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                         btrfs_clear_opt(info->mount_opt, NOSSD);
 553                         break;
 554                 case Opt_ssd_spread:
 555                         btrfs_set_and_info(info, SSD_SPREAD,
 556                                            "use spread ssd allocation scheme");
 557                         btrfs_set_opt(info->mount_opt, SSD);
 558                         btrfs_clear_opt(info->mount_opt, NOSSD);
 559                         break;
 560                 case Opt_nossd:
 561                         btrfs_set_and_info(info, NOSSD,
 562                                              "not using ssd allocation scheme");
 563                         btrfs_clear_opt(info->mount_opt, SSD);
 564                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 565                         break;
 566                 case Opt_barrier:
 567                         btrfs_clear_and_info(info, NOBARRIER,
 568                                              "turning on barriers");
 569                         break;
 570                 case Opt_nobarrier:
 571                         btrfs_set_and_info(info, NOBARRIER,
 572                                            "turning off barriers");
 573                         break;
 574                 case Opt_thread_pool:
 575                         ret = match_int(&args[0], &intarg);
 576                         if (ret) {
 577                                 goto out;
 578                         } else if (intarg > 0) {
 579                                 info->thread_pool_size = intarg;
 580                         } else {
 581                                 ret = -EINVAL;
 582                                 goto out;
 583                         }
 584                         break;
 585                 case Opt_max_inline:
 586                         num = match_strdup(&args[0]);
 587                         if (num) {
 588                                 info->max_inline = memparse(num, NULL);
 589                                 kfree(num);
 590
 591                                 if (info->max_inline) {
 592                                         info->max_inline = min_t(u64,
 593                                                 info->max_inline,
 594                                                 info->sectorsize);
 595                                 }
 596                                 btrfs_info(info, "max_inline at %llu",
 597                                            info->max_inline);
 598                         } else {
 599                                 ret = -ENOMEM;
 600                                 goto out;
 601                         }
 602                         break;
 603                 case Opt_alloc_start:
 604                         num = match_strdup(&args[0]);
 605                         if (num) {
 606                                 mutex_lock(&info->chunk_mutex);
 607                                 info->alloc_start = memparse(num, NULL);
 608                                 mutex_unlock(&info->chunk_mutex);
 609                                 kfree(num);
 610                                 btrfs_info(info, "allocations start at %llu",
 611                                            info->alloc_start);
 612                         } else {
 613                                 ret = -ENOMEM;
 614                                 goto out;
 615                         }
 616                         break;
 617                 case Opt_acl:
 618 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 619                         info->sb->s_flags |= MS_POSIXACL;
 620                         break;
 621 #else
 622                         btrfs_err(info, "support for ACL not compiled in!");
 623                         ret = -EINVAL;
 624                         goto out;
 625 #endif
 626                 case Opt_noacl:
 627                         info->sb->s_flags &= ~MS_POSIXACL;
 628                         break;
 629                 case Opt_notreelog:
 630                         btrfs_set_and_info(info, NOTREELOG,
 631                                            "disabling tree log");
 632                         break;
 633                 case Opt_treelog:
 634                         btrfs_clear_and_info(info, NOTREELOG,
 635                                              "enabling tree log");
 636                         break;
 637                 case Opt_norecovery:
 638                 case Opt_nologreplay:
 639                         btrfs_set_and_info(info, NOLOGREPLAY,
 640                                            "disabling log replay at mount time");
 641                         break;
 642                 case Opt_flushoncommit:
 643                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 644                                            "turning on flush-on-commit");
 645                         break;
 646                 case Opt_noflushoncommit:
 647                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 648                                              "turning off flush-on-commit");
 649                         break;
 650                 case Opt_ratio:
 651                         ret = match_int(&args[0], &intarg);
 652                         if (ret) {
 653                                 goto out;
 654                         } else if (intarg >= 0) {
 655                                 info->metadata_ratio = intarg;
 656                                 btrfs_info(info, "metadata ratio %d",
 657                                            info->metadata_ratio);
 658                         } else {
 659                                 ret = -EINVAL;
 660                                 goto out;
 661                         }
 662                         break;
 663                 case Opt_discard:
 664                         btrfs_set_and_info(info, DISCARD,
 665                                            "turning on discard");
 666                         break;
 667                 case Opt_nodiscard:
 668                         btrfs_clear_and_info(info, DISCARD,
 669                                              "turning off discard");
 670                         break;
 671                 case Opt_space_cache:
 672                 case Opt_space_cache_version:
 673                         if (token == Opt_space_cache ||
 674                             strcmp(args[0].from, "v1") == 0) {
 675                                 btrfs_clear_opt(info->mount_opt,
 676                                                 FREE_SPACE_TREE);
 677                                 btrfs_set_and_info(info, SPACE_CACHE,
 678                                            "enabling disk space caching");
 679                         } else if (strcmp(args[0].from, "v2") == 0) {
