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 #define CREATE_TRACE_POINTS
  65 #include <trace/events/btrfs.h>
  66
  67 static const struct super_operations btrfs_super_ops;
  68 static struct file_system_type btrfs_fs_type;
  69
  70 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  71
  72 const char *btrfs_decode_error(int errno)
  73 {
  74         char *errstr = "unknown";
  75
  76         switch (errno) {
  77         case -EIO:
  78                 errstr = "IO failure";
  79                 break;
  80         case -ENOMEM:
  81                 errstr = "Out of memory";
  82                 break;
  83         case -EROFS:
  84                 errstr = "Readonly filesystem";
  85                 break;
  86         case -EEXIST:
  87                 errstr = "Object already exists";
  88                 break;
  89         case -ENOSPC:
  90                 errstr = "No space left";
  91                 break;
  92         case -ENOENT:
  93                 errstr = "No such entry";
  94                 break;
  95         }
  96
  97         return errstr;
  98 }
  99
 100 /* btrfs handle error by forcing the filesystem readonly */
 101 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 102 {
 103         struct super_block *sb = fs_info->sb;
 104
 105         if (sb->s_flags & MS_RDONLY)
 106                 return;
 107
 108         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 109                 sb->s_flags |= MS_RDONLY;
 110                 btrfs_info(fs_info, "forced readonly");
 111                 /*
 112                  * Note that a running device replace operation is not
 113                  * canceled here although there is no way to update
 114                  * the progress. It would add the risk of a deadlock,
 115                  * therefore the canceling is omitted. The only penalty
 116                  * is that some I/O remains active until the procedure
 117                  * completes. The next time when the filesystem is
 118                  * mounted writeable again, the device replace
 119                  * operation continues.
 120                  */
 121         }
 122 }
 123
 124 /*
 125  * __btrfs_handle_fs_error decodes expected errors from the caller and
 126  * invokes the approciate error response.
 127  */
 128 __cold
 129 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
 130                        unsigned int line, int errno, const char *fmt, ...)
 131 {
 132         struct super_block *sb = fs_info->sb;
 133 #ifdef CONFIG_PRINTK
 134         const char *errstr;
 135 #endif
 136
 137         /*
 138          * Special case: if the error is EROFS, and we're already
 139          * under MS_RDONLY, then it is safe here.
 140          */
 141         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 142                 return;
 143
 144 #ifdef CONFIG_PRINTK
 145         errstr = btrfs_decode_error(errno);
 146         if (fmt) {
 147                 struct va_format vaf;
 148                 va_list args;
 149
 150                 va_start(args, fmt);
 151                 vaf.fmt = fmt;
 152                 vaf.va = &args;
 153
 154                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 155                         sb->s_id, function, line, errno, errstr, &vaf);
 156                 va_end(args);
 157         } else {
 158                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 159                         sb->s_id, function, line, errno, errstr);
 160         }
 161 #endif
 162
 163         /*
 164          * Today we only save the error info to memory.  Long term we'll
 165          * also send it down to the disk
 166          */
 167         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 168
 169         /* Don't go through full error handling during mount */
 170         if (sb->s_flags & MS_BORN)
 171                 btrfs_handle_error(fs_info);
 172 }
 173
 174 #ifdef CONFIG_PRINTK
 175 static const char * const logtypes[] = {
 176         "emergency",
 177         "alert",
 178         "critical",
 179         "error",
 180         "warning",
 181         "notice",
 182         "info",
 183         "debug",
 184 };
 185
 186
 187 /*
 188  * Use one ratelimit state per log level so that a flood of less important
 189  * messages doesn't cause more important ones to be dropped.
 190  */
 191 static struct ratelimit_state printk_limits[] = {
 192         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
 193         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
 194         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
 195         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
 196         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
 197         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
 198         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
 199         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
 200 };
 201
 202 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 203 {
 204         struct super_block *sb = fs_info->sb;
 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, sb->s_id, &vaf);
 231
 232         va_end(args);
 233 }
 234 #endif
 235
 236 /*
 237  * We only mark the transaction aborted and then set the file system read-only.
 238  * This will prevent new transactions from starting or trying to join this
 239  * one.
 240  *
 241  * This means that error recovery at the call site is limited to freeing
 242  * any local memory allocations and passing the error code up without
 243  * further cleanup. The transaction should complete as it normally would
 244  * in the call path but will return -EIO.
 245  *
 246  * We'll complete the cleanup in btrfs_end_transaction and
 247  * btrfs_commit_transaction.
 248  */
 249 __cold
 250 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 251                                const char *function,
 252                                unsigned int line, int errno)
 253 {
 254         struct btrfs_fs_info *fs_info = trans->fs_info;
 255
 256         trans->aborted = errno;
 257         /* Nothing used. The other threads that have joined this
 258          * transaction may be able to continue. */
 259         if (!trans->dirty && list_empty(&trans->new_bgs)) {
 260                 const char *errstr;
 261
 262                 errstr = btrfs_decode_error(errno);
 263                 btrfs_warn(fs_info,
 264                            "%s:%d: Aborting unused transaction(%s).",
 265                            function, line, errstr);
 266                 return;
 267         }
 268         WRITE_ONCE(trans->transaction->aborted, errno);
 269         /* Wake up anybody who may be waiting on this transaction */
 270         wake_up(&fs_info->transaction_wait);
 271         wake_up(&fs_info->transaction_blocked_wait);
 272         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
 273 }
 274 /*
 275  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 276  * issues an alert, and either panics or BUGs, depending on mount options.
 277  */
 278 __cold
 279 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 280                    unsigned int line, int errno, const char *fmt, ...)
 281 {
 282         char *s_id = "<unknown>";
 283         const char *errstr;
 284         struct va_format vaf = { .fmt = fmt };
 285         va_list args;
 286
 287         if (fs_info)
 288                 s_id = fs_info->sb->s_id;
 289
 290         va_start(args, fmt);
 291         vaf.va = &args;
 292
 293         errstr = btrfs_decode_error(errno);
 294         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 295                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 296                         s_id, function, line, &vaf, errno, errstr);
 297
 298         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 299                    function, line, &vaf, errno, errstr);
 300         va_end(args);
 301         /* Caller calls BUG() */
 302 }
 303
 304 static void btrfs_put_super(struct super_block *sb)
 305 {
 306         close_ctree(btrfs_sb(sb));
 307 }
 308
 309 enum {
 310         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 311         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 312         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 313         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 314         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 315         Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
 316         Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
 317         Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
 318         Opt_skip_balance, Opt_check_integrity,
 319         Opt_check_integrity_including_extent_data,
 320         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 321         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
 322         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
 323         Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
 324         Opt_nologreplay, Opt_norecovery,
 325 #ifdef CONFIG_BTRFS_DEBUG
 326         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 327 #endif
 328         Opt_err,
 329 };
 330
 331 static const match_table_t tokens = {
 332         {Opt_degraded, "degraded"},
 333         {Opt_subvol, "subvol=%s"},
 334         {Opt_subvolid, "subvolid=%s"},
 335         {Opt_device, "device=%s"},
 336         {Opt_nodatasum, "nodatasum"},
 337         {Opt_datasum, "datasum"},
 338         {Opt_nodatacow, "nodatacow"},
 339         {Opt_datacow, "datacow"},
 340         {Opt_nobarrier, "nobarrier"},
 341         {Opt_barrier, "barrier"},
 342         {Opt_max_inline, "max_inline=%s"},
 343         {Opt_alloc_start, "alloc_start=%s"},
 344         {Opt_thread_pool, "thread_pool=%d"},
 345         {Opt_compress, "compress"},
 346         {Opt_compress_type, "compress=%s"},
 347         {Opt_compress_force, "compress-force"},
 348         {Opt_compress_force_type, "compress-force=%s"},
 349         {Opt_ssd, "ssd"},
 350         {Opt_ssd_spread, "ssd_spread"},
 351         {Opt_nossd, "nossd"},
 352         {Opt_acl, "acl"},
 353         {Opt_noacl, "noacl"},
 354         {Opt_notreelog, "notreelog"},
 355         {Opt_treelog, "treelog"},
 356         {Opt_nologreplay, "nologreplay"},
 357         {Opt_norecovery, "norecovery"},
 358         {Opt_flushoncommit, "flushoncommit"},
 359         {Opt_noflushoncommit, "noflushoncommit"},
 360         {Opt_ratio, "metadata_ratio=%d"},
 361         {Opt_discard, "discard"},
 362         {Opt_nodiscard, "nodiscard"},
 363         {Opt_space_cache, "space_cache"},
 364         {Opt_space_cache_version, "space_cache=%s"},
 365         {Opt_clear_cache, "clear_cache"},
 366         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 367         {Opt_enospc_debug, "enospc_debug"},
 368         {Opt_noenospc_debug, "noenospc_debug"},
 369         {Opt_subvolrootid, "subvolrootid=%d"},
 370         {Opt_defrag, "autodefrag"},
 371         {Opt_nodefrag, "noautodefrag"},
 372         {Opt_inode_cache, "inode_cache"},
 373         {Opt_noinode_cache, "noinode_cache"},
 374         {Opt_no_space_cache, "nospace_cache"},
 375         {Opt_recovery, "recovery"}, /* deprecated */
 376         {Opt_usebackuproot, "usebackuproot"},
 377         {Opt_skip_balance, "skip_balance"},
 378         {Opt_check_integrity, "check_int"},
 379         {Opt_check_integrity_including_extent_data, "check_int_data"},
 380         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 381         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 382         {Opt_fatal_errors, "fatal_errors=%s"},
 383         {Opt_commit_interval, "commit=%d"},
 384 #ifdef CONFIG_BTRFS_DEBUG
 385         {Opt_fragment_data, "fragment=data"},
 386         {Opt_fragment_metadata, "fragment=metadata"},
 387         {Opt_fragment_all, "fragment=all"},
 388 #endif
 389         {Opt_err, NULL},
 390 };
 391
 392 /*
 393  * Regular mount options parser.  Everything that is needed only when
 394  * reading in a new superblock is parsed here.
