fs/btrfs/super.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/time.h>
  26 #include <linux/init.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mount.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/parser.h>
  37 #include <linux/ctype.h>
  38 #include <linux/namei.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/magic.h>
  41 #include <linux/slab.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/ratelimit.h>
  44 #include <linux/btrfs.h>
  45 #include "delayed-inode.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "hash.h"
  52 #include "props.h"
  53 #include "xattr.h"
  54 #include "volumes.h"
  55 #include "export.h"
  56 #include "compression.h"
  57 #include "rcu-string.h"
  58 #include "dev-replace.h"
  59 #include "free-space-cache.h"
  60 #include "backref.h"
  61 #include "tests/btrfs-tests.h"
  62
  63 #include "qgroup.h"
  64 #include "backref.h"
  65 #define CREATE_TRACE_POINTS
  66 #include <trace/events/btrfs.h>
  67
  68 static const struct super_operations btrfs_super_ops;
  69 static struct file_system_type btrfs_fs_type;
  70
  71 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  72
  73 const char *btrfs_decode_error(int errno)
  74 {
  75         char *errstr = "unknown";
  76
  77         switch (errno) {
  78         case -EIO:
  79                 errstr = "IO failure";
  80                 break;
  81         case -ENOMEM:
  82                 errstr = "Out of memory";
  83                 break;
  84         case -EROFS:
  85                 errstr = "Readonly filesystem";
  86                 break;
  87         case -EEXIST:
  88                 errstr = "Object already exists";
  89                 break;
  90         case -ENOSPC:
  91                 errstr = "No space left";
  92                 break;
  93         case -ENOENT:
  94                 errstr = "No such entry";
  95                 break;
  96         }
  97
  98         return errstr;
  99 }
 100
 101 /* btrfs handle error by forcing the filesystem readonly */
 102 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 103 {
 104         struct super_block *sb = fs_info->sb;
 105
 106         if (sb_rdonly(sb))
 107                 return;
 108
 109         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 110                 sb->s_flags |= MS_RDONLY;
 111                 btrfs_info(fs_info, "forced readonly");
 112                 /*
 113                  * Note that a running device replace operation is not
 114                  * canceled here although there is no way to update
 115                  * the progress. It would add the risk of a deadlock,
 116                  * therefore the canceling is omitted. The only penalty
 117                  * is that some I/O remains active until the procedure
 118                  * completes. The next time when the filesystem is
 119                  * mounted writeable again, the device replace
 120                  * operation continues.
 121                  */
 122         }
 123 }
 124
 125 /*
 126  * __btrfs_handle_fs_error decodes expected errors from the caller and
 127  * invokes the approciate error response.
 128  */
 129 __cold
 130 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
 131                        unsigned int line, int errno, const char *fmt, ...)
 132 {
 133         struct super_block *sb = fs_info->sb;
 134 #ifdef CONFIG_PRINTK
 135         const char *errstr;
 136 #endif
 137
 138         /*
 139          * Special case: if the error is EROFS, and we're already
 140          * under MS_RDONLY, then it is safe here.
 141          */
 142         if (errno == -EROFS && sb_rdonly(sb))
 143                 return;
 144
 145 #ifdef CONFIG_PRINTK
 146         errstr = btrfs_decode_error(errno);
 147         if (fmt) {
 148                 struct va_format vaf;
 149                 va_list args;
 150
 151                 va_start(args, fmt);
 152                 vaf.fmt = fmt;
 153                 vaf.va = &args;
 154
 155                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 156                         sb->s_id, function, line, errno, errstr, &vaf);
 157                 va_end(args);
 158         } else {
 159                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 160                         sb->s_id, function, line, errno, errstr);
 161         }
 162 #endif
 163
 164         /*
 165          * Today we only save the error info to memory.  Long term we'll
 166          * also send it down to the disk
 167          */
 168         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 169
 170         /* Don't go through full error handling during mount */
 171         if (sb->s_flags & MS_BORN)
 172                 btrfs_handle_error(fs_info);
 173 }
 174
 175 #ifdef CONFIG_PRINTK
 176 static const char * const logtypes[] = {
 177         "emergency",
 178         "alert",
 179         "critical",
 180         "error",
 181         "warning",
 182         "notice",
 183         "info",
 184         "debug",
 185 };
 186
 187
 188 /*
 189  * Use one ratelimit state per log level so that a flood of less important
 190  * messages doesn't cause more important ones to be dropped.
 191  */
 192 static struct ratelimit_state printk_limits[] = {
 193         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
 194         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
 195         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
 196         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
 197         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
 198         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
 199         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
 200         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
 201 };
 202
 203 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 204 {
 205         struct super_block *sb = fs_info->sb;
 206         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
 207         struct va_format vaf;
 208         va_list args;
 209         int kern_level;
 210         const char *type = logtypes[4];
 211         struct ratelimit_state *ratelimit = &printk_limits[4];
 212
 213         va_start(args, fmt);
 214
 215         while ((kern_level = printk_get_level(fmt)) != 0) {
 216                 size_t size = printk_skip_level(fmt) - fmt;
 217
 218                 if (kern_level >= '0' && kern_level <= '7') {
 219                         memcpy(lvl, fmt,  size);
 220                         lvl[size] = '\0';
 221                         type = logtypes[kern_level - '0'];
 222                         ratelimit = &printk_limits[kern_level - '0'];
 223                 }
 224                 fmt += size;
 225         }
 226
 227         vaf.fmt = fmt;
 228         vaf.va = &args;
 229
 230         if (__ratelimit(ratelimit))
 231                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 232
 233         va_end(args);
 234 }
 235 #endif
 236
 237 /*
 238  * We only mark the transaction aborted and then set the file system read-only.
 239  * This will prevent new transactions from starting or trying to join this
 240  * one.
 241  *
 242  * This means that error recovery at the call site is limited to freeing
 243  * any local memory allocations and passing the error code up without
 244  * further cleanup. The transaction should complete as it normally would
 245  * in the call path but will return -EIO.
 246  *
 247  * We'll complete the cleanup in btrfs_end_transaction and
 248  * btrfs_commit_transaction.
 249  */
 250 __cold
 251 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 252                                const char *function,
 253                                unsigned int line, int errno)
 254 {
 255         struct btrfs_fs_info *fs_info = trans->fs_info;
 256
 257         trans->aborted = errno;
 258         /* Nothing used. The other threads that have joined this
 259          * transaction may be able to continue. */
 260         if (!trans->dirty && list_empty(&trans->new_bgs)) {
 261                 const char *errstr;
 262
 263                 errstr = btrfs_decode_error(errno);
 264                 btrfs_warn(fs_info,
 265                            "%s:%d: Aborting unused transaction(%s).",
 266                            function, line, errstr);
 267                 return;
 268         }
 269         WRITE_ONCE(trans->transaction->aborted, errno);
 270         /* Wake up anybody who may be waiting on this transaction */
 271         wake_up(&fs_info->transaction_wait);
 272         wake_up(&fs_info->transaction_blocked_wait);
 273         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
 274 }
 275 /*
 276  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 277  * issues an alert, and either panics or BUGs, depending on mount options.
 278  */
 279 __cold
 280 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 281                    unsigned int line, int errno, const char *fmt, ...)
 282 {
 283         char *s_id = "<unknown>";
 284         const char *errstr;
 285         struct va_format vaf = { .fmt = fmt };
 286         va_list args;
 287
 288         if (fs_info)
 289                 s_id = fs_info->sb->s_id;
 290
 291         va_start(args, fmt);
 292         vaf.va = &args;
 293
 294         errstr = btrfs_decode_error(errno);
 295         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 296                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 297                         s_id, function, line, &vaf, errno, errstr);
 298
 299         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 300                    function, line, &vaf, errno, errstr);
 301         va_end(args);
 302         /* Caller calls BUG() */
 303 }
 304
 305 static void btrfs_put_super(struct super_block *sb)
 306 {
 307         close_ctree(btrfs_sb(sb));
 308 }
 309
 310 enum {
 311         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 312         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 313         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 314         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 315         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 316         Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
 317         Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
 318         Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
 319         Opt_skip_balance, Opt_check_integrity,
 320         Opt_check_integrity_including_extent_data,
 321         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 322         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
 323         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
 324         Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
 325         Opt_nologreplay, Opt_norecovery,
 326 #ifdef CONFIG_BTRFS_DEBUG
 327         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 328 #endif
 329         Opt_err,
 330 };
 331
 332 static const match_table_t tokens = {
 333         {Opt_degraded, "degraded"},
 334         {Opt_subvol, "subvol=%s"},
 335         {Opt_subvolid, "subvolid=%s"},
 336         {Opt_device, "device=%s"},
 337         {Opt_nodatasum, "nodatasum"},
 338         {Opt_datasum, "datasum"},
 339         {Opt_nodatacow, "nodatacow"},
 340         {Opt_datacow, "datacow"},
 341         {Opt_nobarrier, "nobarrier"},
 342         {Opt_barrier, "barrier"},
 343         {Opt_max_inline, "max_inline=%s"},
 344         {Opt_alloc_start, "alloc_start=%s"},
 345         {Opt_thread_pool, "thread_pool=%d"},
 346         {Opt_compress, "compress"},
 347         {Opt_compress_type, "compress=%s"},
 348         {Opt_compress_force, "compress-force"},
 349         {Opt_compress_force_type, "compress-force=%s"},
 350         {Opt_ssd, "ssd"},
 351         {Opt_ssd_spread, "ssd_spread"},
 352         {Opt_nossd, "nossd"},
 353         {Opt_acl, "acl"},
 354         {Opt_noacl, "noacl"},
 355         {Opt_notreelog, "notreelog"},
 356         {Opt_treelog, "treelog"},
 357         {Opt_nologreplay, "nologreplay"},
 358         {Opt_norecovery, "norecovery"},
 359         {Opt_flushoncommit, "flushoncommit"},
 360         {Opt_noflushoncommit, "noflushoncommit"},
 361         {Opt_ratio, "metadata_ratio=%d"},
 362         {Opt_discard, "discard"},
 363         {Opt_nodiscard, "nodiscard"},
 364         {Opt_space_cache, "space_cache"},
 365         {Opt_space_cache_version, "space_cache=%s"},
 366         {Opt_clear_cache, "clear_cache"},
 367         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 368         {Opt_enospc_debug, "enospc_debug"},
 369         {Opt_noenospc_debug, "noenospc_debug"},
 370         {Opt_subvolrootid, "subvolrootid=%d"},
 371         {Opt_defrag, "autodefrag"},
 372         {Opt_nodefrag, "noautodefrag"},
 373         {Opt_inode_cache, "inode_cache"},
 374         {Opt_noinode_cache, "noinode_cache"},
 375         {Opt_no_space_cache, "nospace_cache"},
 376         {Opt_recovery, "recovery"}, /* deprecated */
 377         {Opt_usebackuproot, "usebackuproot"},
 378         {Opt_skip_balance, "skip_balance"},
 379         {Opt_check_integrity, "check_int"},
 380         {Opt_check_integrity_including_extent_data, "check_int_data"},
 381         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 382         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 383         {Opt_fatal_errors, "fatal_errors=%s"},
 384         {Opt_commit_interval, "commit=%d"},
 385 #ifdef CONFIG_BTRFS_DEBUG
 386         {Opt_fragment_data, "fragment=data"},
 387         {Opt_fragment_metadata, "fragment=metadata"},
 388         {Opt_fragment_all, "fragment=all"},
 389 #endif
 390         {Opt_err, NULL},
 391 };
 392
 393 /*
 394  * Regular mount options parser.  Everything that is needed only when
 395  * reading in a new superblock is parsed here.