 680                                 btrfs_clear_opt(info->mount_opt,
 681                                                 SPACE_CACHE);
 682                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 683                                                    "enabling free space tree");
 684                         } else {
 685                                 ret = -EINVAL;
 686                                 goto out;
 687                         }
 688                         break;
 689                 case Opt_rescan_uuid_tree:
 690                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 691                         break;
 692                 case Opt_no_space_cache:
 693                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 694                                 btrfs_clear_and_info(info, SPACE_CACHE,
 695                                              "disabling disk space caching");
 696                         }
 697                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 698                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 699                                              "disabling free space tree");
 700                         }
 701                         break;
 702                 case Opt_inode_cache:
 703                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 704                                            "enabling inode map caching");
 705                         break;
 706                 case Opt_noinode_cache:
 707                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 708                                              "disabling inode map caching");
 709                         break;
 710                 case Opt_clear_cache:
 711                         btrfs_set_and_info(info, CLEAR_CACHE,
 712                                            "force clearing of disk cache");
 713                         break;
 714                 case Opt_user_subvol_rm_allowed:
 715                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 716                         break;
 717                 case Opt_enospc_debug:
 718                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 719                         break;
 720                 case Opt_noenospc_debug:
 721                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 722                         break;
 723                 case Opt_defrag:
 724                         btrfs_set_and_info(info, AUTO_DEFRAG,
 725                                            "enabling auto defrag");
 726                         break;
 727                 case Opt_nodefrag:
 728                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 729                                              "disabling auto defrag");
 730                         break;
 731                 case Opt_recovery:
 732                         btrfs_warn(info,
 733                                    "'recovery' is deprecated, use 'usebackuproot' instead");
 734                 case Opt_usebackuproot:
 735                         btrfs_info(info,
 736                                    "trying to use backup root at mount time");
 737                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 738                         break;
 739                 case Opt_skip_balance:
 740                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 741                         break;
 742 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 743                 case Opt_check_integrity_including_extent_data:
 744                         btrfs_info(info,
 745                                    "enabling check integrity including extent data");
 746                         btrfs_set_opt(info->mount_opt,
 747                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 748                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 749                         break;
 750                 case Opt_check_integrity:
 751                         btrfs_info(info, "enabling check integrity");
 752                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 753                         break;
 754                 case Opt_check_integrity_print_mask:
 755                         ret = match_int(&args[0], &intarg);
 756                         if (ret) {
 757                                 goto out;
 758                         } else if (intarg >= 0) {
 759                                 info->check_integrity_print_mask = intarg;
 760                                 btrfs_info(info,
 761                                            "check_integrity_print_mask 0x%x",
 762                                            info->check_integrity_print_mask);
 763                         } else {
 764                                 ret = -EINVAL;
 765                                 goto out;
 766                         }
 767                         break;
 768 #else
 769                 case Opt_check_integrity_including_extent_data:
 770                 case Opt_check_integrity:
 771                 case Opt_check_integrity_print_mask:
 772                         btrfs_err(info,
 773                                   "support for check_integrity* not compiled in!");
 774                         ret = -EINVAL;
 775                         goto out;
 776 #endif
 777                 case Opt_fatal_errors:
 778                         if (strcmp(args[0].from, "panic") == 0)
 779                                 btrfs_set_opt(info->mount_opt,
 780                                               PANIC_ON_FATAL_ERROR);
 781                         else if (strcmp(args[0].from, "bug") == 0)
 782                                 btrfs_clear_opt(info->mount_opt,
 783                                               PANIC_ON_FATAL_ERROR);
 784                         else {
 785                                 ret = -EINVAL;
 786                                 goto out;
 787                         }
 788                         break;
 789                 case Opt_commit_interval:
 790                         intarg = 0;
 791                         ret = match_int(&args[0], &intarg);
 792                         if (ret < 0) {
 793                                 btrfs_err(info, "invalid commit interval");
 794                                 ret = -EINVAL;
 795                                 goto out;
 796                         }
 797                         if (intarg > 0) {
 798                                 if (intarg > 300) {
 799                                         btrfs_warn(info,
 800                                                 "excessive commit interval %d",
 801                                                 intarg);
 802                                 }
 803                                 info->commit_interval = intarg;
 804                         } else {