 395  * XXX JDM: This needs to be cleaned up for remount.
 396  */
 397 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 398                         unsigned long new_flags)
 399 {
 400         substring_t args[MAX_OPT_ARGS];
 401         char *p, *num, *orig = NULL;
 402         u64 cache_gen;
 403         int intarg;
 404         int ret = 0;
 405         char *compress_type;
 406         bool compress_force = false;
 407         enum btrfs_compression_type saved_compress_type;
 408         bool saved_compress_force;
 409         int no_compress = 0;
 410
 411         cache_gen = btrfs_super_cache_generation(info->super_copy);
 412         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 413                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 414         else if (cache_gen)
 415                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 416
 417         /*
 418          * Even the options are empty, we still need to do extra check
 419          * against new flags
 420          */
 421         if (!options)
 422                 goto check;
 423
 424         /*
 425          * strsep changes the string, duplicate it because parse_options
 426          * gets called twice
 427          */
 428         options = kstrdup(options, GFP_NOFS);
 429         if (!options)
 430                 return -ENOMEM;
 431
 432         orig = options;
 433
 434         while ((p = strsep(&options, ",")) != NULL) {
 435                 int token;
 436                 if (!*p)
 437                         continue;
 438
 439                 token = match_token(p, tokens, args);
 440                 switch (token) {
 441                 case Opt_degraded:
 442                         btrfs_info(info, "allowing degraded mounts");
 443                         btrfs_set_opt(info->mount_opt, DEGRADED);
 444                         break;
 445                 case Opt_subvol:
 446                 case Opt_subvolid:
 447                 case Opt_subvolrootid:
 448                 case Opt_device:
 449                         /*
 450                          * These are parsed by btrfs_parse_early_options
 451                          * and can be happily ignored here.
 452                          */
 453                         break;
 454                 case Opt_nodatasum:
 455                         btrfs_set_and_info(info, NODATASUM,
 456                                            "setting nodatasum");
 457                         break;
 458                 case Opt_datasum:
 459                         if (btrfs_test_opt(info, NODATASUM)) {
 460                                 if (btrfs_test_opt(info, NODATACOW))
 461                                         btrfs_info(info,
 462                                                    "setting datasum, datacow enabled");
 463                                 else
 464                                         btrfs_info(info, "setting datasum");
 465                         }
 466                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 467                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 468                         break;
 469                 case Opt_nodatacow:
 470                         if (!btrfs_test_opt(info, NODATACOW)) {
 471                                 if (!btrfs_test_opt(info, COMPRESS) ||
 472                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 473                                         btrfs_info(info,
 474                                                    "setting nodatacow, compression disabled");
 475                                 } else {
 476                                         btrfs_info(info, "setting nodatacow");
 477                                 }
 478                         }
 479                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 480                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 481                         btrfs_set_opt(info->mount_opt, NODATACOW);
 482                         btrfs_set_opt(info->mount_opt, NODATASUM);
 483                         break;
 484                 case Opt_datacow:
 485                         btrfs_clear_and_info(info, NODATACOW,
 486                                              "setting datacow");
 487                         break;
 488                 case Opt_compress_force:
 489                 case Opt_compress_force_type:
 490                         compress_force = true;
 491                         /* Fallthrough */
 492                 case Opt_compress:
 493                 case Opt_compress_type:
 494                         saved_compress_type = btrfs_test_opt(info,
 495                                                              COMPRESS) ?
 496                                 info->compress_type : BTRFS_COMPRESS_NONE;
 497                         saved_compress_force =
 498                                 btrfs_test_opt(info, FORCE_COMPRESS);
 499                         if (token == Opt_compress ||
 500                             token == Opt_compress_force ||
 501                             strcmp(args[0].from, "zlib") == 0) {
 502                                 compress_type = "zlib";
 503                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 504                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 505                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 506                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 507                                 no_compress = 0;
 508                         } else if (strcmp(args[0].from, "lzo") == 0) {
 509                                 compress_type = "lzo";
 510                                 info->compress_type = BTRFS_COMPRESS_LZO;
 511                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 512                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 513                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 514                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 515                                 no_compress = 0;
 516                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 517                                 compress_type = "no";
 518                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 519                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 520                                 compress_force = false;
 521                                 no_compress++;
 522                         } else {
 523                                 ret = -EINVAL;
 524                                 goto out;
 525                         }
 526
 527                         if (compress_force) {
 528                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 529                         } else {
 530                                 /*
 531                                  * If we remount from compress-force=xxx to
 532                                  * compress=xxx, we need clear FORCE_COMPRESS
 533                                  * flag, otherwise, there is no way for users
 534                                  * to disable forcible compression separately.
 535                                  */
 536                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 537                         }
 538                         if ((btrfs_test_opt(info, COMPRESS) &&
 539                              (info->compress_type != saved_compress_type ||
 540                               compress_force != saved_compress_force)) ||
 541                             (!btrfs_test_opt(info, COMPRESS) &&
 542                              no_compress == 1)) {
 543                                 btrfs_info(info, "%s %s compression",
 544                                            (compress_force) ? "force" : "use",
 545                                            compress_type);
 546                         }
 547                         compress_force = false;
 548                         break;
 549                 case Opt_ssd:
 550                         btrfs_set_and_info(info, SSD,
 551                                            "use ssd allocation scheme");
 552                         break;
 553                 case Opt_ssd_spread:
 554                         btrfs_set_and_info(info, SSD_SPREAD,
 555                                            "use spread ssd allocation scheme");
 556                         btrfs_set_opt(info->mount_opt, SSD);
 557                         break;
 558                 case Opt_nossd:
 559                         btrfs_set_and_info(info, NOSSD,
 560                                              "not using ssd allocation scheme");
 561                         btrfs_clear_opt(info->mount_opt, SSD);
 562                         break;
 563                 case Opt_barrier:
 564                         btrfs_clear_and_info(info, NOBARRIER,
 565                                              "turning on barriers");
 566                         break;
 567                 case Opt_nobarrier:
 568                         btrfs_set_and_info(info, NOBARRIER,
 569                                            "turning off barriers");
 570                         break;
 571                 case Opt_thread_pool:
 572                         ret = match_int(&args[0], &intarg);
 573                         if (ret) {
 574                                 goto out;
 575                         } else if (intarg > 0) {
 576                                 info->thread_pool_size = intarg;
 577                         } else {
 578                                 ret = -EINVAL;
 579                                 goto out;
 580                         }
 581                         break;
 582                 case Opt_max_inline:
 583                         num = match_strdup(&args[0]);
 584                         if (num) {
 585                                 info->max_inline = memparse(num, NULL);
 586                                 kfree(num);
 587
 588                                 if (info->max_inline) {
 589                                         info->max_inline = min_t(u64,
 590                                                 info->max_inline,
 591                                                 info->sectorsize);
 592                                 }
 593                                 btrfs_info(info, "max_inline at %llu",
 594                                            info->max_inline);
 595                         } else {
 596                                 ret = -ENOMEM;
 597                                 goto out;
 598                         }
 599                         break;
 600                 case Opt_alloc_start:
 601                         num = match_strdup(&args[0]);
 602                         if (num) {
 603                                 mutex_lock(&info->chunk_mutex);
 604                                 info->alloc_start = memparse(num, NULL);
 605                                 mutex_unlock(&info->chunk_mutex);
 606                                 kfree(num);
 607                                 btrfs_info(info, "allocations start at %llu",
 608                                            info->alloc_start);
 609                         } else {
 610                                 ret = -ENOMEM;
 611                                 goto out;
 612                         }
 613                         break;
 614                 case Opt_acl:
 615 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 616                         info->sb->s_flags |= MS_POSIXACL;
 617                         break;
 618 #else
 619                         btrfs_err(info, "support for ACL not compiled in!");
 620                         ret = -EINVAL;
 621                         goto out;
 622 #endif
 623                 case Opt_noacl:
 624                         info->sb->s_flags &= ~MS_POSIXACL;
 625                         break;
 626                 case Opt_notreelog:
 627                         btrfs_set_and_info(info, NOTREELOG,
 628                                            "disabling tree log");
 629                         break;
 630                 case Opt_treelog:
 631                         btrfs_clear_and_info(info, NOTREELOG,
 632                                              "enabling tree log");
 633                         break;
 634                 case Opt_norecovery:
 635                 case Opt_nologreplay:
 636                         btrfs_set_and_info(info, NOLOGREPLAY,
 637                                            "disabling log replay at mount time");
 638                         break;
 639                 case Opt_flushoncommit:
 640                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 641                                            "turning on flush-on-commit");
 642                         break;
 643                 case Opt_noflushoncommit:
 644                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 645                                              "turning off flush-on-commit");
 646                         break;
 647                 case Opt_ratio:
 648                         ret = match_int(&args[0], &intarg);
 649                         if (ret) {
 650                                 goto out;
 651                         } else if (intarg >= 0) {
 652                                 info->metadata_ratio = intarg;
 653                                 btrfs_info(info, "metadata ratio %d",
 654                                            info->metadata_ratio);
 655                         } else {
 656                                 ret = -EINVAL;
 657                                 goto out;
 658                         }
 659                         break;
 660                 case Opt_discard:
 661                         btrfs_set_and_info(info, DISCARD,
 662                                            "turning on discard");
 663                         break;
 664                 case Opt_nodiscard:
 665                         btrfs_clear_and_info(info, DISCARD,
 666                                              "turning off discard");
 667                         break;
 668                 case Opt_space_cache:
