1203         sb->s_iflags |= SB_I_CGROUPWB;
1204
1205         err = super_setup_bdi(sb);
1206         if (err) {
1207                 btrfs_err(fs_info, "super_setup_bdi failed");
1208                 return err;
1209         }
1210
1211         err = open_ctree(sb, fs_devices, (char *)data);
1212         if (err) {
1213                 btrfs_err(fs_info, "open_ctree failed");
1214                 return err;
1215         }
1216
1217         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1218         key.type = BTRFS_INODE_ITEM_KEY;
1219         key.offset = 0;
1220         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1221         if (IS_ERR(inode)) {
1222                 err = PTR_ERR(inode);
1223                 goto fail_close;
1224         }
1225
1226         sb->s_root = d_make_root(inode);
1227         if (!sb->s_root) {
1228                 err = -ENOMEM;
1229                 goto fail_close;
1230         }
1231
1232         cleancache_init_fs(sb);
1233         sb->s_flags |= SB_ACTIVE;
1234         return 0;
1235
1236 fail_close:
1237         close_ctree(fs_info);
1238         return err;
1239 }
1240
1241 int btrfs_sync_fs(struct super_block *sb, int wait)
1242 {
1243         struct btrfs_trans_handle *trans;
1244         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1245         struct btrfs_root *root = fs_info->tree_root;
1246
1247         trace_btrfs_sync_fs(fs_info, wait);
1248
1249         if (!wait) {
1250                 filemap_flush(fs_info->btree_inode->i_mapping);
1251                 return 0;
1252         }
1253
1254         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1255
1256         trans = btrfs_attach_transaction_barrier(root);
1257         if (IS_ERR(trans)) {
1258                 /* no transaction, don't bother */
1259                 if (PTR_ERR(trans) == -ENOENT) {
1260                         /*
1261                          * Exit unless we have some pending changes
1262                          * that need to go through commit
1263                          */
1264                         if (fs_info->pending_changes == 0)
1265                                 return 0;
1266                         /*
1267                          * A non-blocking test if the fs is frozen. We must not
1268                          * start a new transaction here otherwise a deadlock
1269                          * happens. The pending operations are delayed to the
1270                          * next commit after thawing.
1271                          */
1272                         if (sb_start_write_trylock(sb))
1273                                 sb_end_write(sb);
1274                         else
1275                                 return 0;
1276                         trans = btrfs_start_transaction(root, 0);
1277                 }
1278                 if (IS_ERR(trans))
1279                         return PTR_ERR(trans);
1280         }
1281         return btrfs_commit_transaction(trans);
1282 }
1283
1284 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1285 {
1286         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1287         const char *compress_type;
1288
1289         if (btrfs_test_opt(info, DEGRADED))
1290                 seq_puts(seq, ",degraded");
1291         if (btrfs_test_opt(info, NODATASUM))
1292                 seq_puts(seq, ",nodatasum");
1293         if (btrfs_test_opt(info, NODATACOW))
1294                 seq_puts(seq, ",nodatacow");
1295         if (btrfs_test_opt(info, NOBARRIER))
1296                 seq_puts(seq, ",nobarrier");
1297         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1298                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1299         if (info->thread_pool_size !=  min_t(unsigned long,
1300                                              num_online_cpus() + 2, 8))
1301                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1302         if (btrfs_test_opt(info, COMPRESS)) {
1303                 compress_type = btrfs_compress_type2str(info->compress_type);
1304                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1305                         seq_printf(seq, ",compress-force=%s", compress_type);
1306                 else
1307                         seq_printf(seq, ",compress=%s", compress_type);
1308                 if (info->compress_level)
1309                         seq_printf(seq, ":%d", info->compress_level);
1310         }
1311         if (btrfs_test_opt(info, NOSSD))
1312                 seq_puts(seq, ",nossd");
1313         if (btrfs_test_opt(info, SSD_SPREAD))
1314                 seq_puts(seq, ",ssd_spread");
1315         else if (btrfs_test_opt(info, SSD))
1316                 seq_puts(seq, ",ssd");
1317         if (btrfs_test_opt(info, NOTREELOG))
1318                 seq_puts(seq, ",notreelog");
1319         if (btrfs_test_opt(info, NOLOGREPLAY))
1320                 seq_puts(seq, ",nologreplay");
1321         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1322                 seq_puts(seq, ",flushoncommit");
1323         if (btrfs_test_opt(info, DISCARD))
1324                 seq_puts(seq, ",discard");
1325         if (!(info->sb->s_flags & SB_POSIXACL))
1326                 seq_puts(seq, ",noacl");
1327         if (btrfs_test_opt(info, SPACE_CACHE))
1328                 seq_puts(seq, ",space_cache");
1329         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1330                 seq_puts(seq, ",space_cache=v2");
1331         else
1332                 seq_puts(seq, ",nospace_cache");
1333         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1334                 seq_puts(seq, ",rescan_uuid_tree");
1335         if (btrfs_test_opt(info, CLEAR_CACHE))
1336                 seq_puts(seq, ",clear_cache");
1337         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1338                 seq_puts(seq, ",user_subvol_rm_allowed");
1339         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1340                 seq_puts(seq, ",enospc_debug");
1341         if (btrfs_test_opt(info, AUTO_DEFRAG))
1342                 seq_puts(seq, ",autodefrag");
1343         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1344                 seq_puts(seq, ",inode_cache");
1345         if (btrfs_test_opt(info, SKIP_BALANCE))
1346                 seq_puts(seq, ",skip_balance");
