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