 669                 case Opt_space_cache_version:
 670                         if (token == Opt_space_cache ||
 671                             strcmp(args[0].from, "v1") == 0) {
 672                                 btrfs_clear_opt(info->mount_opt,
 673                                                 FREE_SPACE_TREE);
 674                                 btrfs_set_and_info(info, SPACE_CACHE,
 675                                            "enabling disk space caching");
 676                         } else if (strcmp(args[0].from, "v2") == 0) {
 677                                 btrfs_clear_opt(info->mount_opt,
 678                                                 SPACE_CACHE);
 679                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 680                                                    "enabling free space tree");
 681                         } else {
 682                                 ret = -EINVAL;
 683                                 goto out;
 684                         }
 685                         break;
 686                 case Opt_rescan_uuid_tree:
 687                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 688                         break;
 689                 case Opt_no_space_cache:
 690                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 691                                 btrfs_clear_and_info(info, SPACE_CACHE,
 692                                              "disabling disk space caching");
 693                         }
 694                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 695                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 696                                              "disabling free space tree");
 697                         }
 698                         break;
 699                 case Opt_inode_cache:
 700                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 701                                            "enabling inode map caching");
 702                         break;
 703                 case Opt_noinode_cache:
 704                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 705                                              "disabling inode map caching");
 706                         break;
 707                 case Opt_clear_cache:
 708                         btrfs_set_and_info(info, CLEAR_CACHE,
 709                                            "force clearing of disk cache");
 710                         break;
 711                 case Opt_user_subvol_rm_allowed:
 712                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 713                         break;
 714                 case Opt_enospc_debug:
 715                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 716                         break;
 717                 case Opt_noenospc_debug:
 718                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 719                         break;
 720                 case Opt_defrag:
 721                         btrfs_set_and_info(info, AUTO_DEFRAG,
 722                                            "enabling auto defrag");
 723                         break;
 724                 case Opt_nodefrag:
 725                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 726                                              "disabling auto defrag");
 727                         break;
 728                 case Opt_recovery:
 729                         btrfs_warn(info,
 730                                    "'recovery' is deprecated, use 'usebackuproot' instead");
 731                 case Opt_usebackuproot:
 732                         btrfs_info(info,
 733                                    "trying to use backup root at mount time");
 734                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 735                         break;
 736                 case Opt_skip_balance:
 737                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 738                         break;
 739 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 740                 case Opt_check_integrity_including_extent_data:
 741                         btrfs_info(info,
 742                                    "enabling check integrity including extent data");
 743                         btrfs_set_opt(info->mount_opt,
 744                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 745                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 746                         break;
 747                 case Opt_check_integrity:
 748                         btrfs_info(info, "enabling check integrity");
 749                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 750                         break;
 751                 case Opt_check_integrity_print_mask:
 752                         ret = match_int(&args[0], &intarg);
 753                         if (ret) {
 754                                 goto out;
 755                         } else if (intarg >= 0) {
 756                                 info->check_integrity_print_mask = intarg;
 757                                 btrfs_info(info,
 758                                            "check_integrity_print_mask 0x%x",
 759                                            info->check_integrity_print_mask);
 760                         } else {
 761                                 ret = -EINVAL;
 762                                 goto out;
 763                         }
 764                         break;
 765 #else
 766                 case Opt_check_integrity_including_extent_data:
 767                 case Opt_check_integrity:
 768                 case Opt_check_integrity_print_mask:
 769                         btrfs_err(info,
 770                                   "support for check_integrity* not compiled in!");
 771                         ret = -EINVAL;
 772                         goto out;
 773 #endif
 774                 case Opt_fatal_errors:
 775                         if (strcmp(args[0].from, "panic") == 0)
 776                                 btrfs_set_opt(info->mount_opt,
 777                                               PANIC_ON_FATAL_ERROR);
 778                         else if (strcmp(args[0].from, "bug") == 0)
 779                                 btrfs_clear_opt(info->mount_opt,
 780                                               PANIC_ON_FATAL_ERROR);
 781                         else {
 782                                 ret = -EINVAL;
 783                                 goto out;
 784                         }
 785                         break;
 786                 case Opt_commit_interval:
 787                         intarg = 0;
 788                         ret = match_int(&args[0], &intarg);
 789                         if (ret < 0) {
 790                                 btrfs_err(info, "invalid commit interval");
 791                                 ret = -EINVAL;
 792                                 goto out;
 793                         }
 794                         if (intarg > 0) {
 795                                 if (intarg > 300) {
 796                                         btrfs_warn(info,
 797                                                 "excessive commit interval %d",
 798                                                 intarg);
 799                                 }
 800                                 info->commit_interval = intarg;
 801                         } else {
 802                                 btrfs_info(info,
 803                                            "using default commit interval %ds",
 804                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
 805                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 806                         }
 807                         break;
 808 #ifdef CONFIG_BTRFS_DEBUG
 809                 case Opt_fragment_all:
 810                         btrfs_info(info, "fragmenting all space");
 811                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 812                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 813                         break;
 814                 case Opt_fragment_metadata:
 815                         btrfs_info(info, "fragmenting metadata");
 816                         btrfs_set_opt(info->mount_opt,
 817                                       FRAGMENT_METADATA);
 818                         break;
 819                 case Opt_fragment_data:
 820                         btrfs_info(info, "fragmenting data");
 821                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 822                         break;
 823 #endif
 824                 case Opt_err:
 825                         btrfs_info(info, "unrecognized mount option '%s'", p);
 826                         ret = -EINVAL;
 827                         goto out;
 828                 default:
 829                         break;
 830                 }
 831         }
 832 check:
 833         /*
 834          * Extra check for current option against current flag
 835          */
 836         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
 837                 btrfs_err(info,
 838                           "nologreplay must be used with ro mount option");
 839                 ret = -EINVAL;
 840         }
 841 out:
 842         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
 843             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
 844             !btrfs_test_opt(info, CLEAR_CACHE)) {
 845                 btrfs_err(info, "cannot disable free space tree");
 846                 ret = -EINVAL;
 847
 848         }
 849         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
 850                 btrfs_info(info, "disk space caching is enabled");
 851         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
 852                 btrfs_info(info, "using free space tree");
 853         kfree(orig);
 854         return ret;
 855 }
 856
 857 /*
 858  * Parse mount options that are required early in the mount process.
 859  *
 860  * All other options will be parsed on much later in the mount process and
 861  * only when we need to allocate a new super block.
 862  */
 863 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 864                 void *holder, char **subvol_name, u64 *subvol_objectid,
 865                 struct btrfs_fs_devices **fs_devices)
 866 {
 867         substring_t args[MAX_OPT_ARGS];
 868         char *device_name, *opts, *orig, *p;
 869         char *num = NULL;
 870         int error = 0;
 871
 872         if (!options)
 873                 return 0;
 874
 875         /*
 876          * strsep changes the string, duplicate it because parse_options
 877          * gets called twice
 878          */
 879         opts = kstrdup(options, GFP_KERNEL);
 880         if (!opts)
 881                 return -ENOMEM;
 882         orig = opts;
 883
 884         while ((p = strsep(&opts, ",")) != NULL) {
 885                 int token;
 886                 if (!*p)
 887                         continue;
 888
 889                 token = match_token(p, tokens, args);
 890                 switch (token) {
 891                 case Opt_subvol:
 892                         kfree(*subvol_name);
 893                         *subvol_name = match_strdup(&args[0]);
 894                         if (!*subvol_name) {
 895                                 error = -ENOMEM;
 896                                 goto out;
 897                         }
 898                         break;
 899                 case Opt_subvolid:
 900                         num = match_strdup(&args[0]);
 901                         if (num) {
 902                                 *subvol_objectid = memparse(num, NULL);
 903                                 kfree(num);
 904                                 /* we want the original fs_tree */
 905                                 if (!*subvol_objectid)
 906                                         *subvol_objectid =
 907                                                 BTRFS_FS_TREE_OBJECTID;
 908                         } else {
 909                                 error = -EINVAL;
 910                                 goto out;
 911                         }
 912                         break;
 913                 case Opt_subvolrootid:
 914                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
 915                         break;
 916                 case Opt_device:
 917                         device_name = match_strdup(&args[0]);
 918                         if (!device_name) {
 919                                 error = -ENOMEM;
 920                                 goto out;
 921                         }
 922                         error = btrfs_scan_one_device(device_name,
 923                                         flags, holder, fs_devices);
 924                         kfree(device_name);
 925                         if (error)
 926                                 goto out;
 927                         break;
 928                 default:
 929                         break;
 930                 }
 931         }
 932
 933 out:
 934         kfree(orig);
 935         return error;
 936 }
 937
 938 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 939                                            u64 subvol_objectid)
 940 {
 941         struct btrfs_root *root = fs_info->tree_root;
 942         struct btrfs_root *fs_root;
 943         struct btrfs_root_ref *root_ref;
 944         struct btrfs_inode_ref *inode_ref;
 945         struct btrfs_key key;
 946         struct btrfs_path *path = NULL;
 947         char *name = NULL, *ptr;
 948         u64 dirid;
 949         int len;
 950         int ret;
 951
 952         path = btrfs_alloc_path();
 953         if (!path) {
 954                 ret = -ENOMEM;
 955                 goto err;
 956         }
 957         path->leave_spinning = 1;
 958
 959         name = kmalloc(PATH_MAX, GFP_NOFS);
 960         if (!name) {
 961                 ret = -ENOMEM;
 962                 goto err;
 963         }
 964         ptr = name + PATH_MAX - 1;
 965         ptr[0] = '\0';
 966
 967         /*
 968          * Walk up the subvolume trees in the tree of tree roots by root
 969          * backrefs until we hit the top-level subvolume.