 396  * XXX JDM: This needs to be cleaned up for remount.
 397  */
 398 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 399                         unsigned long new_flags)
 400 {
 401         substring_t args[MAX_OPT_ARGS];
 402         char *p, *num, *orig = NULL;
 403         u64 cache_gen;
 404         int intarg;
 405         int ret = 0;
 406         char *compress_type;
 407         bool compress_force = false;
 408         enum btrfs_compression_type saved_compress_type;
 409         bool saved_compress_force;
 410         int no_compress = 0;
 411
 412         cache_gen = btrfs_super_cache_generation(info->super_copy);
 413         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 414                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 415         else if (cache_gen)
 416                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 417
 418         /*
 419          * Even the options are empty, we still need to do extra check
 420          * against new flags
 421          */
 422         if (!options)
 423                 goto check;
 424
 425         /*
 426          * strsep changes the string, duplicate it because parse_options
 427          * gets called twice
 428          */
 429         options = kstrdup(options, GFP_KERNEL);
 430         if (!options)
 431                 return -ENOMEM;
 432
 433         orig = options;
 434
 435         while ((p = strsep(&options, ",")) != NULL) {
 436                 int token;
 437                 if (!*p)
 438                         continue;
 439
 440                 token = match_token(p, tokens, args);
 441                 switch (token) {
 442                 case Opt_degraded:
 443                         btrfs_info(info, "allowing degraded mounts");
 444                         btrfs_set_opt(info->mount_opt, DEGRADED);
 445                         break;
 446                 case Opt_subvol:
 447                 case Opt_subvolid:
 448                 case Opt_subvolrootid:
 449                 case Opt_device:
 450                         /*
 451                          * These are parsed by btrfs_parse_early_options
 452                          * and can be happily ignored here.
 453                          */
 454                         break;
 455                 case Opt_nodatasum:
 456                         btrfs_set_and_info(info, NODATASUM,
 457                                            "setting nodatasum");
 458                         break;
 459                 case Opt_datasum:
 460                         if (btrfs_test_opt(info, NODATASUM)) {
 461                                 if (btrfs_test_opt(info, NODATACOW))
 462                                         btrfs_info(info,
 463                                                    "setting datasum, datacow enabled");
 464                                 else
 465                                         btrfs_info(info, "setting datasum");
 466                         }
 467                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 468                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 469                         break;
 470                 case Opt_nodatacow:
 471                         if (!btrfs_test_opt(info, NODATACOW)) {
 472                                 if (!btrfs_test_opt(info, COMPRESS) ||
 473                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 474                                         btrfs_info(info,
 475                                                    "setting nodatacow, compression disabled");
 476                                 } else {
 477                                         btrfs_info(info, "setting nodatacow");
 478                                 }
 479                         }
 480                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 481                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 482                         btrfs_set_opt(info->mount_opt, NODATACOW);
 483                         btrfs_set_opt(info->mount_opt, NODATASUM);
 484                         break;
 485                 case Opt_datacow:
 486                         btrfs_clear_and_info(info, NODATACOW,
 487                                              "setting datacow");
 488                         break;
 489                 case Opt_compress_force:
 490                 case Opt_compress_force_type:
 491                         compress_force = true;
 492                         /* Fallthrough */
 493                 case Opt_compress:
 494                 case Opt_compress_type:
 495                         saved_compress_type = btrfs_test_opt(info,
 496                                                              COMPRESS) ?
 497                                 info->compress_type : BTRFS_COMPRESS_NONE;
 498                         saved_compress_force =
 499                                 btrfs_test_opt(info, FORCE_COMPRESS);
 500                         if (token == Opt_compress ||
 501                             token == Opt_compress_force ||
 502                             strncmp(args[0].from, "zlib", 4) == 0) {
 503                                 compress_type = "zlib";
 504                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 505                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 506                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 507                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 508                                 no_compress = 0;
 509                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
 510                                 compress_type = "lzo";
 511                                 info->compress_type = BTRFS_COMPRESS_LZO;
 512                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 513                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 514                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 515                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 516                                 no_compress = 0;
 517                         } else if (strcmp(args[0].from, "zstd") == 0) {
 518                                 compress_type = "zstd";
 519                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
 520                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 521                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 522                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 523                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
 524                                 no_compress = 0;
 525                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 526                                 compress_type = "no";
 527                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 528                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 529                                 compress_force = false;
 530                                 no_compress++;
 531                         } else {
 532                                 ret = -EINVAL;
 533                                 goto out;
 534                         }
 535
 536                         if (compress_force) {
 537                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 538                         } else {
 539                                 /*
 540                                  * If we remount from compress-force=xxx to
 541                                  * compress=xxx, we need clear FORCE_COMPRESS
 542                                  * flag, otherwise, there is no way for users
 543                                  * to disable forcible compression separately.
 544                                  */
 545                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 546                         }
 547                         if ((btrfs_test_opt(info, COMPRESS) &&
 548                              (info->compress_type != saved_compress_type ||
 549                               compress_force != saved_compress_force)) ||
 550                             (!btrfs_test_opt(info, COMPRESS) &&
 551                              no_compress == 1)) {
 552                                 btrfs_info(info, "%s %s compression",
 553                                            (compress_force) ? "force" : "use",
 554                                            compress_type);
 555                         }
 556                         compress_force = false;
 557                         break;
 558                 case Opt_ssd:
 559                         btrfs_set_and_info(info, SSD,
 560                                            "enabling ssd optimizations");
 561                         btrfs_clear_opt(info->mount_opt, NOSSD);
 562                         break;
 563                 case Opt_ssd_spread:
 564                         btrfs_set_and_info(info, SSD,
 565                                            "enabling ssd optimizations");
 566                         btrfs_set_and_info(info, SSD_SPREAD,
 567                                            "using spread ssd allocation scheme");
 568                         btrfs_clear_opt(info->mount_opt, NOSSD);
 569                         break;
 570                 case Opt_nossd:
 571                         btrfs_set_opt(info->mount_opt, NOSSD);
 572                         btrfs_clear_and_info(info, SSD,
 573                                              "not using ssd optimizations");
 574                         btrfs_clear_and_info(info, SSD_SPREAD,
 575                                              "not using spread ssd allocation scheme");
 576                         break;
 577                 case Opt_barrier:
 578                         btrfs_clear_and_info(info, NOBARRIER,
 579                                              "turning on barriers");
 580                         break;
 581                 case Opt_nobarrier:
 582                         btrfs_set_and_info(info, NOBARRIER,
 583                                            "turning off barriers");
 584                         break;
 585                 case Opt_thread_pool:
 586                         ret = match_int(&args[0], &intarg);
 587                         if (ret) {
 588                                 goto out;
 589                         } else if (intarg > 0) {
 590                                 info->thread_pool_size = intarg;
 591                         } else {
 592                                 ret = -EINVAL;
 593                                 goto out;
 594                         }
 595                         break;
 596                 case Opt_max_inline:
 597                         num = match_strdup(&args[0]);
 598                         if (num) {
 599                                 info->max_inline = memparse(num, NULL);
 600                                 kfree(num);
 601
 602                                 if (info->max_inline) {
 603                                         info->max_inline = min_t(u64,
 604                                                 info->max_inline,
 605                                                 info->sectorsize);
 606                                 }
 607                                 btrfs_info(info, "max_inline at %llu",
 608                                            info->max_inline);
 609                         } else {
 610                                 ret = -ENOMEM;
 611                                 goto out;
 612                         }
 613                         break;
 614                 case Opt_alloc_start:
 615                         btrfs_info(info,
 616                                 "option alloc_start is obsolete, ignored");
 617                         break;
 618                 case Opt_acl:
 619 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 620                         info->sb->s_flags |= MS_POSIXACL;
 621                         break;
 622 #else
 623                         btrfs_err(info, "support for ACL not compiled in!");
 624                         ret = -EINVAL;
 625                         goto out;
 626 #endif
 627                 case Opt_noacl:
 628                         info->sb->s_flags &= ~MS_POSIXACL;
 629                         break;
 630                 case Opt_notreelog:
