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         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
 206         struct va_format vaf;
 207         va_list args;
 208         int kern_level;
 209         const char *type = logtypes[4];
 210         struct ratelimit_state *ratelimit = &printk_limits[4];
 211
 212         va_start(args, fmt);
 213
 214         while ((kern_level = printk_get_level(fmt)) != 0) {
 215                 size_t size = printk_skip_level(fmt) - fmt;
 216
 217                 if (kern_level >= '0' && kern_level <= '7') {
 218                         memcpy(lvl, fmt,  size);
 219                         lvl[size] = '\0';
 220                         type = logtypes[kern_level - '0'];
 221                         ratelimit = &printk_limits[kern_level - '0'];
 222                 }
 223                 fmt += size;
 224         }
 225
 226         vaf.fmt = fmt;
 227         vaf.va = &args;
 228
 229         if (__ratelimit(ratelimit))
 230                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type,
 231                         fs_info ? fs_info->sb->s_id : "<unknown>", &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 && (btrfs_test_opt(fs_info, 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 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 330         Opt_ref_verify,
 331 #endif
 332         Opt_err,
 333 };
 334
 335 static const match_table_t tokens = {
 336         {Opt_degraded, "degraded"},
 337         {Opt_subvol, "subvol=%s"},
 338         {Opt_subvolid, "subvolid=%s"},
 339         {Opt_device, "device=%s"},
 340         {Opt_nodatasum, "nodatasum"},
 341         {Opt_datasum, "datasum"},
 342         {Opt_nodatacow, "nodatacow"},
 343         {Opt_datacow, "datacow"},
 344         {Opt_nobarrier, "nobarrier"},
 345         {Opt_barrier, "barrier"},
 346         {Opt_max_inline, "max_inline=%s"},
 347         {Opt_alloc_start, "alloc_start=%s"},
 348         {Opt_thread_pool, "thread_pool=%d"},
 349         {Opt_compress, "compress"},
 350         {Opt_compress_type, "compress=%s"},
 351         {Opt_compress_force, "compress-force"},
 352         {Opt_compress_force_type, "compress-force=%s"},
 353         {Opt_ssd, "ssd"},
 354         {Opt_ssd_spread, "ssd_spread"},
 355         {Opt_nossd, "nossd"},
 356         {Opt_acl, "acl"},
 357         {Opt_noacl, "noacl"},
 358         {Opt_notreelog, "notreelog"},
 359         {Opt_treelog, "treelog"},
 360         {Opt_nologreplay, "nologreplay"},
 361         {Opt_norecovery, "norecovery"},
 362         {Opt_flushoncommit, "flushoncommit"},
 363         {Opt_noflushoncommit, "noflushoncommit"},
 364         {Opt_ratio, "metadata_ratio=%d"},
 365         {Opt_discard, "discard"},
 366         {Opt_nodiscard, "nodiscard"},
 367         {Opt_space_cache, "space_cache"},
 368         {Opt_space_cache_version, "space_cache=%s"},
 369         {Opt_clear_cache, "clear_cache"},
 370         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 371         {Opt_enospc_debug, "enospc_debug"},
 372         {Opt_noenospc_debug, "noenospc_debug"},
 373         {Opt_subvolrootid, "subvolrootid=%d"},
 374         {Opt_defrag, "autodefrag"},
 375         {Opt_nodefrag, "noautodefrag"},
 376         {Opt_inode_cache, "inode_cache"},
 377         {Opt_noinode_cache, "noinode_cache"},
 378         {Opt_no_space_cache, "nospace_cache"},
 379         {Opt_recovery, "recovery"}, /* deprecated */
 380         {Opt_usebackuproot, "usebackuproot"},
 381         {Opt_skip_balance, "skip_balance"},
 382         {Opt_check_integrity, "check_int"},
 383         {Opt_check_integrity_including_extent_data, "check_int_data"},
 384         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 385         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 386         {Opt_fatal_errors, "fatal_errors=%s"},
 387         {Opt_commit_interval, "commit=%d"},
 388 #ifdef CONFIG_BTRFS_DEBUG
 389         {Opt_fragment_data, "fragment=data"},
 390         {Opt_fragment_metadata, "fragment=metadata"},
 391         {Opt_fragment_all, "fragment=all"},
 392 #endif
 393 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 394         {Opt_ref_verify, "ref_verify"},
 395 #endif
 396         {Opt_err, NULL},
 397 };
 398
 399 /*
 400  * Regular mount options parser.  Everything that is needed only when
 401  * reading in a new superblock is parsed here.
 402  * XXX JDM: This needs to be cleaned up for remount.
 403  */
 404 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 405                         unsigned long new_flags)
 406 {
 407         substring_t args[MAX_OPT_ARGS];
 408         char *p, *num, *orig = NULL;
 409         u64 cache_gen;
 410         int intarg;
 411         int ret = 0;
 412         char *compress_type;
 413         bool compress_force = false;
 414         enum btrfs_compression_type saved_compress_type;
 415         bool saved_compress_force;
 416         int no_compress = 0;
 417
 418         cache_gen = btrfs_super_cache_generation(info->super_copy);
 419         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 420                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 421         else if (cache_gen)
 422                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 423
 424         /*
 425          * Even the options are empty, we still need to do extra check
 426          * against new flags
 427          */
 428         if (!options)
 429                 goto check;
 430
 431         /*
 432          * strsep changes the string, duplicate it because parse_options
 433          * gets called twice
 434          */
 435         options = kstrdup(options, GFP_KERNEL);
 436         if (!options)
 437                 return -ENOMEM;
 438
 439         orig = options;
 440
 441         while ((p = strsep(&options, ",")) != NULL) {
 442                 int token;
 443                 if (!*p)
 444                         continue;
 445
 446                 token = match_token(p, tokens, args);
 447                 switch (token) {
 448                 case Opt_degraded:
 449                         btrfs_info(info, "allowing degraded mounts");
 450                         btrfs_set_opt(info->mount_opt, DEGRADED);
 451                         break;
 452                 case Opt_subvol:
 453                 case Opt_subvolid:
 454                 case Opt_subvolrootid:
 455                 case Opt_device:
 456                         /*
 457                          * These are parsed by btrfs_parse_early_options
 458                          * and can be happily ignored here.
 459                          */
 460                         break;
 461                 case Opt_nodatasum:
 462                         btrfs_set_and_info(info, NODATASUM,
 463                                            "setting nodatasum");
 464                         break;
 465                 case Opt_datasum:
 466                         if (btrfs_test_opt(info, NODATASUM)) {
 467                                 if (btrfs_test_opt(info, NODATACOW))
 468                                         btrfs_info(info,
 469                                                    "setting datasum, datacow enabled");
 470                                 else
 471                                         btrfs_info(info, "setting datasum");
 472                         }
 473                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 474                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 475                         break;
 476                 case Opt_nodatacow:
 477                         if (!btrfs_test_opt(info, NODATACOW)) {
 478                                 if (!btrfs_test_opt(info, COMPRESS) ||
 479                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 480                                         btrfs_info(info,
 481                                                    "setting nodatacow, compression disabled");
 482                                 } else {
 483                                         btrfs_info(info, "setting nodatacow");
 484                                 }
 485                         }
 486                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 487                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 488                         btrfs_set_opt(info->mount_opt, NODATACOW);
 489                         btrfs_set_opt(info->mount_opt, NODATASUM);
 490                         break;
 491                 case Opt_datacow:
 492                         btrfs_clear_and_info(info, NODATACOW,
 493                                              "setting datacow");
 494                         break;
 495                 case Opt_compress_force:
 496                 case Opt_compress_force_type:
 497                         compress_force = true;
 498                         /* Fallthrough */
 499                 case Opt_compress:
 500                 case Opt_compress_type:
 501                         saved_compress_type = btrfs_test_opt(info,
 502                                                              COMPRESS) ?
 503                                 info->compress_type : BTRFS_COMPRESS_NONE;
 504                         saved_compress_force =
 505                                 btrfs_test_opt(info, FORCE_COMPRESS);
 506                         if (token == Opt_compress ||
 507                             token == Opt_compress_force ||
 508                             strncmp(args[0].from, "zlib", 4) == 0) {
 509                                 compress_type = "zlib";
 510                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 511                                 info->compress_level =
 512                                         btrfs_compress_str2level(args[0].from);
 513                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 514                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 515                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 516                                 no_compress = 0;
 517                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
 518                                 compress_type = "lzo";
 519                                 info->compress_type = BTRFS_COMPRESS_LZO;
 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_LZO);
 524                                 no_compress = 0;
 525                         } else if (strcmp(args[0].from, "zstd") == 0) {
 526                                 compress_type = "zstd";
 527                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
 528                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 529                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 530                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 531                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
 532                                 no_compress = 0;
 533                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 534                                 compress_type = "no";
 535                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 536                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 537                                 compress_force = false;
 538                                 no_compress++;
 539                         } else {
 540                                 ret = -EINVAL;
 541                                 goto out;
 542                         }
 543
 544                         if (compress_force) {
 545                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 546                         } else {
 547                                 /*
 548                                  * If we remount from compress-force=xxx to
 549                                  * compress=xxx, we need clear FORCE_COMPRESS
 550                                  * flag, otherwise, there is no way for users
 551                                  * to disable forcible compression separately.