1347 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1348         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1349                 seq_puts(seq, ",check_int_data");
1350         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1351                 seq_puts(seq, ",check_int");
1352         if (info->check_integrity_print_mask)
1353                 seq_printf(seq, ",check_int_print_mask=%d",
1354                                 info->check_integrity_print_mask);
1355 #endif
1356         if (info->metadata_ratio)
1357                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1358         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1359                 seq_puts(seq, ",fatal_errors=panic");
1360         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1361                 seq_printf(seq, ",commit=%u", info->commit_interval);
1362 #ifdef CONFIG_BTRFS_DEBUG
1363         if (btrfs_test_opt(info, FRAGMENT_DATA))
1364                 seq_puts(seq, ",fragment=data");
1365         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1366                 seq_puts(seq, ",fragment=metadata");
1367 #endif
1368         if (btrfs_test_opt(info, REF_VERIFY))
1369                 seq_puts(seq, ",ref_verify");
1370         seq_printf(seq, ",subvolid=%llu",
1371                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1372         seq_puts(seq, ",subvol=");
1373         seq_dentry(seq, dentry, " \t\n\\");
1374         return 0;
1375 }
1376
1377 static int btrfs_test_super(struct super_block *s, void *data)
1378 {
1379         struct btrfs_fs_info *p = data;
1380         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1381
1382         return fs_info->fs_devices == p->fs_devices;
1383 }
1384
1385 static int btrfs_set_super(struct super_block *s, void *data)
1386 {
1387         int err = set_anon_super(s, data);
1388         if (!err)
1389                 s->s_fs_info = data;
1390         return err;
1391 }
1392
1393 /*
1394  * subvolumes are identified by ino 256
1395  */
1396 static inline int is_subvolume_inode(struct inode *inode)
1397 {
1398         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1399                 return 1;
1400         return 0;
1401 }
1402
1403 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1404                                    struct vfsmount *mnt)
1405 {
1406         struct dentry *root;
1407         int ret;
1408
1409         if (!subvol_name) {
1410                 if (!subvol_objectid) {
1411                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1412                                                           &subvol_objectid);
1413                         if (ret) {
1414                                 root = ERR_PTR(ret);
1415                                 goto out;
1416                         }
1417                 }
1418                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1419                                                             subvol_objectid);
1420                 if (IS_ERR(subvol_name)) {
1421                         root = ERR_CAST(subvol_name);
1422                         subvol_name = NULL;
1423                         goto out;
1424                 }
1425
1426         }
1427
1428         root = mount_subtree(mnt, subvol_name);
1429         /* mount_subtree() drops our reference on the vfsmount. */
1430         mnt = NULL;
1431
1432         if (!IS_ERR(root)) {
1433                 struct super_block *s = root->d_sb;
1434                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1435                 struct inode *root_inode = d_inode(root);
1436                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1437
1438                 ret = 0;
1439                 if (!is_subvolume_inode(root_inode)) {
1440                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1441                                subvol_name);
1442                         ret = -EINVAL;
1443                 }
1444                 if (subvol_objectid && root_objectid != subvol_objectid) {
1445                         /*
1446                          * This will also catch a race condition where a
1447                          * subvolume which was passed by ID is renamed and
1448                          * another subvolume is renamed over the old location.
1449                          */
1450                         btrfs_err(fs_info,
1451                                   "subvol '%s' does not match subvolid %llu",
1452                                   subvol_name, subvol_objectid);
1453                         ret = -EINVAL;
1454                 }
1455                 if (ret) {
1456                         dput(root);
1457                         root = ERR_PTR(ret);
1458                         deactivate_locked_super(s);
1459                 }
1460         }
1461
1462 out:
1463         mntput(mnt);
1464         kfree(subvol_name);
1465         return root;
1466 }
1467
1468 /*
1469  * Find a superblock for the given device / mount point.
1470  *
1471  * Note: This is based on mount_bdev from fs/super.c with a few additions
1472  *       for multiple device setup.  Make sure to keep it in sync.
1473  */
1474 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1475                 int flags, const char *device_name, void *data)
1476 {
1477         struct block_device *bdev = NULL;
1478         struct super_block *s;
1479         struct btrfs_device *device = NULL;
1480         struct btrfs_fs_devices *fs_devices = NULL;
1481         struct btrfs_fs_info *fs_info = NULL;
1482         void *new_sec_opts = NULL;
1483         fmode_t mode = FMODE_READ;
1484         int error = 0;
1485
1486         if (!(flags & SB_RDONLY))
1487                 mode |= FMODE_WRITE;
1488
1489         if (data) {
1490                 error = security_sb_eat_lsm_opts(data, &new_sec_opts);
1491                 if (error)
1492                         return ERR_PTR(error);
1493         }
1494
1495         /*
1496          * Setup a dummy root and fs_info for test/set super.  This is because
1497          * we don't actually fill this stuff out until open_ctree, but we need
1498          * it for searching for existing supers, so this lets us do that and
1499          * then open_ctree will properly initialize everything later.