 631                         btrfs_set_and_info(info, NOTREELOG,
 632                                            "disabling tree log");
 633                         break;
 634                 case Opt_treelog:
 635                         btrfs_clear_and_info(info, NOTREELOG,
 636                                              "enabling tree log");
 637                         break;
 638                 case Opt_norecovery:
 639                 case Opt_nologreplay:
 640                         btrfs_set_and_info(info, NOLOGREPLAY,
 641                                            "disabling log replay at mount time");
 642                         break;
 643                 case Opt_flushoncommit:
 644                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 645                                            "turning on flush-on-commit");
 646                         break;
 647                 case Opt_noflushoncommit:
 648                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 649                                              "turning off flush-on-commit");
 650                         break;
 651                 case Opt_ratio:
 652                         ret = match_int(&args[0], &intarg);
 653                         if (ret) {
 654                                 goto out;
 655                         } else if (intarg >= 0) {
 656                                 info->metadata_ratio = intarg;
 657                                 btrfs_info(info, "metadata ratio %d",
 658                                            info->metadata_ratio);
 659                         } else {
 660                                 ret = -EINVAL;
 661                                 goto out;
 662                         }
 663                         break;
 664                 case Opt_discard:
 665                         btrfs_set_and_info(info, DISCARD,
 666                                            "turning on discard");
 667                         break;
 668                 case Opt_nodiscard:
 669                         btrfs_clear_and_info(info, DISCARD,
 670                                              "turning off discard");
 671                         break;
 672                 case Opt_space_cache:
 673                 case Opt_space_cache_version:
 674                         if (token == Opt_space_cache ||
 675                             strcmp(args[0].from, "v1") == 0) {
 676                                 btrfs_clear_opt(info->mount_opt,
 677                                                 FREE_SPACE_TREE);
 678                                 btrfs_set_and_info(info, SPACE_CACHE,
 679                                            "enabling disk space caching");
 680                         } else if (strcmp(args[0].from, "v2") == 0) {
 681                                 btrfs_clear_opt(info->mount_opt,
 682                                                 SPACE_CACHE);
 683                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 684                                                    "enabling free space tree");
 685                         } else {
 686                                 ret = -EINVAL;
 687                                 goto out;
 688                         }
 689                         break;
 690                 case Opt_rescan_uuid_tree:
 691                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 692                         break;
 693                 case Opt_no_space_cache:
 694                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 695                                 btrfs_clear_and_info(info, SPACE_CACHE,
 696                                              "disabling disk space caching");
 697                         }
 698                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 699                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 700                                              "disabling free space tree");
 701                         }
 702                         break;
 703                 case Opt_inode_cache:
 704                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 705                                            "enabling inode map caching");
 706                         break;
 707                 case Opt_noinode_cache:
 708                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 709                                              "disabling inode map caching");
 710                         break;
 711                 case Opt_clear_cache:
 712                         btrfs_set_and_info(info, CLEAR_CACHE,
 713                                            "force clearing of disk cache");
 714                         break;
 715                 case Opt_user_subvol_rm_allowed:
 716                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 717                         break;
 718                 case Opt_enospc_debug:
 719                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 720                         break;
 721                 case Opt_noenospc_debug:
 722                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 723                         break;
 724                 case Opt_defrag:
 725                         btrfs_set_and_info(info, AUTO_DEFRAG,
 726                                            "enabling auto defrag");
 727                         break;
 728                 case Opt_nodefrag:
 729                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 730                                              "disabling auto defrag");
 731                         break;
 732                 case Opt_recovery:
 733                         btrfs_warn(info,
 734                                    "'recovery' is deprecated, use 'usebackuproot' instead");
 735                 case Opt_usebackuproot:
 736                         btrfs_info(info,
 737                                    "trying to use backup root at mount time");
 738                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 739                         break;
 740                 case Opt_skip_balance:
 741                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 742                         break;
 743 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 744                 case Opt_check_integrity_including_extent_data:
 745                         btrfs_info(info,
 746                                    "enabling check integrity including extent data");
 747                         btrfs_set_opt(info->mount_opt,
 748                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 749                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 750                         break;
 751                 case Opt_check_integrity:
 752                         btrfs_info(info, "enabling check integrity");
 753                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 754                         break;
 755                 case Opt_check_integrity_print_mask:
 756                         ret = match_int(&args[0], &intarg);
 757                         if (ret) {
 758                                 goto out;
 759                         } else if (intarg >= 0) {
 760                                 info->check_integrity_print_mask = intarg;
 761                                 btrfs_info(info,
 762                                            "check_integrity_print_mask 0x%x",
 763                                            info->check_integrity_print_mask);
 764                         } else {
 765                                 ret = -EINVAL;
 766                                 goto out;
 767                         }
 768                         break;
 769 #else
 770                 case Opt_check_integrity_including_extent_data:
 771                 case Opt_check_integrity:
 772                 case Opt_check_integrity_print_mask:
 773                         btrfs_err(info,
 774                                   "support for check_integrity* not compiled in!");
 775                         ret = -EINVAL;
 776                         goto out;
 777 #endif
 778                 case Opt_fatal_errors:
 779                         if (strcmp(args[0].from, "panic") == 0)
 780                                 btrfs_set_opt(info->mount_opt,
 781                                               PANIC_ON_FATAL_ERROR);
 782                         else if (strcmp(args[0].from, "bug") == 0)
 783                                 btrfs_clear_opt(info->mount_opt,
 784                                               PANIC_ON_FATAL_ERROR);
 785                         else {
 786                                 ret = -EINVAL;
 787                                 goto out;
 788                         }
 789                         break;
 790                 case Opt_commit_interval:
 791                         intarg = 0;
 792                         ret = match_int(&args[0], &intarg);
 793                         if (ret < 0) {
 794                                 btrfs_err(info, "invalid commit interval");
 795                                 ret = -EINVAL;
 796                                 goto out;
 797                         }
 798                         if (intarg > 0) {
 799                                 if (intarg > 300) {
 800                                         btrfs_warn(info,
 801                                                 "excessive commit interval %d",
 802                                                 intarg);
 803                                 }
 804                                 info->commit_interval = intarg;
 805                         } else {
 806                                 btrfs_info(info,
 807                                            "using default commit interval %ds",
 808                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
 809                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 810                         }
 811                         break;
 812 #ifdef CONFIG_BTRFS_DEBUG
 813                 case Opt_fragment_all:
 814                         btrfs_info(info, "fragmenting all space");
 815                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 816                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 817                         break;
 818                 case Opt_fragment_metadata:
 819                         btrfs_info(info, "fragmenting metadata");
 820                         btrfs_set_opt(info->mount_opt,
 821                                       FRAGMENT_METADATA);
 822                         break;
 823                 case Opt_fragment_data:
 824                         btrfs_info(info, "fragmenting data");
 825                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 826                         break;
 827 #endif
 828                 case Opt_err:
 829                         btrfs_info(info, "unrecognized mount option '%s'", p);
 830                         ret = -EINVAL;
 831                         goto out;
 832                 default:
 833                         break;
 834                 }
 835         }
 836 check:
 837         /*
 838          * Extra check for current option against current flag
 839          */
 840         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
 841                 btrfs_err(info,
 842                           "nologreplay must be used with ro mount option");
 843                 ret = -EINVAL;
 844         }
 845 out:
 846         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
 847             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
 848             !btrfs_test_opt(info, CLEAR_CACHE)) {
 849                 btrfs_err(info, "cannot disable free space tree");
 850                 ret = -EINVAL;
 851
 852         }
 853         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
 854                 btrfs_info(info, "disk space caching is enabled");
 855         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
 856                 btrfs_info(info, "using free space tree");
 857         kfree(orig);
 858         return ret;
 859 }
 860
 861 /*
 862  * Parse mount options that are required early in the mount process.
 863  *
 864  * All other options will be parsed on much later in the mount process and
 865  * only when we need to allocate a new super block.