 552                                  */
 553                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 554                         }
 555                         if ((btrfs_test_opt(info, COMPRESS) &&
 556                              (info->compress_type != saved_compress_type ||
 557                               compress_force != saved_compress_force)) ||
 558                             (!btrfs_test_opt(info, COMPRESS) &&
 559                              no_compress == 1)) {
 560                                 btrfs_info(info, "%s %s compression, level %d",
 561                                            (compress_force) ? "force" : "use",
 562                                            compress_type, info->compress_level);
 563                         }
 564                         compress_force = false;
 565                         break;
 566                 case Opt_ssd:
 567                         btrfs_set_and_info(info, SSD,
 568                                            "enabling ssd optimizations");
 569                         btrfs_clear_opt(info->mount_opt, NOSSD);
 570                         break;
 571                 case Opt_ssd_spread:
 572                         btrfs_set_and_info(info, SSD,
 573                                            "enabling ssd optimizations");
 574                         btrfs_set_and_info(info, SSD_SPREAD,
 575                                            "using spread ssd allocation scheme");
 576                         btrfs_clear_opt(info->mount_opt, NOSSD);
 577                         break;
 578                 case Opt_nossd:
 579                         btrfs_set_opt(info->mount_opt, NOSSD);
 580                         btrfs_clear_and_info(info, SSD,
 581                                              "not using ssd optimizations");
 582                         btrfs_clear_and_info(info, SSD_SPREAD,
 583                                              "not using spread ssd allocation scheme");
 584                         break;
 585                 case Opt_barrier:
 586                         btrfs_clear_and_info(info, NOBARRIER,
 587                                              "turning on barriers");
 588                         break;
 589                 case Opt_nobarrier:
 590                         btrfs_set_and_info(info, NOBARRIER,
 591                                            "turning off barriers");
 592                         break;
 593                 case Opt_thread_pool:
 594                         ret = match_int(&args[0], &intarg);
 595                         if (ret) {
 596                                 goto out;
 597                         } else if (intarg > 0) {
 598                                 info->thread_pool_size = intarg;
 599                         } else {
 600                                 ret = -EINVAL;
 601                                 goto out;
 602                         }
 603                         break;
 604                 case Opt_max_inline:
 605                         num = match_strdup(&args[0]);
 606                         if (num) {
 607                                 info->max_inline = memparse(num, NULL);
 608                                 kfree(num);
 609
 610                                 if (info->max_inline) {
 611                                         info->max_inline = min_t(u64,
 612                                                 info->max_inline,
 613                                                 info->sectorsize);
 614                                 }
 615                                 btrfs_info(info, "max_inline at %llu",
 616                                            info->max_inline);
 617                         } else {
 618                                 ret = -ENOMEM;
 619                                 goto out;
 620                         }
 621                         break;
 622                 case Opt_alloc_start:
 623                         btrfs_info(info,
 624                                 "option alloc_start is obsolete, ignored");
 625                         break;
 626                 case Opt_acl:
 627 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 628                         info->sb->s_flags |= MS_POSIXACL;
 629                         break;
 630 #else
 631                         btrfs_err(info, "support for ACL not compiled in!");
 632                         ret = -EINVAL;
 633                         goto out;
 634 #endif
 635                 case Opt_noacl:
 636                         info->sb->s_flags &= ~MS_POSIXACL;
 637                         break;
 638                 case Opt_notreelog:
 639                         btrfs_set_and_info(info, NOTREELOG,
 640                                            "disabling tree log");
 641                         break;
 642                 case Opt_treelog:
 643                         btrfs_clear_and_info(info, NOTREELOG,
 644                                              "enabling tree log");
 645                         break;
 646                 case Opt_norecovery:
 647                 case Opt_nologreplay:
 648                         btrfs_set_and_info(info, NOLOGREPLAY,
 649                                            "disabling log replay at mount time");
 650                         break;
 651                 case Opt_flushoncommit:
 652                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 653                                            "turning on flush-on-commit");
 654                         break;
 655                 case Opt_noflushoncommit:
 656                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 657                                              "turning off flush-on-commit");
 658                         break;
 659                 case Opt_ratio:
 660                         ret = match_int(&args[0], &intarg);
 661                         if (ret) {
 662                                 goto out;
 663                         } else if (intarg >= 0) {
 664                                 info->metadata_ratio = intarg;
 665                                 btrfs_info(info, "metadata ratio %d",
 666                                            info->metadata_ratio);
 667                         } else {
 668                                 ret = -EINVAL;
 669                                 goto out;
 670                         }
 671                         break;
 672                 case Opt_discard:
 673                         btrfs_set_and_info(info, DISCARD,
 674                                            "turning on discard");
 675                         break;
 676                 case Opt_nodiscard:
 677                         btrfs_clear_and_info(info, DISCARD,
 678                                              "turning off discard");
 679                         break;
 680                 case Opt_space_cache:
 681                 case Opt_space_cache_version:
 682                         if (token == Opt_space_cache ||
 683                             strcmp(args[0].from, "v1") == 0) {
 684                                 btrfs_clear_opt(info->mount_opt,
 685                                                 FREE_SPACE_TREE);
 686                                 btrfs_set_and_info(info, SPACE_CACHE,
 687                                            "enabling disk space caching");
 688                         } else if (strcmp(args[0].from, "v2") == 0) {
 689                                 btrfs_clear_opt(info->mount_opt,
 690                                                 SPACE_CACHE);
 691                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 692                                                    "enabling free space tree");
 693                         } else {
 694                                 ret = -EINVAL;
 695                                 goto out;
 696                         }
 697                         break;
 698                 case Opt_rescan_uuid_tree:
 699                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 700                         break;
 701                 case Opt_no_space_cache:
 702                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 703                                 btrfs_clear_and_info(info, SPACE_CACHE,
 704                                              "disabling disk space caching");
 705                         }
 706                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 707                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 708                                              "disabling free space tree");
 709                         }
 710                         break;
 711                 case Opt_inode_cache:
 712                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 713                                            "enabling inode map caching");
 714                         break;
 715                 case Opt_noinode_cache:
 716                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 717                                              "disabling inode map caching");
 718                         break;
 719                 case Opt_clear_cache:
 720                         btrfs_set_and_info(info, CLEAR_CACHE,
 721                                            "force clearing of disk cache");
 722                         break;
 723                 case Opt_user_subvol_rm_allowed:
 724                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 725                         break;
 726                 case Opt_enospc_debug:
 727                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 728                         break;
 729                 case Opt_noenospc_debug:
 730                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 731                         break;
 732                 case Opt_defrag:
 733                         btrfs_set_and_info(info, AUTO_DEFRAG,
 734                                            "enabling auto defrag");
 735                         break;
 736                 case Opt_nodefrag:
 737                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 738                                              "disabling auto defrag");
 739                         break;
 740                 case Opt_recovery:
 741                         btrfs_warn(info,
 742                                    "'recovery' is deprecated, use 'usebackuproot' instead");
 743                 case Opt_usebackuproot:
 744                         btrfs_info(info,
 745                                    "trying to use backup root at mount time");
 746                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 747                         break;
 748                 case Opt_skip_balance:
 749                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 750                         break;
 751 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 752                 case Opt_check_integrity_including_extent_data:
 753                         btrfs_info(info,
 754                                    "enabling check integrity including extent data");
 755                         btrfs_set_opt(info->mount_opt,
 756                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 757                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 758                         break;
 759                 case Opt_check_integrity:
 760                         btrfs_info(info, "enabling check integrity");
 761                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 762                         break;
 763                 case Opt_check_integrity_print_mask:
 764                         ret = match_int(&args[0], &intarg);
 765                         if (ret) {
 766                                 goto out;
 767                         } else if (intarg >= 0) {
 768                                 info->check_integrity_print_mask = intarg;
 769                                 btrfs_info(info,
 770                                            "check_integrity_print_mask 0x%x",
 771                                            info->check_integrity_print_mask);
 772                         } else {
 773                                 ret = -EINVAL;
 774                                 goto out;
 775                         }
 776                         break;
 777 #else
 778                 case Opt_check_integrity_including_extent_data:
 779                 case Opt_check_integrity:
 780                 case Opt_check_integrity_print_mask:
 781                         btrfs_err(info,
 782                                   "support for check_integrity* not compiled in!");
 783                         ret = -EINVAL;
 784                         goto out;
 785 #endif
 786                 case Opt_fatal_errors:
 787                         if (strcmp(args[0].from, "panic") == 0)
 788                                 btrfs_set_opt(info->mount_opt,
 789                                               PANIC_ON_FATAL_ERROR);
 790                         else if (strcmp(args[0].from, "bug") == 0)
 791                                 btrfs_clear_opt(info->mount_opt,
 792                                               PANIC_ON_FATAL_ERROR);
 793                         else {
 794                                 ret = -EINVAL;