1500          */
1501         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1502         if (!fs_info) {
1503                 error = -ENOMEM;
1504                 goto error_sec_opts;
1505         }
1506
1507         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1508         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1509         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1510                 error = -ENOMEM;
1511                 goto error_fs_info;
1512         }
1513
1514         mutex_lock(&uuid_mutex);
1515         error = btrfs_parse_device_options(data, mode, fs_type);
1516         if (error) {
1517                 mutex_unlock(&uuid_mutex);
1518                 goto error_fs_info;
1519         }
1520
1521         device = btrfs_scan_one_device(device_name, mode, fs_type);
1522         if (IS_ERR(device)) {
1523                 mutex_unlock(&uuid_mutex);
1524                 error = PTR_ERR(device);
1525                 goto error_fs_info;
1526         }
1527
1528         fs_devices = device->fs_devices;
1529         fs_info->fs_devices = fs_devices;
1530
1531         error = btrfs_open_devices(fs_devices, mode, fs_type);
1532         mutex_unlock(&uuid_mutex);
1533         if (error)
1534                 goto error_fs_info;
1535
1536         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1537                 error = -EACCES;
1538                 goto error_close_devices;
1539         }
1540
1541         bdev = fs_devices->latest_bdev;
1542         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1543                  fs_info);
1544         if (IS_ERR(s)) {
1545                 error = PTR_ERR(s);
1546                 goto error_close_devices;
1547         }
1548
1549         if (s->s_root) {
1550                 btrfs_close_devices(fs_devices);
1551                 free_fs_info(fs_info);
1552                 if ((flags ^ s->s_flags) & SB_RDONLY)
1553                         error = -EBUSY;
1554         } else {
1555                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1556                 btrfs_sb(s)->bdev_holder = fs_type;
1557                 if (!strstr(crc32c_impl(), "generic"))
1558                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
1559                 error = btrfs_fill_super(s, fs_devices, data);
1560         }
1561         if (!error)
1562                 error = security_sb_set_mnt_opts(s, new_sec_opts, 0, NULL);
1563         security_free_mnt_opts(&new_sec_opts);
1564         if (error) {
1565                 deactivate_locked_super(s);
1566                 return ERR_PTR(error);
1567         }
1568
1569         return dget(s->s_root);
1570
1571 error_close_devices:
1572         btrfs_close_devices(fs_devices);
1573 error_fs_info:
1574         free_fs_info(fs_info);
1575 error_sec_opts:
1576         security_free_mnt_opts(&new_sec_opts);
1577         return ERR_PTR(error);
1578 }
1579
1580 /*
1581  * Mount function which is called by VFS layer.
1582  *
1583  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1584  * which needs vfsmount* of device's root (/).  This means device's root has to
1585  * be mounted internally in any case.
1586  *
1587  * Operation flow:
1588  *   1. Parse subvol id related options for later use in mount_subvol().
1589  *
1590  *   2. Mount device's root (/) by calling vfs_kern_mount().
1591  *
1592  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1593  *      first place. In order to avoid calling btrfs_mount() again, we use
1594  *      different file_system_type which is not registered to VFS by
1595  *      register_filesystem() (btrfs_root_fs_type). As a result,
1596  *      btrfs_mount_root() is called. The return value will be used by
1597  *      mount_subtree() in mount_subvol().
1598  *
1599  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1600  *      "btrfs subvolume set-default", mount_subvol() is called always.
1601  */
1602 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1603                 const char *device_name, void *data)
1604 {
1605         struct vfsmount *mnt_root;
1606         struct dentry *root;
1607         char *subvol_name = NULL;
1608         u64 subvol_objectid = 0;
1609         int error = 0;
1610
1611         error = btrfs_parse_subvol_options(data, &subvol_name,
1612                                         &subvol_objectid);
1613         if (error) {
1614                 kfree(subvol_name);
1615                 return ERR_PTR(error);
1616         }
1617
1618         /* mount device's root (/) */
1619         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1620         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1621                 if (flags & SB_RDONLY) {
1622                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1623                                 flags & ~SB_RDONLY, device_name, data);
1624                 } else {
1625                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1626                                 flags | SB_RDONLY, device_name, data);
1627                         if (IS_ERR(mnt_root)) {
1628                                 root = ERR_CAST(mnt_root);
1629                                 kfree(subvol_name);
1630                                 goto out;
1631                         }
1632
1633                         down_write(&mnt_root->mnt_sb->s_umount);
1634                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1635                         up_write(&mnt_root->mnt_sb->s_umount);
1636                         if (error < 0) {
1637                                 root = ERR_PTR(error);
1638                                 mntput(mnt_root);
1639                                 kfree(subvol_name);
1640                                 goto out;
1641                         }
1642                 }
1643         }
1644         if (IS_ERR(mnt_root)) {
1645                 root = ERR_CAST(mnt_root);
1646                 kfree(subvol_name);
1647                 goto out;
1648         }
1649
1650         /* mount_subvol() will free subvol_name and mnt_root */
1651         root = mount_subvol(subvol_name, subvol_objectid, mnt_root);
1652
1653 out:
1654         return root;
1655 }
1656
1657 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1658                                      u32 new_pool_size, u32 old_pool_size)
1659 {
1660         if (new_pool_size == old_pool_size)
1661                 return;
1662
1663         fs_info->thread_pool_size = new_pool_size;
1664
1665         btrfs_info(fs_info, "resize thread pool %d -> %d",
1666                old_pool_size, new_pool_size);
1667
1668         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1669         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1670         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1671         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1672         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1673         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1674         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1675                                 new_pool_size);
1676         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1677         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1678         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1679         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1680         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1681                                 new_pool_size);
1682 }
1683
1684 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1685 {
1686         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1687 }
1688
1689 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1690                                        unsigned long old_opts, int flags)
1691 {
1692         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1693             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1694              (flags & SB_RDONLY))) {
1695                 /* wait for any defraggers to finish */
1696                 wait_event(fs_info->transaction_wait,
1697                            (atomic_read(&fs_info->defrag_running) == 0));
1698                 if (flags & SB_RDONLY)
1699                         sync_filesystem(fs_info->sb);
1700         }
1701 }
1702
1703 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1704                                          unsigned long old_opts)
1705 {
1706         /*
1707          * We need to cleanup all defragable inodes if the autodefragment is
1708          * close or the filesystem is read only.