 866  */
 867 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 868                 void *holder, char **subvol_name, u64 *subvol_objectid,
 869                 struct btrfs_fs_devices **fs_devices)
 870 {
 871         substring_t args[MAX_OPT_ARGS];
 872         char *device_name, *opts, *orig, *p;
 873         char *num = NULL;
 874         int error = 0;
 875
 876         if (!options)
 877                 return 0;
 878
 879         /*
 880          * strsep changes the string, duplicate it because parse_options
 881          * gets called twice
 882          */
 883         opts = kstrdup(options, GFP_KERNEL);
 884         if (!opts)
 885                 return -ENOMEM;
 886         orig = opts;
 887
 888         while ((p = strsep(&opts, ",")) != NULL) {
 889                 int token;
 890                 if (!*p)
 891                         continue;
 892
 893                 token = match_token(p, tokens, args);
 894                 switch (token) {
 895                 case Opt_subvol:
 896                         kfree(*subvol_name);
 897                         *subvol_name = match_strdup(&args[0]);
 898                         if (!*subvol_name) {
 899                                 error = -ENOMEM;
 900                                 goto out;
 901                         }
 902                         break;
 903                 case Opt_subvolid:
 904                         num = match_strdup(&args[0]);
 905                         if (num) {
 906                                 *subvol_objectid = memparse(num, NULL);
 907                                 kfree(num);
 908                                 /* we want the original fs_tree */
 909                                 if (!*subvol_objectid)
 910                                         *subvol_objectid =
 911                                                 BTRFS_FS_TREE_OBJECTID;
 912                         } else {
 913                                 error = -EINVAL;
 914                                 goto out;
 915                         }
 916                         break;
 917                 case Opt_subvolrootid:
 918                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
 919                         break;
 920                 case Opt_device:
 921                         device_name = match_strdup(&args[0]);
 922                         if (!device_name) {
 923                                 error = -ENOMEM;
 924                                 goto out;
 925                         }
 926                         error = btrfs_scan_one_device(device_name,
 927                                         flags, holder, fs_devices);
 928                         kfree(device_name);
 929                         if (error)
 930                                 goto out;
 931                         break;
 932                 default:
 933                         break;
 934                 }
 935         }
 936
 937 out:
 938         kfree(orig);
 939         return error;
 940 }
 941
 942 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 943                                            u64 subvol_objectid)
 944 {
 945         struct btrfs_root *root = fs_info->tree_root;
 946         struct btrfs_root *fs_root;
 947         struct btrfs_root_ref *root_ref;
 948         struct btrfs_inode_ref *inode_ref;
 949         struct btrfs_key key;
 950         struct btrfs_path *path = NULL;
 951         char *name = NULL, *ptr;
 952         u64 dirid;
 953         int len;
 954         int ret;
 955
 956         path = btrfs_alloc_path();
 957         if (!path) {
 958                 ret = -ENOMEM;
 959                 goto err;
 960         }
 961         path->leave_spinning = 1;
 962
 963         name = kmalloc(PATH_MAX, GFP_KERNEL);
 964         if (!name) {
 965                 ret = -ENOMEM;
 966                 goto err;
 967         }
 968         ptr = name + PATH_MAX - 1;
 969         ptr[0] = '\0';
 970
 971         /*
 972          * Walk up the subvolume trees in the tree of tree roots by root
 973          * backrefs until we hit the top-level subvolume.
 974          */
 975         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 976                 key.objectid = subvol_objectid;
 977                 key.type = BTRFS_ROOT_BACKREF_KEY;
 978                 key.offset = (u64)-1;
 979
 980                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 981                 if (ret < 0) {
 982                         goto err;
 983                 } else if (ret > 0) {
 984                         ret = btrfs_previous_item(root, path, subvol_objectid,
 985                                                   BTRFS_ROOT_BACKREF_KEY);
 986                         if (ret < 0) {
 987                                 goto err;
 988                         } else if (ret > 0) {
 989                                 ret = -ENOENT;
 990                                 goto err;
 991                         }
 992                 }
 993
 994                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 995                 subvol_objectid = key.offset;
 996
 997                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 998                                           struct btrfs_root_ref);
 999                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1000                 ptr -= len + 1;
1001                 if (ptr < name) {
1002                         ret = -ENAMETOOLONG;
1003                         goto err;
1004                 }
1005                 read_extent_buffer(path->nodes[0], ptr + 1,
1006                                    (unsigned long)(root_ref + 1), len);
1007                 ptr[0] = '/';
1008                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1009                 btrfs_release_path(path);
1010
1011                 key.objectid = subvol_objectid;
1012                 key.type = BTRFS_ROOT_ITEM_KEY;
1013                 key.offset = (u64)-1;
1014                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1015                 if (IS_ERR(fs_root)) {
1016                         ret = PTR_ERR(fs_root);
1017                         goto err;
1018                 }
1019
1020                 /*
1021                  * Walk up the filesystem tree by inode refs until we hit the
1022                  * root directory.
1023                  */
1024                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1025                         key.objectid = dirid;
1026                         key.type = BTRFS_INODE_REF_KEY;
1027                         key.offset = (u64)-1;
1028
1029                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1030                         if (ret < 0) {
1031                                 goto err;
1032                         } else if (ret > 0) {
1033                                 ret = btrfs_previous_item(fs_root, path, dirid,
1034                                                           BTRFS_INODE_REF_KEY);
1035                                 if (ret < 0) {
1036                                         goto err;
1037                                 } else if (ret > 0) {
1038                                         ret = -ENOENT;
1039                                         goto err;
1040                                 }
1041                         }
1042
1043                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1044                         dirid = key.offset;
1045
1046                         inode_ref = btrfs_item_ptr(path->nodes[0],
1047                                                    path->slots[0],
1048                                                    struct btrfs_inode_ref);
1049                         len = btrfs_inode_ref_name_len(path->nodes[0],
1050                                                        inode_ref);
1051                         ptr -= len + 1;
1052                         if (ptr < name) {
1053                                 ret = -ENAMETOOLONG;
1054                                 goto err;
1055                         }
1056                         read_extent_buffer(path->nodes[0], ptr + 1,
1057                                            (unsigned long)(inode_ref + 1), len);
1058                         ptr[0] = '/';
1059                         btrfs_release_path(path);
1060                 }
1061         }
1062
1063         btrfs_free_path(path);
1064         if (ptr == name + PATH_MAX - 1) {
1065                 name[0] = '/';
1066                 name[1] = '\0';
1067         } else {
1068                 memmove(name, ptr, name + PATH_MAX - ptr);
1069         }
1070         return name;
1071
1072 err:
1073         btrfs_free_path(path);
1074         kfree(name);
1075         return ERR_PTR(ret);
1076 }
1077
1078 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1079 {
1080         struct btrfs_root *root = fs_info->tree_root;
1081         struct btrfs_dir_item *di;
1082         struct btrfs_path *path;
1083         struct btrfs_key location;
1084         u64 dir_id;
1085
1086         path = btrfs_alloc_path();
1087         if (!path)
1088                 return -ENOMEM;
1089         path->leave_spinning = 1;
1090
1091         /*
1092          * Find the "default" dir item which points to the root item that we
1093          * will mount by default if we haven't been given a specific subvolume
1094          * to mount.
1095          */
1096         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1097         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1098         if (IS_ERR(di)) {
1099                 btrfs_free_path(path);
1100                 return PTR_ERR(di);
1101         }
1102         if (!di) {
1103                 /*
1104                  * Ok the default dir item isn't there.  This is weird since
1105                  * it's always been there, but don't freak out, just try and
1106                  * mount the top-level subvolume.
1107                  */
1108                 btrfs_free_path(path);
1109                 *objectid = BTRFS_FS_TREE_OBJECTID;
1110                 return 0;
1111         }
1112
1113         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1114         btrfs_free_path(path);
1115         *objectid = location.objectid;
1116         return 0;
1117 }
1118
1119 static int btrfs_fill_super(struct super_block *sb,
1120                             struct btrfs_fs_devices *fs_devices,
1121                             void *data)
1122 {
1123         struct inode *inode;
1124         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1125         struct btrfs_key key;
1126         int err;
1127
1128         sb->s_maxbytes = MAX_LFS_FILESIZE;
1129         sb->s_magic = BTRFS_SUPER_MAGIC;
1130         sb->s_op = &btrfs_super_ops;
1131         sb->s_d_op = &btrfs_dentry_operations;
1132         sb->s_export_op = &btrfs_export_ops;
1133         sb->s_xattr = btrfs_xattr_handlers;
1134         sb->s_time_gran = 1;
1135 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1136         sb->s_flags |= MS_POSIXACL;
1137 #endif
1138         sb->s_flags |= SB_I_VERSION;
1139         sb->s_iflags |= SB_I_CGROUPWB;
1140
1141         err = super_setup_bdi(sb);
1142         if (err) {
1143                 btrfs_err(fs_info, "super_setup_bdi failed");
1144                 return err;
1145         }
1146
1147         err = open_ctree(sb, fs_devices, (char *)data);
1148         if (err) {
1149                 btrfs_err(fs_info, "open_ctree failed");
1150                 return err;
1151         }
1152
1153         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1154         key.type = BTRFS_INODE_ITEM_KEY;
1155         key.offset = 0;
1156         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1157         if (IS_ERR(inode)) {
1158                 err = PTR_ERR(inode);
1159                 goto fail_close;
1160         }
1161
1162         sb->s_root = d_make_root(inode);
1163         if (!sb->s_root) {
1164                 err = -ENOMEM;
1165                 goto fail_close;
1166         }
1167
1168         cleancache_init_fs(sb);
1169         sb->s_flags |= MS_ACTIVE;
1170         return 0;
1171
1172 fail_close:
1173         close_ctree(fs_info);
1174         return err;
1175 }
1176
1177 int btrfs_sync_fs(struct super_block *sb, int wait)
1178 {
1179         struct btrfs_trans_handle *trans;
1180         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1181         struct btrfs_root *root = fs_info->tree_root;
1182
1183         trace_btrfs_sync_fs(fs_info, wait);
1184
1185         if (!wait) {
1186                 filemap_flush(fs_info->btree_inode->i_mapping);
1187                 return 0;
1188         }
1189
1190         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1191
1192         trans = btrfs_attach_transaction_barrier(root);
1193         if (IS_ERR(trans)) {
1194                 /* no transaction, don't bother */
1195                 if (PTR_ERR(trans) == -ENOENT) {
1196                         /*
1197                          * Exit unless we have some pending changes
1198                          * that need to go through commit
1199                          */
1200                         if (fs_info->pending_changes == 0)
1201                                 return 0;
1202                         /*
1203                          * A non-blocking test if the fs is frozen. We must not
1204                          * start a new transaction here otherwise a deadlock
1205                          * happens. The pending operations are delayed to the
1206                          * next commit after thawing.