 805                                 btrfs_info(info,
 806                                            "using default commit interval %ds",
 807                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
 808                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 809                         }
 810                         break;
 811 #ifdef CONFIG_BTRFS_DEBUG
 812                 case Opt_fragment_all:
 813                         btrfs_info(info, "fragmenting all space");
 814                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 815                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 816                         break;
 817                 case Opt_fragment_metadata:
 818                         btrfs_info(info, "fragmenting metadata");
 819                         btrfs_set_opt(info->mount_opt,
 820                                       FRAGMENT_METADATA);
 821                         break;
 822                 case Opt_fragment_data:
 823                         btrfs_info(info, "fragmenting data");
 824                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 825                         break;
 826 #endif
 827                 case Opt_err:
 828                         btrfs_info(info, "unrecognized mount option '%s'", p);
 829                         ret = -EINVAL;
 830                         goto out;
 831                 default:
 832                         break;
 833                 }
 834         }
 835 check:
 836         /*
 837          * Extra check for current option against current flag
 838          */
 839         if (btrfs_test_opt(info, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
 840                 btrfs_err(info,
 841                           "nologreplay must be used with ro mount option");
 842                 ret = -EINVAL;
 843         }
 844 out:
 845         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
 846             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
 847             !btrfs_test_opt(info, CLEAR_CACHE)) {
 848                 btrfs_err(info, "cannot disable free space tree");
 849                 ret = -EINVAL;
 850
 851         }
 852         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
 853                 btrfs_info(info, "disk space caching is enabled");
 854         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
 855                 btrfs_info(info, "using free space tree");
 856         kfree(orig);
 857         return ret;
 858 }
 859
 860 /*
 861  * Parse mount options that are required early in the mount process.
 862  *
 863  * All other options will be parsed on much later in the mount process and
 864  * only when we need to allocate a new super block.
 865  */
 866 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 867                 void *holder, char **subvol_name, u64 *subvol_objectid,
 868                 struct btrfs_fs_devices **fs_devices)
 869 {
 870         substring_t args[MAX_OPT_ARGS];
 871         char *device_name, *opts, *orig, *p;
 872         char *num = NULL;
 873         int error = 0;
 874
 875         if (!options)
 876                 return 0;
 877
 878         /*
 879          * strsep changes the string, duplicate it because parse_options
 880          * gets called twice
 881          */
 882         opts = kstrdup(options, GFP_KERNEL);
 883         if (!opts)
 884                 return -ENOMEM;
 885         orig = opts;
 886
 887         while ((p = strsep(&opts, ",")) != NULL) {
 888                 int token;
 889                 if (!*p)
 890                         continue;
 891
 892                 token = match_token(p, tokens, args);
 893                 switch (token) {
 894                 case Opt_subvol:
 895                         kfree(*subvol_name);
 896                         *subvol_name = match_strdup(&args[0]);
 897                         if (!*subvol_name) {
 898                                 error = -ENOMEM;
 899                                 goto out;
 900                         }
 901                         break;
 902                 case Opt_subvolid:
 903                         num = match_strdup(&args[0]);
 904                         if (num) {
 905                                 *subvol_objectid = memparse(num, NULL);
 906                                 kfree(num);
 907                                 /* we want the original fs_tree */
 908                                 if (!*subvol_objectid)
 909                                         *subvol_objectid =
 910                                                 BTRFS_FS_TREE_OBJECTID;
 911                         } else {
 912                                 error = -EINVAL;
 913                                 goto out;
 914                         }
 915                         break;
 916                 case Opt_subvolrootid:
 917                         pr_warn("BTRFS: 'subvolrootid' mount option is deprecated and has no effect\n");
 918                         break;
 919                 case Opt_device:
 920                         device_name = match_strdup(&args[0]);
 921                         if (!device_name) {
 922                                 error = -ENOMEM;
 923                                 goto out;
 924                         }
 925                         error = btrfs_scan_one_device(device_name,
 926                                         flags, holder, fs_devices);
 927                         kfree(device_name);
 928                         if (error)
 929                                 goto out;
 930                         break;
 931                 default:
 932                         break;
 933                 }
 934         }
 935
 936 out:
 937         kfree(orig);
 938         return error;
 939 }
 940
 941 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 942                                            u64 subvol_objectid)
 943 {
 944         struct btrfs_root *root = fs_info->tree_root;
 945         struct btrfs_root *fs_root;
 946         struct btrfs_root_ref *root_ref;
 947         struct btrfs_inode_ref *inode_ref;
 948         struct btrfs_key key;
 949         struct btrfs_path *path = NULL;
 950         char *name = NULL, *ptr;
 951         u64 dirid;
 952         int len;
 953         int ret;
 954
 955         path = btrfs_alloc_path();
 956         if (!path) {
 957                 ret = -ENOMEM;
 958                 goto err;
 959         }
 960         path->leave_spinning = 1;
 961
 962         name = kmalloc(PATH_MAX, GFP_NOFS);
 963         if (!name) {
 964                 ret = -ENOMEM;
 965                 goto err;
 966         }
 967         ptr = name + PATH_MAX - 1;
 968         ptr[0] = '\0';
 969
 970         /*
 971          * Walk up the subvolume trees in the tree of tree roots by root
 972          * backrefs until we hit the top-level subvolume.
