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/ratelimit.h>
  27 #include <linux/crc32c.h>
  28 #include <linux/btrfs.h>
  29 #include "delayed-inode.h"
  30 #include "ctree.h"
  31 #include "disk-io.h"
  32 #include "transaction.h"
  33 #include "btrfs_inode.h"
  34 #include "print-tree.h"
  35 #include "props.h"
  36 #include "xattr.h"
  37 #include "volumes.h"
  38 #include "export.h"
  39 #include "compression.h"
  40 #include "rcu-string.h"
  41 #include "dev-replace.h"
  42 #include "free-space-cache.h"
  43 #include "backref.h"
  44 #include "space-info.h"
  45 #include "sysfs.h"
  46 #include "zoned.h"
  47 #include "tests/btrfs-tests.h"
  48 #include "block-group.h"
  49 #include "discard.h"
  50 #include "qgroup.h"
  51 #define CREATE_TRACE_POINTS
  52 #include <trace/events/btrfs.h>
  53
  54 static const struct super_operations btrfs_super_ops;
  55
  56 /*
  57  * Types for mounting the default subvolume and a subvolume explicitly
  58  * requested by subvol=/path. That way the callchain is straightforward and we
  59  * don't have to play tricks with the mount options and recursive calls to
  60  * btrfs_mount.
  61  *
  62  * The new btrfs_root_fs_type also servers as a tag for the bdev_holder.
  63  */
  64 static struct file_system_type btrfs_fs_type;
  65 static struct file_system_type btrfs_root_fs_type;
  66
  67 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  68
  69 #ifdef CONFIG_PRINTK
  70
  71 #define STATE_STRING_PREFACE    ": state "
  72 #define STATE_STRING_BUF_LEN    (sizeof(STATE_STRING_PREFACE) + BTRFS_FS_STATE_COUNT)
  73
  74 /*
  75  * Characters to print to indicate error conditions or uncommon filesystem sate.
  76  * RO is not an error.
  77  */
  78 static const char fs_state_chars[] = {
  79         [BTRFS_FS_STATE_ERROR]                  = 'E',
  80         [BTRFS_FS_STATE_REMOUNTING]             = 'M',
  81         [BTRFS_FS_STATE_RO]                     = 0,
  82         [BTRFS_FS_STATE_TRANS_ABORTED]          = 'A',
  83         [BTRFS_FS_STATE_DEV_REPLACING]          = 'R',
  84         [BTRFS_FS_STATE_DUMMY_FS_INFO]          = 0,
  85         [BTRFS_FS_STATE_NO_CSUMS]               = 'C',
  86         [BTRFS_FS_STATE_LOG_CLEANUP_ERROR]      = 'L',
  87 };
  88
  89 static void btrfs_state_to_string(const struct btrfs_fs_info *info, char *buf)
  90 {
  91         unsigned int bit;
  92         bool states_printed = false;
  93         unsigned long fs_state = READ_ONCE(info->fs_state);
  94         char *curr = buf;
  95
  96         memcpy(curr, STATE_STRING_PREFACE, sizeof(STATE_STRING_PREFACE));
  97         curr += sizeof(STATE_STRING_PREFACE) - 1;
  98
  99         for_each_set_bit(bit, &fs_state, sizeof(fs_state)) {
 100                 WARN_ON_ONCE(bit >= BTRFS_FS_STATE_COUNT);
 101                 if ((bit < BTRFS_FS_STATE_COUNT) && fs_state_chars[bit]) {
 102                         *curr++ = fs_state_chars[bit];
 103                         states_printed = true;
 104                 }
 105         }
 106
 107         /* If no states were printed, reset the buffer */
 108         if (!states_printed)
 109                 curr = buf;
 110
 111         *curr++ = 0;
 112 }
 113 #endif
 114
 115 /*
 116  * Generally the error codes correspond to their respective errors, but there
 117  * are a few special cases.
 118  *
 119  * EUCLEAN: Any sort of corruption that we encounter.  The tree-checker for
 120  *          instance will return EUCLEAN if any of the blocks are corrupted in
 121  *          a way that is problematic.  We want to reserve EUCLEAN for these
 122  *          sort of corruptions.
 123  *
 124  * EROFS: If we check BTRFS_FS_STATE_ERROR and fail out with a return error, we
 125  *        need to use EROFS for this case.  We will have no idea of the
 126  *        original failure, that will have been reported at the time we tripped
 127  *        over the error.  Each subsequent error that doesn't have any context
 128  *        of the original error should use EROFS when handling BTRFS_FS_STATE_ERROR.
 129  */
 130 const char * __attribute_const__ btrfs_decode_error(int errno)
 131 {
 132         char *errstr = "unknown";
 133
 134         switch (errno) {
 135         case -ENOENT:           /* -2 */
 136                 errstr = "No such entry";
 137                 break;
 138         case -EIO:              /* -5 */
 139                 errstr = "IO failure";
 140                 break;
 141         case -ENOMEM:           /* -12*/
 142                 errstr = "Out of memory";
 143                 break;
 144         case -EEXIST:           /* -17 */
 145                 errstr = "Object already exists";
 146                 break;
 147         case -ENOSPC:           /* -28 */
 148                 errstr = "No space left";
 149                 break;
 150         case -EROFS:            /* -30 */
 151                 errstr = "Readonly filesystem";
 152                 break;
 153         case -EOPNOTSUPP:       /* -95 */
 154                 errstr = "Operation not supported";
 155                 break;
 156         case -EUCLEAN:          /* -117 */
 157                 errstr = "Filesystem corrupted";
 158                 break;
 159         case -EDQUOT:           /* -122 */
 160                 errstr = "Quota exceeded";
 161                 break;
 162         }
 163
 164         return errstr;
 165 }
 166
 167 /*
 168  * __btrfs_handle_fs_error decodes expected errors from the caller and
 169  * invokes the appropriate error response.
 170  */
 171 __cold
 172 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
 173                        unsigned int line, int errno, const char *fmt, ...)
 174 {
 175         struct super_block *sb = fs_info->sb;
 176 #ifdef CONFIG_PRINTK
 177         char statestr[STATE_STRING_BUF_LEN];
 178         const char *errstr;
 179 #endif
 180
 181         /*
 182          * Special case: if the error is EROFS, and we're already
 183          * under SB_RDONLY, then it is safe here.
 184          */
 185         if (errno == -EROFS && sb_rdonly(sb))
 186                 return;
 187
 188 #ifdef CONFIG_PRINTK
 189         errstr = btrfs_decode_error(errno);
 190         btrfs_state_to_string(fs_info, statestr);
 191         if (fmt) {
 192                 struct va_format vaf;
 193                 va_list args;
 194
 195                 va_start(args, fmt);
 196                 vaf.fmt = fmt;
 197                 vaf.va = &args;
 198
 199                 pr_crit("BTRFS: error (device %s%s) in %s:%d: errno=%d %s (%pV)\n",
 200                         sb->s_id, statestr, function, line, errno, errstr, &vaf);
 201                 va_end(args);
 202         } else {
 203                 pr_crit("BTRFS: error (device %s%s) in %s:%d: errno=%d %s\n",
 204                         sb->s_id, statestr, function, line, errno, errstr);
 205         }
 206 #endif
 207
 208         /*
 209          * Today we only save the error info to memory.  Long term we'll
 210          * also send it down to the disk
 211          */
 212         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 213
 214         /* Don't go through full error handling during mount */
 215         if (!(sb->s_flags & SB_BORN))
 216                 return;
 217
 218         if (sb_rdonly(sb))
 219                 return;
 220
 221         btrfs_discard_stop(fs_info);
 222
 223         /* btrfs handle error by forcing the filesystem readonly */
 224         btrfs_set_sb_rdonly(sb);
 225         btrfs_info(fs_info, "forced readonly");
 226         /*
 227          * Note that a running device replace operation is not canceled here
 228          * although there is no way to update the progress. It would add the
 229          * risk of a deadlock, therefore the canceling is omitted. The only
 230          * penalty is that some I/O remains active until the procedure
 231          * completes. The next time when the filesystem is mounted writable
 232          * again, the device replace operation continues.
 233          */
 234 }
 235
 236 #ifdef CONFIG_PRINTK
 237 static const char * const logtypes[] = {
 238         "emergency",
 239         "alert",
 240         "critical",
 241         "error",
 242         "warning",
 243         "notice",
 244         "info",
 245         "debug",
 246 };
 247
 248
 249 /*
 250  * Use one ratelimit state per log level so that a flood of less important
 251  * messages doesn't cause more important ones to be dropped.
 252  */
 253 static struct ratelimit_state printk_limits[] = {
 254         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
 255         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
 256         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
 257         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
 258         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
 259         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
 260         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
 261         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
 262 };
 263
 264 void __cold btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 265 {
 266         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
 267         struct va_format vaf;
 268         va_list args;
 269         int kern_level;
 270         const char *type = logtypes[4];
 271         struct ratelimit_state *ratelimit = &printk_limits[4];
 272
 273         va_start(args, fmt);
 274
 275         while ((kern_level = printk_get_level(fmt)) != 0) {
 276                 size_t size = printk_skip_level(fmt) - fmt;
 277
 278                 if (kern_level >= '0' && kern_level <= '7') {
 279                         memcpy(lvl, fmt,  size);
 280                         lvl[size] = '\0';
 281                         type = logtypes[kern_level - '0'];
 282                         ratelimit = &printk_limits[kern_level - '0'];
 283                 }
 284                 fmt += size;
 285         }
 286
 287         vaf.fmt = fmt;
 288         vaf.va = &args;
 289
 290         if (__ratelimit(ratelimit)) {
 291                 if (fs_info) {
 292                         char statestr[STATE_STRING_BUF_LEN];
 293
 294                         btrfs_state_to_string(fs_info, statestr);
 295                         printk("%sBTRFS %s (device %s%s): %pV\n", lvl, type,
 296                                 fs_info->sb->s_id, statestr, &vaf);
 297                 } else {
 298                         printk("%sBTRFS %s: %pV\n", lvl, type, &vaf);
 299                 }
 300         }
 301
 302         va_end(args);
 303 }
 304 #endif
 305
 306 #if BITS_PER_LONG == 32
 307 void __cold btrfs_warn_32bit_limit(struct btrfs_fs_info *fs_info)
 308 {
 309         if (!test_and_set_bit(BTRFS_FS_32BIT_WARN, &fs_info->flags)) {
 310                 btrfs_warn(fs_info, "reaching 32bit limit for logical addresses");
 311                 btrfs_warn(fs_info,
 312 "due to page cache limit on 32bit systems, btrfs can't access metadata at or beyond %lluT",
 313                            BTRFS_32BIT_MAX_FILE_SIZE >> 40);
 314                 btrfs_warn(fs_info,
 315                            "please consider upgrading to 64bit kernel/hardware");
 316         }
 317 }
 318
 319 void __cold btrfs_err_32bit_limit(struct btrfs_fs_info *fs_info)
 320 {
 321         if (!test_and_set_bit(BTRFS_FS_32BIT_ERROR, &fs_info->flags)) {
 322                 btrfs_err(fs_info, "reached 32bit limit for logical addresses");
 323                 btrfs_err(fs_info,
 324 "due to page cache limit on 32bit systems, metadata beyond %lluT can't be accessed",
 325                           BTRFS_32BIT_MAX_FILE_SIZE >> 40);
 326                 btrfs_err(fs_info,
 327                            "please consider upgrading to 64bit kernel/hardware");
 328         }
 329 }
 330 #endif
 331
 332 /*
 333  * We only mark the transaction aborted and then set the file system read-only.
 334  * This will prevent new transactions from starting or trying to join this
 335  * one.
