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