 973          */
 974         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 975                 key.objectid = subvol_objectid;
 976                 key.type = BTRFS_ROOT_BACKREF_KEY;
 977                 key.offset = (u64)-1;
 978
 979                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 980                 if (ret < 0) {
 981                         goto err;
 982                 } else if (ret > 0) {
 983                         ret = btrfs_previous_item(root, path, subvol_objectid,
 984                                                   BTRFS_ROOT_BACKREF_KEY);
 985                         if (ret < 0) {
 986                                 goto err;
 987                         } else if (ret > 0) {
 988                                 ret = -ENOENT;
 989                                 goto err;
 990                         }
 991                 }
 992
 993                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 994                 subvol_objectid = key.offset;
 995
 996                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 997                                           struct btrfs_root_ref);
 998                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 999                 ptr -= len + 1;
1000                 if (ptr < name) {
1001                         ret = -ENAMETOOLONG;
1002                         goto err;
1003                 }
1004                 read_extent_buffer(path->nodes[0], ptr + 1,
1005                                    (unsigned long)(root_ref + 1), len);
1006                 ptr[0] = '/';
1007                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1008                 btrfs_release_path(path);
1009
1010                 key.objectid = subvol_objectid;
1011                 key.type = BTRFS_ROOT_ITEM_KEY;
1012                 key.offset = (u64)-1;
1013                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1014                 if (IS_ERR(fs_root)) {
1015                         ret = PTR_ERR(fs_root);
1016                         goto err;
1017                 }
1018
1019                 /*
1020                  * Walk up the filesystem tree by inode refs until we hit the
1021                  * root directory.
1022                  */
1023                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1024                         key.objectid = dirid;
1025                         key.type = BTRFS_INODE_REF_KEY;
1026                         key.offset = (u64)-1;
1027
1028                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1029                         if (ret < 0) {
1030                                 goto err;
1031                         } else if (ret > 0) {
1032                                 ret = btrfs_previous_item(fs_root, path, dirid,
1033                                                           BTRFS_INODE_REF_KEY);
1034                                 if (ret < 0) {
1035                                         goto err;
1036                                 } else if (ret > 0) {
1037                                         ret = -ENOENT;
1038                                         goto err;
1039                                 }
1040                         }
1041
1042                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1043                         dirid = key.offset;
1044
1045                         inode_ref = btrfs_item_ptr(path->nodes[0],
1046                                                    path->slots[0],
1047                                                    struct btrfs_inode_ref);
1048                         len = btrfs_inode_ref_name_len(path->nodes[0],
1049                                                        inode_ref);
1050                         ptr -= len + 1;
1051                         if (ptr < name) {
1052                                 ret = -ENAMETOOLONG;
1053                                 goto err;
1054                         }
1055                         read_extent_buffer(path->nodes[0], ptr + 1,
1056                                            (unsigned long)(inode_ref + 1), len);
1057                         ptr[0] = '/';
1058                         btrfs_release_path(path);
1059                 }
1060         }
1061
1062         btrfs_free_path(path);
1063         if (ptr == name + PATH_MAX - 1) {
1064                 name[0] = '/';
1065                 name[1] = '\0';
1066         } else {
1067                 memmove(name, ptr, name + PATH_MAX - ptr);
1068         }
1069         return name;
1070
1071 err:
1072         btrfs_free_path(path);
1073         kfree(name);
1074         return ERR_PTR(ret);
1075 }
1076
1077 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1078 {
1079         struct btrfs_root *root = fs_info->tree_root;
1080         struct btrfs_dir_item *di;
1081         struct btrfs_path *path;
1082         struct btrfs_key location;
1083         u64 dir_id;
1084
1085         path = btrfs_alloc_path();
1086         if (!path)
1087                 return -ENOMEM;
1088         path->leave_spinning = 1;
1089
1090         /*
1091          * Find the "default" dir item which points to the root item that we
1092          * will mount by default if we haven't been given a specific subvolume
1093          * to mount.