 336  *
 337  * This means that error recovery at the call site is limited to freeing
 338  * any local memory allocations and passing the error code up without
 339  * further cleanup. The transaction should complete as it normally would
 340  * in the call path but will return -EIO.
 341  *
 342  * We'll complete the cleanup in btrfs_end_transaction and
 343  * btrfs_commit_transaction.
 344  */
 345 __cold
 346 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 347                                const char *function,
 348                                unsigned int line, int errno)
 349 {
 350         struct btrfs_fs_info *fs_info = trans->fs_info;
 351
 352         WRITE_ONCE(trans->aborted, errno);
 353         WRITE_ONCE(trans->transaction->aborted, errno);
 354         /* Wake up anybody who may be waiting on this transaction */
 355         wake_up(&fs_info->transaction_wait);
 356         wake_up(&fs_info->transaction_blocked_wait);
 357         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
 358 }
 359 /*
 360  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 361  * issues an alert, and either panics or BUGs, depending on mount options.
 362  */
 363 __cold
 364 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 365                    unsigned int line, int errno, const char *fmt, ...)
 366 {
 367         char *s_id = "<unknown>";
 368         const char *errstr;
 369         struct va_format vaf = { .fmt = fmt };
 370         va_list args;
 371
 372         if (fs_info)
 373                 s_id = fs_info->sb->s_id;
 374
 375         va_start(args, fmt);
 376         vaf.va = &args;
 377
 378         errstr = btrfs_decode_error(errno);
 379         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
 380                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 381                         s_id, function, line, &vaf, errno, errstr);
 382
 383         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 384                    function, line, &vaf, errno, errstr);
 385         va_end(args);
 386         /* Caller calls BUG() */
 387 }
 388
 389 static void btrfs_put_super(struct super_block *sb)
 390 {
 391         close_ctree(btrfs_sb(sb));
 392 }
 393
 394 enum {
 395         Opt_acl, Opt_noacl,
 396         Opt_clear_cache,
 397         Opt_commit_interval,
 398         Opt_compress,
 399         Opt_compress_force,
 400         Opt_compress_force_type,
 401         Opt_compress_type,
 402         Opt_degraded,
 403         Opt_device,
 404         Opt_fatal_errors,
 405         Opt_flushoncommit, Opt_noflushoncommit,
 406         Opt_max_inline,
 407         Opt_barrier, Opt_nobarrier,
 408         Opt_datacow, Opt_nodatacow,
 409         Opt_datasum, Opt_nodatasum,
 410         Opt_defrag, Opt_nodefrag,
 411         Opt_discard, Opt_nodiscard,
 412         Opt_discard_mode,
 413         Opt_norecovery,
 414         Opt_ratio,
 415         Opt_rescan_uuid_tree,
 416         Opt_skip_balance,
 417         Opt_space_cache, Opt_no_space_cache,
 418         Opt_space_cache_version,
 419         Opt_ssd, Opt_nossd,
 420         Opt_ssd_spread, Opt_nossd_spread,
 421         Opt_subvol,
 422         Opt_subvol_empty,
 423         Opt_subvolid,
 424         Opt_thread_pool,
 425         Opt_treelog, Opt_notreelog,
 426         Opt_user_subvol_rm_allowed,
 427
 428         /* Rescue options */
 429         Opt_rescue,
 430         Opt_usebackuproot,
 431         Opt_nologreplay,
 432         Opt_ignorebadroots,
 433         Opt_ignoredatacsums,
 434         Opt_rescue_all,
 435
 436         /* Deprecated options */
 437         Opt_recovery,
 438         Opt_inode_cache, Opt_noinode_cache,
 439
 440         /* Debugging options */
 441         Opt_check_integrity,
 442         Opt_check_integrity_including_extent_data,
 443         Opt_check_integrity_print_mask,
 444         Opt_enospc_debug, Opt_noenospc_debug,
 445 #ifdef CONFIG_BTRFS_DEBUG
 446         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 447 #endif
 448 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 449         Opt_ref_verify,
 450 #endif
 451         Opt_err,
 452 };
 453
 454 static const match_table_t tokens = {
 455         {Opt_acl, "acl"},
 456         {Opt_noacl, "noacl"},
 457         {Opt_clear_cache, "clear_cache"},
 458         {Opt_commit_interval, "commit=%u"},
 459         {Opt_compress, "compress"},
 460         {Opt_compress_type, "compress=%s"},
 461         {Opt_compress_force, "compress-force"},
 462         {Opt_compress_force_type, "compress-force=%s"},
 463         {Opt_degraded, "degraded"},
 464         {Opt_device, "device=%s"},
 465         {Opt_fatal_errors, "fatal_errors=%s"},
 466         {Opt_flushoncommit, "flushoncommit"},
 467         {Opt_noflushoncommit, "noflushoncommit"},
 468         {Opt_inode_cache, "inode_cache"},
 469         {Opt_noinode_cache, "noinode_cache"},
 470         {Opt_max_inline, "max_inline=%s"},
 471         {Opt_barrier, "barrier"},
 472         {Opt_nobarrier, "nobarrier"},
 473         {Opt_datacow, "datacow"},
 474         {Opt_nodatacow, "nodatacow"},
 475         {Opt_datasum, "datasum"},
 476         {Opt_nodatasum, "nodatasum"},
 477         {Opt_defrag, "autodefrag"},
 478         {Opt_nodefrag, "noautodefrag"},
 479         {Opt_discard, "discard"},
 480         {Opt_discard_mode, "discard=%s"},
 481         {Opt_nodiscard, "nodiscard"},
 482         {Opt_norecovery, "norecovery"},
 483         {Opt_ratio, "metadata_ratio=%u"},
 484         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 485         {Opt_skip_balance, "skip_balance"},
 486         {Opt_space_cache, "space_cache"},
 487         {Opt_no_space_cache, "nospace_cache"},
 488         {Opt_space_cache_version, "space_cache=%s"},
 489         {Opt_ssd, "ssd"},
 490         {Opt_nossd, "nossd"},
 491         {Opt_ssd_spread, "ssd_spread"},
 492         {Opt_nossd_spread, "nossd_spread"},
 493         {Opt_subvol, "subvol=%s"},
 494         {Opt_subvol_empty, "subvol="},
 495         {Opt_subvolid, "subvolid=%s"},
 496         {Opt_thread_pool, "thread_pool=%u"},
 497         {Opt_treelog, "treelog"},
 498         {Opt_notreelog, "notreelog"},
 499         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 500
 501         /* Rescue options */
 502         {Opt_rescue, "rescue=%s"},
 503         /* Deprecated, with alias rescue=nologreplay */
 504         {Opt_nologreplay, "nologreplay"},
 505         /* Deprecated, with alias rescue=usebackuproot */
 506         {Opt_usebackuproot, "usebackuproot"},
 507
 508         /* Deprecated options */
 509         {Opt_recovery, "recovery"},
 510
 511         /* Debugging options */
 512         {Opt_check_integrity, "check_int"},
 513         {Opt_check_integrity_including_extent_data, "check_int_data"},
 514         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
 515         {Opt_enospc_debug, "enospc_debug"},
 516         {Opt_noenospc_debug, "noenospc_debug"},
 517 #ifdef CONFIG_BTRFS_DEBUG
 518         {Opt_fragment_data, "fragment=data"},
 519         {Opt_fragment_metadata, "fragment=metadata"},
 520         {Opt_fragment_all, "fragment=all"},
 521 #endif
 522 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 523         {Opt_ref_verify, "ref_verify"},
 524 #endif
 525         {Opt_err, NULL},
 526 };
 527
 528 static const match_table_t rescue_tokens = {
 529         {Opt_usebackuproot, "usebackuproot"},
 530         {Opt_nologreplay, "nologreplay"},
 531         {Opt_ignorebadroots, "ignorebadroots"},
 532         {Opt_ignorebadroots, "ibadroots"},
 533         {Opt_ignoredatacsums, "ignoredatacsums"},
 534         {Opt_ignoredatacsums, "idatacsums"},
 535         {Opt_rescue_all, "all"},
 536         {Opt_err, NULL},
 537 };
 538
 539 static bool check_ro_option(struct btrfs_fs_info *fs_info, unsigned long opt,
 540                             const char *opt_name)
 541 {
 542         if (fs_info->mount_opt & opt) {
 543                 btrfs_err(fs_info, "%s must be used with ro mount option",
 544                           opt_name);
 545                 return true;
 546         }
 547         return false;
 548 }
 549
 550 static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)
 551 {
 552         char *opts;
 553         char *orig;
 554         char *p;
 555         substring_t args[MAX_OPT_ARGS];
 556         int ret = 0;
 557
 558         opts = kstrdup(options, GFP_KERNEL);
 559         if (!opts)
 560                 return -ENOMEM;
 561         orig = opts;
 562
 563         while ((p = strsep(&opts, ":")) != NULL) {
 564                 int token;
 565
 566                 if (!*p)
 567                         continue;
 568                 token = match_token(p, rescue_tokens, args);
 569                 switch (token){
 570                 case Opt_usebackuproot:
 571                         btrfs_info(info,
 572                                    "trying to use backup root at mount time");
 573                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 574                         break;
 575                 case Opt_nologreplay:
 576                         btrfs_set_and_info(info, NOLOGREPLAY,
 577                                            "disabling log replay at mount time");
 578                         break;
 579                 case Opt_ignorebadroots:
 580                         btrfs_set_and_info(info, IGNOREBADROOTS,
 581                                            "ignoring bad roots");
 582                         break;
 583                 case Opt_ignoredatacsums:
 584                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 585                                            "ignoring data csums");
 586                         break;
 587                 case Opt_rescue_all:
 588                         btrfs_info(info, "enabling all of the rescue options");
 589                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 590                                            "ignoring data csums");
 591                         btrfs_set_and_info(info, IGNOREBADROOTS,
 592                                            "ignoring bad roots");
 593                         btrfs_set_and_info(info, NOLOGREPLAY,
 594                                            "disabling log replay at mount time");
 595                         break;
 596                 case Opt_err:
 597                         btrfs_info(info, "unrecognized rescue option '%s'", p);
 598                         ret = -EINVAL;
 599                         goto out;
 600                 default:
 601                         break;
 602                 }
 603
 604         }
 605 out:
 606         kfree(orig);
 607         return ret;
 608 }
 609
 610 /*
 611  * Regular mount options parser.  Everything that is needed only when
 612  * reading in a new superblock is parsed here.
 613  * XXX JDM: This needs to be cleaned up for remount.
 614  */
 615 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 616                         unsigned long new_flags)
 617 {
 618         substring_t args[MAX_OPT_ARGS];
 619         char *p, *num;
 620         int intarg;
 621         int ret = 0;
 622         char *compress_type;
 623         bool compress_force = false;
 624         enum btrfs_compression_type saved_compress_type;
 625         int saved_compress_level;
 626         bool saved_compress_force;
 627         int no_compress = 0;
 628
 629         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 630                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 631         else if (btrfs_free_space_cache_v1_active(info)) {
 632                 if (btrfs_is_zoned(info)) {
 633                         btrfs_info(info,
 634                         "zoned: clearing existing space cache");
 635                         btrfs_set_super_cache_generation(info->super_copy, 0);
 636                 } else {
 637                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 638                 }
 639         }
 640
 641         /*
 642          * Even the options are empty, we still need to do extra check
 643          * against new flags
 644          */
 645         if (!options)
 646                 goto check;
 647
 648         while ((p = strsep(&options, ",")) != NULL) {
 649                 int token;
 650                 if (!*p)
 651                         continue;
 652
 653                 token = match_token(p, tokens, args);
 654                 switch (token) {
 655                 case Opt_degraded:
 656                         btrfs_info(info, "allowing degraded mounts");
 657                         btrfs_set_opt(info->mount_opt, DEGRADED);
 658                         break;
 659                 case Opt_subvol:
 660                 case Opt_subvol_empty:
 661                 case Opt_subvolid:
 662                 case Opt_device:
 663                         /*
 664                          * These are parsed by btrfs_parse_subvol_options or
 665                          * btrfs_parse_device_options and can be ignored here.