1207                          */
1208                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1209                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1210                         else
1211                                 return 0;
1212                         trans = btrfs_start_transaction(root, 0);
1213                 }
1214                 if (IS_ERR(trans))
1215                         return PTR_ERR(trans);
1216         }
1217         return btrfs_commit_transaction(trans);
1218 }
1219
1220 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1221 {
1222         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1223         char *compress_type;
1224
1225         if (btrfs_test_opt(info, DEGRADED))
1226                 seq_puts(seq, ",degraded");
1227         if (btrfs_test_opt(info, NODATASUM))
1228                 seq_puts(seq, ",nodatasum");
1229         if (btrfs_test_opt(info, NODATACOW))
1230                 seq_puts(seq, ",nodatacow");
1231         if (btrfs_test_opt(info, NOBARRIER))
1232                 seq_puts(seq, ",nobarrier");
1233         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1234                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1235         if (info->thread_pool_size !=  min_t(unsigned long,
1236                                              num_online_cpus() + 2, 8))
1237                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1238         if (btrfs_test_opt(info, COMPRESS)) {
1239                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1240                         compress_type = "zlib";
1241                 else if (info->compress_type == BTRFS_COMPRESS_LZO)
1242                         compress_type = "lzo";
1243                 else
1244                         compress_type = "zstd";
1245                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1246                         seq_printf(seq, ",compress-force=%s", compress_type);
1247                 else
1248                         seq_printf(seq, ",compress=%s", compress_type);
1249         }
1250         if (btrfs_test_opt(info, NOSSD))
1251                 seq_puts(seq, ",nossd");
1252         if (btrfs_test_opt(info, SSD_SPREAD))
1253                 seq_puts(seq, ",ssd_spread");
1254         else if (btrfs_test_opt(info, SSD))
1255                 seq_puts(seq, ",ssd");
1256         if (btrfs_test_opt(info, NOTREELOG))
1257                 seq_puts(seq, ",notreelog");
1258         if (btrfs_test_opt(info, NOLOGREPLAY))
1259                 seq_puts(seq, ",nologreplay");
1260         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1261                 seq_puts(seq, ",flushoncommit");
1262         if (btrfs_test_opt(info, DISCARD))
1263                 seq_puts(seq, ",discard");
1264         if (!(info->sb->s_flags & MS_POSIXACL))
1265                 seq_puts(seq, ",noacl");
1266         if (btrfs_test_opt(info, SPACE_CACHE))
1267                 seq_puts(seq, ",space_cache");
1268         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1269                 seq_puts(seq, ",space_cache=v2");
1270         else
1271                 seq_puts(seq, ",nospace_cache");
1272         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1273                 seq_puts(seq, ",rescan_uuid_tree");
1274         if (btrfs_test_opt(info, CLEAR_CACHE))
1275                 seq_puts(seq, ",clear_cache");
1276         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1277                 seq_puts(seq, ",user_subvol_rm_allowed");
1278         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1279                 seq_puts(seq, ",enospc_debug");
1280         if (btrfs_test_opt(info, AUTO_DEFRAG))
1281                 seq_puts(seq, ",autodefrag");
1282         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1283                 seq_puts(seq, ",inode_cache");
1284         if (btrfs_test_opt(info, SKIP_BALANCE))
1285                 seq_puts(seq, ",skip_balance");
1286 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1287         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1288                 seq_puts(seq, ",check_int_data");
1289         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1290                 seq_puts(seq, ",check_int");
1291         if (info->check_integrity_print_mask)
1292                 seq_printf(seq, ",check_int_print_mask=%d",
1293                                 info->check_integrity_print_mask);
1294 #endif
1295         if (info->metadata_ratio)
1296                 seq_printf(seq, ",metadata_ratio=%d",
1297                                 info->metadata_ratio);
1298         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1299                 seq_puts(seq, ",fatal_errors=panic");
1300         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1301                 seq_printf(seq, ",commit=%d", info->commit_interval);
1302 #ifdef CONFIG_BTRFS_DEBUG
1303         if (btrfs_test_opt(info, FRAGMENT_DATA))
1304                 seq_puts(seq, ",fragment=data");
1305         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1306                 seq_puts(seq, ",fragment=metadata");
1307 #endif
1308         seq_printf(seq, ",subvolid=%llu",
1309                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1310         seq_puts(seq, ",subvol=");
1311         seq_dentry(seq, dentry, " \t\n\\");
1312         return 0;
1313 }
1314
1315 static int btrfs_test_super(struct super_block *s, void *data)
1316 {
1317         struct btrfs_fs_info *p = data;
1318         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1319
1320         return fs_info->fs_devices == p->fs_devices;
1321 }
1322
1323 static int btrfs_set_super(struct super_block *s, void *data)
1324 {
1325         int err = set_anon_super(s, data);
1326         if (!err)
1327                 s->s_fs_info = data;
1328         return err;
1329 }
1330
1331 /*
1332  * subvolumes are identified by ino 256
1333  */
1334 static inline int is_subvolume_inode(struct inode *inode)
1335 {
1336         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1337                 return 1;
1338         return 0;
1339 }
1340
1341 /*
1342  * This will add subvolid=0 to the argument string while removing any subvol=
1343  * and subvolid= arguments to make sure we get the top-level root for path
1344  * walking to the subvol we want.
1345  */
1346 static char *setup_root_args(char *args)
1347 {
1348         char *buf, *dst, *sep;
1349
1350         if (!args)
1351                 return kstrdup("subvolid=0", GFP_KERNEL);
1352
1353         /* The worst case is that we add ",subvolid=0" to the end. */
1354         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1,
1355                         GFP_KERNEL);
1356         if (!buf)
1357                 return NULL;
1358
1359         while (1) {
1360                 sep = strchrnul(args, ',');
1361                 if (!strstarts(args, "subvol=") &&
1362                     !strstarts(args, "subvolid=")) {
1363                         memcpy(dst, args, sep - args);
1364                         dst += sep - args;
1365                         *dst++ = ',';
1366                 }
1367                 if (*sep)
1368                         args = sep + 1;
1369                 else
1370                         break;
1371         }
1372         strcpy(dst, "subvolid=0");
1373
1374         return buf;
1375 }
1376
1377 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1378                                    int flags, const char *device_name,
1379                                    char *data)
1380 {
1381         struct dentry *root;
1382         struct vfsmount *mnt = NULL;
1383         char *newargs;
1384         int ret;
1385
1386         newargs = setup_root_args(data);
1387         if (!newargs) {
1388                 root = ERR_PTR(-ENOMEM);
1389                 goto out;
1390         }
1391
1392         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1393         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1394                 if (flags & MS_RDONLY) {
1395                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1396                                              device_name, newargs);
1397                 } else {
1398                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1399                                              device_name, newargs);
1400                         if (IS_ERR(mnt)) {
1401                                 root = ERR_CAST(mnt);
1402                                 mnt = NULL;
1403                                 goto out;
1404                         }
1405
1406                         down_write(&mnt->mnt_sb->s_umount);
1407                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1408                         up_write(&mnt->mnt_sb->s_umount);
1409                         if (ret < 0) {
1410                                 root = ERR_PTR(ret);
1411                                 goto out;
1412                         }
1413                 }
1414         }
1415         if (IS_ERR(mnt)) {
1416                 root = ERR_CAST(mnt);
1417                 mnt = NULL;
1418                 goto out;
1419         }
1420
1421         if (!subvol_name) {
1422                 if (!subvol_objectid) {
1423                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1424                                                           &subvol_objectid);
1425                         if (ret) {
1426                                 root = ERR_PTR(ret);
1427                                 goto out;
1428                         }
1429                 }
1430                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1431                                                             subvol_objectid);
1432                 if (IS_ERR(subvol_name)) {
1433                         root = ERR_CAST(subvol_name);
1434                         subvol_name = NULL;
1435                         goto out;
1436                 }
1437
1438         }
1439
1440         root = mount_subtree(mnt, subvol_name);
1441         /* mount_subtree() drops our reference on the vfsmount. */
1442         mnt = NULL;
1443
1444         if (!IS_ERR(root)) {
1445                 struct super_block *s = root->d_sb;
1446                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1447                 struct inode *root_inode = d_inode(root);
1448                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1449
1450                 ret = 0;
1451                 if (!is_subvolume_inode(root_inode)) {
1452                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1453                                subvol_name);
1454                         ret = -EINVAL;
1455                 }
1456                 if (subvol_objectid && root_objectid != subvol_objectid) {
1457                         /*
1458                          * This will also catch a race condition where a
1459                          * subvolume which was passed by ID is renamed and
1460                          * another subvolume is renamed over the old location.
1461                          */
1462                         btrfs_err(fs_info,
1463                                   "subvol '%s' does not match subvolid %llu",
1464                                   subvol_name, subvol_objectid);
1465                         ret = -EINVAL;
1466                 }
1467                 if (ret) {
1468                         dput(root);
1469                         root = ERR_PTR(ret);
1470                         deactivate_locked_super(s);
1471                 }
1472         }
1473
1474 out:
1475         mntput(mnt);
1476         kfree(newargs);
1477         kfree(subvol_name);
1478         return root;
1479 }
1480
1481 static int parse_security_options(char *orig_opts,
1482                                   struct security_mnt_opts *sec_opts)
1483 {
1484         char *secdata = NULL;
1485         int ret = 0;
1486
1487         secdata = alloc_secdata();
1488         if (!secdata)
1489                 return -ENOMEM;
1490         ret = security_sb_copy_data(orig_opts, secdata);
1491         if (ret) {
1492                 free_secdata(secdata);
1493                 return ret;
1494         }
1495         ret = security_sb_parse_opts_str(secdata, sec_opts);
1496         free_secdata(secdata);
1497         return ret;
1498 }
1499
1500 static int setup_security_options(struct btrfs_fs_info *fs_info,
1501                                   struct super_block *sb,
1502                                   struct security_mnt_opts *sec_opts)
1503 {
1504         int ret = 0;
1505
1506         /*
1507          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1508          * is valid.