1094          */
1095         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1096         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1097         if (IS_ERR(di)) {
1098                 btrfs_free_path(path);
1099                 return PTR_ERR(di);
1100         }
1101         if (!di) {
1102                 /*
1103                  * Ok the default dir item isn't there.  This is weird since
1104                  * it's always been there, but don't freak out, just try and
1105                  * mount the top-level subvolume.
1106                  */
1107                 btrfs_free_path(path);
1108                 *objectid = BTRFS_FS_TREE_OBJECTID;
1109                 return 0;
1110         }
1111
1112         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1113         btrfs_free_path(path);
1114         *objectid = location.objectid;
1115         return 0;
1116 }
1117
1118 static int btrfs_fill_super(struct super_block *sb,
1119                             struct btrfs_fs_devices *fs_devices,
1120                             void *data)
1121 {
1122         struct inode *inode;
1123         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1124         struct btrfs_key key;
1125         int err;
1126
1127         sb->s_maxbytes = MAX_LFS_FILESIZE;
1128         sb->s_magic = BTRFS_SUPER_MAGIC;
1129         sb->s_op = &btrfs_super_ops;
1130         sb->s_d_op = &btrfs_dentry_operations;
1131         sb->s_export_op = &btrfs_export_ops;
1132         sb->s_xattr = btrfs_xattr_handlers;
1133         sb->s_time_gran = 1;
1134 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1135         sb->s_flags |= MS_POSIXACL;
1136 #endif
1137         sb->s_flags |= MS_I_VERSION;
1138         sb->s_iflags |= SB_I_CGROUPWB;
1139         err = open_ctree(sb, fs_devices, (char *)data);
1140         if (err) {
1141                 btrfs_err(fs_info, "open_ctree failed");
1142                 return err;
1143         }
1144
1145         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1146         key.type = BTRFS_INODE_ITEM_KEY;
1147         key.offset = 0;
1148         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1149         if (IS_ERR(inode)) {
1150                 err = PTR_ERR(inode);
1151                 goto fail_close;
1152         }
1153
1154         sb->s_root = d_make_root(inode);
1155         if (!sb->s_root) {
1156                 err = -ENOMEM;
1157                 goto fail_close;
1158         }
1159
1160         save_mount_options(sb, data);
1161         cleancache_init_fs(sb);
1162         sb->s_flags |= MS_ACTIVE;
1163         return 0;
1164
1165 fail_close:
1166         close_ctree(fs_info);
1167         return err;
1168 }
1169
1170 int btrfs_sync_fs(struct super_block *sb, int wait)
1171 {
1172         struct btrfs_trans_handle *trans;
1173         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1174         struct btrfs_root *root = fs_info->tree_root;
1175
1176         trace_btrfs_sync_fs(fs_info, wait);
1177
1178         if (!wait) {
1179                 filemap_flush(fs_info->btree_inode->i_mapping);
1180                 return 0;
1181         }
1182
1183         btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
1184
1185         trans = btrfs_attach_transaction_barrier(root);
1186         if (IS_ERR(trans)) {
1187                 /* no transaction, don't bother */
1188                 if (PTR_ERR(trans) == -ENOENT) {
1189                         /*
1190                          * Exit unless we have some pending changes
1191                          * that need to go through commit
1192                          */
1193                         if (fs_info->pending_changes == 0)
1194                                 return 0;
1195                         /*
1196                          * A non-blocking test if the fs is frozen. We must not
1197                          * start a new transaction here otherwise a deadlock
1198                          * happens. The pending operations are delayed to the
1199                          * next commit after thawing.