 666                          */
 667                         break;
 668                 case Opt_nodatasum:
 669                         btrfs_set_and_info(info, NODATASUM,
 670                                            "setting nodatasum");
 671                         break;
 672                 case Opt_datasum:
 673                         if (btrfs_test_opt(info, NODATASUM)) {
 674                                 if (btrfs_test_opt(info, NODATACOW))
 675                                         btrfs_info(info,
 676                                                    "setting datasum, datacow enabled");
 677                                 else
 678                                         btrfs_info(info, "setting datasum");
 679                         }
 680                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 681                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 682                         break;
 683                 case Opt_nodatacow:
 684                         if (!btrfs_test_opt(info, NODATACOW)) {
 685                                 if (!btrfs_test_opt(info, COMPRESS) ||
 686                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 687                                         btrfs_info(info,
 688                                                    "setting nodatacow, compression disabled");
 689                                 } else {
 690                                         btrfs_info(info, "setting nodatacow");
 691                                 }
 692                         }
 693                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 694                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 695                         btrfs_set_opt(info->mount_opt, NODATACOW);
 696                         btrfs_set_opt(info->mount_opt, NODATASUM);
 697                         break;
 698                 case Opt_datacow:
 699                         btrfs_clear_and_info(info, NODATACOW,
 700                                              "setting datacow");
 701                         break;
 702                 case Opt_compress_force:
 703                 case Opt_compress_force_type:
 704                         compress_force = true;
 705                         fallthrough;
 706                 case Opt_compress:
 707                 case Opt_compress_type:
 708                         saved_compress_type = btrfs_test_opt(info,
 709                                                              COMPRESS) ?
 710                                 info->compress_type : BTRFS_COMPRESS_NONE;
 711                         saved_compress_force =
 712                                 btrfs_test_opt(info, FORCE_COMPRESS);
 713                         saved_compress_level = info->compress_level;
 714                         if (token == Opt_compress ||
 715                             token == Opt_compress_force ||
 716                             strncmp(args[0].from, "zlib", 4) == 0) {
 717                                 compress_type = "zlib";
 718
 719                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 720                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 721                                 /*
 722                                  * args[0] contains uninitialized data since
 723                                  * for these tokens we don't expect any
 724                                  * parameter.
 725                                  */
 726                                 if (token != Opt_compress &&
 727                                     token != Opt_compress_force)
 728                                         info->compress_level =
 729                                           btrfs_compress_str2level(
 730                                                         BTRFS_COMPRESS_ZLIB,
 731                                                         args[0].from + 4);
 732                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 733                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 734                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 735                                 no_compress = 0;
 736                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
 737                                 compress_type = "lzo";
 738                                 info->compress_type = BTRFS_COMPRESS_LZO;
 739                                 info->compress_level = 0;
 740                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 741                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 742                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 743                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 744                                 no_compress = 0;
 745                         } else if (strncmp(args[0].from, "zstd", 4) == 0) {
 746                                 compress_type = "zstd";
 747                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
 748                                 info->compress_level =
 749                                         btrfs_compress_str2level(
 750                                                          BTRFS_COMPRESS_ZSTD,
 751                                                          args[0].from + 4);
 752                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 753                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 754                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 755                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
 756                                 no_compress = 0;
 757                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 758                                 compress_type = "no";
 759                                 info->compress_level = 0;
 760                                 info->compress_type = 0;
 761                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 762                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 763                                 compress_force = false;
 764                                 no_compress++;
 765                         } else {
 766                                 ret = -EINVAL;
 767                                 goto out;
 768                         }
 769
 770                         if (compress_force) {
 771                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 772                         } else {
 773                                 /*
 774                                  * If we remount from compress-force=xxx to
 775                                  * compress=xxx, we need clear FORCE_COMPRESS
 776                                  * flag, otherwise, there is no way for users
 777                                  * to disable forcible compression separately.
 778                                  */
 779                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 780                         }
 781                         if (no_compress == 1) {
 782                                 btrfs_info(info, "use no compression");
 783                         } else if ((info->compress_type != saved_compress_type) ||
 784                                    (compress_force != saved_compress_force) ||
 785                                    (info->compress_level != saved_compress_level)) {
 786                                 btrfs_info(info, "%s %s compression, level %d",
 787                                            (compress_force) ? "force" : "use",
 788                                            compress_type, info->compress_level);
 789                         }
 790                         compress_force = false;
 791                         break;
 792                 case Opt_ssd:
 793                         btrfs_set_and_info(info, SSD,
 794                                            "enabling ssd optimizations");
 795                         btrfs_clear_opt(info->mount_opt, NOSSD);
 796                         break;
 797                 case Opt_ssd_spread:
 798                         btrfs_set_and_info(info, SSD,
 799                                            "enabling ssd optimizations");
 800                         btrfs_set_and_info(info, SSD_SPREAD,
 801                                            "using spread ssd allocation scheme");
 802                         btrfs_clear_opt(info->mount_opt, NOSSD);
 803                         break;
 804                 case Opt_nossd:
 805                         btrfs_set_opt(info->mount_opt, NOSSD);
 806                         btrfs_clear_and_info(info, SSD,
 807                                              "not using ssd optimizations");
 808                         fallthrough;
 809                 case Opt_nossd_spread:
 810                         btrfs_clear_and_info(info, SSD_SPREAD,
 811                                              "not using spread ssd allocation scheme");
 812                         break;
 813                 case Opt_barrier:
 814                         btrfs_clear_and_info(info, NOBARRIER,
 815                                              "turning on barriers");
 816                         break;
 817                 case Opt_nobarrier:
 818                         btrfs_set_and_info(info, NOBARRIER,
 819                                            "turning off barriers");
 820                         break;
 821                 case Opt_thread_pool:
 822                         ret = match_int(&args[0], &intarg);
 823                         if (ret) {
 824                                 goto out;
 825                         } else if (intarg == 0) {
 826                                 ret = -EINVAL;
 827                                 goto out;
 828                         }
 829                         info->thread_pool_size = intarg;
 830                         break;
 831                 case Opt_max_inline:
 832                         num = match_strdup(&args[0]);
 833                         if (num) {
 834                                 info->max_inline = memparse(num, NULL);
 835                                 kfree(num);
 836
 837                                 if (info->max_inline) {
 838                                         info->max_inline = min_t(u64,
 839                                                 info->max_inline,
 840                                                 info->sectorsize);
 841                                 }
 842                                 btrfs_info(info, "max_inline at %llu",
 843                                            info->max_inline);
 844                         } else {
 845                                 ret = -ENOMEM;
 846                                 goto out;
 847                         }
 848                         break;
 849                 case Opt_acl:
 850 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 851                         info->sb->s_flags |= SB_POSIXACL;
 852                         break;
 853 #else
 854                         btrfs_err(info, "support for ACL not compiled in!");
 855                         ret = -EINVAL;
 856                         goto out;
 857 #endif
 858                 case Opt_noacl:
 859                         info->sb->s_flags &= ~SB_POSIXACL;
 860                         break;
 861                 case Opt_notreelog:
 862                         btrfs_set_and_info(info, NOTREELOG,
 863                                            "disabling tree log");
 864                         break;
 865                 case Opt_treelog:
 866                         btrfs_clear_and_info(info, NOTREELOG,
 867                                              "enabling tree log");
 868                         break;
 869                 case Opt_norecovery:
 870                 case Opt_nologreplay:
 871                         btrfs_warn(info,
 872                 "'nologreplay' is deprecated, use 'rescue=nologreplay' instead");
 873                         btrfs_set_and_info(info, NOLOGREPLAY,
 874                                            "disabling log replay at mount time");
 875                         break;
 876                 case Opt_flushoncommit:
 877                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 878                                            "turning on flush-on-commit");
 879                         break;
 880                 case Opt_noflushoncommit:
 881                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 882                                              "turning off flush-on-commit");
 883                         break;
 884                 case Opt_ratio:
 885                         ret = match_int(&args[0], &intarg);
 886                         if (ret)
 887                                 goto out;
 888                         info->metadata_ratio = intarg;
 889                         btrfs_info(info, "metadata ratio %u",
 890                                    info->metadata_ratio);
 891                         break;
 892                 case Opt_discard:
 893                 case Opt_discard_mode:
 894                         if (token == Opt_discard ||
 895                             strcmp(args[0].from, "sync") == 0) {
 896                                 btrfs_clear_opt(info->mount_opt, DISCARD_ASYNC);
 897                                 btrfs_set_and_info(info, DISCARD_SYNC,
 898                                                    "turning on sync discard");
 899                         } else if (strcmp(args[0].from, "async") == 0) {
 900                                 btrfs_clear_opt(info->mount_opt, DISCARD_SYNC);
 901                                 btrfs_set_and_info(info, DISCARD_ASYNC,
 902                                                    "turning on async discard");
 903                         } else {
 904                                 ret = -EINVAL;
 905                                 goto out;
 906                         }
 907                         break;
 908                 case Opt_nodiscard:
 909                         btrfs_clear_and_info(info, DISCARD_SYNC,
 910                                              "turning off discard");
 911                         btrfs_clear_and_info(info, DISCARD_ASYNC,
 912                                              "turning off async discard");
 913                         break;
 914                 case Opt_space_cache:
 915                 case Opt_space_cache_version:
 916                         /*
 917                          * We already set FREE_SPACE_TREE above because we have
 918                          * compat_ro(FREE_SPACE_TREE) set, and we aren't going
 919                          * to allow v1 to be set for extent tree v2, simply
 920                          * ignore this setting if we're extent tree v2.
 921                          */
 922                         if (btrfs_fs_incompat(info, EXTENT_TREE_V2))
 923                                 break;
 924                         if (token == Opt_space_cache ||
 925                             strcmp(args[0].from, "v1") == 0) {
 926                                 btrfs_clear_opt(info->mount_opt,
 927                                                 FREE_SPACE_TREE);
 928                                 btrfs_set_and_info(info, SPACE_CACHE,
 929                                            "enabling disk space caching");
 930                         } else if (strcmp(args[0].from, "v2") == 0) {
 931                                 btrfs_clear_opt(info->mount_opt,
 932                                                 SPACE_CACHE);
 933                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 934                                                    "enabling free space tree");
 935                         } else {
 936                                 ret = -EINVAL;
 937                                 goto out;
 938                         }
 939                         break;
 940                 case Opt_rescan_uuid_tree:
 941                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 942                         break;
 943                 case Opt_no_space_cache:
 944                         /*
 945                          * We cannot operate without the free space tree with
 946                          * extent tree v2, ignore this option.
 947                          */
 948                         if (btrfs_fs_incompat(info, EXTENT_TREE_V2))
 949                                 break;
 950                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 951                                 btrfs_clear_and_info(info, SPACE_CACHE,
 952                                              "disabling disk space caching");
 953                         }
 954                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 955                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 956                                              "disabling free space tree");
 957                         }
 958                         break;
 959                 case Opt_inode_cache:
 960                 case Opt_noinode_cache:
 961                         btrfs_warn(info,
 962         "the 'inode_cache' option is deprecated and has no effect since 5.11");
 963                         break;
 964                 case Opt_clear_cache:
 965                         /*
 966                          * We cannot clear the free space tree with extent tree
 967                          * v2, ignore this option.