1709          */
1710         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1711             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1712                 btrfs_cleanup_defrag_inodes(fs_info);
1713         }
1714
1715         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1716 }
1717
1718 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1719 {
1720         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1721         struct btrfs_root *root = fs_info->tree_root;
1722         unsigned old_flags = sb->s_flags;
1723         unsigned long old_opts = fs_info->mount_opt;
1724         unsigned long old_compress_type = fs_info->compress_type;
1725         u64 old_max_inline = fs_info->max_inline;
1726         u32 old_thread_pool_size = fs_info->thread_pool_size;
1727         u32 old_metadata_ratio = fs_info->metadata_ratio;
1728         int ret;
1729
1730         sync_filesystem(sb);
1731         btrfs_remount_prepare(fs_info);
1732
1733         if (data) {
1734                 void *new_sec_opts = NULL;
1735
1736                 ret = security_sb_eat_lsm_opts(data, &new_sec_opts);
1737                 if (!ret)
1738                         ret = security_sb_remount(sb, new_sec_opts);
1739                 security_free_mnt_opts(&new_sec_opts);
1740                 if (ret)
1741                         goto restore;
1742         }
1743
1744         ret = btrfs_parse_options(fs_info, data, *flags);
1745         if (ret)
1746                 goto restore;
1747
1748         btrfs_remount_begin(fs_info, old_opts, *flags);
1749         btrfs_resize_thread_pool(fs_info,
1750                 fs_info->thread_pool_size, old_thread_pool_size);
1751
1752         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1753                 goto out;
1754
1755         if (*flags & SB_RDONLY) {
1756                 /*
1757                  * this also happens on 'umount -rf' or on shutdown, when
1758                  * the filesystem is busy.
1759                  */
1760                 cancel_work_sync(&fs_info->async_reclaim_work);
1761
1762                 /* wait for the uuid_scan task to finish */
1763                 down(&fs_info->uuid_tree_rescan_sem);
1764                 /* avoid complains from lockdep et al. */
1765                 up(&fs_info->uuid_tree_rescan_sem);
1766
1767                 sb->s_flags |= SB_RDONLY;
1768
1769                 /*
1770                  * Setting SB_RDONLY will put the cleaner thread to
1771                  * sleep at the next loop if it's already active.
1772                  * If it's already asleep, we'll leave unused block
1773                  * groups on disk until we're mounted read-write again
1774                  * unless we clean them up here.
1775                  */
1776                 btrfs_delete_unused_bgs(fs_info);
1777
1778                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1779                 btrfs_scrub_cancel(fs_info);
1780                 btrfs_pause_balance(fs_info);
1781
1782                 ret = btrfs_commit_super(fs_info);
1783                 if (ret)
1784                         goto restore;
1785         } else {
1786                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1787                         btrfs_err(fs_info,
1788                                 "Remounting read-write after error is not allowed");
1789                         ret = -EINVAL;
1790                         goto restore;
1791                 }
1792                 if (fs_info->fs_devices->rw_devices == 0) {
1793                         ret = -EACCES;
1794                         goto restore;
1795                 }
1796
1797                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1798                         btrfs_warn(fs_info,
1799                 "too many missing devices, writable remount is not allowed");
1800                         ret = -EACCES;
1801                         goto restore;
1802                 }
1803
1804                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1805                         ret = -EINVAL;
1806                         goto restore;
1807                 }
1808
1809                 ret = btrfs_cleanup_fs_roots(fs_info);
1810                 if (ret)
1811                         goto restore;
1812
1813                 /* recover relocation */
1814                 mutex_lock(&fs_info->cleaner_mutex);
1815                 ret = btrfs_recover_relocation(root);
1816                 mutex_unlock(&fs_info->cleaner_mutex);
1817                 if (ret)
1818                         goto restore;
1819
1820                 ret = btrfs_resume_balance_async(fs_info);
1821                 if (ret)
1822                         goto restore;
1823
1824                 ret = btrfs_resume_dev_replace_async(fs_info);
1825                 if (ret) {
1826                         btrfs_warn(fs_info, "failed to resume dev_replace");
1827                         goto restore;
1828                 }
1829
1830                 btrfs_qgroup_rescan_resume(fs_info);
1831
1832                 if (!fs_info->uuid_root) {
1833                         btrfs_info(fs_info, "creating UUID tree");
1834                         ret = btrfs_create_uuid_tree(fs_info);
1835                         if (ret) {
1836                                 btrfs_warn(fs_info,
1837                                            "failed to create the UUID tree %d",