1200                          */
1201                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1202                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1203                         else
1204                                 return 0;
1205                         trans = btrfs_start_transaction(root, 0);
1206                 }
1207                 if (IS_ERR(trans))
1208                         return PTR_ERR(trans);
1209         }
1210         return btrfs_commit_transaction(trans);
1211 }
1212
1213 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1214 {
1215         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1216         char *compress_type;
1217
1218         if (btrfs_test_opt(info, DEGRADED))
1219                 seq_puts(seq, ",degraded");
1220         if (btrfs_test_opt(info, NODATASUM))
1221                 seq_puts(seq, ",nodatasum");
1222         if (btrfs_test_opt(info, NODATACOW))
1223                 seq_puts(seq, ",nodatacow");
1224         if (btrfs_test_opt(info, NOBARRIER))
1225                 seq_puts(seq, ",nobarrier");
1226         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1227                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1228         if (info->alloc_start != 0)
1229                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1230         if (info->thread_pool_size !=  min_t(unsigned long,
1231                                              num_online_cpus() + 2, 8))
1232                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1233         if (btrfs_test_opt(info, COMPRESS)) {
1234                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1235                         compress_type = "zlib";
1236                 else
1237                         compress_type = "lzo";
1238                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1239                         seq_printf(seq, ",compress-force=%s", compress_type);
1240                 else
1241                         seq_printf(seq, ",compress=%s", compress_type);
1242         }
1243         if (btrfs_test_opt(info, NOSSD))
1244                 seq_puts(seq, ",nossd");
1245         if (btrfs_test_opt(info, SSD_SPREAD))
1246                 seq_puts(seq, ",ssd_spread");
1247         else if (btrfs_test_opt(info, SSD))
1248                 seq_puts(seq, ",ssd");
1249         if (btrfs_test_opt(info, NOTREELOG))
1250                 seq_puts(seq, ",notreelog");
1251         if (btrfs_test_opt(info, NOLOGREPLAY))
1252                 seq_puts(seq, ",nologreplay");
1253         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1254                 seq_puts(seq, ",flushoncommit");
1255         if (btrfs_test_opt(info, DISCARD))
1256                 seq_puts(seq, ",discard");
1257         if (!(info->sb->s_flags & MS_POSIXACL))
1258                 seq_puts(seq, ",noacl");
1259         if (btrfs_test_opt(info, SPACE_CACHE))
1260                 seq_puts(seq, ",space_cache");
1261         else if (btrfs_test_opt(info, FREE_SPACE_TREE))
1262                 seq_puts(seq, ",space_cache=v2");
1263         else
1264                 seq_puts(seq, ",nospace_cache");
1265         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1266                 seq_puts(seq, ",rescan_uuid_tree");
1267         if (btrfs_test_opt(info, CLEAR_CACHE))
1268                 seq_puts(seq, ",clear_cache");
1269         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1270                 seq_puts(seq, ",user_subvol_rm_allowed");
1271         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1272                 seq_puts(seq, ",enospc_debug");
1273         if (btrfs_test_opt(info, AUTO_DEFRAG))
1274                 seq_puts(seq, ",autodefrag");
1275         if (btrfs_test_opt(info, INODE_MAP_CACHE))
1276                 seq_puts(seq, ",inode_cache");
1277         if (btrfs_test_opt(info, SKIP_BALANCE))
1278                 seq_puts(seq, ",skip_balance");
1279 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1280         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1281                 seq_puts(seq, ",check_int_data");
1282         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1283                 seq_puts(seq, ",check_int");
1284         if (info->check_integrity_print_mask)
1285                 seq_printf(seq, ",check_int_print_mask=%d",
1286                                 info->check_integrity_print_mask);
1287 #endif
1288         if (info->metadata_ratio)
1289                 seq_printf(seq, ",metadata_ratio=%d",
1290                                 info->metadata_ratio);
1291         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1292                 seq_puts(seq, ",fatal_errors=panic");
1293         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1294                 seq_printf(seq, ",commit=%d", info->commit_interval);
1295 #ifdef CONFIG_BTRFS_DEBUG
1296         if (btrfs_test_opt(info, FRAGMENT_DATA))
1297                 seq_puts(seq, ",fragment=data");
1298         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1299                 seq_puts(seq, ",fragment=metadata");
1300 #endif
1301         seq_printf(seq, ",subvolid=%llu",
1302                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1303         seq_puts(seq, ",subvol=");
1304         seq_dentry(seq, dentry, " \t\n\\");
1305         return 0;
1306 }
1307
1308 static int btrfs_test_super(struct super_block *s, void *data)
1309 {
1310         struct btrfs_fs_info *p = data;
1311         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1312
1313         return fs_info->fs_devices == p->fs_devices;
1314 }
1315
1316 static int btrfs_set_super(struct super_block *s, void *data)
1317 {
1318         int err = set_anon_super(s, data);
1319         if (!err)
1320                 s->s_fs_info = data;
1321         return err;
1322 }
1323
1324 /*
1325  * subvolumes are identified by ino 256
1326  */
1327 static inline int is_subvolume_inode(struct inode *inode)
1328 {
1329         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1330                 return 1;
1331         return 0;
1332 }
1333
1334 /*
1335  * This will add subvolid=0 to the argument string while removing any subvol=
1336  * and subvolid= arguments to make sure we get the top-level root for path
1337  * walking to the subvol we want.