 968                          */
 969                         if (btrfs_fs_incompat(info, EXTENT_TREE_V2))
 970                                 break;
 971                         btrfs_set_and_info(info, CLEAR_CACHE,
 972                                            "force clearing of disk cache");
 973                         break;
 974                 case Opt_user_subvol_rm_allowed:
 975                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 976                         break;
 977                 case Opt_enospc_debug:
 978                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 979                         break;
 980                 case Opt_noenospc_debug:
 981                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 982                         break;
 983                 case Opt_defrag:
 984                         btrfs_set_and_info(info, AUTO_DEFRAG,
 985                                            "enabling auto defrag");
 986                         break;
 987                 case Opt_nodefrag:
 988                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 989                                              "disabling auto defrag");
 990                         break;
 991                 case Opt_recovery:
 992                 case Opt_usebackuproot:
 993                         btrfs_warn(info,
 994                         "'%s' is deprecated, use 'rescue=usebackuproot' instead",
 995                                    token == Opt_recovery ? "recovery" :
 996                                    "usebackuproot");
 997                         btrfs_info(info,
 998                                    "trying to use backup root at mount time");
 999                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
1000                         break;
1001                 case Opt_skip_balance:
1002                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
1003                         break;
1004 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1005                 case Opt_check_integrity_including_extent_data:
1006                         btrfs_info(info,
1007                                    "enabling check integrity including extent data");
1008                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY_DATA);
1009                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
1010                         break;
1011                 case Opt_check_integrity:
1012                         btrfs_info(info, "enabling check integrity");
1013                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
1014                         break;
1015                 case Opt_check_integrity_print_mask:
1016                         ret = match_int(&args[0], &intarg);
1017                         if (ret)
1018                                 goto out;
1019                         info->check_integrity_print_mask = intarg;
1020                         btrfs_info(info, "check_integrity_print_mask 0x%x",
1021                                    info->check_integrity_print_mask);
1022                         break;
1023 #else
1024                 case Opt_check_integrity_including_extent_data:
1025                 case Opt_check_integrity:
1026                 case Opt_check_integrity_print_mask:
1027                         btrfs_err(info,
1028                                   "support for check_integrity* not compiled in!");
1029                         ret = -EINVAL;
1030                         goto out;
1031 #endif
1032                 case Opt_fatal_errors:
1033                         if (strcmp(args[0].from, "panic") == 0)
1034                                 btrfs_set_opt(info->mount_opt,
1035                                               PANIC_ON_FATAL_ERROR);
1036                         else if (strcmp(args[0].from, "bug") == 0)
1037                                 btrfs_clear_opt(info->mount_opt,
1038                                               PANIC_ON_FATAL_ERROR);
1039                         else {
1040                                 ret = -EINVAL;
1041                                 goto out;
1042                         }
1043                         break;
1044                 case Opt_commit_interval:
1045                         intarg = 0;
1046                         ret = match_int(&args[0], &intarg);
1047                         if (ret)
1048                                 goto out;
1049                         if (intarg == 0) {
1050                                 btrfs_info(info,
1051                                            "using default commit interval %us",
1052                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
1053                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
1054                         } else if (intarg > 300) {
1055                                 btrfs_warn(info, "excessive commit interval %d",
1056                                            intarg);
1057                         }
1058                         info->commit_interval = intarg;
1059                         break;
1060                 case Opt_rescue:
1061                         ret = parse_rescue_options(info, args[0].from);
1062                         if (ret < 0)
1063                                 goto out;
1064                         break;
1065 #ifdef CONFIG_BTRFS_DEBUG
1066                 case Opt_fragment_all:
1067                         btrfs_info(info, "fragmenting all space");
1068                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
1069                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
1070                         break;
1071                 case Opt_fragment_metadata:
1072                         btrfs_info(info, "fragmenting metadata");
1073                         btrfs_set_opt(info->mount_opt,
1074                                       FRAGMENT_METADATA);
1075                         break;
1076                 case Opt_fragment_data:
1077                         btrfs_info(info, "fragmenting data");
1078                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
1079                         break;
1080 #endif
1081 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
1082                 case Opt_ref_verify:
1083                         btrfs_info(info, "doing ref verification");
1084                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
1085                         break;
1086 #endif
1087                 case Opt_err:
1088                         btrfs_err(info, "unrecognized mount option '%s'", p);
1089                         ret = -EINVAL;
1090                         goto out;
1091                 default:
1092                         break;
1093                 }
1094         }
1095 check:
1096         /* We're read-only, don't have to check. */
1097         if (new_flags & SB_RDONLY)
1098                 goto out;
1099
1100         if (check_ro_option(info, BTRFS_MOUNT_NOLOGREPLAY, "nologreplay") ||
1101             check_ro_option(info, BTRFS_MOUNT_IGNOREBADROOTS, "ignorebadroots") ||
1102             check_ro_option(info, BTRFS_MOUNT_IGNOREDATACSUMS, "ignoredatacsums"))
1103                 ret = -EINVAL;
1104 out:
1105         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
1106             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
1107             !btrfs_test_opt(info, CLEAR_CACHE)) {
1108                 btrfs_err(info, "cannot disable free space tree");
1109                 ret = -EINVAL;
1110
1111         }
1112         if (!ret)
1113                 ret = btrfs_check_mountopts_zoned(info);
1114         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
1115                 btrfs_info(info, "disk space caching is enabled");
1116         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
1117                 btrfs_info(info, "using free space tree");
1118         return ret;
1119 }
1120
1121 /*
1122  * Parse mount options that are required early in the mount process.
1123  *
1124  * All other options will be parsed on much later in the mount process and
1125  * only when we need to allocate a new super block.
1126  */
1127 static int btrfs_parse_device_options(const char *options, fmode_t flags,
1128                                       void *holder)
1129 {
1130         substring_t args[MAX_OPT_ARGS];
1131         char *device_name, *opts, *orig, *p;
1132         struct btrfs_device *device = NULL;
1133         int error = 0;
1134
1135         lockdep_assert_held(&uuid_mutex);
1136
1137         if (!options)
1138                 return 0;
1139
1140         /*
1141          * strsep changes the string, duplicate it because btrfs_parse_options
1142          * gets called later
1143          */
1144         opts = kstrdup(options, GFP_KERNEL);
1145         if (!opts)
1146                 return -ENOMEM;
1147         orig = opts;
1148
1149         while ((p = strsep(&opts, ",")) != NULL) {
1150                 int token;
1151
1152                 if (!*p)
1153                         continue;
1154
1155                 token = match_token(p, tokens, args);
1156                 if (token == Opt_device) {
1157                         device_name = match_strdup(&args[0]);
1158                         if (!device_name) {
1159                                 error = -ENOMEM;
1160                                 goto out;
1161                         }
1162                         device = btrfs_scan_one_device(device_name, flags,
1163                                         holder);
1164                         kfree(device_name);
1165                         if (IS_ERR(device)) {
1166                                 error = PTR_ERR(device);
1167                                 goto out;
1168                         }
1169                 }
1170         }
1171
1172 out:
1173         kfree(orig);
1174         return error;
1175 }
1176
1177 /*
1178  * Parse mount options that are related to subvolume id
1179  *
1180  * The value is later passed to mount_subvol()
1181  */
1182 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
1183                 u64 *subvol_objectid)
1184 {
1185         substring_t args[MAX_OPT_ARGS];
1186         char *opts, *orig, *p;
1187         int error = 0;
1188         u64 subvolid;
1189
1190         if (!options)
1191                 return 0;
1192
1193         /*
1194          * strsep changes the string, duplicate it because
1195          * btrfs_parse_device_options gets called later
1196          */
1197         opts = kstrdup(options, GFP_KERNEL);
1198         if (!opts)
1199                 return -ENOMEM;
1200         orig = opts;
1201
1202         while ((p = strsep(&opts, ",")) != NULL) {
1203                 int token;
1204                 if (!*p)
1205                         continue;
1206
1207                 token = match_token(p, tokens, args);
1208                 switch (token) {
1209                 case Opt_subvol:
1210                         kfree(*subvol_name);
1211                         *subvol_name = match_strdup(&args[0]);
1212                         if (!*subvol_name) {
1213                                 error = -ENOMEM;
1214                                 goto out;
1215                         }
1216                         break;
1217                 case Opt_subvolid:
1218                         error = match_u64(&args[0], &subvolid);
1219                         if (error)
1220                                 goto out;
1221
1222                         /* we want the original fs_tree */
1223                         if (subvolid == 0)
1224                                 subvolid = BTRFS_FS_TREE_OBJECTID;
1225
1226                         *subvol_objectid = subvolid;
1227                         break;
1228                 default:
1229                         break;
1230                 }
1231         }
1232
1233 out:
1234         kfree(orig);
1235         return error;
1236 }
1237
1238 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
1239                                           u64 subvol_objectid)
1240 {
1241         struct btrfs_root *root = fs_info->tree_root;
1242         struct btrfs_root *fs_root = NULL;
1243         struct btrfs_root_ref *root_ref;
1244         struct btrfs_inode_ref *inode_ref;
1245         struct btrfs_key key;
1246         struct btrfs_path *path = NULL;
1247         char *name = NULL, *ptr;
1248         u64 dirid;
1249         int len;
1250         int ret;
1251
1252         path = btrfs_alloc_path();
1253         if (!path) {
1254                 ret = -ENOMEM;
1255                 goto err;
1256         }
1257
1258         name = kmalloc(PATH_MAX, GFP_KERNEL);
1259         if (!name) {
1260                 ret = -ENOMEM;
1261                 goto err;
1262         }
1263         ptr = name + PATH_MAX - 1;
1264         ptr[0] = '\0';
1265
1266         /*
1267          * Walk up the subvolume trees in the tree of tree roots by root
1268          * backrefs until we hit the top-level subvolume.
1269          */
1270         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1271                 key.objectid = subvol_objectid;
1272                 key.type = BTRFS_ROOT_BACKREF_KEY;
1273                 key.offset = (u64)-1;
1274
1275                 ret = btrfs_search_backwards(root, &key, path);
1276                 if (ret < 0) {
1277                         goto err;
1278                 } else if (ret > 0) {
1279                         ret = -ENOENT;
1280                         goto err;
1281                 }
1282
1283                 subvol_objectid = key.offset;
1284
1285                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1286                                           struct btrfs_root_ref);
1287                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1288                 ptr -= len + 1;
1289                 if (ptr < name) {
1290                         ret = -ENAMETOOLONG;
1291                         goto err;
1292                 }
1293                 read_extent_buffer(path->nodes[0], ptr + 1,
1294                                    (unsigned long)(root_ref + 1), len);
1295                 ptr[0] = '/';
1296                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1297                 btrfs_release_path(path);
1298
1299                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
1300                 if (IS_ERR(fs_root)) {
1301                         ret = PTR_ERR(fs_root);
1302                         fs_root = NULL;
1303                         goto err;
1304                 }
1305
1306                 /*
1307                  * Walk up the filesystem tree by inode refs until we hit the
1308                  * root directory.