 795                                 goto out;
 796                         }
 797                         break;
 798                 case Opt_commit_interval:
 799                         intarg = 0;
 800                         ret = match_int(&args[0], &intarg);
 801                         if (ret < 0) {
 802                                 btrfs_err(info, "invalid commit interval");
 803                                 ret = -EINVAL;
 804                                 goto out;
 805                         }
 806                         if (intarg > 0) {
 807                                 if (intarg > 300) {
 808                                         btrfs_warn(info,
 809                                                 "excessive commit interval %d",
 810                                                 intarg);
 811                                 }
 812                                 info->commit_interval = intarg;
 813                         } else {
 814                                 btrfs_info(info,
 815                                            "using default commit interval %ds",
 816                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
 817                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 818                         }
 819                         break;
 820 #ifdef CONFIG_BTRFS_DEBUG
 821                 case Opt_fragment_all:
 822                         btrfs_info(info, "fragmenting all space");
 823                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 824                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 825                         break;
 826                 case Opt_fragment_metadata:
 827                         btrfs_info(info, "fragmenting metadata");
 828                         btrfs_set_opt(info->mount_opt,
 829                                       FRAGMENT_METADATA);
 830                         break;
 831                 case Opt_fragment_data:
 832                         btrfs_info(info, "fragmenting data");
 833                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 834                         break;
 835 #endif
 836 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 837                 case Opt_ref_verify:
 838                         btrfs_info(info, "doing ref verification");
 839                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
 840                         break;
 841 #endif
 842                 case Opt_err:
 843                         btrfs_info(info, "unrecognized mount option '%s'", p);
 844                         ret = -EINVAL;
 845                         goto out;
 846                 default:
 847                         break;
 848                 }
 849         }
 850 check:
 851         /*
 852          * Extra check for current option against current flag
 853          */
 854         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
 855                 btrfs_err(info,
 856                           "nologreplay must be used with ro mount option");
 857                 ret = -EINVAL;
 858         }
 859 out:
 860         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
 861             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
 862             !btrfs_test_opt(info, CLEAR_CACHE)) {
 863                 btrfs_err(info, "cannot disable free space tree");
 864                 ret = -EINVAL;
 865
 866         }
 867         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
 868                 btrfs_info(info, "disk space caching is enabled");
 869         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
 870                 btrfs_info(info, "using free space tree");
 871         kfree(orig);
 872         return ret;
 873 }
 874
 875 /*
 876  * Parse mount options that are required early in the mount process.
 877  *
 878  * All other options will be parsed on much later in the mount process and
 879  * only when we need to allocate a new super block.
 880  */
 881 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 882                 void *holder, char **subvol_name, u64 *subvol_objectid,
 883                 struct btrfs_fs_devices **fs_devices)
 884 {
 885         substring_t args[MAX_OPT_ARGS];
 886         char *device_name, *opts, *orig, *p;
 887         char *num = NULL;
 888         int error = 0;
 889
 890         if (!options)
 891                 return 0;
 892
 893         /*
 894          * strsep changes the string, duplicate it because parse_options
 895          * gets called twice
 896          */
 897         opts = kstrdup(options, GFP_KERNEL);
 898         if (!opts)
 899                 return -ENOMEM;
 900         orig = opts;
 901
 902         while ((p = strsep(&opts, ",")) != NULL) {
 903                 int token;
 904                 if (!*p)
 905                         continue;
 906
 907                 token = match_token(p, tokens, args);
 908                 switch (token) {
 909                 case Opt_subvol:
 910                         kfree(*subvol_name);
 911                         *subvol_name = match_strdup(&args[0]);
 912                         if (!*subvol_name) {
 913                                 error = -ENOMEM;
 914                                 goto out;
 915                         }
 916                         break;
 917                 case Opt_subvolid:
 918                         num = match_strdup(&args[0]);
 919                         if (num) {
 920                                 *subvol_objectid = memparse(num, NULL);
 921                                 kfree(num);
 922                                 /* we want the original fs_tree */
 923                                 if (!*subvol_objectid)
 924                                         *subvol_objectid =
 925                                                 BTRFS_FS_TREE_OBJECTID;
 926                         } else {
 927                                 error = -EINVAL;
 928                                 goto out;
 929                         }
 930                         break;
 931                 case Opt_subvolrootid:
 932                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
 933                         break;
 934                 case Opt_device:
 935                         device_name = match_strdup(&args[0]);
 936                         if (!device_name) {
 937                                 error = -ENOMEM;
 938                                 goto out;
 939                         }
 940                         error = btrfs_scan_one_device(device_name,
 941                                         flags, holder, fs_devices);
 942                         kfree(device_name);
 943                         if (error)
 944                                 goto out;
 945                         break;
 946                 default:
 947                         break;
 948                 }
 949         }
 950
 951 out:
 952         kfree(orig);
 953         return error;
 954 }
 955
 956 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 957                                            u64 subvol_objectid)
 958 {
 959         struct btrfs_root *root = fs_info->tree_root;
 960         struct btrfs_root *fs_root;
 961         struct btrfs_root_ref *root_ref;
 962         struct btrfs_inode_ref *inode_ref;
 963         struct btrfs_key key;
 964         struct btrfs_path *path = NULL;
 965         char *name = NULL, *ptr;
 966         u64 dirid;
 967         int len;
 968         int ret;
 969
 970         path = btrfs_alloc_path();
 971         if (!path) {
 972                 ret = -ENOMEM;
 973                 goto err;
 974         }
 975         path->leave_spinning = 1;
 976
 977         name = kmalloc(PATH_MAX, GFP_KERNEL);
 978         if (!name) {
 979                 ret = -ENOMEM;
 980                 goto err;
 981         }
 982         ptr = name + PATH_MAX - 1;
 983         ptr[0] = '\0';
 984
 985         /*
 986          * Walk up the subvolume trees in the tree of tree roots by root
 987          * backrefs until we hit the top-level subvolume.
 988          */
 989         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 990                 key.objectid = subvol_objectid;
 991                 key.type = BTRFS_ROOT_BACKREF_KEY;
 992                 key.offset = (u64)-1;
 993
 994                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 995                 if (ret < 0) {
 996                         goto err;
 997                 } else if (ret > 0) {
 998                         ret = btrfs_previous_item(root, path, subvol_objectid,
 999                                                   BTRFS_ROOT_BACKREF_KEY);
1000                         if (ret < 0) {
1001                                 goto err;
1002                         } else if (ret > 0) {
1003                                 ret = -ENOENT;
1004                                 goto err;
1005                         }
1006                 }
1007
1008                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1009                 subvol_objectid = key.offset;
1010
1011                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1012                                           struct btrfs_root_ref);
1013                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1014                 ptr -= len + 1;
1015                 if (ptr < name) {
1016                         ret = -ENAMETOOLONG;
1017                         goto err;
1018                 }
1019                 read_extent_buffer(path->nodes[0], ptr + 1,
1020                                    (unsigned long)(root_ref + 1), len);
1021                 ptr[0] = '/';
1022                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1023                 btrfs_release_path(path);
1024
1025                 key.objectid = subvol_objectid;
1026                 key.type = BTRFS_ROOT_ITEM_KEY;
1027                 key.offset = (u64)-1;
1028                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1029                 if (IS_ERR(fs_root)) {
1030                         ret = PTR_ERR(fs_root);
1031                         goto err;
1032                 }
1033
1034                 /*
1035                  * Walk up the filesystem tree by inode refs until we hit the
1036                  * root directory.
1037                  */
1038                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1039                         key.objectid = dirid;
1040                         key.type = BTRFS_INODE_REF_KEY;
1041                         key.offset = (u64)-1;
1042
1043                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1044                         if (ret < 0) {
1045                                 goto err;
1046                         } else if (ret > 0) {
1047                                 ret = btrfs_previous_item(fs_root, path, dirid,
1048                                                           BTRFS_INODE_REF_KEY);
1049                                 if (ret < 0) {
1050                                         goto err;
1051                                 } else if (ret > 0) {
1052                                         ret = -ENOENT;
1053                                         goto err;
1054                                 }
1055                         }
1056
1057                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1058                         dirid = key.offset;
1059
1060                         inode_ref = btrfs_item_ptr(path->nodes[0],
1061                                                    path->slots[0],
1062                                                    struct btrfs_inode_ref);
1063                         len = btrfs_inode_ref_name_len(path->nodes[0],
1064                                                        inode_ref);
1065                         ptr -= len + 1;
1066                         if (ptr < name) {
1067                                 ret = -ENAMETOOLONG;
1068                                 goto err;
1069                         }
1070                         read_extent_buffer(path->nodes[0], ptr + 1,
1071                                            (unsigned long)(inode_ref + 1), len);
1072                         ptr[0] = '/';
1073                         btrfs_release_path(path);
1074                 }
1075         }
1076
1077         btrfs_free_path(path);
1078         if (ptr == name + PATH_MAX - 1) {
1079                 name[0] = '/';
1080                 name[1] = '\0';
1081         } else {
1082                 memmove(name, ptr, name + PATH_MAX - ptr);
1083         }
1084         return name;
1085
1086 err:
1087         btrfs_free_path(path);
1088         kfree(name);
1089         return ERR_PTR(ret);
1090 }
1091
1092 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1093 {
1094         struct btrfs_root *root = fs_info->tree_root;
1095         struct btrfs_dir_item *di;
1096         struct btrfs_path *path;
1097         struct btrfs_key location;
1098         u64 dir_id;
1099
1100         path = btrfs_alloc_path();
1101         if (!path)
1102                 return -ENOMEM;
1103         path->leave_spinning = 1;
1104
1105         /*
1106          * Find the "default" dir item which points to the root item that we
1107          * will mount by default if we haven't been given a specific subvolume
1108          * to mount.