1838                                            ret);
1839                                 goto restore;
1840                         }
1841                 }
1842                 sb->s_flags &= ~SB_RDONLY;
1843
1844                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1845         }
1846 out:
1847         wake_up_process(fs_info->transaction_kthread);
1848         btrfs_remount_cleanup(fs_info, old_opts);
1849         return 0;
1850
1851 restore:
1852         /* We've hit an error - don't reset SB_RDONLY */
1853         if (sb_rdonly(sb))
1854                 old_flags |= SB_RDONLY;
1855         sb->s_flags = old_flags;
1856         fs_info->mount_opt = old_opts;
1857         fs_info->compress_type = old_compress_type;
1858         fs_info->max_inline = old_max_inline;
1859         btrfs_resize_thread_pool(fs_info,
1860                 old_thread_pool_size, fs_info->thread_pool_size);
1861         fs_info->metadata_ratio = old_metadata_ratio;
1862         btrfs_remount_cleanup(fs_info, old_opts);
1863         return ret;
1864 }
1865
1866 /* Used to sort the devices by max_avail(descending sort) */
1867 static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
1868                                        const void *dev_info2)
1869 {
1870         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1871             ((struct btrfs_device_info *)dev_info2)->max_avail)
1872                 return -1;
1873         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1874                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1875                 return 1;
1876         else
1877         return 0;
1878 }
1879
1880 /*
1881  * sort the devices by max_avail, in which max free extent size of each device
1882  * is stored.(Descending Sort)
1883  */
1884 static inline void btrfs_descending_sort_devices(
1885                                         struct btrfs_device_info *devices,
1886                                         size_t nr_devices)
1887 {
1888         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1889              btrfs_cmp_device_free_bytes, NULL);
1890 }
1891
1892 /*
1893  * The helper to calc the free space on the devices that can be used to store
1894  * file data.
1895  */
1896 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1897                                               u64 *free_bytes)
1898 {
1899         struct btrfs_device_info *devices_info;
1900         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1901         struct btrfs_device *device;
1902         u64 skip_space;
1903         u64 type;
1904         u64 avail_space;
1905         u64 min_stripe_size;
1906         int min_stripes, num_stripes = 1;
1907         int i = 0, nr_devices;
1908         const struct btrfs_raid_attr *rattr;
1909
1910         /*
1911          * We aren't under the device list lock, so this is racy-ish, but good
1912          * enough for our purposes.
1913          */
1914         nr_devices = fs_info->fs_devices->open_devices;
1915         if (!nr_devices) {
1916                 smp_mb();
1917                 nr_devices = fs_info->fs_devices->open_devices;
1918                 ASSERT(nr_devices);
1919                 if (!nr_devices) {
1920                         *free_bytes = 0;
1921                         return 0;
1922                 }
1923         }
1924
1925         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1926                                GFP_KERNEL);
1927         if (!devices_info)
1928                 return -ENOMEM;
1929
1930         /* calc min stripe number for data space allocation */
1931         type = btrfs_data_alloc_profile(fs_info);
1932         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
1933         min_stripes = rattr->devs_min;
1934
1935         if (type & BTRFS_BLOCK_GROUP_RAID0)
1936                 num_stripes = nr_devices;
1937         else if (type & BTRFS_BLOCK_GROUP_RAID1)
1938                 num_stripes = 2;
1939         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1940                 num_stripes = 4;
1941
1942         /* Adjust for more than 1 stripe per device */
1943         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
1944
1945         rcu_read_lock();
1946         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1947                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1948                                                 &device->dev_state) ||
1949                     !device->bdev ||
1950                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1951                         continue;
1952
1953                 if (i >= nr_devices)
1954                         break;
1955
1956                 avail_space = device->total_bytes - device->bytes_used;
1957
1958                 /* align with stripe_len */
1959                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1960                 avail_space *= BTRFS_STRIPE_LEN;
1961
1962                 /*
1963                  * In order to avoid overwriting the superblock on the drive,
1964                  * btrfs starts at an offset of at least 1MB when doing chunk
1965                  * allocation.
1966                  */
1967                 skip_space = SZ_1M;
1968
1969                 /*
1970                  * we can use the free space in [0, skip_space - 1], subtract
1971                  * it from the total.