1309                  */
1310                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1311                         key.objectid = dirid;
1312                         key.type = BTRFS_INODE_REF_KEY;
1313                         key.offset = (u64)-1;
1314
1315                         ret = btrfs_search_backwards(fs_root, &key, path);
1316                         if (ret < 0) {
1317                                 goto err;
1318                         } else if (ret > 0) {
1319                                 ret = -ENOENT;
1320                                 goto err;
1321                         }
1322
1323                         dirid = key.offset;
1324
1325                         inode_ref = btrfs_item_ptr(path->nodes[0],
1326                                                    path->slots[0],
1327                                                    struct btrfs_inode_ref);
1328                         len = btrfs_inode_ref_name_len(path->nodes[0],
1329                                                        inode_ref);
1330                         ptr -= len + 1;
1331                         if (ptr < name) {
1332                                 ret = -ENAMETOOLONG;
1333                                 goto err;
1334                         }
1335                         read_extent_buffer(path->nodes[0], ptr + 1,
1336                                            (unsigned long)(inode_ref + 1), len);
1337                         ptr[0] = '/';
1338                         btrfs_release_path(path);
1339                 }
1340                 btrfs_put_root(fs_root);
1341                 fs_root = NULL;
1342         }
1343
1344         btrfs_free_path(path);
1345         if (ptr == name + PATH_MAX - 1) {
1346                 name[0] = '/';
1347                 name[1] = '\0';
1348         } else {
1349                 memmove(name, ptr, name + PATH_MAX - ptr);
1350         }
1351         return name;
1352
1353 err:
1354         btrfs_put_root(fs_root);
1355         btrfs_free_path(path);
1356         kfree(name);
1357         return ERR_PTR(ret);
1358 }
1359
1360 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1361 {
1362         struct btrfs_root *root = fs_info->tree_root;
1363         struct btrfs_dir_item *di;
1364         struct btrfs_path *path;
1365         struct btrfs_key location;
1366         u64 dir_id;
1367
1368         path = btrfs_alloc_path();
1369         if (!path)
1370                 return -ENOMEM;
1371
1372         /*
1373          * Find the "default" dir item which points to the root item that we
1374          * will mount by default if we haven't been given a specific subvolume
1375          * to mount.
1376          */
1377         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1378         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1379         if (IS_ERR(di)) {
1380                 btrfs_free_path(path);
1381                 return PTR_ERR(di);
1382         }
1383         if (!di) {
1384                 /*
1385                  * Ok the default dir item isn't there.  This is weird since
1386                  * it's always been there, but don't freak out, just try and
1387                  * mount the top-level subvolume.
1388                  */
1389                 btrfs_free_path(path);
1390                 *objectid = BTRFS_FS_TREE_OBJECTID;
1391                 return 0;
1392         }
1393
1394         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1395         btrfs_free_path(path);
1396         *objectid = location.objectid;
1397         return 0;
1398 }
1399
1400 static int btrfs_fill_super(struct super_block *sb,
1401                             struct btrfs_fs_devices *fs_devices,
1402                             void *data)
1403 {
1404         struct inode *inode;
1405         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1406         int err;
1407
1408         sb->s_maxbytes = MAX_LFS_FILESIZE;
1409         sb->s_magic = BTRFS_SUPER_MAGIC;
1410         sb->s_op = &btrfs_super_ops;
1411         sb->s_d_op = &btrfs_dentry_operations;
1412         sb->s_export_op = &btrfs_export_ops;
1413 #ifdef CONFIG_FS_VERITY
1414         sb->s_vop = &btrfs_verityops;
1415 #endif
1416         sb->s_xattr = btrfs_xattr_handlers;
1417         sb->s_time_gran = 1;
1418 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1419         sb->s_flags |= SB_POSIXACL;
1420 #endif
1421         sb->s_flags |= SB_I_VERSION;
1422         sb->s_iflags |= SB_I_CGROUPWB;
1423
1424         err = super_setup_bdi(sb);
1425         if (err) {
1426                 btrfs_err(fs_info, "super_setup_bdi failed");
1427                 return err;
1428         }
1429
1430         err = open_ctree(sb, fs_devices, (char *)data);
1431         if (err) {
1432                 btrfs_err(fs_info, "open_ctree failed");
1433                 return err;
1434         }
1435
1436         inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
1437         if (IS_ERR(inode)) {
1438                 err = PTR_ERR(inode);
1439                 goto fail_close;
1440         }
1441
1442         sb->s_root = d_make_root(inode);
1443         if (!sb->s_root) {
1444                 err = -ENOMEM;
1445                 goto fail_close;
1446         }
1447
1448         sb->s_flags |= SB_ACTIVE;
1449         return 0;
1450
1451 fail_close:
1452         close_ctree(fs_info);
1453         return err;
1454 }
1455
1456 int btrfs_sync_fs(struct super_block *sb, int wait)
1457 {
1458         struct btrfs_trans_handle *trans;
1459         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1460         struct btrfs_root *root = fs_info->tree_root;
1461
1462         trace_btrfs_sync_fs(fs_info, wait);
1463
1464         if (!wait) {
1465                 filemap_flush(fs_info->btree_inode->i_mapping);
1466                 return 0;
1467         }
1468
1469         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1470
1471         trans = btrfs_attach_transaction_barrier(root);
1472         if (IS_ERR(trans)) {
1473                 /* no transaction, don't bother */
1474                 if (PTR_ERR(trans) == -ENOENT) {
1475                         /*
1476                          * Exit unless we have some pending changes
1477                          * that need to go through commit
1478                          */
1479                         if (fs_info->pending_changes == 0)
1480                                 return 0;
1481                         /*
1482                          * A non-blocking test if the fs is frozen. We must not
1483                          * start a new transaction here otherwise a deadlock
1484                          * happens. The pending operations are delayed to the
1485                          * next commit after thawing.
1486                          */
1487                         if (sb_start_write_trylock(sb))
1488                                 sb_end_write(sb);
1489                         else
1490                                 return 0;
1491                         trans = btrfs_start_transaction(root, 0);
1492                 }
1493                 if (IS_ERR(trans))
1494                         return PTR_ERR(trans);
1495         }
1496         return btrfs_commit_transaction(trans);
1497 }
1498
1499 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1500 {
1501         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1502         *printed = true;
1503 }
1504
1505 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1506 {
1507         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1508         const char *compress_type;
1509         const char *subvol_name;
1510         bool printed = false;
1511
1512         if (btrfs_test_opt(info, DEGRADED))
1513                 seq_puts(seq, ",degraded");
1514         if (btrfs_test_opt(info, NODATASUM))
1515                 seq_puts(seq, ",nodatasum");
1516         if (btrfs_test_opt(info, NODATACOW))
1517                 seq_puts(seq, ",nodatacow");
1518         if (btrfs_test_opt(info, NOBARRIER))
1519                 seq_puts(seq, ",nobarrier");
1520         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1521                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1522         if (info->thread_pool_size !=  min_t(unsigned long,
1523                                              num_online_cpus() + 2, 8))
1524                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1525         if (btrfs_test_opt(info, COMPRESS)) {
1526                 compress_type = btrfs_compress_type2str(info->compress_type);
1527                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1528                         seq_printf(seq, ",compress-force=%s", compress_type);
1529                 else
1530                         seq_printf(seq, ",compress=%s", compress_type);
1531                 if (info->compress_level)
1532                         seq_printf(seq, ":%d", info->compress_level);
1533         }
1534         if (btrfs_test_opt(info, NOSSD))
1535                 seq_puts(seq, ",nossd");
1536         if (btrfs_test_opt(info, SSD_SPREAD))
1537                 seq_puts(seq, ",ssd_spread");
1538         else if (btrfs_test_opt(info, SSD))
1539                 seq_puts(seq, ",ssd");
1540         if (btrfs_test_opt(info, NOTREELOG))
1541                 seq_puts(seq, ",notreelog");
1542         if (btrfs_test_opt(info, NOLOGREPLAY))
1543                 print_rescue_option(seq, "nologreplay", &printed);
1544         if (btrfs_test_opt(info, USEBACKUPROOT))
1545                 print_rescue_option(seq, "usebackuproot", &printed);
1546         if (btrfs_test_opt(info, IGNOREBADROOTS))
1547                 print_rescue_option(seq, "ignorebadroots", &printed);
1548         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1549                 print_rescue_option(seq, "ignoredatacsums", &printed);
1550         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1551                 seq_puts(seq, ",flushoncommit");
1552         if (btrfs_test_opt(info, DISCARD_SYNC))
1553                 seq_puts(seq, ",discard");
1554         if (btrfs_test_opt(info, DISCARD_ASYNC))
1555                 seq_puts(seq, ",discard=async");
1556         if (!(info->sb->s_flags & SB_POSIXACL))
1557                 seq_puts(seq, ",noacl");
1558         if (btrfs_free_space_cache_v1_active(info))
1559                 seq_puts(seq, ",space_cache");
1560         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1561                 seq_puts(seq, ",space_cache=v2");
1562         else
1563                 seq_puts(seq, ",nospace_cache");
1564         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1565                 seq_puts(seq, ",rescan_uuid_tree");
1566         if (btrfs_test_opt(info, CLEAR_CACHE))
1567                 seq_puts(seq, ",clear_cache");
1568         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1569                 seq_puts(seq, ",user_subvol_rm_allowed");
1570         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1571                 seq_puts(seq, ",enospc_debug");
1572         if (btrfs_test_opt(info, AUTO_DEFRAG))
1573                 seq_puts(seq, ",autodefrag");
1574         if (btrfs_test_opt(info, SKIP_BALANCE))
1575                 seq_puts(seq, ",skip_balance");
1576 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1577         if (btrfs_test_opt(info, CHECK_INTEGRITY_DATA))
1578                 seq_puts(seq, ",check_int_data");
1579         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1580                 seq_puts(seq, ",check_int");
1581         if (info->check_integrity_print_mask)
1582                 seq_printf(seq, ",check_int_print_mask=%d",
1583                                 info->check_integrity_print_mask);
1584 #endif
1585         if (info->metadata_ratio)
1586                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1587         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1588                 seq_puts(seq, ",fatal_errors=panic");
1589         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1590                 seq_printf(seq, ",commit=%u", info->commit_interval);
1591 #ifdef CONFIG_BTRFS_DEBUG
1592         if (btrfs_test_opt(info, FRAGMENT_DATA))
1593                 seq_puts(seq, ",fragment=data");
1594         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1595                 seq_puts(seq, ",fragment=metadata");
1596 #endif
1597         if (btrfs_test_opt(info, REF_VERIFY))
1598                 seq_puts(seq, ",ref_verify");
1599         seq_printf(seq, ",subvolid=%llu",
1600                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1601         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1602                         BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1603         if (!IS_ERR(subvol_name)) {
1604                 seq_puts(seq, ",subvol=");
1605                 seq_escape(seq, subvol_name, " \t\n\\");
1606                 kfree(subvol_name);
1607         }
1608         return 0;
1609 }
1610
1611 static int btrfs_test_super(struct super_block *s, void *data)
1612 {
1613         struct btrfs_fs_info *p = data;
1614         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1615
1616         return fs_info->fs_devices == p->fs_devices;
1617 }
1618
1619 static int btrfs_set_super(struct super_block *s, void *data)
1620 {
1621         int err = set_anon_super(s, data);
1622         if (!err)
1623                 s->s_fs_info = data;
1624         return err;
1625 }
1626
1627 /*
1628  * subvolumes are identified by ino 256
1629  */
1630 static inline int is_subvolume_inode(struct inode *inode)
1631 {
1632         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1633                 return 1;
1634         return 0;
1635 }
1636
1637 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1638                                    struct vfsmount *mnt)
1639 {
1640         struct dentry *root;
1641         int ret;
1642
1643         if (!subvol_name) {
1644                 if (!subvol_objectid) {
1645                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1646                                                           &subvol_objectid);
1647                         if (ret) {
1648                                 root = ERR_PTR(ret);
1649                                 goto out;
1650                         }
1651                 }
1652                 subvol_name = btrfs_get_subvol_name_from_objectid(
1653                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1654                 if (IS_ERR(subvol_name)) {
1655                         root = ERR_CAST(subvol_name);
1656                         subvol_name = NULL;
1657                         goto out;
1658                 }
1659
1660         }
1661
1662         root = mount_subtree(mnt, subvol_name);
1663         /* mount_subtree() drops our reference on the vfsmount. */
1664         mnt = NULL;
1665
1666         if (!IS_ERR(root)) {
1667                 struct super_block *s = root->d_sb;
1668                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1669                 struct inode *root_inode = d_inode(root);
1670                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1671
1672                 ret = 0;
1673                 if (!is_subvolume_inode(root_inode)) {
1674                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1675                                subvol_name);
1676                         ret = -EINVAL;
1677                 }
1678                 if (subvol_objectid && root_objectid != subvol_objectid) {
1679                         /*
1680                          * This will also catch a race condition where a
1681                          * subvolume which was passed by ID is renamed and
1682                          * another subvolume is renamed over the old location.