 970          */
 971         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 972                 key.objectid = subvol_objectid;
 973                 key.type = BTRFS_ROOT_BACKREF_KEY;
 974                 key.offset = (u64)-1;
 975
 976                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 977                 if (ret < 0) {
 978                         goto err;
 979                 } else if (ret > 0) {
 980                         ret = btrfs_previous_item(root, path, subvol_objectid,
 981                                                   BTRFS_ROOT_BACKREF_KEY);
 982                         if (ret < 0) {
 983                                 goto err;
 984                         } else if (ret > 0) {
 985                                 ret = -ENOENT;
 986                                 goto err;
 987                         }
 988                 }
 989
 990                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 991                 subvol_objectid = key.offset;
 992
 993                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 994                                           struct btrfs_root_ref);
 995                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 996                 ptr -= len + 1;
 997                 if (ptr < name) {
 998                         ret = -ENAMETOOLONG;
 999                         goto err;
1000                 }
1001                 read_extent_buffer(path->nodes[0], ptr + 1,
1002                                    (unsigned long)(root_ref + 1), len);
1003                 ptr[0] = '/';
1004                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1005                 btrfs_release_path(path);
1006
1007                 key.objectid = subvol_objectid;
1008                 key.type = BTRFS_ROOT_ITEM_KEY;
1009                 key.offset = (u64)-1;
1010                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1011                 if (IS_ERR(fs_root)) {
1012                         ret = PTR_ERR(fs_root);
1013                         goto err;
1014                 }
1015
1016                 /*
1017                  * Walk up the filesystem tree by inode refs until we hit the
1018                  * root directory.
1019                  */
1020                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1021                         key.objectid = dirid;
1022                         key.type = BTRFS_INODE_REF_KEY;
1023                         key.offset = (u64)-1;
1024
1025                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1026                         if (ret < 0) {
1027                                 goto err;
1028                         } else if (ret > 0) {
1029                                 ret = btrfs_previous_item(fs_root, path, dirid,
1030                                                           BTRFS_INODE_REF_KEY);
1031                                 if (ret < 0) {
1032                                         goto err;
1033                                 } else if (ret > 0) {
1034                                         ret = -ENOENT;
1035                                         goto err;
1036                                 }
1037                         }
1038
1039                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1040                         dirid = key.offset;
1041
1042                         inode_ref = btrfs_item_ptr(path->nodes[0],
1043                                                    path->slots[0],
1044                                                    struct btrfs_inode_ref);
1045                         len = btrfs_inode_ref_name_len(path->nodes[0],
1046                                                        inode_ref);
1047                         ptr -= len + 1;
1048                         if (ptr < name) {
1049                                 ret = -ENAMETOOLONG;
1050                                 goto err;
1051                         }
1052                         read_extent_buffer(path->nodes[0], ptr + 1,
1053                                            (unsigned long)(inode_ref + 1), len);
1054                         ptr[0] = '/';
1055                         btrfs_release_path(path);
1056                 }
1057         }
1058
1059         btrfs_free_path(path);
1060         if (ptr == name + PATH_MAX - 1) {
1061                 name[0] = '/';
1062                 name[1] = '\0';
1063         } else {
1064                 memmove(name, ptr, name + PATH_MAX - ptr);
1065         }
1066         return name;
1067
1068 err:
1069         btrfs_free_path(path);
1070         kfree(name);
1071         return ERR_PTR(ret);
1072 }
1073
1074 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1075 {
1076         struct btrfs_root *root = fs_info->tree_root;
1077         struct btrfs_dir_item *di;
1078         struct btrfs_path *path;
1079         struct btrfs_key location;
1080         u64 dir_id;
1081
1082         path = btrfs_alloc_path();
1083         if (!path)
1084                 return -ENOMEM;
1085         path->leave_spinning = 1;
1086
1087         /*
1088          * Find the "default" dir item which points to the root item that we
1089          * will mount by default if we haven't been given a specific subvolume
1090          * to mount.
1091          */
1092         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1093         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1094         if (IS_ERR(di)) {
1095                 btrfs_free_path(path);
1096                 return PTR_ERR(di);
1097         }
1098         if (!di) {
1099                 /*
1100                  * Ok the default dir item isn't there.  This is weird since
1101                  * it's always been there, but don't freak out, just try and
1102                  * mount the top-level subvolume.
1103                  */
1104                 btrfs_free_path(path);
1105                 *objectid = BTRFS_FS_TREE_OBJECTID;
1106                 return 0;
1107         }
1108
1109         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1110         btrfs_free_path(path);
1111         *objectid = location.objectid;
1112         return 0;
1113 }
1114
1115 static int btrfs_fill_super(struct super_block *sb,
1116                             struct btrfs_fs_devices *fs_devices,
1117                             void *data)
1118 {
1119         struct inode *inode;
1120         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1121         struct btrfs_key key;
1122         int err;
1123
1124         sb->s_maxbytes = MAX_LFS_FILESIZE;
1125         sb->s_magic = BTRFS_SUPER_MAGIC;
1126         sb->s_op = &btrfs_super_ops;
1127         sb->s_d_op = &btrfs_dentry_operations;
1128         sb->s_export_op = &btrfs_export_ops;
1129         sb->s_xattr = btrfs_xattr_handlers;
1130         sb->s_time_gran = 1;
1131 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1132         sb->s_flags |= MS_POSIXACL;
1133 #endif
1134         sb->s_flags |= MS_I_VERSION;
1135         sb->s_iflags |= SB_I_CGROUPWB;
1136         err = open_ctree(sb, fs_devices, (char *)data);
1137         if (err) {
1138                 btrfs_err(fs_info, "open_ctree failed");
1139                 return err;
1140         }
1141
1142         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1143         key.type = BTRFS_INODE_ITEM_KEY;
1144         key.offset = 0;
1145         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1146         if (IS_ERR(inode)) {
1147                 err = PTR_ERR(inode);
1148                 goto fail_close;
1149         }
1150
1151         sb->s_root = d_make_root(inode);
1152         if (!sb->s_root) {
1153                 err = -ENOMEM;
1154                 goto fail_close;
1155         }
1156
1157         save_mount_options(sb, data);
1158         cleancache_init_fs(sb);
1159         sb->s_flags |= MS_ACTIVE;
1160         return 0;
1161
1162 fail_close:
1163         close_ctree(fs_info);
1164         return err;
1165 }
1166
1167 int btrfs_sync_fs(struct super_block *sb, int wait)
1168 {
1169         struct btrfs_trans_handle *trans;
1170         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1171         struct btrfs_root *root = fs_info->tree_root;
1172
1173         trace_btrfs_sync_fs(fs_info, wait);
1174
1175         if (!wait) {
1176                 filemap_flush(fs_info->btree_inode->i_mapping);
1177                 return 0;
1178         }
1179
1180         btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
1181
1182         trans = btrfs_attach_transaction_barrier(root);
1183         if (IS_ERR(trans)) {
1184                 /* no transaction, don't bother */
1185                 if (PTR_ERR(trans) == -ENOENT) {
1186                         /*
1187                          * Exit unless we have some pending changes
1188                          * that need to go through commit
1189                          */
1190                         if (fs_info->pending_changes == 0)
1191                                 return 0;
1192                         /*
1193                          * A non-blocking test if the fs is frozen. We must not
1194                          * start a new transaction here otherwise a deadlock
1195                          * happens. The pending operations are delayed to the
1196                          * next commit after thawing.