1509          */
1510         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1511         if (ret)
1512                 return ret;
1513
1514 #ifdef CONFIG_SECURITY
1515         if (!fs_info->security_opts.num_mnt_opts) {
1516                 /* first time security setup, copy sec_opts to fs_info */
1517                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1518         } else {
1519                 /*
1520                  * Since SELinux (the only one supporting security_mnt_opts)
1521                  * does NOT support changing context during remount/mount of
1522                  * the same sb, this must be the same or part of the same
1523                  * security options, just free it.
1524                  */
1525                 security_free_mnt_opts(sec_opts);
1526         }
1527 #endif
1528         return ret;
1529 }
1530
1531 /*
1532  * Find a superblock for the given device / mount point.
1533  *
1534  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1535  *        for multiple device setup.  Make sure to keep it in sync.
1536  */
1537 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1538                 const char *device_name, void *data)
1539 {
1540         struct block_device *bdev = NULL;
1541         struct super_block *s;
1542         struct btrfs_fs_devices *fs_devices = NULL;
1543         struct btrfs_fs_info *fs_info = NULL;
1544         struct security_mnt_opts new_sec_opts;
1545         fmode_t mode = FMODE_READ;
1546         char *subvol_name = NULL;
1547         u64 subvol_objectid = 0;
1548         int error = 0;
1549
1550         if (!(flags & MS_RDONLY))
1551                 mode |= FMODE_WRITE;
1552
1553         error = btrfs_parse_early_options(data, mode, fs_type,
1554                                           &subvol_name, &subvol_objectid,
1555                                           &fs_devices);
1556         if (error) {
1557                 kfree(subvol_name);
1558                 return ERR_PTR(error);
1559         }
1560
1561         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1562                 /* mount_subvol() will free subvol_name. */
1563                 return mount_subvol(subvol_name, subvol_objectid, flags,
1564                                     device_name, data);
1565         }
1566
1567         security_init_mnt_opts(&new_sec_opts);
1568         if (data) {
1569                 error = parse_security_options(data, &new_sec_opts);
1570                 if (error)
1571                         return ERR_PTR(error);
1572         }
1573
1574         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1575         if (error)
1576                 goto error_sec_opts;
1577
1578         /*
1579          * Setup a dummy root and fs_info for test/set super.  This is because
1580          * we don't actually fill this stuff out until open_ctree, but we need
1581          * it for searching for existing supers, so this lets us do that and
1582          * then open_ctree will properly initialize everything later.
1583          */
1584         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1585         if (!fs_info) {
1586                 error = -ENOMEM;
1587                 goto error_sec_opts;
1588         }
1589
1590         fs_info->fs_devices = fs_devices;
1591
1592         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1593         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1594         security_init_mnt_opts(&fs_info->security_opts);
1595         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1596                 error = -ENOMEM;
1597                 goto error_fs_info;
1598         }
1599
1600         error = btrfs_open_devices(fs_devices, mode, fs_type);
1601         if (error)
1602                 goto error_fs_info;
1603
1604         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1605                 error = -EACCES;
1606                 goto error_close_devices;
1607         }
1608
1609         bdev = fs_devices->latest_bdev;
1610         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1611                  fs_info);
1612         if (IS_ERR(s)) {
1613                 error = PTR_ERR(s);
1614                 goto error_close_devices;
1615         }
1616
1617         if (s->s_root) {
1618                 btrfs_close_devices(fs_devices);
1619                 free_fs_info(fs_info);
1620                 if ((flags ^ s->s_flags) & MS_RDONLY)
1621                         error = -EBUSY;
1622         } else {
1623                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1624                 btrfs_sb(s)->bdev_holder = fs_type;
1625                 error = btrfs_fill_super(s, fs_devices, data);
1626         }
1627         if (error) {
1628                 deactivate_locked_super(s);
1629                 goto error_sec_opts;
1630         }
1631
1632         fs_info = btrfs_sb(s);
1633         error = setup_security_options(fs_info, s, &new_sec_opts);
1634         if (error) {
1635                 deactivate_locked_super(s);
1636                 goto error_sec_opts;
1637         }
1638
1639         return dget(s->s_root);
1640
1641 error_close_devices:
1642         btrfs_close_devices(fs_devices);
1643 error_fs_info:
1644         free_fs_info(fs_info);
1645 error_sec_opts:
1646         security_free_mnt_opts(&new_sec_opts);
1647         return ERR_PTR(error);
1648 }
1649
1650 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1651                                      int new_pool_size, int old_pool_size)
1652 {
1653         if (new_pool_size == old_pool_size)
1654                 return;
1655
1656         fs_info->thread_pool_size = new_pool_size;
1657
1658         btrfs_info(fs_info, "resize thread pool %d -> %d",
1659                old_pool_size, new_pool_size);
1660
1661         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1662         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1663         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1664         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1665         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1666         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1667         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1668                                 new_pool_size);
1669         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1670         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1671         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1672         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1673         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1674                                 new_pool_size);
1675 }
1676
1677 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1678 {
1679         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1680 }
1681
1682 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1683                                        unsigned long old_opts, int flags)
1684 {
1685         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1686             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1687              (flags & MS_RDONLY))) {
1688                 /* wait for any defraggers to finish */
1689                 wait_event(fs_info->transaction_wait,
1690                            (atomic_read(&fs_info->defrag_running) == 0));
1691                 if (flags & MS_RDONLY)
1692                         sync_filesystem(fs_info->sb);
1693         }
1694 }
1695
1696 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1697                                          unsigned long old_opts)
1698 {
1699         /*
1700          * We need to cleanup all defragable inodes if the autodefragment is
1701          * close or the filesystem is read only.
1702          */
1703         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1704             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1705                 btrfs_cleanup_defrag_inodes(fs_info);
1706         }
1707
1708         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1709 }
1710
1711 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1712 {
1713         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1714         struct btrfs_root *root = fs_info->tree_root;
1715         unsigned old_flags = sb->s_flags;
1716         unsigned long old_opts = fs_info->mount_opt;
1717         unsigned long old_compress_type = fs_info->compress_type;
1718         u64 old_max_inline = fs_info->max_inline;
1719         int old_thread_pool_size = fs_info->thread_pool_size;
1720         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1721         int ret;
1722
1723         sync_filesystem(sb);
1724         btrfs_remount_prepare(fs_info);
1725
1726         if (data) {
1727                 struct security_mnt_opts new_sec_opts;
1728
1729                 security_init_mnt_opts(&new_sec_opts);
1730                 ret = parse_security_options(data, &new_sec_opts);
1731                 if (ret)
1732                         goto restore;
1733                 ret = setup_security_options(fs_info, sb,
1734                                              &new_sec_opts);
1735                 if (ret) {
1736                         security_free_mnt_opts(&new_sec_opts);
1737                         goto restore;
1738                 }
1739         }
1740
1741         ret = btrfs_parse_options(fs_info, data, *flags);
1742         if (ret) {
1743                 ret = -EINVAL;
1744                 goto restore;
1745         }
1746
1747         btrfs_remount_begin(fs_info, old_opts, *flags);
1748         btrfs_resize_thread_pool(fs_info,
1749                 fs_info->thread_pool_size, old_thread_pool_size);
1750
1751         if ((bool)(*flags & MS_RDONLY) == sb_rdonly(sb))
1752                 goto out;
1753
1754         if (*flags & MS_RDONLY) {
1755                 /*
1756                  * this also happens on 'umount -rf' or on shutdown, when
1757                  * the filesystem is busy.
1758                  */
1759                 cancel_work_sync(&fs_info->async_reclaim_work);
1760
1761                 /* wait for the uuid_scan task to finish */
1762                 down(&fs_info->uuid_tree_rescan_sem);
1763                 /* avoid complains from lockdep et al. */
1764                 up(&fs_info->uuid_tree_rescan_sem);
1765
1766                 sb->s_flags |= MS_RDONLY;
1767
1768                 /*
1769                  * Setting MS_RDONLY will put the cleaner thread to
1770                  * sleep at the next loop if it's already active.
1771                  * If it's already asleep, we'll leave unused block
1772                  * groups on disk until we're mounted read-write again
1773                  * unless we clean them up here.