1338  */
1339 static char *setup_root_args(char *args)
1340 {
1341         char *buf, *dst, *sep;
1342
1343         if (!args)
1344                 return kstrdup("subvolid=0", GFP_NOFS);
1345
1346         /* The worst case is that we add ",subvolid=0" to the end. */
1347         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1348         if (!buf)
1349                 return NULL;
1350
1351         while (1) {
1352                 sep = strchrnul(args, ',');
1353                 if (!strstarts(args, "subvol=") &&
1354                     !strstarts(args, "subvolid=")) {
1355                         memcpy(dst, args, sep - args);
1356                         dst += sep - args;
1357                         *dst++ = ',';
1358                 }
1359                 if (*sep)
1360                         args = sep + 1;
1361                 else
1362                         break;
1363         }
1364         strcpy(dst, "subvolid=0");
1365
1366         return buf;
1367 }
1368
1369 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1370                                    int flags, const char *device_name,
1371                                    char *data)
1372 {
1373         struct dentry *root;
1374         struct vfsmount *mnt = NULL;
1375         char *newargs;
1376         int ret;
1377
1378         newargs = setup_root_args(data);
1379         if (!newargs) {
1380                 root = ERR_PTR(-ENOMEM);
1381                 goto out;
1382         }
1383
1384         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1385         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1386                 if (flags & MS_RDONLY) {
1387                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1388                                              device_name, newargs);
1389                 } else {
1390                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1391                                              device_name, newargs);
1392                         if (IS_ERR(mnt)) {
1393                                 root = ERR_CAST(mnt);
1394                                 mnt = NULL;
1395                                 goto out;
1396                         }
1397
1398                         down_write(&mnt->mnt_sb->s_umount);
1399                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1400                         up_write(&mnt->mnt_sb->s_umount);
1401                         if (ret < 0) {
1402                                 root = ERR_PTR(ret);
1403                                 goto out;
1404                         }
1405                 }
1406         }
1407         if (IS_ERR(mnt)) {
1408                 root = ERR_CAST(mnt);
1409                 mnt = NULL;
1410                 goto out;
1411         }
1412
1413         if (!subvol_name) {
1414                 if (!subvol_objectid) {
1415                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1416                                                           &subvol_objectid);
1417                         if (ret) {
1418                                 root = ERR_PTR(ret);
1419                                 goto out;
1420                         }
1421                 }
1422                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1423                                                             subvol_objectid);
1424                 if (IS_ERR(subvol_name)) {
1425                         root = ERR_CAST(subvol_name);
1426                         subvol_name = NULL;
1427                         goto out;
1428                 }
1429
1430         }
1431
1432         root = mount_subtree(mnt, subvol_name);
1433         /* mount_subtree() drops our reference on the vfsmount. */
1434         mnt = NULL;
1435
1436         if (!IS_ERR(root)) {
1437                 struct super_block *s = root->d_sb;
1438                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1439                 struct inode *root_inode = d_inode(root);
1440                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1441
1442                 ret = 0;
1443                 if (!is_subvolume_inode(root_inode)) {
1444                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1445                                subvol_name);
1446                         ret = -EINVAL;
1447                 }
1448                 if (subvol_objectid && root_objectid != subvol_objectid) {
1449                         /*
1450                          * This will also catch a race condition where a
1451                          * subvolume which was passed by ID is renamed and
1452                          * another subvolume is renamed over the old location.