1972                  */
1973                 if (avail_space && avail_space >= skip_space)
1974                         avail_space -= skip_space;
1975                 else
1976                         avail_space = 0;
1977
1978                 if (avail_space < min_stripe_size)
1979                         continue;
1980
1981                 devices_info[i].dev = device;
1982                 devices_info[i].max_avail = avail_space;
1983
1984                 i++;
1985         }
1986         rcu_read_unlock();
1987
1988         nr_devices = i;
1989
1990         btrfs_descending_sort_devices(devices_info, nr_devices);
1991
1992         i = nr_devices - 1;
1993         avail_space = 0;
1994         while (nr_devices >= min_stripes) {
1995                 if (num_stripes > nr_devices)
1996                         num_stripes = nr_devices;
1997
1998                 if (devices_info[i].max_avail >= min_stripe_size) {
1999                         int j;
2000                         u64 alloc_size;
2001
2002                         avail_space += devices_info[i].max_avail * num_stripes;
2003                         alloc_size = devices_info[i].max_avail;
2004                         for (j = i + 1 - num_stripes; j <= i; j++)
2005                                 devices_info[j].max_avail -= alloc_size;
2006                 }
2007                 i--;
2008                 nr_devices--;
2009         }
2010
2011         kfree(devices_info);
2012         *free_bytes = avail_space;
2013         return 0;
2014 }
2015
2016 /*
2017  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2018  *
2019  * If there's a redundant raid level at DATA block groups, use the respective
2020  * multiplier to scale the sizes.
2021  *
2022  * Unused device space usage is based on simulating the chunk allocator
2023  * algorithm that respects the device sizes and order of allocations.  This is
2024  * a close approximation of the actual use but there are other factors that may
2025  * change the result (like a new metadata chunk).
2026  *
2027  * If metadata is exhausted, f_bavail will be 0.
2028  */
2029 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2030 {
2031         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2032         struct btrfs_super_block *disk_super = fs_info->super_copy;
2033         struct list_head *head = &fs_info->space_info;
2034         struct btrfs_space_info *found;
2035         u64 total_used = 0;
2036         u64 total_free_data = 0;
2037         u64 total_free_meta = 0;
2038         int bits = dentry->d_sb->s_blocksize_bits;
2039         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
2040         unsigned factor = 1;
2041         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2042         int ret;
2043         u64 thresh = 0;
2044         int mixed = 0;
2045
2046         rcu_read_lock();
2047         list_for_each_entry_rcu(found, head, list) {
2048                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2049                         int i;
2050
2051                         total_free_data += found->disk_total - found->disk_used;
2052                         total_free_data -=
2053                                 btrfs_account_ro_block_groups_free_space(found);
2054
2055                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2056                                 if (!list_empty(&found->block_groups[i]))
2057                                         factor = btrfs_bg_type_to_factor(
2058                                                 btrfs_raid_array[i].bg_flag);
2059                         }
2060                 }
2061
2062                 /*
2063                  * Metadata in mixed block goup profiles are accounted in data
2064                  */
2065                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2066                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2067                                 mixed = 1;
2068                         else
2069                                 total_free_meta += found->disk_total -
2070                                         found->disk_used;
2071                 }
2072
2073                 total_used += found->disk_used;
2074         }
2075
2076         rcu_read_unlock();
2077
2078         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2079         buf->f_blocks >>= bits;
2080         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2081
2082         /* Account global block reserve as used, it's in logical size already */
2083         spin_lock(&block_rsv->lock);
2084         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2085         if (buf->f_bfree >= block_rsv->size >> bits)
2086                 buf->f_bfree -= block_rsv->size >> bits;
2087         else
2088                 buf->f_bfree = 0;
2089         spin_unlock(&block_rsv->lock);
2090
2091         buf->f_bavail = div_u64(total_free_data, factor);
2092         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2093         if (ret)
2094                 return ret;
2095         buf->f_bavail += div_u64(total_free_data, factor);
2096         buf->f_bavail = buf->f_bavail >> bits;
2097
2098         /*
2099          * We calculate the remaining metadata space minus global reserve. If
2100          * this is (supposedly) smaller than zero, there's no space. But this
2101          * does not hold in practice, the exhausted state happens where's still
2102          * some positive delta. So we apply some guesswork and compare the
2103          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2104          *
2105          * We probably cannot calculate the exact threshold value because this
2106          * depends on the internal reservations requested by various
2107          * operations, so some operations that consume a few metadata will
2108          * succeed even if the Avail is zero. But this is better than the other
2109          * way around.
2110          */
2111         thresh = SZ_4M;
2112
2113         if (!mixed && total_free_meta - thresh < block_rsv->size)
2114                 buf->f_bavail = 0;
2115
2116         buf->f_type = BTRFS_SUPER_MAGIC;
2117         buf->f_bsize = dentry->d_sb->s_blocksize;
2118         buf->f_namelen = BTRFS_NAME_LEN;
2119
2120         /* We treat it as constant endianness (it doesn't matter _which_)
2121            because we want the fsid to come out the same whether mounted
2122            on a big-endian or little-endian host */
2123         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2124         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2125         /* Mask in the root object ID too, to disambiguate subvols */
2126         buf->f_fsid.val[0] ^=
2127                 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
2128         buf->f_fsid.val[1] ^=
2129                 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
2130
2131         return 0;
2132 }
2133
2134 static void btrfs_kill_super(struct super_block *sb)
2135 {
2136         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2137         kill_anon_super(sb);
2138         free_fs_info(fs_info);
2139 }
2140
2141 static struct file_system_type btrfs_fs_type = {
2142         .owner          = THIS_MODULE,
2143         .name           = "btrfs",
2144         .mount          = btrfs_mount,
2145         .kill_sb        = btrfs_kill_super,
2146         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2147 };
2148
2149 static struct file_system_type btrfs_root_fs_type = {
2150         .owner          = THIS_MODULE,
2151         .name           = "btrfs",
2152         .mount          = btrfs_mount_root,
2153         .kill_sb        = btrfs_kill_super,
2154         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2155 };
2156
2157 MODULE_ALIAS_FS("btrfs");
2158
2159 static int btrfs_control_open(struct inode *inode, struct file *file)
2160 {
2161         /*
2162          * The control file's private_data is used to hold the
2163          * transaction when it is started and is used to keep
2164          * track of whether a transaction is already in progress.