1197                          */
1198                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1199                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1200                         else
1201                                 return 0;
1202                         trans = btrfs_start_transaction(root, 0);
1203                 }
1204                 if (IS_ERR(trans))
1205                         return PTR_ERR(trans);
1206         }
1207         return btrfs_commit_transaction(trans);
1208 }
1209
1210 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1211 {
1212         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1213         char *compress_type;
1214
1215         if (btrfs_test_opt(info, DEGRADED))
1216                 seq_puts(seq, ",degraded");
1217         if (btrfs_test_opt(info, NODATASUM))
1218                 seq_puts(seq, ",nodatasum");
1219         if (btrfs_test_opt(info, NODATACOW))
1220                 seq_puts(seq, ",nodatacow");
1221         if (btrfs_test_opt(info, NOBARRIER))
1222                 seq_puts(seq, ",nobarrier");
1223         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1224                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1225         if (info->alloc_start != 0)
1226                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1227         if (info->thread_pool_size !=  min_t(unsigned long,
1228                                              num_online_cpus() + 2, 8))
1229                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1230         if (btrfs_test_opt(info, COMPRESS)) {
1231                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1232                         compress_type = "zlib";
1233                 else
1234                         compress_type = "lzo";
1235                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1236                         seq_printf(seq, ",compress-force=%s", compress_type);
1237                 else
1238                         seq_printf(seq, ",compress=%s", compress_type);
1239         }
1240         if (btrfs_test_opt(info, NOSSD))
1241                 seq_puts(seq, ",nossd");
1242         if (btrfs_test_opt(info, SSD_SPREAD))
1243                 seq_puts(seq, ",ssd_spread");
1244         else if (btrfs_test_opt(info, SSD))
1245                 seq_puts(seq, ",ssd");
1246         if (btrfs_test_opt(info, NOTREELOG))
1247                 seq_puts(seq, ",notreelog");
1248         if (btrfs_test_opt(info, NOLOGREPLAY))
1249                 seq_puts(seq, ",nologreplay");
1250         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1251                 seq_puts(seq, ",flushoncommit");
1252         if (btrfs_test_opt(info, DISCARD))
1253                 seq_puts(seq, ",discard");
1254         if (!(info->sb->s_flags & MS_POSIXACL))
1255                 seq_puts(seq, ",noacl");
1256         if (btrfs_test_opt(info, SPACE_CACHE))
1257                 seq_puts(seq, ",space_cache");
1258         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1259                 seq_puts(seq, ",space_cache=v2");
1260         else
1261                 seq_puts(seq, ",nospace_cache");
1262         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1263                 seq_puts(seq, ",rescan_uuid_tree");
1264         if (btrfs_test_opt(info, CLEAR_CACHE))
1265                 seq_puts(seq, ",clear_cache");
1266         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1267                 seq_puts(seq, ",user_subvol_rm_allowed");
1268         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1269                 seq_puts(seq, ",enospc_debug");
1270         if (btrfs_test_opt(info, AUTO_DEFRAG))
1271                 seq_puts(seq, ",autodefrag");
1272         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1273                 seq_puts(seq, ",inode_cache");
1274         if (btrfs_test_opt(info, SKIP_BALANCE))
1275                 seq_puts(seq, ",skip_balance");
1276 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1277         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1278                 seq_puts(seq, ",check_int_data");
1279         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1280                 seq_puts(seq, ",check_int");
1281         if (info->check_integrity_print_mask)
1282                 seq_printf(seq, ",check_int_print_mask=%d",
1283                                 info->check_integrity_print_mask);
1284 #endif
1285         if (info->metadata_ratio)
1286                 seq_printf(seq, ",metadata_ratio=%d",
1287                                 info->metadata_ratio);
1288         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1289                 seq_puts(seq, ",fatal_errors=panic");
1290         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1291                 seq_printf(seq, ",commit=%d", info->commit_interval);
1292 #ifdef CONFIG_BTRFS_DEBUG
1293         if (btrfs_test_opt(info, FRAGMENT_DATA))
1294                 seq_puts(seq, ",fragment=data");
1295         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1296                 seq_puts(seq, ",fragment=metadata");
1297 #endif
1298         seq_printf(seq, ",subvolid=%llu",
1299                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1300         seq_puts(seq, ",subvol=");
1301         seq_dentry(seq, dentry, " \t\n\\");
1302         return 0;
1303 }
1304
1305 static int btrfs_test_super(struct super_block *s, void *data)
1306 {
1307         struct btrfs_fs_info *p = data;
1308         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1309
1310         return fs_info->fs_devices == p->fs_devices;
1311 }
1312
1313 static int btrfs_set_super(struct super_block *s, void *data)
1314 {
1315         int err = set_anon_super(s, data);
1316         if (!err)
1317                 s->s_fs_info = data;
1318         return err;
1319 }
1320
1321 /*
1322  * subvolumes are identified by ino 256
1323  */
1324 static inline int is_subvolume_inode(struct inode *inode)
1325 {
1326         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1327                 return 1;
1328         return 0;
1329 }
1330
1331 /*
1332  * This will add subvolid=0 to the argument string while removing any subvol=
1333  * and subvolid= arguments to make sure we get the top-level root for path
1334  * walking to the subvol we want.
1335  */
1336 static char *setup_root_args(char *args)
1337 {
1338         char *buf, *dst, *sep;
1339
1340         if (!args)
1341                 return kstrdup("subvolid=0", GFP_NOFS);
1342
1343         /* The worst case is that we add ",subvolid=0" to the end. */
1344         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1345         if (!buf)
1346                 return NULL;
1347
1348         while (1) {
1349                 sep = strchrnul(args, ',');
1350                 if (!strstarts(args, "subvol=") &&
1351                     !strstarts(args, "subvolid=")) {
1352                         memcpy(dst, args, sep - args);
1353                         dst += sep - args;
1354                         *dst++ = ',';
1355                 }
1356                 if (*sep)
1357                         args = sep + 1;
1358                 else
1359                         break;
1360         }
1361         strcpy(dst, "subvolid=0");
1362
1363         return buf;
1364 }
1365
1366 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1367                                    int flags, const char *device_name,
1368                                    char *data)
1369 {
1370         struct dentry *root;
1371         struct vfsmount *mnt = NULL;
1372         char *newargs;
1373         int ret;
1374
1375         newargs = setup_root_args(data);
1376         if (!newargs) {
1377                 root = ERR_PTR(-ENOMEM);
1378                 goto out;
1379         }
1380
1381         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1382         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1383                 if (flags & MS_RDONLY) {
1384                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1385                                              device_name, newargs);
1386                 } else {
1387                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1388                                              device_name, newargs);
1389                         if (IS_ERR(mnt)) {
1390                                 root = ERR_CAST(mnt);
1391                                 mnt = NULL;
1392                                 goto out;
1393                         }
1394
1395                         down_write(&mnt->mnt_sb->s_umount);
1396                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1397                         up_write(&mnt->mnt_sb->s_umount);
1398                         if (ret < 0) {
1399                                 root = ERR_PTR(ret);
1400                                 goto out;
1401                         }
1402                 }
1403         }
1404         if (IS_ERR(mnt)) {
1405                 root = ERR_CAST(mnt);
1406                 mnt = NULL;
1407                 goto out;
1408         }
1409
1410         if (!subvol_name) {
1411                 if (!subvol_objectid) {
1412                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1413                                                           &subvol_objectid);
1414                         if (ret) {
1415                                 root = ERR_PTR(ret);
1416                                 goto out;
1417                         }
1418                 }
1419                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1420                                                             subvol_objectid);
1421                 if (IS_ERR(subvol_name)) {
1422                         root = ERR_CAST(subvol_name);
1423                         subvol_name = NULL;
1424                         goto out;
1425                 }
1426
1427         }
1428
1429         root = mount_subtree(mnt, subvol_name);
1430         /* mount_subtree() drops our reference on the vfsmount. */
1431         mnt = NULL;
1432
1433         if (!IS_ERR(root)) {
1434                 struct super_block *s = root->d_sb;
1435                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1436                 struct inode *root_inode = d_inode(root);
1437                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1438
1439                 ret = 0;
1440                 if (!is_subvolume_inode(root_inode)) {
1441                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1442                                subvol_name);
1443                         ret = -EINVAL;
1444                 }
1445                 if (subvol_objectid && root_objectid != subvol_objectid) {
1446                         /*
1447                          * This will also catch a race condition where a
1448                          * subvolume which was passed by ID is renamed and
1449                          * another subvolume is renamed over the old location.
1450                          */
1451                         btrfs_err(fs_info,
1452                                   "subvol '%s' does not match subvolid %llu",
1453                                   subvol_name, subvol_objectid);
1454                         ret = -EINVAL;
1455                 }
1456                 if (ret) {
1457                         dput(root);
1458                         root = ERR_PTR(ret);
1459                         deactivate_locked_super(s);
1460                 }
1461         }
1462
1463 out:
1464         mntput(mnt);
1465         kfree(newargs);
1466         kfree(subvol_name);
1467         return root;
1468 }
1469
1470 static int parse_security_options(char *orig_opts,
1471                                   struct security_mnt_opts *sec_opts)
1472 {
1473         char *secdata = NULL;
1474         int ret = 0;
1475
1476         secdata = alloc_secdata();
1477         if (!secdata)
1478                 return -ENOMEM;
1479         ret = security_sb_copy_data(orig_opts, secdata);
1480         if (ret) {
1481                 free_secdata(secdata);
1482                 return ret;
1483         }
1484         ret = security_sb_parse_opts_str(secdata, sec_opts);
1485         free_secdata(secdata);
1486         return ret;
1487 }
1488
1489 static int setup_security_options(struct btrfs_fs_info *fs_info,
1490                                   struct super_block *sb,
1491                                   struct security_mnt_opts *sec_opts)
1492 {
1493         int ret = 0;
1494
1495         /*
1496          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1497          * is valid.
1498          */
1499         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1500         if (ret)
1501                 return ret;
1502
1503 #ifdef CONFIG_SECURITY
1504         if (!fs_info->security_opts.num_mnt_opts) {
1505                 /* first time security setup, copy sec_opts to fs_info */
1506                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1507         } else {
1508                 /*
1509                  * Since SELinux (the only one supporting security_mnt_opts)
1510                  * does NOT support changing context during remount/mount of
1511                  * the same sb, this must be the same or part of the same
1512                  * security options, just free it.
1513                  */
1514                 security_free_mnt_opts(sec_opts);
1515         }
1516 #endif
1517         return ret;
1518 }
1519
1520 /*
1521  * Find a superblock for the given device / mount point.
1522  *
1523  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1524  *        for multiple device setup.  Make sure to keep it in sync.
1525  */
1526 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1527                 const char *device_name, void *data)
1528 {
1529         struct block_device *bdev = NULL;
1530         struct super_block *s;
1531         struct btrfs_fs_devices *fs_devices = NULL;
1532         struct btrfs_fs_info *fs_info = NULL;
1533         struct security_mnt_opts new_sec_opts;
1534         fmode_t mode = FMODE_READ;
1535         char *subvol_name = NULL;
1536         u64 subvol_objectid = 0;
1537         int error = 0;
1538
1539         if (!(flags & MS_RDONLY))
1540                 mode |= FMODE_WRITE;
1541
1542         error = btrfs_parse_early_options(data, mode, fs_type,
1543                                           &subvol_name, &subvol_objectid,
1544                                           &fs_devices);
1545         if (error) {
1546                 kfree(subvol_name);
1547                 return ERR_PTR(error);
1548         }
1549
1550         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1551                 /* mount_subvol() will free subvol_name. */
1552                 return mount_subvol(subvol_name, subvol_objectid, flags,
1553                                     device_name, data);
1554         }
1555
1556         security_init_mnt_opts(&new_sec_opts);
1557         if (data) {
1558                 error = parse_security_options(data, &new_sec_opts);
1559                 if (error)
1560                         return ERR_PTR(error);
1561         }
1562
1563         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1564         if (error)
1565                 goto error_sec_opts;
1566
1567         /*
1568          * Setup a dummy root and fs_info for test/set super.  This is because
1569          * we don't actually fill this stuff out until open_ctree, but we need
1570          * it for searching for existing supers, so this lets us do that and
1571          * then open_ctree will properly initialize everything later.