1109          */
1110         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1111         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1112         if (IS_ERR(di)) {
1113                 btrfs_free_path(path);
1114                 return PTR_ERR(di);
1115         }
1116         if (!di) {
1117                 /*
1118                  * Ok the default dir item isn't there.  This is weird since
1119                  * it's always been there, but don't freak out, just try and
1120                  * mount the top-level subvolume.
1121                  */
1122                 btrfs_free_path(path);
1123                 *objectid = BTRFS_FS_TREE_OBJECTID;
1124                 return 0;
1125         }
1126
1127         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1128         btrfs_free_path(path);
1129         *objectid = location.objectid;
1130         return 0;
1131 }
1132
1133 static int btrfs_fill_super(struct super_block *sb,
1134                             struct btrfs_fs_devices *fs_devices,
1135                             void *data)
1136 {
1137         struct inode *inode;
1138         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1139         struct btrfs_key key;
1140         int err;
1141
1142         sb->s_maxbytes = MAX_LFS_FILESIZE;
1143         sb->s_magic = BTRFS_SUPER_MAGIC;
1144         sb->s_op = &btrfs_super_ops;
1145         sb->s_d_op = &btrfs_dentry_operations;
1146         sb->s_export_op = &btrfs_export_ops;
1147         sb->s_xattr = btrfs_xattr_handlers;
1148         sb->s_time_gran = 1;
1149 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1150         sb->s_flags |= MS_POSIXACL;
1151 #endif
1152         sb->s_flags |= SB_I_VERSION;
1153         sb->s_iflags |= SB_I_CGROUPWB;
1154
1155         err = super_setup_bdi(sb);
1156         if (err) {
1157                 btrfs_err(fs_info, "super_setup_bdi failed");
1158                 return err;
1159         }
1160
1161         err = open_ctree(sb, fs_devices, (char *)data);
1162         if (err) {
1163                 btrfs_err(fs_info, "open_ctree failed");
1164                 return err;
1165         }
1166
1167         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1168         key.type = BTRFS_INODE_ITEM_KEY;
1169         key.offset = 0;
1170         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1171         if (IS_ERR(inode)) {
1172                 err = PTR_ERR(inode);
1173                 goto fail_close;
1174         }
1175
1176         sb->s_root = d_make_root(inode);
1177         if (!sb->s_root) {
1178                 err = -ENOMEM;
1179                 goto fail_close;
1180         }
1181
1182         cleancache_init_fs(sb);
1183         sb->s_flags |= MS_ACTIVE;
1184         return 0;
1185
1186 fail_close:
1187         close_ctree(fs_info);
1188         return err;
1189 }
1190
1191 int btrfs_sync_fs(struct super_block *sb, int wait)
1192 {
1193         struct btrfs_trans_handle *trans;
1194         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1195         struct btrfs_root *root = fs_info->tree_root;
1196
1197         trace_btrfs_sync_fs(fs_info, wait);
1198
1199         if (!wait) {
1200                 filemap_flush(fs_info->btree_inode->i_mapping);
1201                 return 0;
1202         }
1203
1204         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1205
1206         trans = btrfs_attach_transaction_barrier(root);
1207         if (IS_ERR(trans)) {
1208                 /* no transaction, don't bother */
1209                 if (PTR_ERR(trans) == -ENOENT) {
1210                         /*
1211                          * Exit unless we have some pending changes
1212                          * that need to go through commit
1213                          */
1214                         if (fs_info->pending_changes == 0)
1215                                 return 0;
1216                         /*
1217                          * A non-blocking test if the fs is frozen. We must not
1218                          * start a new transaction here otherwise a deadlock
1219                          * happens. The pending operations are delayed to the
1220                          * next commit after thawing.
1221                          */
1222                         if (sb_start_write_trylock(sb))
1223                                 sb_end_write(sb);
1224                         else
1225                                 return 0;
1226                         trans = btrfs_start_transaction(root, 0);
1227                 }
1228                 if (IS_ERR(trans))
1229                         return PTR_ERR(trans);
1230         }
1231         return btrfs_commit_transaction(trans);
1232 }
1233
1234 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1235 {
1236         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1237         char *compress_type;
1238
1239         if (btrfs_test_opt(info, DEGRADED))
1240                 seq_puts(seq, ",degraded");
1241         if (btrfs_test_opt(info, NODATASUM))
1242                 seq_puts(seq, ",nodatasum");
1243         if (btrfs_test_opt(info, NODATACOW))
1244                 seq_puts(seq, ",nodatacow");
1245         if (btrfs_test_opt(info, NOBARRIER))
1246                 seq_puts(seq, ",nobarrier");
1247         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1248                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1249         if (info->thread_pool_size !=  min_t(unsigned long,
1250                                              num_online_cpus() + 2, 8))
1251                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1252         if (btrfs_test_opt(info, COMPRESS)) {
1253                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1254                         compress_type = "zlib";
1255                 else if (info->compress_type == BTRFS_COMPRESS_LZO)
1256                         compress_type = "lzo";
1257                 else
1258                         compress_type = "zstd";
1259                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1260                         seq_printf(seq, ",compress-force=%s", compress_type);
1261                 else
1262                         seq_printf(seq, ",compress=%s", compress_type);
1263                 if (info->compress_level)
1264                         seq_printf(seq, ":%d", info->compress_level);
1265         }
1266         if (btrfs_test_opt(info, NOSSD))
1267                 seq_puts(seq, ",nossd");
1268         if (btrfs_test_opt(info, SSD_SPREAD))
1269                 seq_puts(seq, ",ssd_spread");
1270         else if (btrfs_test_opt(info, SSD))
1271                 seq_puts(seq, ",ssd");
1272         if (btrfs_test_opt(info, NOTREELOG))
1273                 seq_puts(seq, ",notreelog");
1274         if (btrfs_test_opt(info, NOLOGREPLAY))
1275                 seq_puts(seq, ",nologreplay");
1276         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1277                 seq_puts(seq, ",flushoncommit");
1278         if (btrfs_test_opt(info, DISCARD))
1279                 seq_puts(seq, ",discard");
1280         if (!(info->sb->s_flags & MS_POSIXACL))
1281                 seq_puts(seq, ",noacl");
1282         if (btrfs_test_opt(info, SPACE_CACHE))
1283                 seq_puts(seq, ",space_cache");
1284         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1285                 seq_puts(seq, ",space_cache=v2");
1286         else
1287                 seq_puts(seq, ",nospace_cache");
1288         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1289                 seq_puts(seq, ",rescan_uuid_tree");
1290         if (btrfs_test_opt(info, CLEAR_CACHE))
1291                 seq_puts(seq, ",clear_cache");
1292         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1293                 seq_puts(seq, ",user_subvol_rm_allowed");
1294         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1295                 seq_puts(seq, ",enospc_debug");
1296         if (btrfs_test_opt(info, AUTO_DEFRAG))
1297                 seq_puts(seq, ",autodefrag");
1298         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1299                 seq_puts(seq, ",inode_cache");
1300         if (btrfs_test_opt(info, SKIP_BALANCE))
1301                 seq_puts(seq, ",skip_balance");
1302 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1303         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1304                 seq_puts(seq, ",check_int_data");
1305         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1306                 seq_puts(seq, ",check_int");
1307         if (info->check_integrity_print_mask)
1308                 seq_printf(seq, ",check_int_print_mask=%d",
1309                                 info->check_integrity_print_mask);
1310 #endif
1311         if (info->metadata_ratio)
1312                 seq_printf(seq, ",metadata_ratio=%d",
1313                                 info->metadata_ratio);
1314         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1315                 seq_puts(seq, ",fatal_errors=panic");
1316         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1317                 seq_printf(seq, ",commit=%d", info->commit_interval);
1318 #ifdef CONFIG_BTRFS_DEBUG
1319         if (btrfs_test_opt(info, FRAGMENT_DATA))
1320                 seq_puts(seq, ",fragment=data");
1321         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1322                 seq_puts(seq, ",fragment=metadata");
1323 #endif
1324         if (btrfs_test_opt(info, REF_VERIFY))
1325                 seq_puts(seq, ",ref_verify");
1326         seq_printf(seq, ",subvolid=%llu",
1327                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1328         seq_puts(seq, ",subvol=");
1329         seq_dentry(seq, dentry, " \t\n\\");
1330         return 0;
1331 }
1332
1333 static int btrfs_test_super(struct super_block *s, void *data)
1334 {
1335         struct btrfs_fs_info *p = data;
1336         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1337
1338         return fs_info->fs_devices == p->fs_devices;
1339 }
1340
1341 static int btrfs_set_super(struct super_block *s, void *data)
1342 {
1343         int err = set_anon_super(s, data);
1344         if (!err)
1345                 s->s_fs_info = data;
1346         return err;
1347 }
1348
1349 /*
1350  * subvolumes are identified by ino 256
1351  */
1352 static inline int is_subvolume_inode(struct inode *inode)
1353 {
1354         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1355                 return 1;
1356         return 0;
1357 }
1358
1359 /*
1360  * This will add subvolid=0 to the argument string while removing any subvol=
1361  * and subvolid= arguments to make sure we get the top-level root for path
1362  * walking to the subvol we want.