2165          */
2166         file->private_data = NULL;
2167         return 0;
2168 }
2169
2170 /*
2171  * used by btrfsctl to scan devices when no FS is mounted
2172  */
2173 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2174                                 unsigned long arg)
2175 {
2176         struct btrfs_ioctl_vol_args *vol;
2177         struct btrfs_device *device = NULL;
2178         int ret = -ENOTTY;
2179
2180         if (!capable(CAP_SYS_ADMIN))
2181                 return -EPERM;
2182
2183         vol = memdup_user((void __user *)arg, sizeof(*vol));
2184         if (IS_ERR(vol))
2185                 return PTR_ERR(vol);
2186         vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2187
2188         switch (cmd) {
2189         case BTRFS_IOC_SCAN_DEV:
2190                 mutex_lock(&uuid_mutex);
2191                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2192                                                &btrfs_root_fs_type);
2193                 ret = PTR_ERR_OR_ZERO(device);
2194                 mutex_unlock(&uuid_mutex);
2195                 break;
2196         case BTRFS_IOC_FORGET_DEV:
2197                 ret = btrfs_forget_devices(vol->name);
2198                 break;
2199         case BTRFS_IOC_DEVICES_READY:
2200                 mutex_lock(&uuid_mutex);
2201                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2202                                                &btrfs_root_fs_type);
2203                 if (IS_ERR(device)) {
2204                         mutex_unlock(&uuid_mutex);
2205                         ret = PTR_ERR(device);
2206                         break;
2207                 }
2208                 ret = !(device->fs_devices->num_devices ==
2209                         device->fs_devices->total_devices);
2210                 mutex_unlock(&uuid_mutex);
2211                 break;
2212         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2213                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2214                 break;
2215         }
2216
2217         kfree(vol);
2218         return ret;
2219 }
2220
2221 static int btrfs_freeze(struct super_block *sb)
2222 {
2223         struct btrfs_trans_handle *trans;
2224         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2225         struct btrfs_root *root = fs_info->tree_root;
2226
2227         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2228         /*
2229          * We don't need a barrier here, we'll wait for any transaction that
2230          * could be in progress on other threads (and do delayed iputs that
2231          * we want to avoid on a frozen filesystem), or do the commit
2232          * ourselves.
2233          */
2234         trans = btrfs_attach_transaction_barrier(root);
2235         if (IS_ERR(trans)) {
2236                 /* no transaction, don't bother */
2237                 if (PTR_ERR(trans) == -ENOENT)
2238                         return 0;
2239                 return PTR_ERR(trans);
2240         }
2241         return btrfs_commit_transaction(trans);
2242 }
2243
2244 static int btrfs_unfreeze(struct super_block *sb)
2245 {
2246         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2247
2248         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2249         return 0;
2250 }
2251
2252 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2253 {
2254         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2255         struct btrfs_fs_devices *cur_devices;
2256         struct btrfs_device *dev, *first_dev = NULL;
2257         struct list_head *head;
2258
2259         /*
2260          * Lightweight locking of the devices. We should not need
2261          * device_list_mutex here as we only read the device data and the list
2262          * is protected by RCU.  Even if a device is deleted during the list
2263          * traversals, we'll get valid data, the freeing callback will wait at
2264          * least until the rcu_read_unlock.