1774                  */
1775                 btrfs_delete_unused_bgs(fs_info);
1776
1777                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1778                 btrfs_scrub_cancel(fs_info);
1779                 btrfs_pause_balance(fs_info);
1780
1781                 ret = btrfs_commit_super(fs_info);
1782                 if (ret)
1783                         goto restore;
1784         } else {
1785                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1786                         btrfs_err(fs_info,
1787                                 "Remounting read-write after error is not allowed");
1788                         ret = -EINVAL;
1789                         goto restore;
1790                 }
1791                 if (fs_info->fs_devices->rw_devices == 0) {
1792                         ret = -EACCES;
1793                         goto restore;
1794                 }
1795
1796                 if (!btrfs_check_rw_degradable(fs_info)) {
1797                         btrfs_warn(fs_info,
1798                                 "too many missing devices, writeable remount is not allowed");
1799                         ret = -EACCES;
1800                         goto restore;
1801                 }
1802
1803                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1804                         ret = -EINVAL;
1805                         goto restore;
1806                 }
1807
1808                 ret = btrfs_cleanup_fs_roots(fs_info);
1809                 if (ret)
1810                         goto restore;
1811
1812                 /* recover relocation */
1813                 mutex_lock(&fs_info->cleaner_mutex);
1814                 ret = btrfs_recover_relocation(root);
1815                 mutex_unlock(&fs_info->cleaner_mutex);
1816                 if (ret)
1817                         goto restore;
1818
1819                 ret = btrfs_resume_balance_async(fs_info);
1820                 if (ret)
1821                         goto restore;
1822
1823                 ret = btrfs_resume_dev_replace_async(fs_info);
1824                 if (ret) {
1825                         btrfs_warn(fs_info, "failed to resume dev_replace");
1826                         goto restore;
1827                 }
1828
1829                 btrfs_qgroup_rescan_resume(fs_info);
1830
1831                 if (!fs_info->uuid_root) {
1832                         btrfs_info(fs_info, "creating UUID tree");
1833                         ret = btrfs_create_uuid_tree(fs_info);
1834                         if (ret) {
1835                                 btrfs_warn(fs_info,
1836                                            "failed to create the UUID tree %d",
1837                                            ret);
1838                                 goto restore;
1839                         }
1840                 }
1841                 sb->s_flags &= ~MS_RDONLY;
1842
1843                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1844         }
1845 out:
1846         wake_up_process(fs_info->transaction_kthread);
1847         btrfs_remount_cleanup(fs_info, old_opts);
1848         return 0;
1849
1850 restore:
1851         /* We've hit an error - don't reset MS_RDONLY */
1852         if (sb_rdonly(sb))
1853                 old_flags |= MS_RDONLY;
1854         sb->s_flags = old_flags;
1855         fs_info->mount_opt = old_opts;
1856         fs_info->compress_type = old_compress_type;
1857         fs_info->max_inline = old_max_inline;
1858         btrfs_resize_thread_pool(fs_info,
1859                 old_thread_pool_size, fs_info->thread_pool_size);
1860         fs_info->metadata_ratio = old_metadata_ratio;
1861         btrfs_remount_cleanup(fs_info, old_opts);
1862         return ret;
1863 }
1864
1865 /* Used to sort the devices by max_avail(descending sort) */
1866 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1867                                        const void *dev_info2)
1868 {
1869         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1870             ((struct btrfs_device_info *)dev_info2)->max_avail)
1871                 return -1;
1872         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1873                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1874                 return 1;
1875         else
1876         return 0;
1877 }
1878
1879 /*
1880  * sort the devices by max_avail, in which max free extent size of each device
1881  * is stored.(Descending Sort)
1882  */
1883 static inline void btrfs_descending_sort_devices(
1884                                         struct btrfs_device_info *devices,
1885                                         size_t nr_devices)
1886 {
1887         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1888              btrfs_cmp_device_free_bytes, NULL);
1889 }
1890
1891 /*
1892  * The helper to calc the free space on the devices that can be used to store
1893  * file data.
1894  */
1895 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1896                                        u64 *free_bytes)
1897 {
1898         struct btrfs_device_info *devices_info;
1899         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1900         struct btrfs_device *device;
1901         u64 skip_space;
1902         u64 type;
1903         u64 avail_space;
1904         u64 min_stripe_size;
1905         int min_stripes = 1, num_stripes = 1;
1906         int i = 0, nr_devices;
1907
1908         /*
1909          * We aren't under the device list lock, so this is racy-ish, but good
1910          * enough for our purposes.
1911          */
1912         nr_devices = fs_info->fs_devices->open_devices;
1913         if (!nr_devices) {
1914                 smp_mb();
1915                 nr_devices = fs_info->fs_devices->open_devices;
1916                 ASSERT(nr_devices);
1917                 if (!nr_devices) {
1918                         *free_bytes = 0;
1919                         return 0;
1920                 }
1921         }
1922
1923         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1924                                GFP_KERNEL);
1925         if (!devices_info)
1926                 return -ENOMEM;
1927
1928         /* calc min stripe number for data space allocation */
1929         type = btrfs_data_alloc_profile(fs_info);
1930         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1931                 min_stripes = 2;
1932                 num_stripes = nr_devices;
1933         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1934                 min_stripes = 2;
1935                 num_stripes = 2;
1936         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1937                 min_stripes = 4;
1938                 num_stripes = 4;
1939         }
1940
1941         if (type & BTRFS_BLOCK_GROUP_DUP)
1942                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1943         else
1944                 min_stripe_size = BTRFS_STRIPE_LEN;
1945
1946         rcu_read_lock();
1947         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1948                 if (!device->in_fs_metadata || !device->bdev ||
1949                     device->is_tgtdev_for_dev_replace)
1950                         continue;
1951
1952                 if (i >= nr_devices)
1953                         break;
1954
1955                 avail_space = device->total_bytes - device->bytes_used;
1956
1957                 /* align with stripe_len */
1958                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1959                 avail_space *= BTRFS_STRIPE_LEN;
1960
1961                 /*
1962                  * In order to avoid overwriting the superblock on the drive,
1963                  * btrfs starts at an offset of at least 1MB when doing chunk
1964                  * allocation.
1965                  */
1966                 skip_space = SZ_1M;
1967
1968                 /*
1969                  * we can use the free space in [0, skip_space - 1], subtract
1970                  * it from the total.
1971                  */
1972                 if (avail_space && avail_space >= skip_space)
1973                         avail_space -= skip_space;
1974                 else
1975                         avail_space = 0;
1976
1977                 if (avail_space < min_stripe_size)
1978                         continue;
1979
1980                 devices_info[i].dev = device;
1981                 devices_info[i].max_avail = avail_space;
1982
1983                 i++;
1984         }
1985         rcu_read_unlock();
1986
1987         nr_devices = i;
1988
1989         btrfs_descending_sort_devices(devices_info, nr_devices);
1990
1991         i = nr_devices - 1;
1992         avail_space = 0;
1993         while (nr_devices >= min_stripes) {
1994                 if (num_stripes > nr_devices)
1995                         num_stripes = nr_devices;
1996
1997                 if (devices_info[i].max_avail >= min_stripe_size) {
1998                         int j;
1999                         u64 alloc_size;
2000
2001                         avail_space += devices_info[i].max_avail * num_stripes;
2002                         alloc_size = devices_info[i].max_avail;
2003                         for (j = i + 1 - num_stripes; j <= i; j++)
2004                                 devices_info[j].max_avail -= alloc_size;
2005                 }
2006                 i--;
2007                 nr_devices--;
2008         }
2009
2010         kfree(devices_info);
2011         *free_bytes = avail_space;
2012         return 0;
2013 }
2014
2015 /*
2016  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2017  *
2018  * If there's a redundant raid level at DATA block groups, use the respective
2019  * multiplier to scale the sizes.
2020  *
2021  * Unused device space usage is based on simulating the chunk allocator
2022  * algorithm that respects the device sizes and order of allocations.  This is
2023  * a close approximation of the actual use but there are other factors that may
2024  * change the result (like a new metadata chunk).
2025  *
2026  * If metadata is exhausted, f_bavail will be 0.
2027  */
2028 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2029 {
2030         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2031         struct btrfs_super_block *disk_super = fs_info->super_copy;
2032         struct list_head *head = &fs_info->space_info;
2033         struct btrfs_space_info *found;
2034         u64 total_used = 0;
2035         u64 total_free_data = 0;
2036         u64 total_free_meta = 0;
2037         int bits = dentry->d_sb->s_blocksize_bits;
2038         __be32 *fsid = (__be32 *)fs_info->fsid;
2039         unsigned factor = 1;
2040         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2041         int ret;
2042         u64 thresh = 0;
2043         int mixed = 0;
2044
2045         rcu_read_lock();
2046         list_for_each_entry_rcu(found, head, list) {
2047                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2048                         int i;
2049
2050                         total_free_data += found->disk_total - found->disk_used;
2051                         total_free_data -=
2052                                 btrfs_account_ro_block_groups_free_space(found);
2053
2054                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2055                                 if (!list_empty(&found->block_groups[i])) {
2056                                         switch (i) {
2057                                         case BTRFS_RAID_DUP:
2058                                         case BTRFS_RAID_RAID1:
2059                                         case BTRFS_RAID_RAID10:
2060                                                 factor = 2;
2061                                         }
2062                                 }
2063                         }
2064                 }
2065
2066                 /*
2067                  * Metadata in mixed block goup profiles are accounted in data
2068                  */
2069                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2070                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2071                                 mixed = 1;
2072                         else
2073                                 total_free_meta += found->disk_total -
2074                                         found->disk_used;
2075                 }
2076
2077                 total_used += found->disk_used;
2078         }
2079
2080         rcu_read_unlock();
2081
2082         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2083         buf->f_blocks >>= bits;
2084         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2085
2086         /* Account global block reserve as used, it's in logical size already */
2087         spin_lock(&block_rsv->lock);
2088         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2089         if (buf->f_bfree >= block_rsv->size >> bits)
2090                 buf->f_bfree -= block_rsv->size >> bits;
2091         else
2092                 buf->f_bfree = 0;
2093         spin_unlock(&block_rsv->lock);
2094
2095         buf->f_bavail = div_u64(total_free_data, factor);
2096         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2097         if (ret)
2098                 return ret;
2099         buf->f_bavail += div_u64(total_free_data, factor);
2100         buf->f_bavail = buf->f_bavail >> bits;
2101
2102         /*
2103          * We calculate the remaining metadata space minus global reserve. If
2104          * this is (supposedly) smaller than zero, there's no space. But this
2105          * does not hold in practice, the exhausted state happens where's still
2106          * some positive delta. So we apply some guesswork and compare the
2107          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2108          *
2109          * We probably cannot calculate the exact threshold value because this
2110          * depends on the internal reservations requested by various
2111          * operations, so some operations that consume a few metadata will
2112          * succeed even if the Avail is zero. But this is better than the other
2113          * way around.