1363  */
1364 static char *setup_root_args(char *args)
1365 {
1366         char *buf, *dst, *sep;
1367
1368         if (!args)
1369                 return kstrdup("subvolid=0", GFP_KERNEL);
1370
1371         /* The worst case is that we add ",subvolid=0" to the end. */
1372         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1,
1373                         GFP_KERNEL);
1374         if (!buf)
1375                 return NULL;
1376
1377         while (1) {
1378                 sep = strchrnul(args, ',');
1379                 if (!strstarts(args, "subvol=") &&
1380                     !strstarts(args, "subvolid=")) {
1381                         memcpy(dst, args, sep - args);
1382                         dst += sep - args;
1383                         *dst++ = ',';
1384                 }
1385                 if (*sep)
1386                         args = sep + 1;
1387                 else
1388                         break;
1389         }
1390         strcpy(dst, "subvolid=0");
1391
1392         return buf;
1393 }
1394
1395 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1396                                    int flags, const char *device_name,
1397                                    char *data)
1398 {
1399         struct dentry *root;
1400         struct vfsmount *mnt = NULL;
1401         char *newargs;
1402         int ret;
1403
1404         newargs = setup_root_args(data);
1405         if (!newargs) {
1406                 root = ERR_PTR(-ENOMEM);
1407                 goto out;
1408         }
1409
1410         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1411         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1412                 if (flags & MS_RDONLY) {
1413                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1414                                              device_name, newargs);
1415                 } else {
1416                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1417                                              device_name, newargs);
1418                         if (IS_ERR(mnt)) {
1419                                 root = ERR_CAST(mnt);
1420                                 mnt = NULL;
1421                                 goto out;
1422                         }
1423
1424                         down_write(&mnt->mnt_sb->s_umount);
1425                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1426                         up_write(&mnt->mnt_sb->s_umount);
1427                         if (ret < 0) {
1428                                 root = ERR_PTR(ret);
1429                                 goto out;
1430                         }
1431                 }
1432         }
1433         if (IS_ERR(mnt)) {
1434                 root = ERR_CAST(mnt);
1435                 mnt = NULL;
1436                 goto out;
1437         }
1438
1439         if (!subvol_name) {
1440                 if (!subvol_objectid) {
1441                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1442                                                           &subvol_objectid);
1443                         if (ret) {
1444                                 root = ERR_PTR(ret);
1445                                 goto out;
1446                         }
1447                 }
1448                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1449                                                             subvol_objectid);
1450                 if (IS_ERR(subvol_name)) {
1451                         root = ERR_CAST(subvol_name);
1452                         subvol_name = NULL;
1453                         goto out;
1454                 }
1455
1456         }
1457
1458         root = mount_subtree(mnt, subvol_name);
1459         /* mount_subtree() drops our reference on the vfsmount. */
1460         mnt = NULL;
1461
1462         if (!IS_ERR(root)) {
1463                 struct super_block *s = root->d_sb;
1464                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1465                 struct inode *root_inode = d_inode(root);
1466                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1467
1468                 ret = 0;
1469                 if (!is_subvolume_inode(root_inode)) {
1470                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1471                                subvol_name);
1472                         ret = -EINVAL;
1473                 }
1474                 if (subvol_objectid && root_objectid != subvol_objectid) {
1475                         /*
1476                          * This will also catch a race condition where a
1477                          * subvolume which was passed by ID is renamed and
1478                          * another subvolume is renamed over the old location.
1479                          */
1480                         btrfs_err(fs_info,
1481                                   "subvol '%s' does not match subvolid %llu",
1482                                   subvol_name, subvol_objectid);
1483                         ret = -EINVAL;
1484                 }
1485                 if (ret) {
1486                         dput(root);
1487                         root = ERR_PTR(ret);
1488                         deactivate_locked_super(s);
1489                 }
1490         }
1491
1492 out:
1493         mntput(mnt);
1494         kfree(newargs);
1495         kfree(subvol_name);
1496         return root;
1497 }
1498
1499 static int parse_security_options(char *orig_opts,
1500                                   struct security_mnt_opts *sec_opts)
1501 {
1502         char *secdata = NULL;
1503         int ret = 0;
1504
1505         secdata = alloc_secdata();
1506         if (!secdata)
1507                 return -ENOMEM;
1508         ret = security_sb_copy_data(orig_opts, secdata);
1509         if (ret) {
1510                 free_secdata(secdata);
1511                 return ret;
1512         }
1513         ret = security_sb_parse_opts_str(secdata, sec_opts);
1514         free_secdata(secdata);
1515         return ret;
1516 }
1517
1518 static int setup_security_options(struct btrfs_fs_info *fs_info,
1519                                   struct super_block *sb,
1520                                   struct security_mnt_opts *sec_opts)
1521 {
1522         int ret = 0;
1523
1524         /*
1525          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1526          * is valid.
1527          */
1528         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1529         if (ret)
1530                 return ret;
1531
1532 #ifdef CONFIG_SECURITY
1533         if (!fs_info->security_opts.num_mnt_opts) {
1534                 /* first time security setup, copy sec_opts to fs_info */
1535                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1536         } else {
1537                 /*
1538                  * Since SELinux (the only one supporting security_mnt_opts)
1539                  * does NOT support changing context during remount/mount of
1540                  * the same sb, this must be the same or part of the same
1541                  * security options, just free it.
1542                  */
1543                 security_free_mnt_opts(sec_opts);
1544         }
1545 #endif
1546         return ret;
1547 }
1548
1549 /*
1550  * Find a superblock for the given device / mount point.
1551  *
1552  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1553  *        for multiple device setup.  Make sure to keep it in sync.
1554  */
1555 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1556                 const char *device_name, void *data)
1557 {
1558         struct block_device *bdev = NULL;
1559         struct super_block *s;
1560         struct btrfs_fs_devices *fs_devices = NULL;
1561         struct btrfs_fs_info *fs_info = NULL;
1562         struct security_mnt_opts new_sec_opts;
1563         fmode_t mode = FMODE_READ;
1564         char *subvol_name = NULL;
1565         u64 subvol_objectid = 0;
1566         int error = 0;
1567
1568         if (!(flags & MS_RDONLY))
1569                 mode |= FMODE_WRITE;
1570
1571         error = btrfs_parse_early_options(data, mode, fs_type,
1572                                           &subvol_name, &subvol_objectid,
1573                                           &fs_devices);
1574         if (error) {
1575                 kfree(subvol_name);
1576                 return ERR_PTR(error);
1577         }
1578
1579         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1580                 /* mount_subvol() will free subvol_name. */
1581                 return mount_subvol(subvol_name, subvol_objectid, flags,
1582                                     device_name, data);
1583         }
1584
1585         security_init_mnt_opts(&new_sec_opts);
1586         if (data) {
1587                 error = parse_security_options(data, &new_sec_opts);
1588                 if (error)
1589                         return ERR_PTR(error);
1590         }
1591
1592         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1593         if (error)
1594                 goto error_sec_opts;
1595
1596         /*
1597          * Setup a dummy root and fs_info for test/set super.  This is because
1598          * we don't actually fill this stuff out until open_ctree, but we need
1599          * it for searching for existing supers, so this lets us do that and
1600          * then open_ctree will properly initialize everything later.