2265          */
2266         rcu_read_lock();
2267         cur_devices = fs_info->fs_devices;
2268         while (cur_devices) {
2269                 head = &cur_devices->devices;
2270                 list_for_each_entry_rcu(dev, head, dev_list) {
2271                         if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2272                                 continue;
2273                         if (!dev->name)
2274                                 continue;
2275                         if (!first_dev || dev->devid < first_dev->devid)
2276                                 first_dev = dev;
2277                 }
2278                 cur_devices = cur_devices->seed;
2279         }
2280
2281         if (first_dev)
2282                 seq_escape(m, rcu_str_deref(first_dev->name), " \t\n\\");
2283         else
2284                 WARN_ON(1);
2285         rcu_read_unlock();
2286         return 0;
2287 }
2288
2289 static const struct super_operations btrfs_super_ops = {
2290         .drop_inode     = btrfs_drop_inode,
2291         .evict_inode    = btrfs_evict_inode,
2292         .put_super      = btrfs_put_super,
2293         .sync_fs        = btrfs_sync_fs,
2294         .show_options   = btrfs_show_options,
2295         .show_devname   = btrfs_show_devname,
2296         .alloc_inode    = btrfs_alloc_inode,
2297         .destroy_inode  = btrfs_destroy_inode,
2298         .free_inode     = btrfs_free_inode,
2299         .statfs         = btrfs_statfs,
2300         .remount_fs     = btrfs_remount,
2301         .freeze_fs      = btrfs_freeze,
2302         .unfreeze_fs    = btrfs_unfreeze,
2303 };
2304
2305 static const struct file_operations btrfs_ctl_fops = {
2306         .open = btrfs_control_open,
2307         .unlocked_ioctl  = btrfs_control_ioctl,
2308         .compat_ioctl = btrfs_control_ioctl,
2309         .owner   = THIS_MODULE,
2310         .llseek = noop_llseek,
2311 };
2312
2313 static struct miscdevice btrfs_misc = {
2314         .minor          = BTRFS_MINOR,
2315         .name           = "btrfs-control",
2316         .fops           = &btrfs_ctl_fops
2317 };
2318
2319 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2320 MODULE_ALIAS("devname:btrfs-control");
2321
2322 static int __init btrfs_interface_init(void)
2323 {
2324         return misc_register(&btrfs_misc);
2325 }
2326
2327 static __cold void btrfs_interface_exit(void)
2328 {
2329         misc_deregister(&btrfs_misc);
2330 }
2331
2332 static void __init btrfs_print_mod_info(void)
2333 {
2334         static const char options[] = ""
2335 #ifdef CONFIG_BTRFS_DEBUG
2336                         ", debug=on"
2337 #endif
2338 #ifdef CONFIG_BTRFS_ASSERT
2339                         ", assert=on"
2340 #endif
2341 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2342                         ", integrity-checker=on"
2343 #endif
2344 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2345                         ", ref-verify=on"
2346 #endif
2347                         ;
2348         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2349 }
2350
2351 static int __init init_btrfs_fs(void)
2352 {
2353         int err;
2354
2355         btrfs_props_init();
2356
2357         err = btrfs_init_sysfs();
2358         if (err)
2359                 return err;
2360
2361         btrfs_init_compress();
2362
2363         err = btrfs_init_cachep();
2364         if (err)
2365                 goto free_compress;
2366
2367         err = extent_io_init();
2368         if (err)
2369                 goto free_cachep;
2370
2371         err = extent_map_init();
2372         if (err)
2373                 goto free_extent_io;
2374
2375         err = ordered_data_init();
2376         if (err)
2377                 goto free_extent_map;
2378
2379         err = btrfs_delayed_inode_init();
2380         if (err)
2381                 goto free_ordered_data;
2382
2383         err = btrfs_auto_defrag_init();
2384         if (err)
2385                 goto free_delayed_inode;
2386
2387         err = btrfs_delayed_ref_init();
2388         if (err)
2389                 goto free_auto_defrag;
2390
2391         err = btrfs_prelim_ref_init();
2392         if (err)
2393                 goto free_delayed_ref;
2394
2395         err = btrfs_end_io_wq_init();
2396         if (err)
2397                 goto free_prelim_ref;
2398
2399         err = btrfs_interface_init();
2400         if (err)
2401                 goto free_end_io_wq;
2402
2403         btrfs_init_lockdep();
2404
2405         btrfs_print_mod_info();
2406
2407         err = btrfs_run_sanity_tests();
2408         if (err)
2409                 goto unregister_ioctl;
2410
2411         err = register_filesystem(&btrfs_fs_type);
2412         if (err)
2413                 goto unregister_ioctl;
2414
2415         return 0;
2416
2417 unregister_ioctl:
2418         btrfs_interface_exit();
2419 free_end_io_wq:
2420         btrfs_end_io_wq_exit();
2421 free_prelim_ref:
2422         btrfs_prelim_ref_exit();
2423 free_delayed_ref:
2424         btrfs_delayed_ref_exit();
2425 free_auto_defrag:
2426         btrfs_auto_defrag_exit();
2427 free_delayed_inode:
2428         btrfs_delayed_inode_exit();
2429 free_ordered_data:
2430         ordered_data_exit();
2431 free_extent_map:
2432         extent_map_exit();
2433 free_extent_io:
2434         extent_io_exit();
2435 free_cachep:
2436         btrfs_destroy_cachep();
2437 free_compress:
2438         btrfs_exit_compress();
2439         btrfs_exit_sysfs();
2440
2441         return err;
2442 }
2443
2444 static void __exit exit_btrfs_fs(void)
2445 {
2446         btrfs_destroy_cachep();
2447         btrfs_delayed_ref_exit();
2448         btrfs_auto_defrag_exit();
2449         btrfs_delayed_inode_exit();
2450         btrfs_prelim_ref_exit();
2451         ordered_data_exit();
2452         extent_map_exit();
2453         extent_io_exit();
2454         btrfs_interface_exit();
2455         btrfs_end_io_wq_exit();
2456         unregister_filesystem(&btrfs_fs_type);
2457         btrfs_exit_sysfs();
2458         btrfs_cleanup_fs_uuids();
2459         btrfs_exit_compress();
2460 }
2461
2462 late_initcall(init_btrfs_fs);
2463 module_exit(exit_btrfs_fs)
2464
2465 MODULE_LICENSE("GPL");
2466 MODULE_SOFTDEP("pre: crc32c");