1572          */
1573         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1574         if (!fs_info) {
1575                 error = -ENOMEM;
1576                 goto error_sec_opts;
1577         }
1578
1579         fs_info->fs_devices = fs_devices;
1580
1581         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1582         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1583         security_init_mnt_opts(&fs_info->security_opts);
1584         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1585                 error = -ENOMEM;
1586                 goto error_fs_info;
1587         }
1588
1589         error = btrfs_open_devices(fs_devices, mode, fs_type);
1590         if (error)
1591                 goto error_fs_info;
1592
1593         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1594                 error = -EACCES;
1595                 goto error_close_devices;
1596         }
1597
1598         bdev = fs_devices->latest_bdev;
1599         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1600                  fs_info);
1601         if (IS_ERR(s)) {
1602                 error = PTR_ERR(s);
1603                 goto error_close_devices;
1604         }
1605
1606         if (s->s_root) {
1607                 btrfs_close_devices(fs_devices);
1608                 free_fs_info(fs_info);
1609                 if ((flags ^ s->s_flags) & MS_RDONLY)
1610                         error = -EBUSY;
1611         } else {
1612                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1613                 btrfs_sb(s)->bdev_holder = fs_type;
1614                 error = btrfs_fill_super(s, fs_devices, data);
1615         }
1616         if (error) {
1617                 deactivate_locked_super(s);
1618                 goto error_sec_opts;
1619         }
1620
1621         fs_info = btrfs_sb(s);
1622         error = setup_security_options(fs_info, s, &new_sec_opts);
1623         if (error) {
1624                 deactivate_locked_super(s);
1625                 goto error_sec_opts;
1626         }
1627
1628         return dget(s->s_root);
1629
1630 error_close_devices:
1631         btrfs_close_devices(fs_devices);
1632 error_fs_info:
1633         free_fs_info(fs_info);
1634 error_sec_opts:
1635         security_free_mnt_opts(&new_sec_opts);
1636         return ERR_PTR(error);
1637 }
1638
1639 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1640                                      int new_pool_size, int old_pool_size)
1641 {
1642         if (new_pool_size == old_pool_size)
1643                 return;
1644
1645         fs_info->thread_pool_size = new_pool_size;
1646
1647         btrfs_info(fs_info, "resize thread pool %d -> %d",
1648                old_pool_size, new_pool_size);
1649
1650         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1651         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1652         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1653         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1654         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1655         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1656         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1657                                 new_pool_size);
1658         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1659         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1660         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1661         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1662         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1663                                 new_pool_size);
1664 }
1665
1666 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1667 {
1668         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1669 }
1670
1671 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1672                                        unsigned long old_opts, int flags)
1673 {
1674         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1675             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1676              (flags & MS_RDONLY))) {
1677                 /* wait for any defraggers to finish */
1678                 wait_event(fs_info->transaction_wait,
1679                            (atomic_read(&fs_info->defrag_running) == 0));
1680                 if (flags & MS_RDONLY)
1681                         sync_filesystem(fs_info->sb);
1682         }
1683 }
1684
1685 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1686                                          unsigned long old_opts)
1687 {
1688         /*
1689          * We need to cleanup all defragable inodes if the autodefragment is
1690          * close or the filesystem is read only.
1691          */
1692         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1693             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1694              (fs_info->sb->s_flags & MS_RDONLY))) {
1695                 btrfs_cleanup_defrag_inodes(fs_info);
1696         }
1697
1698         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1699 }
1700
1701 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1702 {
1703         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1704         struct btrfs_root *root = fs_info->tree_root;
1705         unsigned old_flags = sb->s_flags;
1706         unsigned long old_opts = fs_info->mount_opt;
1707         unsigned long old_compress_type = fs_info->compress_type;
1708         u64 old_max_inline = fs_info->max_inline;
1709         u64 old_alloc_start = fs_info->alloc_start;
1710         int old_thread_pool_size = fs_info->thread_pool_size;
1711         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1712         int ret;
1713
1714         sync_filesystem(sb);
1715         btrfs_remount_prepare(fs_info);
1716
1717         if (data) {
1718                 struct security_mnt_opts new_sec_opts;
1719
1720                 security_init_mnt_opts(&new_sec_opts);
1721                 ret = parse_security_options(data, &new_sec_opts);
1722                 if (ret)
1723                         goto restore;
1724                 ret = setup_security_options(fs_info, sb,
1725                                              &new_sec_opts);
1726                 if (ret) {
1727                         security_free_mnt_opts(&new_sec_opts);
1728                         goto restore;
1729                 }
1730         }
1731
1732         ret = btrfs_parse_options(fs_info, data, *flags);
1733         if (ret) {
1734                 ret = -EINVAL;
1735                 goto restore;
1736         }
1737
1738         btrfs_remount_begin(fs_info, old_opts, *flags);
1739         btrfs_resize_thread_pool(fs_info,
1740                 fs_info->thread_pool_size, old_thread_pool_size);
1741
1742         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1743                 goto out;
1744
1745         if (*flags & MS_RDONLY) {
1746                 /*
1747                  * this also happens on 'umount -rf' or on shutdown, when
1748                  * the filesystem is busy.
1749                  */
1750                 cancel_work_sync(&fs_info->async_reclaim_work);
1751
1752                 /* wait for the uuid_scan task to finish */
1753                 down(&fs_info->uuid_tree_rescan_sem);
1754                 /* avoid complains from lockdep et al. */
1755                 up(&fs_info->uuid_tree_rescan_sem);
1756
1757                 sb->s_flags |= MS_RDONLY;
1758
1759                 /*
1760                  * Setting MS_RDONLY will put the cleaner thread to
1761                  * sleep at the next loop if it's already active.
1762                  * If it's already asleep, we'll leave unused block
1763                  * groups on disk until we're mounted read-write again
1764                  * unless we clean them up here.
1765                  */
1766                 btrfs_delete_unused_bgs(fs_info);
1767
1768                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1769                 btrfs_scrub_cancel(fs_info);
1770                 btrfs_pause_balance(fs_info);
1771
1772                 ret = btrfs_commit_super(fs_info);
1773                 if (ret)
1774                         goto restore;
1775         } else {
1776                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1777                         btrfs_err(fs_info,
1778                                 "Remounting read-write after error is not allowed");
1779                         ret = -EINVAL;
1780                         goto restore;
1781                 }
1782                 if (fs_info->fs_devices->rw_devices == 0) {
1783                         ret = -EACCES;
1784                         goto restore;
1785                 }
1786
1787                 if (fs_info->fs_devices->missing_devices >
1788                      fs_info->num_tolerated_disk_barrier_failures &&
1789                     !(*flags & MS_RDONLY)) {
1790                         btrfs_warn(fs_info,
1791                                 "too many missing devices, writeable remount is not allowed");
1792                         ret = -EACCES;
1793                         goto restore;
1794                 }
1795
1796                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1797                         ret = -EINVAL;
1798                         goto restore;
1799                 }
1800
1801                 ret = btrfs_cleanup_fs_roots(fs_info);
1802                 if (ret)
1803                         goto restore;
1804
1805                 /* recover relocation */
1806                 mutex_lock(&fs_info->cleaner_mutex);
1807                 ret = btrfs_recover_relocation(root);
1808                 mutex_unlock(&fs_info->cleaner_mutex);
1809                 if (ret)
1810                         goto restore;
1811
1812                 ret = btrfs_resume_balance_async(fs_info);
1813                 if (ret)
1814                         goto restore;
1815
1816                 ret = btrfs_resume_dev_replace_async(fs_info);
1817                 if (ret) {
1818                         btrfs_warn(fs_info, "failed to resume dev_replace");
1819                         goto restore;
1820                 }
1821
1822                 if (!fs_info->uuid_root) {
1823                         btrfs_info(fs_info, "creating UUID tree");
1824                         ret = btrfs_create_uuid_tree(fs_info);
1825                         if (ret) {
1826                                 btrfs_warn(fs_info,
1827                                            "failed to create the UUID tree %d",
1828                                            ret);
1829                                 goto restore;
1830                         }
1831                 }
1832                 sb->s_flags &= ~MS_RDONLY;
1833
1834                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1835         }
1836 out:
1837         wake_up_process(fs_info->transaction_kthread);
1838         btrfs_remount_cleanup(fs_info, old_opts);
1839         return 0;
1840
1841 restore:
1842         /* We've hit an error - don't reset MS_RDONLY */
1843         if (sb->s_flags & MS_RDONLY)
1844                 old_flags |= MS_RDONLY;
1845         sb->s_flags = old_flags;
1846         fs_info->mount_opt = old_opts;
1847         fs_info->compress_type = old_compress_type;
1848         fs_info->max_inline = old_max_inline;
1849         mutex_lock(&fs_info->chunk_mutex);
1850         fs_info->alloc_start = old_alloc_start;
1851         mutex_unlock(&fs_info->chunk_mutex);
1852         btrfs_resize_thread_pool(fs_info,
1853                 old_thread_pool_size, fs_info->thread_pool_size);
1854         fs_info->metadata_ratio = old_metadata_ratio;
1855         btrfs_remount_cleanup(fs_info, old_opts);
1856         return ret;
1857 }
1858
1859 /* Used to sort the devices by max_avail(descending sort) */
1860 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1861                                        const void *dev_info2)
1862 {
1863         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1864             ((struct btrfs_device_info *)dev_info2)->max_avail)
1865                 return -1;
1866         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1867                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1868                 return 1;
1869         else
1870         return 0;
1871 }
1872
1873 /*
1874  * sort the devices by max_avail, in which max free extent size of each device
1875  * is stored.(Descending Sort)
1876  */
1877 static inline void btrfs_descending_sort_devices(
1878                                         struct btrfs_device_info *devices,
1879                                         size_t nr_devices)
1880 {
1881         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1882              btrfs_cmp_device_free_bytes, NULL);
1883 }
1884
1885 /*
1886  * The helper to calc the free space on the devices that can be used to store
1887  * file data.