1601          */
1602         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1603         if (!fs_info) {
1604                 error = -ENOMEM;
1605                 goto error_sec_opts;
1606         }
1607
1608         fs_info->fs_devices = fs_devices;
1609
1610         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1611         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1612         security_init_mnt_opts(&fs_info->security_opts);
1613         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1614                 error = -ENOMEM;
1615                 goto error_fs_info;
1616         }
1617
1618         error = btrfs_open_devices(fs_devices, mode, fs_type);
1619         if (error)
1620                 goto error_fs_info;
1621
1622         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1623                 error = -EACCES;
1624                 goto error_close_devices;
1625         }
1626
1627         bdev = fs_devices->latest_bdev;
1628         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1629                  fs_info);
1630         if (IS_ERR(s)) {
1631                 error = PTR_ERR(s);
1632                 goto error_close_devices;
1633         }
1634
1635         if (s->s_root) {
1636                 btrfs_close_devices(fs_devices);
1637                 free_fs_info(fs_info);
1638                 if ((flags ^ s->s_flags) & MS_RDONLY)
1639                         error = -EBUSY;
1640         } else {
1641                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1642                 btrfs_sb(s)->bdev_holder = fs_type;
1643                 error = btrfs_fill_super(s, fs_devices, data);
1644         }
1645         if (error) {
1646                 deactivate_locked_super(s);
1647                 goto error_sec_opts;
1648         }
1649
1650         fs_info = btrfs_sb(s);
1651         error = setup_security_options(fs_info, s, &new_sec_opts);
1652         if (error) {
1653                 deactivate_locked_super(s);
1654                 goto error_sec_opts;
1655         }
1656
1657         return dget(s->s_root);
1658
1659 error_close_devices:
1660         btrfs_close_devices(fs_devices);
1661 error_fs_info:
1662         free_fs_info(fs_info);
1663 error_sec_opts:
1664         security_free_mnt_opts(&new_sec_opts);
1665         return ERR_PTR(error);
1666 }
1667
1668 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1669                                      int new_pool_size, int old_pool_size)
1670 {
1671         if (new_pool_size == old_pool_size)
1672                 return;
1673
1674         fs_info->thread_pool_size = new_pool_size;
1675
1676         btrfs_info(fs_info, "resize thread pool %d -> %d",
1677                old_pool_size, new_pool_size);
1678
1679         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1680         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1681         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1682         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1683         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1684         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1685         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1686                                 new_pool_size);
1687         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1688         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1689         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1690         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1691         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1692                                 new_pool_size);
1693 }
1694
1695 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1696 {
1697         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1698 }
1699
1700 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1701                                        unsigned long old_opts, int flags)
1702 {
1703         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1704             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1705              (flags & MS_RDONLY))) {
1706                 /* wait for any defraggers to finish */
1707                 wait_event(fs_info->transaction_wait,
1708                            (atomic_read(&fs_info->defrag_running) == 0));
1709                 if (flags & MS_RDONLY)
1710                         sync_filesystem(fs_info->sb);
1711         }
1712 }
1713
1714 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1715                                          unsigned long old_opts)
1716 {
1717         /*
1718          * We need to cleanup all defragable inodes if the autodefragment is
1719          * close or the filesystem is read only.
1720          */
1721         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1722             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1723                 btrfs_cleanup_defrag_inodes(fs_info);
1724         }
1725
1726         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1727 }
1728
1729 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1730 {
1731         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1732         struct btrfs_root *root = fs_info->tree_root;
1733         unsigned old_flags = sb->s_flags;
1734         unsigned long old_opts = fs_info->mount_opt;
1735         unsigned long old_compress_type = fs_info->compress_type;
1736         u64 old_max_inline = fs_info->max_inline;
1737         int old_thread_pool_size = fs_info->thread_pool_size;
1738         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1739         int ret;
1740
1741         sync_filesystem(sb);
1742         btrfs_remount_prepare(fs_info);
1743
1744         if (data) {
1745                 struct security_mnt_opts new_sec_opts;
1746
1747                 security_init_mnt_opts(&new_sec_opts);
1748                 ret = parse_security_options(data, &new_sec_opts);
1749                 if (ret)
1750                         goto restore;
1751                 ret = setup_security_options(fs_info, sb,
1752                                              &new_sec_opts);
1753                 if (ret) {
1754                         security_free_mnt_opts(&new_sec_opts);
1755                         goto restore;
1756                 }
1757         }
1758
1759         ret = btrfs_parse_options(fs_info, data, *flags);
1760         if (ret) {
1761                 ret = -EINVAL;
1762                 goto restore;
1763         }
1764
1765         btrfs_remount_begin(fs_info, old_opts, *flags);
1766         btrfs_resize_thread_pool(fs_info,
1767                 fs_info->thread_pool_size, old_thread_pool_size);
1768
1769         if ((bool)(*flags & MS_RDONLY) == sb_rdonly(sb))
1770                 goto out;
1771
1772         if (*flags & MS_RDONLY) {
1773                 /*
1774                  * this also happens on 'umount -rf' or on shutdown, when
1775                  * the filesystem is busy.
1776                  */
1777                 cancel_work_sync(&fs_info->async_reclaim_work);
1778
1779                 /* wait for the uuid_scan task to finish */
1780                 down(&fs_info->uuid_tree_rescan_sem);
1781                 /* avoid complains from lockdep et al. */
1782                 up(&fs_info->uuid_tree_rescan_sem);
1783
1784                 sb->s_flags |= MS_RDONLY;
1785
1786                 /*
1787                  * Setting MS_RDONLY will put the cleaner thread to
1788                  * sleep at the next loop if it's already active.
1789                  * If it's already asleep, we'll leave unused block
1790                  * groups on disk until we're mounted read-write again
1791                  * unless we clean them up here.
1792                  */
1793                 btrfs_delete_unused_bgs(fs_info);
1794
1795                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1796                 btrfs_scrub_cancel(fs_info);
1797                 btrfs_pause_balance(fs_info);
1798
1799                 ret = btrfs_commit_super(fs_info);
1800                 if (ret)
1801                         goto restore;
1802         } else {
1803                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1804                         btrfs_err(fs_info,
1805                                 "Remounting read-write after error is not allowed");
1806                         ret = -EINVAL;
1807                         goto restore;
1808                 }
1809                 if (fs_info->fs_devices->rw_devices == 0) {
1810                         ret = -EACCES;
1811                         goto restore;
1812                 }
1813
1814                 if (!btrfs_check_rw_degradable(fs_info)) {
1815                         btrfs_warn(fs_info,
1816                                 "too many missing devices, writeable remount is not allowed");
1817                         ret = -EACCES;
1818                         goto restore;
1819                 }
1820
1821                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1822                         ret = -EINVAL;
1823                         goto restore;
1824                 }
1825
1826                 ret = btrfs_cleanup_fs_roots(fs_info);
1827                 if (ret)
1828                         goto restore;
1829
1830                 /* recover relocation */
1831                 mutex_lock(&fs_info->cleaner_mutex);
1832                 ret = btrfs_recover_relocation(root);
1833                 mutex_unlock(&fs_info->cleaner_mutex);
1834                 if (ret)
1835                         goto restore;
1836
1837                 ret = btrfs_resume_balance_async(fs_info);
1838                 if (ret)
1839                         goto restore;
1840
1841                 ret = btrfs_resume_dev_replace_async(fs_info);
1842                 if (ret) {
1843                         btrfs_warn(fs_info, "failed to resume dev_replace");
1844                         goto restore;
1845                 }
1846
1847                 btrfs_qgroup_rescan_resume(fs_info);
1848
1849                 if (!fs_info->uuid_root) {
1850                         btrfs_info(fs_info, "creating UUID tree");
1851                         ret = btrfs_create_uuid_tree(fs_info);
1852                         if (ret) {
1853                                 btrfs_warn(fs_info,
1854                                            "failed to create the UUID tree %d",
1855                                            ret);
1856                                 goto restore;
1857                         }
1858                 }
1859                 sb->s_flags &= ~MS_RDONLY;
1860
1861                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1862         }
1863 out:
1864         wake_up_process(fs_info->transaction_kthread);
1865         btrfs_remount_cleanup(fs_info, old_opts);
1866         return 0;
1867
1868 restore:
1869         /* We've hit an error - don't reset MS_RDONLY */
1870         if (sb_rdonly(sb))
1871                 old_flags |= MS_RDONLY;
1872         sb->s_flags = old_flags;
1873         fs_info->mount_opt = old_opts;
1874         fs_info->compress_type = old_compress_type;
1875         fs_info->max_inline = old_max_inline;
1876         btrfs_resize_thread_pool(fs_info,
1877                 old_thread_pool_size, fs_info->thread_pool_size);
1878         fs_info->metadata_ratio = old_metadata_ratio;
1879         btrfs_remount_cleanup(fs_info, old_opts);
1880         return ret;
1881 }
1882
1883 /* Used to sort the devices by max_avail(descending sort) */
1884 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1885                                        const void *dev_info2)
1886 {
1887         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1888             ((struct btrfs_device_info *)dev_info2)->max_avail)
1889                 return -1;
1890         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1891                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1892                 return 1;
1893         else
1894         return 0;
1895 }
1896
1897 /*
1898  * sort the devices by max_avail, in which max free extent size of each device
1899  * is stored.(Descending Sort)
1900  */
1901 static inline void btrfs_descending_sort_devices(
1902                                         struct btrfs_device_info *devices,
1903                                         size_t nr_devices)
1904 {
1905         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1906              btrfs_cmp_device_free_bytes, NULL);
1907 }
1908
1909 /*
1910  * The helper to calc the free space on the devices that can be used to store
1911  * file data.