1683                          */
1684                         btrfs_err(fs_info,
1685                                   "subvol '%s' does not match subvolid %llu",
1686                                   subvol_name, subvol_objectid);
1687                         ret = -EINVAL;
1688                 }
1689                 if (ret) {
1690                         dput(root);
1691                         root = ERR_PTR(ret);
1692                         deactivate_locked_super(s);
1693                 }
1694         }
1695
1696 out:
1697         mntput(mnt);
1698         kfree(subvol_name);
1699         return root;
1700 }
1701
1702 /*
1703  * Find a superblock for the given device / mount point.
1704  *
1705  * Note: This is based on mount_bdev from fs/super.c with a few additions
1706  *       for multiple device setup.  Make sure to keep it in sync.
1707  */
1708 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1709                 int flags, const char *device_name, void *data)
1710 {
1711         struct block_device *bdev = NULL;
1712         struct super_block *s;
1713         struct btrfs_device *device = NULL;
1714         struct btrfs_fs_devices *fs_devices = NULL;
1715         struct btrfs_fs_info *fs_info = NULL;
1716         void *new_sec_opts = NULL;
1717         fmode_t mode = FMODE_READ;
1718         int error = 0;
1719
1720         if (!(flags & SB_RDONLY))
1721                 mode |= FMODE_WRITE;
1722
1723         if (data) {
1724                 error = security_sb_eat_lsm_opts(data, &new_sec_opts);
1725                 if (error)
1726                         return ERR_PTR(error);
1727         }
1728
1729         /*
1730          * Setup a dummy root and fs_info for test/set super.  This is because
1731          * we don't actually fill this stuff out until open_ctree, but we need
1732          * then open_ctree will properly initialize the file system specific
1733          * settings later.  btrfs_init_fs_info initializes the static elements
1734          * of the fs_info (locks and such) to make cleanup easier if we find a
1735          * superblock with our given fs_devices later on at sget() time.
1736          */
1737         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1738         if (!fs_info) {
1739                 error = -ENOMEM;
1740                 goto error_sec_opts;
1741         }
1742         btrfs_init_fs_info(fs_info);
1743
1744         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1745         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1746         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1747                 error = -ENOMEM;
1748                 goto error_fs_info;
1749         }
1750
1751         mutex_lock(&uuid_mutex);
1752         error = btrfs_parse_device_options(data, mode, fs_type);
1753         if (error) {
1754                 mutex_unlock(&uuid_mutex);
1755                 goto error_fs_info;
1756         }
1757
1758         device = btrfs_scan_one_device(device_name, mode, fs_type);
1759         if (IS_ERR(device)) {
1760                 mutex_unlock(&uuid_mutex);
1761                 error = PTR_ERR(device);
1762                 goto error_fs_info;
1763         }
1764
1765         fs_devices = device->fs_devices;
1766         fs_info->fs_devices = fs_devices;
1767
1768         error = btrfs_open_devices(fs_devices, mode, fs_type);
1769         mutex_unlock(&uuid_mutex);
1770         if (error)
1771                 goto error_fs_info;
1772
1773         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1774                 error = -EACCES;
1775                 goto error_close_devices;
1776         }
1777
1778         bdev = fs_devices->latest_dev->bdev;
1779         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1780                  fs_info);
1781         if (IS_ERR(s)) {
1782                 error = PTR_ERR(s);
1783                 goto error_close_devices;
1784         }
1785
1786         if (s->s_root) {
1787                 btrfs_close_devices(fs_devices);
1788                 btrfs_free_fs_info(fs_info);
1789                 if ((flags ^ s->s_flags) & SB_RDONLY)
1790                         error = -EBUSY;
1791         } else {
1792                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1793                 btrfs_sb(s)->bdev_holder = fs_type;
1794                 if (!strstr(crc32c_impl(), "generic"))
1795                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
1796                 error = btrfs_fill_super(s, fs_devices, data);
1797         }
1798         if (!error)
1799                 error = security_sb_set_mnt_opts(s, new_sec_opts, 0, NULL);
1800         security_free_mnt_opts(&new_sec_opts);
1801         if (error) {
1802                 deactivate_locked_super(s);
1803                 return ERR_PTR(error);
1804         }
1805
1806         return dget(s->s_root);
1807
1808 error_close_devices:
1809         btrfs_close_devices(fs_devices);
1810 error_fs_info:
1811         btrfs_free_fs_info(fs_info);
1812 error_sec_opts:
1813         security_free_mnt_opts(&new_sec_opts);
1814         return ERR_PTR(error);
1815 }
1816
1817 /*
1818  * Mount function which is called by VFS layer.
1819  *
1820  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1821  * which needs vfsmount* of device's root (/).  This means device's root has to
1822  * be mounted internally in any case.
1823  *
1824  * Operation flow:
1825  *   1. Parse subvol id related options for later use in mount_subvol().
1826  *
1827  *   2. Mount device's root (/) by calling vfs_kern_mount().
1828  *
1829  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1830  *      first place. In order to avoid calling btrfs_mount() again, we use
1831  *      different file_system_type which is not registered to VFS by
1832  *      register_filesystem() (btrfs_root_fs_type). As a result,
1833  *      btrfs_mount_root() is called. The return value will be used by
1834  *      mount_subtree() in mount_subvol().
1835  *
1836  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1837  *      "btrfs subvolume set-default", mount_subvol() is called always.
1838  */
1839 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1840                 const char *device_name, void *data)
1841 {
1842         struct vfsmount *mnt_root;
1843         struct dentry *root;
1844         char *subvol_name = NULL;
1845         u64 subvol_objectid = 0;
1846         int error = 0;
1847
1848         error = btrfs_parse_subvol_options(data, &subvol_name,
1849                                         &subvol_objectid);
1850         if (error) {
1851                 kfree(subvol_name);
1852                 return ERR_PTR(error);
1853         }
1854
1855         /* mount device's root (/) */
1856         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1857         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1858                 if (flags & SB_RDONLY) {
1859                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1860                                 flags & ~SB_RDONLY, device_name, data);
1861                 } else {
1862                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1863                                 flags | SB_RDONLY, device_name, data);
1864                         if (IS_ERR(mnt_root)) {
1865                                 root = ERR_CAST(mnt_root);
1866                                 kfree(subvol_name);
1867                                 goto out;
1868                         }
1869
1870                         down_write(&mnt_root->mnt_sb->s_umount);
1871                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1872                         up_write(&mnt_root->mnt_sb->s_umount);
1873                         if (error < 0) {
1874                                 root = ERR_PTR(error);
1875                                 mntput(mnt_root);
1876                                 kfree(subvol_name);
1877                                 goto out;
1878                         }
1879                 }
1880         }
1881         if (IS_ERR(mnt_root)) {
1882                 root = ERR_CAST(mnt_root);
1883                 kfree(subvol_name);
1884                 goto out;
1885         }
1886
1887         /* mount_subvol() will free subvol_name and mnt_root */
1888         root = mount_subvol(subvol_name, subvol_objectid, mnt_root);
1889
1890 out:
1891         return root;
1892 }
1893
1894 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1895                                      u32 new_pool_size, u32 old_pool_size)
1896 {
1897         if (new_pool_size == old_pool_size)
1898                 return;
1899
1900         fs_info->thread_pool_size = new_pool_size;
1901
1902         btrfs_info(fs_info, "resize thread pool %d -> %d",
1903                old_pool_size, new_pool_size);
1904
1905         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1906         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1907         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1908         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1909         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1910         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1911                                 new_pool_size);
1912         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1913         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1914         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1915         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1916                                 new_pool_size);
1917 }
1918
1919 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1920                                        unsigned long old_opts, int flags)
1921 {
1922         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1923             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1924              (flags & SB_RDONLY))) {
1925                 /* wait for any defraggers to finish */
1926                 wait_event(fs_info->transaction_wait,
1927                            (atomic_read(&fs_info->defrag_running) == 0));
1928                 if (flags & SB_RDONLY)
1929                         sync_filesystem(fs_info->sb);
1930         }
1931 }
1932
1933 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1934                                          unsigned long old_opts)
1935 {
1936         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1937
1938         /*
1939          * We need to cleanup all defragable inodes if the autodefragment is
1940          * close or the filesystem is read only.
1941          */
1942         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1943             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1944                 btrfs_cleanup_defrag_inodes(fs_info);
1945         }
1946
1947         /* If we toggled discard async */
1948         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1949             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1950                 btrfs_discard_resume(fs_info);
1951         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1952                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1953                 btrfs_discard_cleanup(fs_info);
1954
1955         /* If we toggled space cache */
1956         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1957                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1958 }
1959
1960 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1961 {
1962         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1963         unsigned old_flags = sb->s_flags;
1964         unsigned long old_opts = fs_info->mount_opt;
1965         unsigned long old_compress_type = fs_info->compress_type;
1966         u64 old_max_inline = fs_info->max_inline;
1967         u32 old_thread_pool_size = fs_info->thread_pool_size;
1968         u32 old_metadata_ratio = fs_info->metadata_ratio;
1969         int ret;
1970
1971         sync_filesystem(sb);
1972         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1973
1974         if (data) {
1975                 void *new_sec_opts = NULL;
1976
1977                 ret = security_sb_eat_lsm_opts(data, &new_sec_opts);
1978                 if (!ret)
1979                         ret = security_sb_remount(sb, new_sec_opts);
1980                 security_free_mnt_opts(&new_sec_opts);
1981                 if (ret)
1982                         goto restore;
1983         }
1984
1985         ret = btrfs_parse_options(fs_info, data, *flags);
1986         if (ret)
1987                 goto restore;
1988
1989         btrfs_remount_begin(fs_info, old_opts, *flags);
1990         btrfs_resize_thread_pool(fs_info,
1991                 fs_info->thread_pool_size, old_thread_pool_size);
1992
1993         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
1994             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
1995             (!sb_rdonly(sb) || (*flags & SB_RDONLY))) {
1996                 btrfs_warn(fs_info,
1997                 "remount supports changing free space tree only from ro to rw");
1998                 /* Make sure free space cache options match the state on disk */
1999                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2000                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
2001                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
2002                 }
2003                 if (btrfs_free_space_cache_v1_active(fs_info)) {
2004                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
2005                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
2006                 }
2007         }
2008
2009         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
2010                 goto out;
2011
2012         if (*flags & SB_RDONLY) {
2013                 /*
2014                  * this also happens on 'umount -rf' or on shutdown, when
2015                  * the filesystem is busy.
2016                  */
2017                 cancel_work_sync(&fs_info->async_reclaim_work);
2018                 cancel_work_sync(&fs_info->async_data_reclaim_work);
2019
2020                 btrfs_discard_cleanup(fs_info);
2021
2022                 /* wait for the uuid_scan task to finish */
2023                 down(&fs_info->uuid_tree_rescan_sem);
2024                 /* avoid complains from lockdep et al. */
2025                 up(&fs_info->uuid_tree_rescan_sem);
2026
2027                 btrfs_set_sb_rdonly(sb);
2028
2029                 /*
2030                  * Setting SB_RDONLY will put the cleaner thread to
2031                  * sleep at the next loop if it's already active.
2032                  * If it's already asleep, we'll leave unused block
2033                  * groups on disk until we're mounted read-write again
2034                  * unless we clean them up here.
2035                  */
2036                 btrfs_delete_unused_bgs(fs_info);
2037
2038                 /*
2039                  * The cleaner task could be already running before we set the
2040                  * flag BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).