1888  */
1889 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1890                                        u64 *free_bytes)
1891 {
1892         struct btrfs_root *root = fs_info->tree_root;
1893         struct btrfs_device_info *devices_info;
1894         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1895         struct btrfs_device *device;
1896         u64 skip_space;
1897         u64 type;
1898         u64 avail_space;
1899         u64 used_space;
1900         u64 min_stripe_size;
1901         int min_stripes = 1, num_stripes = 1;
1902         int i = 0, nr_devices;
1903         int ret;
1904
1905         /*
1906          * We aren't under the device list lock, so this is racy-ish, but good
1907          * enough for our purposes.
1908          */
1909         nr_devices = fs_info->fs_devices->open_devices;
1910         if (!nr_devices) {
1911                 smp_mb();
1912                 nr_devices = fs_info->fs_devices->open_devices;
1913                 ASSERT(nr_devices);
1914                 if (!nr_devices) {
1915                         *free_bytes = 0;
1916                         return 0;
1917                 }
1918         }
1919
1920         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1921                                GFP_NOFS);
1922         if (!devices_info)
1923                 return -ENOMEM;
1924
1925         /* calc min stripe number for data space allocation */
1926         type = btrfs_get_alloc_profile(root, 1);
1927         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1928                 min_stripes = 2;
1929                 num_stripes = nr_devices;
1930         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1931                 min_stripes = 2;
1932                 num_stripes = 2;
1933         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1934                 min_stripes = 4;
1935                 num_stripes = 4;
1936         }
1937
1938         if (type & BTRFS_BLOCK_GROUP_DUP)
1939                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1940         else
1941                 min_stripe_size = BTRFS_STRIPE_LEN;
1942
1943         if (fs_info->alloc_start)
1944                 mutex_lock(&fs_devices->device_list_mutex);
1945         rcu_read_lock();
1946         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1947                 if (!device->in_fs_metadata || !device->bdev ||
1948                     device->is_tgtdev_for_dev_replace)
1949                         continue;
1950
1951                 if (i >= nr_devices)
1952                         break;
1953
1954                 avail_space = device->total_bytes - device->bytes_used;
1955
1956                 /* align with stripe_len */
1957                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1958                 avail_space *= BTRFS_STRIPE_LEN;
1959
1960                 /*
1961                  * In order to avoid overwriting the superblock on the drive,
1962                  * btrfs starts at an offset of at least 1MB when doing chunk
1963                  * allocation.
1964                  */
1965                 skip_space = SZ_1M;
1966
1967                 /* user can set the offset in fs_info->alloc_start. */
1968                 if (fs_info->alloc_start &&
1969                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1970                     device->total_bytes) {
1971                         rcu_read_unlock();
1972                         skip_space = max(fs_info->alloc_start, skip_space);
1973
1974                         /*
1975                          * btrfs can not use the free space in
1976                          * [0, skip_space - 1], we must subtract it from the
1977                          * total. In order to implement it, we account the used
1978                          * space in this range first.
1979                          */
1980                         ret = btrfs_account_dev_extents_size(device, 0,
1981                                                              skip_space - 1,
1982                                                              &used_space);
1983                         if (ret) {
1984                                 kfree(devices_info);
1985                                 mutex_unlock(&fs_devices->device_list_mutex);
1986                                 return ret;
1987                         }
1988
1989                         rcu_read_lock();
1990
1991                         /* calc the free space in [0, skip_space - 1] */
1992                         skip_space -= used_space;
1993                 }
1994
1995                 /*
1996                  * we can use the free space in [0, skip_space - 1], subtract
1997                  * it from the total.
1998                  */
1999                 if (avail_space && avail_space >= skip_space)
2000                         avail_space -= skip_space;
2001                 else
2002                         avail_space = 0;
2003
2004                 if (avail_space < min_stripe_size)
2005                         continue;
2006
2007                 devices_info[i].dev = device;
2008                 devices_info[i].max_avail = avail_space;
2009
2010                 i++;
2011         }
2012         rcu_read_unlock();
2013         if (fs_info->alloc_start)
2014                 mutex_unlock(&fs_devices->device_list_mutex);
2015
2016         nr_devices = i;
2017
2018         btrfs_descending_sort_devices(devices_info, nr_devices);
2019
2020         i = nr_devices - 1;
2021         avail_space = 0;
2022         while (nr_devices >= min_stripes) {
2023                 if (num_stripes > nr_devices)
2024                         num_stripes = nr_devices;
2025
2026                 if (devices_info[i].max_avail >= min_stripe_size) {
2027                         int j;
2028                         u64 alloc_size;
2029
2030                         avail_space += devices_info[i].max_avail * num_stripes;
2031                         alloc_size = devices_info[i].max_avail;
2032                         for (j = i + 1 - num_stripes; j <= i; j++)
2033                                 devices_info[j].max_avail -= alloc_size;
2034                 }
2035                 i--;
2036                 nr_devices--;
2037         }
2038
2039         kfree(devices_info);
2040         *free_bytes = avail_space;
2041         return 0;
2042 }
2043
2044 /*
2045  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2046  *
2047  * If there's a redundant raid level at DATA block groups, use the respective
2048  * multiplier to scale the sizes.
2049  *
2050  * Unused device space usage is based on simulating the chunk allocator
2051  * algorithm that respects the device sizes, order of allocations and the
2052  * 'alloc_start' value, this is a close approximation of the actual use but
2053  * there are other factors that may change the result (like a new metadata
2054  * chunk).
2055  *
2056  * If metadata is exhausted, f_bavail will be 0.
2057  */
2058 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2059 {
2060         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2061         struct btrfs_super_block *disk_super = fs_info->super_copy;
2062         struct list_head *head = &fs_info->space_info;
2063         struct btrfs_space_info *found;
2064         u64 total_used = 0;
2065         u64 total_free_data = 0;
2066         u64 total_free_meta = 0;
2067         int bits = dentry->d_sb->s_blocksize_bits;
2068         __be32 *fsid = (__be32 *)fs_info->fsid;
2069         unsigned factor = 1;
2070         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2071         int ret;
2072         u64 thresh = 0;
2073         int mixed = 0;
2074
2075         rcu_read_lock();
2076         list_for_each_entry_rcu(found, head, list) {
2077                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2078                         int i;
2079
2080                         total_free_data += found->disk_total - found->disk_used;
2081                         total_free_data -=
2082                                 btrfs_account_ro_block_groups_free_space(found);
2083
2084                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2085                                 if (!list_empty(&found->block_groups[i])) {
2086                                         switch (i) {
2087                                         case BTRFS_RAID_DUP:
2088                                         case BTRFS_RAID_RAID1:
2089                                         case BTRFS_RAID_RAID10:
2090                                                 factor = 2;
2091                                         }
2092                                 }
2093                         }
2094                 }
2095
2096                 /*
2097                  * Metadata in mixed block goup profiles are accounted in data
2098                  */
2099                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2100                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2101                                 mixed = 1;
2102                         else
2103                                 total_free_meta += found->disk_total -
2104                                         found->disk_used;
2105                 }
2106
2107                 total_used += found->disk_used;
2108         }
2109
2110         rcu_read_unlock();
2111
2112         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2113         buf->f_blocks >>= bits;
2114         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2115
2116         /* Account global block reserve as used, it's in logical size already */
2117         spin_lock(&block_rsv->lock);
2118         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2119         if (buf->f_bfree >= block_rsv->size >> bits)
2120                 buf->f_bfree -= block_rsv->size >> bits;
2121         else
2122                 buf->f_bfree = 0;
2123         spin_unlock(&block_rsv->lock);
2124
2125         buf->f_bavail = div_u64(total_free_data, factor);
2126         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2127         if (ret)
2128                 return ret;
2129         buf->f_bavail += div_u64(total_free_data, factor);
2130         buf->f_bavail = buf->f_bavail >> bits;
2131
2132         /*
2133          * We calculate the remaining metadata space minus global reserve. If
2134          * this is (supposedly) smaller than zero, there's no space. But this
2135          * does not hold in practice, the exhausted state happens where's still
2136          * some positive delta. So we apply some guesswork and compare the
2137          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2138          *
2139          * We probably cannot calculate the exact threshold value because this
2140          * depends on the internal reservations requested by various
2141          * operations, so some operations that consume a few metadata will
2142          * succeed even if the Avail is zero. But this is better than the other
2143          * way around.