1912  */
1913 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1914                                        u64 *free_bytes)
1915 {
1916         struct btrfs_device_info *devices_info;
1917         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1918         struct btrfs_device *device;
1919         u64 skip_space;
1920         u64 type;
1921         u64 avail_space;
1922         u64 min_stripe_size;
1923         int min_stripes = 1, num_stripes = 1;
1924         int i = 0, nr_devices;
1925
1926         /*
1927          * We aren't under the device list lock, so this is racy-ish, but good
1928          * enough for our purposes.
1929          */
1930         nr_devices = fs_info->fs_devices->open_devices;
1931         if (!nr_devices) {
1932                 smp_mb();
1933                 nr_devices = fs_info->fs_devices->open_devices;
1934                 ASSERT(nr_devices);
1935                 if (!nr_devices) {
1936                         *free_bytes = 0;
1937                         return 0;
1938                 }
1939         }
1940
1941         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1942                                GFP_KERNEL);
1943         if (!devices_info)
1944                 return -ENOMEM;
1945
1946         /* calc min stripe number for data space allocation */
1947         type = btrfs_data_alloc_profile(fs_info);
1948         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1949                 min_stripes = 2;
1950                 num_stripes = nr_devices;
1951         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1952                 min_stripes = 2;
1953                 num_stripes = 2;
1954         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1955                 min_stripes = 4;
1956                 num_stripes = 4;
1957         }
1958
1959         if (type & BTRFS_BLOCK_GROUP_DUP)
1960                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1961         else
1962                 min_stripe_size = BTRFS_STRIPE_LEN;
1963
1964         rcu_read_lock();
1965         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1966                 if (!device->in_fs_metadata || !device->bdev ||
1967                     device->is_tgtdev_for_dev_replace)
1968                         continue;
1969
1970                 if (i >= nr_devices)
1971                         break;
1972
1973                 avail_space = device->total_bytes - device->bytes_used;
1974
1975                 /* align with stripe_len */
1976                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1977                 avail_space *= BTRFS_STRIPE_LEN;
1978
1979                 /*
1980                  * In order to avoid overwriting the superblock on the drive,
1981                  * btrfs starts at an offset of at least 1MB when doing chunk
1982                  * allocation.
1983                  */
1984                 skip_space = SZ_1M;
1985
1986                 /*
1987                  * we can use the free space in [0, skip_space - 1], subtract
1988                  * it from the total.
1989                  */
1990                 if (avail_space && avail_space >= skip_space)
1991                         avail_space -= skip_space;
1992                 else
1993                         avail_space = 0;
1994
1995                 if (avail_space < min_stripe_size)
1996                         continue;
1997
1998                 devices_info[i].dev = device;
1999                 devices_info[i].max_avail = avail_space;
2000
2001                 i++;
2002         }
2003         rcu_read_unlock();
2004
2005         nr_devices = i;
2006
2007         btrfs_descending_sort_devices(devices_info, nr_devices);
2008
2009         i = nr_devices - 1;
2010         avail_space = 0;
2011         while (nr_devices >= min_stripes) {
2012                 if (num_stripes > nr_devices)
2013                         num_stripes = nr_devices;
2014
2015                 if (devices_info[i].max_avail >= min_stripe_size) {
2016                         int j;
2017                         u64 alloc_size;
2018
2019                         avail_space += devices_info[i].max_avail * num_stripes;
2020                         alloc_size = devices_info[i].max_avail;
2021                         for (j = i + 1 - num_stripes; j <= i; j++)
2022                                 devices_info[j].max_avail -= alloc_size;
2023                 }
2024                 i--;
2025                 nr_devices--;
2026         }
2027
2028         kfree(devices_info);
2029         *free_bytes = avail_space;
2030         return 0;
2031 }
2032
2033 /*
2034  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2035  *
2036  * If there's a redundant raid level at DATA block groups, use the respective
2037  * multiplier to scale the sizes.
2038  *
2039  * Unused device space usage is based on simulating the chunk allocator
2040  * algorithm that respects the device sizes and order of allocations.  This is
2041  * a close approximation of the actual use but there are other factors that may
2042  * change the result (like a new metadata chunk).
2043  *
2044  * If metadata is exhausted, f_bavail will be 0.
2045  */
2046 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2047 {
2048         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2049         struct btrfs_super_block *disk_super = fs_info->super_copy;
2050         struct list_head *head = &fs_info->space_info;
2051         struct btrfs_space_info *found;
2052         u64 total_used = 0;
2053         u64 total_free_data = 0;
2054         u64 total_free_meta = 0;
2055         int bits = dentry->d_sb->s_blocksize_bits;
2056         __be32 *fsid = (__be32 *)fs_info->fsid;
2057         unsigned factor = 1;
2058         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2059         int ret;
2060         u64 thresh = 0;
2061         int mixed = 0;
2062
2063         rcu_read_lock();
2064         list_for_each_entry_rcu(found, head, list) {
2065                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2066                         int i;
2067
2068                         total_free_data += found->disk_total - found->disk_used;
2069                         total_free_data -=
2070                                 btrfs_account_ro_block_groups_free_space(found);
2071
2072                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2073                                 if (!list_empty(&found->block_groups[i])) {
2074                                         switch (i) {
2075                                         case BTRFS_RAID_DUP:
2076                                         case BTRFS_RAID_RAID1:
2077                                         case BTRFS_RAID_RAID10:
2078                                                 factor = 2;
2079                                         }
2080                                 }
2081                         }
2082                 }
2083
2084                 /*
2085                  * Metadata in mixed block goup profiles are accounted in data
2086                  */
2087                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2088                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2089                                 mixed = 1;
2090                         else
2091                                 total_free_meta += found->disk_total -
2092                                         found->disk_used;
2093                 }
2094
2095                 total_used += found->disk_used;
2096         }
2097
2098         rcu_read_unlock();
2099
2100         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2101         buf->f_blocks >>= bits;
2102         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2103
2104         /* Account global block reserve as used, it's in logical size already */
2105         spin_lock(&block_rsv->lock);
2106         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2107         if (buf->f_bfree >= block_rsv->size >> bits)
2108                 buf->f_bfree -= block_rsv->size >> bits;
2109         else
2110                 buf->f_bfree = 0;
2111         spin_unlock(&block_rsv->lock);
2112
2113         buf->f_bavail = div_u64(total_free_data, factor);
2114         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2115         if (ret)
2116                 return ret;
2117         buf->f_bavail += div_u64(total_free_data, factor);
2118         buf->f_bavail = buf->f_bavail >> bits;
2119
2120         /*
2121          * We calculate the remaining metadata space minus global reserve. If
2122          * this is (supposedly) smaller than zero, there's no space. But this
2123          * does not hold in practice, the exhausted state happens where's still
2124          * some positive delta. So we apply some guesswork and compare the
2125          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2126          *
2127          * We probably cannot calculate the exact threshold value because this
2128          * depends on the internal reservations requested by various
2129          * operations, so some operations that consume a few metadata will
2130          * succeed even if the Avail is zero. But this is better than the other
2131          * way around.
2132          */
2133         thresh = SZ_4M;
2134
2135         if (!mixed && total_free_meta - thresh < block_rsv->size)
2136                 buf->f_bavail = 0;
2137
2138         buf->f_type = BTRFS_SUPER_MAGIC;
2139         buf->f_bsize = dentry->d_sb->s_blocksize;
2140         buf->f_namelen = BTRFS_NAME_LEN;
2141
2142         /* We treat it as constant endianness (it doesn't matter _which_)
2143            because we want the fsid to come out the same whether mounted
2144            on a big-endian or little-endian host */
2145         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2146         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2147         /* Mask in the root object ID too, to disambiguate subvols */
2148         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2149         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2150
2151         return 0;
2152 }
2153
2154 static void btrfs_kill_super(struct super_block *sb)
2155 {
2156         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2157         kill_anon_super(sb);
2158         free_fs_info(fs_info);
2159 }
2160
2161 static struct file_system_type btrfs_fs_type = {
2162         .owner          = THIS_MODULE,
2163         .name           = "btrfs",
2164         .mount          = btrfs_mount,
2165         .kill_sb        = btrfs_kill_super,
2166         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2167 };
2168 MODULE_ALIAS_FS("btrfs");
2169
2170 static int btrfs_control_open(struct inode *inode, struct file *file)
2171 {
2172         /*
2173          * The control file's private_data is used to hold the
2174          * transaction when it is started and is used to keep
2175          * track of whether a transaction is already in progress.