2041                  * We must make sure that after we finish the remount, i.e. after
2042                  * we call btrfs_commit_super(), the cleaner can no longer start
2043                  * a transaction - either because it was dropping a dead root,
2044                  * running delayed iputs or deleting an unused block group (the
2045                  * cleaner picked a block group from the list of unused block
2046                  * groups before we were able to in the previous call to
2047                  * btrfs_delete_unused_bgs()).
2048                  */
2049                 wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING,
2050                             TASK_UNINTERRUPTIBLE);
2051
2052                 /*
2053                  * We've set the superblock to RO mode, so we might have made
2054                  * the cleaner task sleep without running all pending delayed
2055                  * iputs. Go through all the delayed iputs here, so that if an
2056                  * unmount happens without remounting RW we don't end up at
2057                  * finishing close_ctree() with a non-empty list of delayed
2058                  * iputs.
2059                  */
2060                 btrfs_run_delayed_iputs(fs_info);
2061
2062                 btrfs_dev_replace_suspend_for_unmount(fs_info);
2063                 btrfs_scrub_cancel(fs_info);
2064                 btrfs_pause_balance(fs_info);
2065
2066                 /*
2067                  * Pause the qgroup rescan worker if it is running. We don't want
2068                  * it to be still running after we are in RO mode, as after that,
2069                  * by the time we unmount, it might have left a transaction open,
2070                  * so we would leak the transaction and/or crash.
2071                  */
2072                 btrfs_qgroup_wait_for_completion(fs_info, false);
2073
2074                 ret = btrfs_commit_super(fs_info);
2075                 if (ret)
2076                         goto restore;
2077         } else {
2078                 if (BTRFS_FS_ERROR(fs_info)) {
2079                         btrfs_err(fs_info,
2080                                 "Remounting read-write after error is not allowed");
2081                         ret = -EINVAL;
2082                         goto restore;
2083                 }
2084                 if (fs_info->fs_devices->rw_devices == 0) {
2085                         ret = -EACCES;
2086                         goto restore;
2087                 }
2088
2089                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
2090                         btrfs_warn(fs_info,
2091                 "too many missing devices, writable remount is not allowed");
2092                         ret = -EACCES;
2093                         goto restore;
2094                 }
2095
2096                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
2097                         btrfs_warn(fs_info,
2098                 "mount required to replay tree-log, cannot remount read-write");
2099                         ret = -EINVAL;
2100                         goto restore;
2101                 }
2102
2103                 /*
2104                  * NOTE: when remounting with a change that does writes, don't
2105                  * put it anywhere above this point, as we are not sure to be
2106                  * safe to write until we pass the above checks.
2107                  */
2108                 ret = btrfs_start_pre_rw_mount(fs_info);
2109                 if (ret)
2110                         goto restore;
2111
2112                 btrfs_clear_sb_rdonly(sb);
2113
2114                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
2115         }
2116 out:
2117         /*
2118          * We need to set SB_I_VERSION here otherwise it'll get cleared by VFS,
2119          * since the absence of the flag means it can be toggled off by remount.
2120          */
2121         *flags |= SB_I_VERSION;
2122
2123         wake_up_process(fs_info->transaction_kthread);
2124         btrfs_remount_cleanup(fs_info, old_opts);
2125         btrfs_clear_oneshot_options(fs_info);
2126         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2127
2128         return 0;
2129
2130 restore:
2131         /* We've hit an error - don't reset SB_RDONLY */
2132         if (sb_rdonly(sb))
2133                 old_flags |= SB_RDONLY;
2134         if (!(old_flags & SB_RDONLY))
2135                 clear_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2136         sb->s_flags = old_flags;
2137         fs_info->mount_opt = old_opts;
2138         fs_info->compress_type = old_compress_type;
2139         fs_info->max_inline = old_max_inline;
2140         btrfs_resize_thread_pool(fs_info,
2141                 old_thread_pool_size, fs_info->thread_pool_size);
2142         fs_info->metadata_ratio = old_metadata_ratio;
2143         btrfs_remount_cleanup(fs_info, old_opts);
2144         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2145
2146         return ret;
2147 }
2148
2149 /* Used to sort the devices by max_avail(descending sort) */
2150 static int btrfs_cmp_device_free_bytes(const void *a, const void *b)
2151 {
2152         const struct btrfs_device_info *dev_info1 = a;
2153         const struct btrfs_device_info *dev_info2 = b;
2154
2155         if (dev_info1->max_avail > dev_info2->max_avail)
2156                 return -1;
2157         else if (dev_info1->max_avail < dev_info2->max_avail)
2158                 return 1;
2159         return 0;
2160 }
2161
2162 /*
2163  * sort the devices by max_avail, in which max free extent size of each device
2164  * is stored.(Descending Sort)
2165  */
2166 static inline void btrfs_descending_sort_devices(
2167                                         struct btrfs_device_info *devices,
2168                                         size_t nr_devices)
2169 {
2170         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
2171              btrfs_cmp_device_free_bytes, NULL);
2172 }
2173
2174 /*
2175  * The helper to calc the free space on the devices that can be used to store
2176  * file data.
2177  */
2178 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
2179                                               u64 *free_bytes)
2180 {
2181         struct btrfs_device_info *devices_info;
2182         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2183         struct btrfs_device *device;
2184         u64 type;
2185         u64 avail_space;
2186         u64 min_stripe_size;
2187         int num_stripes = 1;
2188         int i = 0, nr_devices;
2189         const struct btrfs_raid_attr *rattr;
2190
2191         /*
2192          * We aren't under the device list lock, so this is racy-ish, but good
2193          * enough for our purposes.
2194          */
2195         nr_devices = fs_info->fs_devices->open_devices;
2196         if (!nr_devices) {
2197                 smp_mb();
2198                 nr_devices = fs_info->fs_devices->open_devices;
2199                 ASSERT(nr_devices);
2200                 if (!nr_devices) {
2201                         *free_bytes = 0;
2202                         return 0;
2203                 }
2204         }
2205
2206         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
2207                                GFP_KERNEL);
2208         if (!devices_info)
2209                 return -ENOMEM;
2210
2211         /* calc min stripe number for data space allocation */
2212         type = btrfs_data_alloc_profile(fs_info);
2213         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
2214
2215         if (type & BTRFS_BLOCK_GROUP_RAID0)
2216                 num_stripes = nr_devices;
2217         else if (type & BTRFS_BLOCK_GROUP_RAID1)
2218                 num_stripes = 2;
2219         else if (type & BTRFS_BLOCK_GROUP_RAID1C3)
2220                 num_stripes = 3;
2221         else if (type & BTRFS_BLOCK_GROUP_RAID1C4)
2222                 num_stripes = 4;
2223         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2224                 num_stripes = 4;
2225
2226         /* Adjust for more than 1 stripe per device */
2227         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
2228
2229         rcu_read_lock();
2230         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2231                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2232                                                 &device->dev_state) ||
2233                     !device->bdev ||
2234                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2235                         continue;
2236
2237                 if (i >= nr_devices)
2238                         break;
2239
2240                 avail_space = device->total_bytes - device->bytes_used;
2241
2242                 /* align with stripe_len */
2243                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
2244
2245                 /*
2246                  * In order to avoid overwriting the superblock on the drive,
2247                  * btrfs starts at an offset of at least 1MB when doing chunk
2248                  * allocation.
2249                  *
2250                  * This ensures we have at least min_stripe_size free space
2251                  * after excluding 1MB.
2252                  */
2253                 if (avail_space <= SZ_1M + min_stripe_size)
2254                         continue;
2255
2256                 avail_space -= SZ_1M;
2257
2258                 devices_info[i].dev = device;
2259                 devices_info[i].max_avail = avail_space;
2260
2261                 i++;
2262         }
2263         rcu_read_unlock();
2264
2265         nr_devices = i;
2266
2267         btrfs_descending_sort_devices(devices_info, nr_devices);
2268
2269         i = nr_devices - 1;
2270         avail_space = 0;
2271         while (nr_devices >= rattr->devs_min) {
2272                 num_stripes = min(num_stripes, nr_devices);
2273
2274                 if (devices_info[i].max_avail >= min_stripe_size) {
2275                         int j;
2276                         u64 alloc_size;
2277
2278                         avail_space += devices_info[i].max_avail * num_stripes;
2279                         alloc_size = devices_info[i].max_avail;
2280                         for (j = i + 1 - num_stripes; j <= i; j++)
2281                                 devices_info[j].max_avail -= alloc_size;
2282                 }
2283                 i--;
2284                 nr_devices--;
2285         }
2286
2287         kfree(devices_info);
2288         *free_bytes = avail_space;
2289         return 0;
2290 }
2291
2292 /*
2293  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2294  *
2295  * If there's a redundant raid level at DATA block groups, use the respective
2296  * multiplier to scale the sizes.
2297  *
2298  * Unused device space usage is based on simulating the chunk allocator
2299  * algorithm that respects the device sizes and order of allocations.  This is
2300  * a close approximation of the actual use but there are other factors that may
2301  * change the result (like a new metadata chunk).
2302  *
2303  * If metadata is exhausted, f_bavail will be 0.
2304  */
2305 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2306 {
2307         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2308         struct btrfs_super_block *disk_super = fs_info->super_copy;
2309         struct btrfs_space_info *found;
2310         u64 total_used = 0;
2311         u64 total_free_data = 0;
2312         u64 total_free_meta = 0;
2313         u32 bits = fs_info->sectorsize_bits;
2314         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
2315         unsigned factor = 1;
2316         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2317         int ret;
2318         u64 thresh = 0;
2319         int mixed = 0;
2320
2321         list_for_each_entry(found, &fs_info->space_info, list) {
2322                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2323                         int i;
2324
2325                         total_free_data += found->disk_total - found->disk_used;
2326                         total_free_data -=
2327                                 btrfs_account_ro_block_groups_free_space(found);
2328
2329                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2330                                 if (!list_empty(&found->block_groups[i]))
2331                                         factor = btrfs_bg_type_to_factor(
2332                                                 btrfs_raid_array[i].bg_flag);
2333                         }
2334                 }
2335
2336                 /*
2337                  * Metadata in mixed block goup profiles are accounted in data
2338                  */
2339                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2340                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2341                                 mixed = 1;
2342                         else
2343                                 total_free_meta += found->disk_total -
2344                                         found->disk_used;
2345                 }
2346
2347                 total_used += found->disk_used;
2348         }
2349
2350         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2351         buf->f_blocks >>= bits;
2352         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2353
2354         /* Account global block reserve as used, it's in logical size already */
2355         spin_lock(&block_rsv->lock);
2356         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2357         if (buf->f_bfree >= block_rsv->size >> bits)
2358                 buf->f_bfree -= block_rsv->size >> bits;
2359         else
2360                 buf->f_bfree = 0;
2361         spin_unlock(&block_rsv->lock);
2362
2363         buf->f_bavail = div_u64(total_free_data, factor);
2364         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2365         if (ret)
2366                 return ret;
2367         buf->f_bavail += div_u64(total_free_data, factor);
2368         buf->f_bavail = buf->f_bavail >> bits;
2369
2370         /*
2371          * We calculate the remaining metadata space minus global reserve. If
2372          * this is (supposedly) smaller than zero, there's no space. But this
2373          * does not hold in practice, the exhausted state happens where's still
2374          * some positive delta. So we apply some guesswork and compare the
2375          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2376          *
2377          * We probably cannot calculate the exact threshold value because this
2378          * depends on the internal reservations requested by various
2379          * operations, so some operations that consume a few metadata will
2380          * succeed even if the Avail is zero. But this is better than the other
2381          * way around.