2114          */
2115         thresh = 4 * 1024 * 1024;
2116
2117         if (!mixed && total_free_meta - thresh < block_rsv->size)
2118                 buf->f_bavail = 0;
2119
2120         buf->f_type = BTRFS_SUPER_MAGIC;
2121         buf->f_bsize = dentry->d_sb->s_blocksize;
2122         buf->f_namelen = BTRFS_NAME_LEN;
2123
2124         /* We treat it as constant endianness (it doesn't matter _which_)
2125            because we want the fsid to come out the same whether mounted
2126            on a big-endian or little-endian host */
2127         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2128         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2129         /* Mask in the root object ID too, to disambiguate subvols */
2130         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2131         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2132
2133         return 0;
2134 }
2135
2136 static void btrfs_kill_super(struct super_block *sb)
2137 {
2138         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2139         kill_anon_super(sb);
2140         free_fs_info(fs_info);
2141 }
2142
2143 static struct file_system_type btrfs_fs_type = {
2144         .owner          = THIS_MODULE,
2145         .name           = "btrfs",
2146         .mount          = btrfs_mount,
2147         .kill_sb        = btrfs_kill_super,
2148         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2149 };
2150 MODULE_ALIAS_FS("btrfs");
2151
2152 static int btrfs_control_open(struct inode *inode, struct file *file)
2153 {
2154         /*
2155          * The control file's private_data is used to hold the
2156          * transaction when it is started and is used to keep
2157          * track of whether a transaction is already in progress.
2158          */
2159         file->private_data = NULL;
2160         return 0;
2161 }
2162
2163 /*
2164  * used by btrfsctl to scan devices when no FS is mounted
2165  */
2166 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2167                                 unsigned long arg)
2168 {
2169         struct btrfs_ioctl_vol_args *vol;
2170         struct btrfs_fs_devices *fs_devices;
2171         int ret = -ENOTTY;
2172
2173         if (!capable(CAP_SYS_ADMIN))
2174                 return -EPERM;
2175
2176         vol = memdup_user((void __user *)arg, sizeof(*vol));
2177         if (IS_ERR(vol))
2178                 return PTR_ERR(vol);
2179
2180         switch (cmd) {
2181         case BTRFS_IOC_SCAN_DEV:
2182                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2183                                             &btrfs_fs_type, &fs_devices);
2184                 break;
2185         case BTRFS_IOC_DEVICES_READY:
2186                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2187                                             &btrfs_fs_type, &fs_devices);
2188                 if (ret)
2189                         break;
2190                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2191                 break;
2192         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2193                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2194                 break;
2195         }
2196
2197         kfree(vol);
2198         return ret;
2199 }
2200
2201 static int btrfs_freeze(struct super_block *sb)
2202 {
2203         struct btrfs_trans_handle *trans;
2204         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2205         struct btrfs_root *root = fs_info->tree_root;
2206
2207         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2208         /*
2209          * We don't need a barrier here, we'll wait for any transaction that
2210          * could be in progress on other threads (and do delayed iputs that
2211          * we want to avoid on a frozen filesystem), or do the commit
2212          * ourselves.
2213          */
2214         trans = btrfs_attach_transaction_barrier(root);
2215         if (IS_ERR(trans)) {
2216                 /* no transaction, don't bother */
2217                 if (PTR_ERR(trans) == -ENOENT)
2218                         return 0;
2219                 return PTR_ERR(trans);
2220         }
2221         return btrfs_commit_transaction(trans);
2222 }
2223
2224 static int btrfs_unfreeze(struct super_block *sb)
2225 {
2226         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2227
2228         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2229         return 0;
2230 }
2231
2232 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2233 {
2234         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2235         struct btrfs_fs_devices *cur_devices;
2236         struct btrfs_device *dev, *first_dev = NULL;
2237         struct list_head *head;
2238         struct rcu_string *name;
2239
2240         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2241         cur_devices = fs_info->fs_devices;
2242         while (cur_devices) {
2243                 head = &cur_devices->devices;
2244                 list_for_each_entry(dev, head, dev_list) {
2245                         if (dev->missing)
2246                                 continue;
2247                         if (!dev->name)
2248                                 continue;
2249                         if (!first_dev || dev->devid < first_dev->devid)
2250                                 first_dev = dev;
2251                 }
2252                 cur_devices = cur_devices->seed;
2253         }
2254
2255         if (first_dev) {
2256                 rcu_read_lock();
2257                 name = rcu_dereference(first_dev->name);
2258                 seq_escape(m, name->str, " \t\n\\");
2259                 rcu_read_unlock();
2260         } else {
2261                 WARN_ON(1);
2262         }
2263         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2264         return 0;
2265 }
2266
2267 static const struct super_operations btrfs_super_ops = {
2268         .drop_inode     = btrfs_drop_inode,
2269         .evict_inode    = btrfs_evict_inode,
2270         .put_super      = btrfs_put_super,
2271         .sync_fs        = btrfs_sync_fs,
2272         .show_options   = btrfs_show_options,
2273         .show_devname   = btrfs_show_devname,
2274         .write_inode    = btrfs_write_inode,
2275         .alloc_inode    = btrfs_alloc_inode,
2276         .destroy_inode  = btrfs_destroy_inode,
2277         .statfs         = btrfs_statfs,
2278         .remount_fs     = btrfs_remount,
2279         .freeze_fs      = btrfs_freeze,
2280         .unfreeze_fs    = btrfs_unfreeze,
2281 };
2282
2283 static const struct file_operations btrfs_ctl_fops = {
2284         .open = btrfs_control_open,
2285         .unlocked_ioctl  = btrfs_control_ioctl,
2286         .compat_ioctl = btrfs_control_ioctl,
2287         .owner   = THIS_MODULE,
2288         .llseek = noop_llseek,
2289 };
2290
2291 static struct miscdevice btrfs_misc = {
2292         .minor          = BTRFS_MINOR,
2293         .name           = "btrfs-control",
2294         .fops           = &btrfs_ctl_fops
2295 };
2296
2297 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2298 MODULE_ALIAS("devname:btrfs-control");
2299
2300 static int btrfs_interface_init(void)
2301 {
2302         return misc_register(&btrfs_misc);
2303 }
2304
2305 static void btrfs_interface_exit(void)
2306 {
2307         misc_deregister(&btrfs_misc);
2308 }
2309
2310 static void btrfs_print_mod_info(void)
2311 {
2312         pr_info("Btrfs loaded, crc32c=%s"
2313 #ifdef CONFIG_BTRFS_DEBUG
2314                         ", debug=on"
2315 #endif
2316 #ifdef CONFIG_BTRFS_ASSERT
2317                         ", assert=on"
2318 #endif
2319 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2320                         ", integrity-checker=on"
2321 #endif
2322                         "\n",
2323                         btrfs_crc32c_impl());
2324 }
2325
2326 static int __init init_btrfs_fs(void)
2327 {
2328         int err;
2329
2330         err = btrfs_hash_init();
2331         if (err)
2332                 return err;
2333
2334         btrfs_props_init();
2335
2336         err = btrfs_init_sysfs();
2337         if (err)
2338                 goto free_hash;
2339
2340         btrfs_init_compress();
2341
2342         err = btrfs_init_cachep();
2343         if (err)
2344                 goto free_compress;
2345
2346         err = extent_io_init();
2347         if (err)
2348                 goto free_cachep;
2349
2350         err = extent_map_init();
2351         if (err)
2352                 goto free_extent_io;
2353
2354         err = ordered_data_init();
2355         if (err)
2356                 goto free_extent_map;
2357
2358         err = btrfs_delayed_inode_init();
2359         if (err)
2360                 goto free_ordered_data;
2361
2362         err = btrfs_auto_defrag_init();
2363         if (err)
2364                 goto free_delayed_inode;
2365
2366         err = btrfs_delayed_ref_init();
2367         if (err)
2368                 goto free_auto_defrag;
2369
2370         err = btrfs_prelim_ref_init();
2371         if (err)
2372                 goto free_delayed_ref;
2373
2374         err = btrfs_end_io_wq_init();
2375         if (err)
2376                 goto free_prelim_ref;
2377
2378         err = btrfs_interface_init();
2379         if (err)
2380                 goto free_end_io_wq;
2381
2382         btrfs_init_lockdep();
2383
2384         btrfs_print_mod_info();
2385
2386         err = btrfs_run_sanity_tests();
2387         if (err)
2388                 goto unregister_ioctl;
2389
2390         err = register_filesystem(&btrfs_fs_type);
2391         if (err)
2392                 goto unregister_ioctl;
2393
2394         return 0;
2395
2396 unregister_ioctl:
2397         btrfs_interface_exit();
2398 free_end_io_wq:
2399         btrfs_end_io_wq_exit();
2400 free_prelim_ref:
2401         btrfs_prelim_ref_exit();
2402 free_delayed_ref:
2403         btrfs_delayed_ref_exit();
2404 free_auto_defrag:
2405         btrfs_auto_defrag_exit();
2406 free_delayed_inode:
2407         btrfs_delayed_inode_exit();
2408 free_ordered_data:
2409         ordered_data_exit();
2410 free_extent_map:
2411         extent_map_exit();
2412 free_extent_io:
2413         extent_io_exit();
2414 free_cachep:
2415         btrfs_destroy_cachep();
2416 free_compress:
2417         btrfs_exit_compress();
2418         btrfs_exit_sysfs();
2419 free_hash:
2420         btrfs_hash_exit();
2421         return err;
2422 }
2423
2424 static void __exit exit_btrfs_fs(void)
2425 {
2426         btrfs_destroy_cachep();
2427         btrfs_delayed_ref_exit();
2428         btrfs_auto_defrag_exit();
2429         btrfs_delayed_inode_exit();
2430         btrfs_prelim_ref_exit();
2431         ordered_data_exit();
2432         extent_map_exit();
2433         extent_io_exit();
2434         btrfs_interface_exit();
2435         btrfs_end_io_wq_exit();
2436         unregister_filesystem(&btrfs_fs_type);
2437         btrfs_exit_sysfs();
2438         btrfs_cleanup_fs_uuids();
2439         btrfs_exit_compress();
2440         btrfs_hash_exit();
2441 }
2442
2443 late_initcall(init_btrfs_fs);
2444 module_exit(exit_btrfs_fs)
2445
2446 MODULE_LICENSE("GPL");