2176          */
2177         file->private_data = NULL;
2178         return 0;
2179 }
2180
2181 /*
2182  * used by btrfsctl to scan devices when no FS is mounted
2183  */
2184 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2185                                 unsigned long arg)
2186 {
2187         struct btrfs_ioctl_vol_args *vol;
2188         struct btrfs_fs_devices *fs_devices;
2189         int ret = -ENOTTY;
2190
2191         if (!capable(CAP_SYS_ADMIN))
2192                 return -EPERM;
2193
2194         vol = memdup_user((void __user *)arg, sizeof(*vol));
2195         if (IS_ERR(vol))
2196                 return PTR_ERR(vol);
2197
2198         switch (cmd) {
2199         case BTRFS_IOC_SCAN_DEV:
2200                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2201                                             &btrfs_fs_type, &fs_devices);
2202                 break;
2203         case BTRFS_IOC_DEVICES_READY:
2204                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2205                                             &btrfs_fs_type, &fs_devices);
2206                 if (ret)
2207                         break;
2208                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2209                 break;
2210         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2211                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2212                 break;
2213         }
2214
2215         kfree(vol);
2216         return ret;
2217 }
2218
2219 static int btrfs_freeze(struct super_block *sb)
2220 {
2221         struct btrfs_trans_handle *trans;
2222         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2223         struct btrfs_root *root = fs_info->tree_root;
2224
2225         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2226         /*
2227          * We don't need a barrier here, we'll wait for any transaction that
2228          * could be in progress on other threads (and do delayed iputs that
2229          * we want to avoid on a frozen filesystem), or do the commit
2230          * ourselves.
2231          */
2232         trans = btrfs_attach_transaction_barrier(root);
2233         if (IS_ERR(trans)) {
2234                 /* no transaction, don't bother */
2235                 if (PTR_ERR(trans) == -ENOENT)
2236                         return 0;
2237                 return PTR_ERR(trans);
2238         }
2239         return btrfs_commit_transaction(trans);
2240 }
2241
2242 static int btrfs_unfreeze(struct super_block *sb)
2243 {
2244         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2245
2246         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2247         return 0;
2248 }
2249
2250 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2251 {
2252         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2253         struct btrfs_fs_devices *cur_devices;
2254         struct btrfs_device *dev, *first_dev = NULL;
2255         struct list_head *head;
2256         struct rcu_string *name;
2257
2258         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2259         cur_devices = fs_info->fs_devices;
2260         while (cur_devices) {
2261                 head = &cur_devices->devices;
2262                 list_for_each_entry(dev, head, dev_list) {
2263                         if (dev->missing)
2264                                 continue;
2265                         if (!dev->name)
2266                                 continue;
2267                         if (!first_dev || dev->devid < first_dev->devid)
2268                                 first_dev = dev;
2269                 }
2270                 cur_devices = cur_devices->seed;
2271         }
2272
2273         if (first_dev) {
2274                 rcu_read_lock();
2275                 name = rcu_dereference(first_dev->name);
2276                 seq_escape(m, name->str, " \t\n\\");
2277                 rcu_read_unlock();
2278         } else {
2279                 WARN_ON(1);
2280         }
2281         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2282         return 0;
2283 }
2284
2285 static const struct super_operations btrfs_super_ops = {
2286         .drop_inode     = btrfs_drop_inode,
2287         .evict_inode    = btrfs_evict_inode,
2288         .put_super      = btrfs_put_super,
2289         .sync_fs        = btrfs_sync_fs,
2290         .show_options   = btrfs_show_options,
2291         .show_devname   = btrfs_show_devname,
2292         .write_inode    = btrfs_write_inode,
2293         .alloc_inode    = btrfs_alloc_inode,
2294         .destroy_inode  = btrfs_destroy_inode,
2295         .statfs         = btrfs_statfs,
2296         .remount_fs     = btrfs_remount,
2297         .freeze_fs      = btrfs_freeze,
2298         .unfreeze_fs    = btrfs_unfreeze,
2299 };
2300
2301 static const struct file_operations btrfs_ctl_fops = {
2302         .open = btrfs_control_open,
2303         .unlocked_ioctl  = btrfs_control_ioctl,
2304         .compat_ioctl = btrfs_control_ioctl,
2305         .owner   = THIS_MODULE,
2306         .llseek = noop_llseek,
2307 };
2308
2309 static struct miscdevice btrfs_misc = {
2310         .minor          = BTRFS_MINOR,
2311         .name           = "btrfs-control",
2312         .fops           = &btrfs_ctl_fops
2313 };
2314
2315 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2316 MODULE_ALIAS("devname:btrfs-control");
2317
2318 static int btrfs_interface_init(void)
2319 {
2320         return misc_register(&btrfs_misc);
2321 }
2322
2323 static void btrfs_interface_exit(void)
2324 {
2325         misc_deregister(&btrfs_misc);
2326 }
2327
2328 static void btrfs_print_mod_info(void)
2329 {
2330         pr_info("Btrfs loaded, crc32c=%s"
2331 #ifdef CONFIG_BTRFS_DEBUG
2332                         ", debug=on"
2333 #endif
2334 #ifdef CONFIG_BTRFS_ASSERT
2335                         ", assert=on"
2336 #endif
2337 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2338                         ", integrity-checker=on"
2339 #endif
2340 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2341                         ", ref-verify=on"
2342 #endif
2343                         "\n",
2344                         btrfs_crc32c_impl());
2345 }
2346
2347 static int __init init_btrfs_fs(void)
2348 {
2349         int err;
2350
2351         err = btrfs_hash_init();
2352         if (err)
2353                 return err;
2354
2355         btrfs_props_init();
2356
2357         err = btrfs_init_sysfs();
2358         if (err)
2359                 goto free_hash;
2360
2361         btrfs_init_compress();
2362
2363         err = btrfs_init_cachep();
2364         if (err)
2365                 goto free_compress;
2366
2367         err = extent_io_init();
2368         if (err)
2369                 goto free_cachep;
2370
2371         err = extent_map_init();
2372         if (err)
2373                 goto free_extent_io;
2374
2375         err = ordered_data_init();
2376         if (err)
2377                 goto free_extent_map;
2378
2379         err = btrfs_delayed_inode_init();
2380         if (err)
2381                 goto free_ordered_data;
2382
2383         err = btrfs_auto_defrag_init();
2384         if (err)
2385                 goto free_delayed_inode;
2386
2387         err = btrfs_delayed_ref_init();
2388         if (err)
2389                 goto free_auto_defrag;
2390
2391         err = btrfs_prelim_ref_init();
2392         if (err)
2393                 goto free_delayed_ref;
2394
2395         err = btrfs_end_io_wq_init();
2396         if (err)
2397                 goto free_prelim_ref;
2398
2399         err = btrfs_interface_init();
2400         if (err)
2401                 goto free_end_io_wq;
2402
2403         btrfs_init_lockdep();
2404
2405         btrfs_print_mod_info();
2406
2407         err = btrfs_run_sanity_tests();
2408         if (err)
2409                 goto unregister_ioctl;
2410
2411         err = register_filesystem(&btrfs_fs_type);
2412         if (err)
2413                 goto unregister_ioctl;
2414
2415         return 0;
2416
2417 unregister_ioctl:
2418         btrfs_interface_exit();
2419 free_end_io_wq:
2420         btrfs_end_io_wq_exit();
2421 free_prelim_ref:
2422         btrfs_prelim_ref_exit();
2423 free_delayed_ref:
2424         btrfs_delayed_ref_exit();
2425 free_auto_defrag:
2426         btrfs_auto_defrag_exit();
2427 free_delayed_inode:
2428         btrfs_delayed_inode_exit();
2429 free_ordered_data:
2430         ordered_data_exit();
2431 free_extent_map:
2432         extent_map_exit();
2433 free_extent_io:
2434         extent_io_exit();
2435 free_cachep:
2436         btrfs_destroy_cachep();
2437 free_compress:
2438         btrfs_exit_compress();
2439         btrfs_exit_sysfs();
2440 free_hash:
2441         btrfs_hash_exit();
2442         return err;
2443 }
2444
2445 static void __exit exit_btrfs_fs(void)
2446 {
2447         btrfs_destroy_cachep();
2448         btrfs_delayed_ref_exit();
2449         btrfs_auto_defrag_exit();
2450         btrfs_delayed_inode_exit();
2451         btrfs_prelim_ref_exit();
2452         ordered_data_exit();
2453         extent_map_exit();
2454         extent_io_exit();
2455         btrfs_interface_exit();
2456         btrfs_end_io_wq_exit();
2457         unregister_filesystem(&btrfs_fs_type);
2458         btrfs_exit_sysfs();
2459         btrfs_cleanup_fs_uuids();
2460         btrfs_exit_compress();
2461         btrfs_hash_exit();
2462 }
2463
2464 late_initcall(init_btrfs_fs);
2465 module_exit(exit_btrfs_fs)
2466
2467 MODULE_LICENSE("GPL");