2144          */
2145         thresh = 4 * 1024 * 1024;
2146
2147         if (!mixed && total_free_meta - thresh < block_rsv->size)
2148                 buf->f_bavail = 0;
2149
2150         buf->f_type = BTRFS_SUPER_MAGIC;
2151         buf->f_bsize = dentry->d_sb->s_blocksize;
2152         buf->f_namelen = BTRFS_NAME_LEN;
2153
2154         /* We treat it as constant endianness (it doesn't matter _which_)
2155            because we want the fsid to come out the same whether mounted
2156            on a big-endian or little-endian host */
2157         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2158         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2159         /* Mask in the root object ID too, to disambiguate subvols */
2160         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2161         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2162
2163         return 0;
2164 }
2165
2166 static void btrfs_kill_super(struct super_block *sb)
2167 {
2168         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2169         kill_anon_super(sb);
2170         free_fs_info(fs_info);
2171 }
2172
2173 static struct file_system_type btrfs_fs_type = {
2174         .owner          = THIS_MODULE,
2175         .name           = "btrfs",
2176         .mount          = btrfs_mount,
2177         .kill_sb        = btrfs_kill_super,
2178         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2179 };
2180 MODULE_ALIAS_FS("btrfs");
2181
2182 static int btrfs_control_open(struct inode *inode, struct file *file)
2183 {
2184         /*
2185          * The control file's private_data is used to hold the
2186          * transaction when it is started and is used to keep
2187          * track of whether a transaction is already in progress.
2188          */
2189         file->private_data = NULL;
2190         return 0;
2191 }
2192
2193 /*
2194  * used by btrfsctl to scan devices when no FS is mounted
2195  */
2196 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2197                                 unsigned long arg)
2198 {
2199         struct btrfs_ioctl_vol_args *vol;
2200         struct btrfs_fs_devices *fs_devices;
2201         int ret = -ENOTTY;
2202
2203         if (!capable(CAP_SYS_ADMIN))
2204                 return -EPERM;
2205
2206         vol = memdup_user((void __user *)arg, sizeof(*vol));
2207         if (IS_ERR(vol))
2208                 return PTR_ERR(vol);
2209
2210         switch (cmd) {
2211         case BTRFS_IOC_SCAN_DEV:
2212                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2213                                             &btrfs_fs_type, &fs_devices);
2214                 break;
2215         case BTRFS_IOC_DEVICES_READY:
2216                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2217                                             &btrfs_fs_type, &fs_devices);
2218                 if (ret)
2219                         break;
2220                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2221                 break;
2222         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2223                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2224                 break;
2225         }
2226
2227         kfree(vol);
2228         return ret;
2229 }
2230
2231 static int btrfs_freeze(struct super_block *sb)
2232 {
2233         struct btrfs_trans_handle *trans;
2234         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2235         struct btrfs_root *root = fs_info->tree_root;
2236
2237         fs_info->fs_frozen = 1;
2238         /*
2239          * We don't need a barrier here, we'll wait for any transaction that
2240          * could be in progress on other threads (and do delayed iputs that
2241          * we want to avoid on a frozen filesystem), or do the commit
2242          * ourselves.
2243          */
2244         trans = btrfs_attach_transaction_barrier(root);
2245         if (IS_ERR(trans)) {
2246                 /* no transaction, don't bother */
2247                 if (PTR_ERR(trans) == -ENOENT)
2248                         return 0;
2249                 return PTR_ERR(trans);
2250         }
2251         return btrfs_commit_transaction(trans);
2252 }
2253
2254 static int btrfs_unfreeze(struct super_block *sb)
2255 {
2256         btrfs_sb(sb)->fs_frozen = 0;
2257         return 0;
2258 }
2259
2260 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2261 {
2262         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2263         struct btrfs_fs_devices *cur_devices;
2264         struct btrfs_device *dev, *first_dev = NULL;
2265         struct list_head *head;
2266         struct rcu_string *name;
2267
2268         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2269         cur_devices = fs_info->fs_devices;
2270         while (cur_devices) {
2271                 head = &cur_devices->devices;
2272                 list_for_each_entry(dev, head, dev_list) {
2273                         if (dev->missing)
2274                                 continue;
2275                         if (!dev->name)
2276                                 continue;
2277                         if (!first_dev || dev->devid < first_dev->devid)
2278                                 first_dev = dev;
2279                 }
2280                 cur_devices = cur_devices->seed;
2281         }
2282
2283         if (first_dev) {
2284                 rcu_read_lock();
2285                 name = rcu_dereference(first_dev->name);
2286                 seq_escape(m, name->str, " \t\n\\");
2287                 rcu_read_unlock();
2288         } else {
2289                 WARN_ON(1);
2290         }
2291         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2292         return 0;
2293 }
2294
2295 static const struct super_operations btrfs_super_ops = {
2296         .drop_inode     = btrfs_drop_inode,
2297         .evict_inode    = btrfs_evict_inode,
2298         .put_super      = btrfs_put_super,
2299         .sync_fs        = btrfs_sync_fs,
2300         .show_options   = btrfs_show_options,
2301         .show_devname   = btrfs_show_devname,
2302         .write_inode    = btrfs_write_inode,
2303         .alloc_inode    = btrfs_alloc_inode,
2304         .destroy_inode  = btrfs_destroy_inode,
2305         .statfs         = btrfs_statfs,
2306         .remount_fs     = btrfs_remount,
2307         .freeze_fs      = btrfs_freeze,
2308         .unfreeze_fs    = btrfs_unfreeze,
2309 };
2310
2311 static const struct file_operations btrfs_ctl_fops = {
2312         .open = btrfs_control_open,
2313         .unlocked_ioctl  = btrfs_control_ioctl,
2314         .compat_ioctl = btrfs_control_ioctl,
2315         .owner   = THIS_MODULE,
2316         .llseek = noop_llseek,
2317 };
2318
2319 static struct miscdevice btrfs_misc = {
2320         .minor          = BTRFS_MINOR,
2321         .name           = "btrfs-control",
2322         .fops           = &btrfs_ctl_fops
2323 };
2324
2325 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2326 MODULE_ALIAS("devname:btrfs-control");
2327
2328 static int btrfs_interface_init(void)
2329 {
2330         return misc_register(&btrfs_misc);
2331 }
2332
2333 static void btrfs_interface_exit(void)
2334 {
2335         misc_deregister(&btrfs_misc);
2336 }
2337
2338 static void btrfs_print_mod_info(void)
2339 {
2340         pr_info("Btrfs loaded, crc32c=%s"
2341 #ifdef CONFIG_BTRFS_DEBUG
2342                         ", debug=on"
2343 #endif
2344 #ifdef CONFIG_BTRFS_ASSERT
2345                         ", assert=on"
2346 #endif
2347 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2348                         ", integrity-checker=on"
2349 #endif
2350                         "\n",
2351                         btrfs_crc32c_impl());
2352 }
2353
2354 static int __init init_btrfs_fs(void)
2355 {
2356         int err;
2357
2358         err = btrfs_hash_init();
2359         if (err)
2360                 return err;
2361
2362         btrfs_props_init();
2363
2364         err = btrfs_init_sysfs();
2365         if (err)
2366                 goto free_hash;
2367
2368         btrfs_init_compress();
2369
2370         err = btrfs_init_cachep();
2371         if (err)
2372                 goto free_compress;
2373
2374         err = extent_io_init();
2375         if (err)
2376                 goto free_cachep;
2377
2378         err = extent_map_init();
2379         if (err)
2380                 goto free_extent_io;
2381
2382         err = ordered_data_init();
2383         if (err)
2384                 goto free_extent_map;
2385
2386         err = btrfs_delayed_inode_init();
2387         if (err)
2388                 goto free_ordered_data;
2389
2390         err = btrfs_auto_defrag_init();
2391         if (err)
2392                 goto free_delayed_inode;
2393
2394         err = btrfs_delayed_ref_init();
2395         if (err)
2396                 goto free_auto_defrag;
2397
2398         err = btrfs_prelim_ref_init();
2399         if (err)
2400                 goto free_delayed_ref;
2401
2402         err = btrfs_end_io_wq_init();
2403         if (err)
2404                 goto free_prelim_ref;
2405
2406         err = btrfs_interface_init();
2407         if (err)
2408                 goto free_end_io_wq;
2409
2410         btrfs_init_lockdep();
2411
2412         btrfs_print_mod_info();
2413
2414         err = btrfs_run_sanity_tests();
2415         if (err)
2416                 goto unregister_ioctl;
2417
2418         err = register_filesystem(&btrfs_fs_type);
2419         if (err)
2420                 goto unregister_ioctl;
2421
2422         return 0;
2423
2424 unregister_ioctl:
2425         btrfs_interface_exit();
2426 free_end_io_wq:
2427         btrfs_end_io_wq_exit();
2428 free_prelim_ref:
2429         btrfs_prelim_ref_exit();
2430 free_delayed_ref:
2431         btrfs_delayed_ref_exit();
2432 free_auto_defrag:
2433         btrfs_auto_defrag_exit();
2434 free_delayed_inode:
2435         btrfs_delayed_inode_exit();
2436 free_ordered_data:
2437         ordered_data_exit();
2438 free_extent_map:
2439         extent_map_exit();
2440 free_extent_io:
2441         extent_io_exit();
2442 free_cachep:
2443         btrfs_destroy_cachep();
2444 free_compress:
2445         btrfs_exit_compress();
2446         btrfs_exit_sysfs();
2447 free_hash:
2448         btrfs_hash_exit();
2449         return err;
2450 }
2451
2452 static void __exit exit_btrfs_fs(void)
2453 {
2454         btrfs_destroy_cachep();
2455         btrfs_delayed_ref_exit();
2456         btrfs_auto_defrag_exit();
2457         btrfs_delayed_inode_exit();
2458         btrfs_prelim_ref_exit();
2459         ordered_data_exit();
2460         extent_map_exit();
2461         extent_io_exit();
2462         btrfs_interface_exit();
2463         btrfs_end_io_wq_exit();
2464         unregister_filesystem(&btrfs_fs_type);
2465         btrfs_exit_sysfs();
2466         btrfs_cleanup_fs_uuids();
2467         btrfs_exit_compress();
2468         btrfs_hash_exit();
2469 }
2470
2471 late_initcall(init_btrfs_fs);
2472 module_exit(exit_btrfs_fs)
2473
2474 MODULE_LICENSE("GPL");