2382          */
2383         thresh = SZ_4M;
2384
2385         /*
2386          * We only want to claim there's no available space if we can no longer
2387          * allocate chunks for our metadata profile and our global reserve will
2388          * not fit in the free metadata space.  If we aren't ->full then we
2389          * still can allocate chunks and thus are fine using the currently
2390          * calculated f_bavail.
2391          */
2392         if (!mixed && block_rsv->space_info->full &&
2393             total_free_meta - thresh < block_rsv->size)
2394                 buf->f_bavail = 0;
2395
2396         buf->f_type = BTRFS_SUPER_MAGIC;
2397         buf->f_bsize = dentry->d_sb->s_blocksize;
2398         buf->f_namelen = BTRFS_NAME_LEN;
2399
2400         /* We treat it as constant endianness (it doesn't matter _which_)
2401            because we want the fsid to come out the same whether mounted
2402            on a big-endian or little-endian host */
2403         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2404         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2405         /* Mask in the root object ID too, to disambiguate subvols */
2406         buf->f_fsid.val[0] ^=
2407                 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
2408         buf->f_fsid.val[1] ^=
2409                 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
2410
2411         return 0;
2412 }
2413
2414 static void btrfs_kill_super(struct super_block *sb)
2415 {
2416         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2417         kill_anon_super(sb);
2418         btrfs_free_fs_info(fs_info);
2419 }
2420
2421 static struct file_system_type btrfs_fs_type = {
2422         .owner          = THIS_MODULE,
2423         .name           = "btrfs",
2424         .mount          = btrfs_mount,
2425         .kill_sb        = btrfs_kill_super,
2426         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2427 };
2428
2429 static struct file_system_type btrfs_root_fs_type = {
2430         .owner          = THIS_MODULE,
2431         .name           = "btrfs",
2432         .mount          = btrfs_mount_root,
2433         .kill_sb        = btrfs_kill_super,
2434         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA | FS_ALLOW_IDMAP,
2435 };
2436
2437 MODULE_ALIAS_FS("btrfs");
2438
2439 static int btrfs_control_open(struct inode *inode, struct file *file)
2440 {
2441         /*
2442          * The control file's private_data is used to hold the
2443          * transaction when it is started and is used to keep
2444          * track of whether a transaction is already in progress.
2445          */
2446         file->private_data = NULL;
2447         return 0;
2448 }
2449
2450 /*
2451  * Used by /dev/btrfs-control for devices ioctls.
2452  */
2453 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2454                                 unsigned long arg)
2455 {
2456         struct btrfs_ioctl_vol_args *vol;
2457         struct btrfs_device *device = NULL;
2458         dev_t devt = 0;
2459         int ret = -ENOTTY;
2460
2461         if (!capable(CAP_SYS_ADMIN))
2462                 return -EPERM;
2463
2464         vol = memdup_user((void __user *)arg, sizeof(*vol));
2465         if (IS_ERR(vol))
2466                 return PTR_ERR(vol);
2467         vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2468
2469         switch (cmd) {
2470         case BTRFS_IOC_SCAN_DEV:
2471                 mutex_lock(&uuid_mutex);
2472                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2473                                                &btrfs_root_fs_type);
2474                 ret = PTR_ERR_OR_ZERO(device);
2475                 mutex_unlock(&uuid_mutex);
2476                 break;
2477         case BTRFS_IOC_FORGET_DEV:
2478                 if (vol->name[0] != 0) {
2479                         ret = lookup_bdev(vol->name, &devt);
2480                         if (ret)
2481                                 break;
2482                 }
2483                 ret = btrfs_forget_devices(devt);
2484                 break;
2485         case BTRFS_IOC_DEVICES_READY:
2486                 mutex_lock(&uuid_mutex);
2487                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2488                                                &btrfs_root_fs_type);
2489                 if (IS_ERR(device)) {
2490                         mutex_unlock(&uuid_mutex);
2491                         ret = PTR_ERR(device);
2492                         break;
2493                 }
2494                 ret = !(device->fs_devices->num_devices ==
2495                         device->fs_devices->total_devices);
2496                 mutex_unlock(&uuid_mutex);
2497                 break;
2498         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2499                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2500                 break;
2501         }
2502
2503         kfree(vol);
2504         return ret;
2505 }
2506
2507 static int btrfs_freeze(struct super_block *sb)
2508 {
2509         struct btrfs_trans_handle *trans;
2510         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2511         struct btrfs_root *root = fs_info->tree_root;
2512
2513         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2514         /*
2515          * We don't need a barrier here, we'll wait for any transaction that
2516          * could be in progress on other threads (and do delayed iputs that
2517          * we want to avoid on a frozen filesystem), or do the commit
2518          * ourselves.
2519          */
2520         trans = btrfs_attach_transaction_barrier(root);
2521         if (IS_ERR(trans)) {
2522                 /* no transaction, don't bother */
2523                 if (PTR_ERR(trans) == -ENOENT)
2524                         return 0;
2525                 return PTR_ERR(trans);
2526         }
2527         return btrfs_commit_transaction(trans);
2528 }
2529
2530 static int btrfs_unfreeze(struct super_block *sb)
2531 {
2532         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2533
2534         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2535         return 0;
2536 }
2537
2538 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2539 {
2540         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2541
2542         /*
2543          * There should be always a valid pointer in latest_dev, it may be stale
2544          * for a short moment in case it's being deleted but still valid until
2545          * the end of RCU grace period.
2546          */
2547         rcu_read_lock();
2548         seq_escape(m, rcu_str_deref(fs_info->fs_devices->latest_dev->name), " \t\n\\");
2549         rcu_read_unlock();
2550
2551         return 0;
2552 }
2553
2554 static const struct super_operations btrfs_super_ops = {
2555         .drop_inode     = btrfs_drop_inode,
2556         .evict_inode    = btrfs_evict_inode,
2557         .put_super      = btrfs_put_super,
2558         .sync_fs        = btrfs_sync_fs,
2559         .show_options   = btrfs_show_options,
2560         .show_devname   = btrfs_show_devname,
2561         .alloc_inode    = btrfs_alloc_inode,
2562         .destroy_inode  = btrfs_destroy_inode,
2563         .free_inode     = btrfs_free_inode,
2564         .statfs         = btrfs_statfs,
2565         .remount_fs     = btrfs_remount,
2566         .freeze_fs      = btrfs_freeze,
2567         .unfreeze_fs    = btrfs_unfreeze,
2568 };
2569
2570 static const struct file_operations btrfs_ctl_fops = {
2571         .open = btrfs_control_open,
2572         .unlocked_ioctl  = btrfs_control_ioctl,
2573         .compat_ioctl = compat_ptr_ioctl,
2574         .owner   = THIS_MODULE,
2575         .llseek = noop_llseek,
2576 };
2577
2578 static struct miscdevice btrfs_misc = {
2579         .minor          = BTRFS_MINOR,
2580         .name           = "btrfs-control",
2581         .fops           = &btrfs_ctl_fops
2582 };
2583
2584 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2585 MODULE_ALIAS("devname:btrfs-control");
2586
2587 static int __init btrfs_interface_init(void)
2588 {
2589         return misc_register(&btrfs_misc);
2590 }
2591
2592 static __cold void btrfs_interface_exit(void)
2593 {
2594         misc_deregister(&btrfs_misc);
2595 }
2596
2597 static void __init btrfs_print_mod_info(void)
2598 {
2599         static const char options[] = ""
2600 #ifdef CONFIG_BTRFS_DEBUG
2601                         ", debug=on"
2602 #endif
2603 #ifdef CONFIG_BTRFS_ASSERT
2604                         ", assert=on"
2605 #endif
2606 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2607                         ", integrity-checker=on"
2608 #endif
2609 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2610                         ", ref-verify=on"
2611 #endif
2612 #ifdef CONFIG_BLK_DEV_ZONED
2613                         ", zoned=yes"
2614 #else
2615                         ", zoned=no"
2616 #endif
2617 #ifdef CONFIG_FS_VERITY
2618                         ", fsverity=yes"
2619 #else
2620                         ", fsverity=no"
2621 #endif
2622                         ;
2623         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2624 }
2625
2626 static int __init init_btrfs_fs(void)
2627 {
2628         int err;
2629
2630         btrfs_props_init();
2631
2632         err = btrfs_init_sysfs();
2633         if (err)
2634                 return err;
2635
2636         btrfs_init_compress();
2637
2638         err = btrfs_init_cachep();
2639         if (err)
2640                 goto free_compress;
2641
2642         err = extent_io_init();
2643         if (err)
2644                 goto free_cachep;
2645
2646         err = extent_state_cache_init();
2647         if (err)
2648                 goto free_extent_io;
2649
2650         err = extent_map_init();
2651         if (err)
2652                 goto free_extent_state_cache;
2653
2654         err = ordered_data_init();
2655         if (err)
2656                 goto free_extent_map;
2657
2658         err = btrfs_delayed_inode_init();
2659         if (err)
2660                 goto free_ordered_data;
2661
2662         err = btrfs_auto_defrag_init();
2663         if (err)
2664                 goto free_delayed_inode;
2665
2666         err = btrfs_delayed_ref_init();
2667         if (err)
2668                 goto free_auto_defrag;
2669
2670         err = btrfs_prelim_ref_init();
2671         if (err)
2672                 goto free_delayed_ref;
2673
2674         err = btrfs_end_io_wq_init();
2675         if (err)
2676                 goto free_prelim_ref;
2677
2678         err = btrfs_interface_init();
2679         if (err)
2680                 goto free_end_io_wq;
2681
2682         btrfs_print_mod_info();
2683
2684         err = btrfs_run_sanity_tests();
2685         if (err)
2686                 goto unregister_ioctl;
2687
2688         err = register_filesystem(&btrfs_fs_type);
2689         if (err)
2690                 goto unregister_ioctl;
2691
2692         return 0;
2693
2694 unregister_ioctl:
2695         btrfs_interface_exit();
2696 free_end_io_wq:
2697         btrfs_end_io_wq_exit();
2698 free_prelim_ref:
2699         btrfs_prelim_ref_exit();
2700 free_delayed_ref:
2701         btrfs_delayed_ref_exit();
2702 free_auto_defrag:
2703         btrfs_auto_defrag_exit();
2704 free_delayed_inode:
2705         btrfs_delayed_inode_exit();
2706 free_ordered_data:
2707         ordered_data_exit();
2708 free_extent_map:
2709         extent_map_exit();
2710 free_extent_state_cache:
2711         extent_state_cache_exit();
2712 free_extent_io:
2713         extent_io_exit();
2714 free_cachep:
2715         btrfs_destroy_cachep();
2716 free_compress:
2717         btrfs_exit_compress();
2718         btrfs_exit_sysfs();
2719
2720         return err;
2721 }
2722
2723 static void __exit exit_btrfs_fs(void)
2724 {
2725         btrfs_destroy_cachep();
2726         btrfs_delayed_ref_exit();
2727         btrfs_auto_defrag_exit();
2728         btrfs_delayed_inode_exit();
2729         btrfs_prelim_ref_exit();
2730         ordered_data_exit();
2731         extent_map_exit();
2732         extent_state_cache_exit();
2733         extent_io_exit();
2734         btrfs_interface_exit();
2735         btrfs_end_io_wq_exit();
2736         unregister_filesystem(&btrfs_fs_type);
2737         btrfs_exit_sysfs();
2738         btrfs_cleanup_fs_uuids();
2739         btrfs_exit_compress();
2740 }
2741
2742 late_initcall(init_btrfs_fs);
2743 module_exit(exit_btrfs_fs)
2744
2745 MODULE_LICENSE("GPL");
2746 MODULE_SOFTDEP("pre: crc32c");
2747 MODULE_SOFTDEP("pre: xxhash64");
2748 MODULE_SOFTDEP("pre: sha256");
2749 MODULE_SOFTDEP("pre: blake2b-256");