1453                          */
1454                         btrfs_err(fs_info,
1455                                   "subvol '%s' does not match subvolid %llu",
1456                                   subvol_name, subvol_objectid);
1457                         ret = -EINVAL;
1458                 }
1459                 if (ret) {
1460                         dput(root);
1461                         root = ERR_PTR(ret);
1462                         deactivate_locked_super(s);
1463                 }
1464         }
1465
1466 out:
1467         mntput(mnt);
1468         kfree(newargs);
1469         kfree(subvol_name);
1470         return root;
1471 }
1472
1473 static int parse_security_options(char *orig_opts,
1474                                   struct security_mnt_opts *sec_opts)
1475 {
1476         char *secdata = NULL;
1477         int ret = 0;
1478
1479         secdata = alloc_secdata();
1480         if (!secdata)
1481                 return -ENOMEM;
1482         ret = security_sb_copy_data(orig_opts, secdata);
1483         if (ret) {
1484                 free_secdata(secdata);
1485                 return ret;
1486         }
1487         ret = security_sb_parse_opts_str(secdata, sec_opts);
1488         free_secdata(secdata);
1489         return ret;
1490 }
1491
1492 static int setup_security_options(struct btrfs_fs_info *fs_info,
1493                                   struct super_block *sb,
1494                                   struct security_mnt_opts *sec_opts)
1495 {
1496         int ret = 0;
1497
1498         /*
1499          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1500          * is valid.
1501          */
1502         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1503         if (ret)
1504                 return ret;
1505
1506 #ifdef CONFIG_SECURITY
1507         if (!fs_info->security_opts.num_mnt_opts) {
1508                 /* first time security setup, copy sec_opts to fs_info */
1509                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1510         } else {
1511                 /*
1512                  * Since SELinux (the only one supporting security_mnt_opts)
1513                  * does NOT support changing context during remount/mount of
1514                  * the same sb, this must be the same or part of the same
1515                  * security options, just free it.
1516                  */
1517                 security_free_mnt_opts(sec_opts);
1518         }
1519 #endif
1520         return ret;
1521 }
1522
1523 /*
1524  * Find a superblock for the given device / mount point.
1525  *
1526  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1527  *        for multiple device setup.  Make sure to keep it in sync.
1528  */
1529 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1530                 const char *device_name, void *data)
1531 {
1532         struct block_device *bdev = NULL;
1533         struct super_block *s;
1534         struct btrfs_fs_devices *fs_devices = NULL;
1535         struct btrfs_fs_info *fs_info = NULL;
1536         struct security_mnt_opts new_sec_opts;
1537         fmode_t mode = FMODE_READ;
1538         char *subvol_name = NULL;
1539         u64 subvol_objectid = 0;
1540         int error = 0;
1541
1542         if (!(flags & MS_RDONLY))
1543                 mode |= FMODE_WRITE;
1544
1545         error = btrfs_parse_early_options(data, mode, fs_type,
1546                                           &subvol_name, &subvol_objectid,
1547                                           &fs_devices);
1548         if (error) {
1549                 kfree(subvol_name);
1550                 return ERR_PTR(error);
1551         }
1552
1553         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1554                 /* mount_subvol() will free subvol_name. */
1555                 return mount_subvol(subvol_name, subvol_objectid, flags,
1556                                     device_name, data);
1557         }
1558
1559         security_init_mnt_opts(&new_sec_opts);
1560         if (data) {
1561                 error = parse_security_options(data, &new_sec_opts);
1562                 if (error)
1563                         return ERR_PTR(error);
1564         }
1565
1566         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1567         if (error)
1568                 goto error_sec_opts;
1569
1570         /*
1571          * Setup a dummy root and fs_info for test/set super.  This is because
1572          * we don't actually fill this stuff out until open_ctree, but we need
1573          * it for searching for existing supers, so this lets us do that and
1574          * then open_ctree will properly initialize everything later.
1575          */
1576         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1577         if (!fs_info) {
1578                 error = -ENOMEM;
1579                 goto error_sec_opts;
1580         }
1581
1582         fs_info->fs_devices = fs_devices;
1583
1584         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1585         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1586         security_init_mnt_opts(&fs_info->security_opts);
1587         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1588                 error = -ENOMEM;
1589                 goto error_fs_info;
1590         }
1591
1592         error = btrfs_open_devices(fs_devices, mode, fs_type);
1593         if (error)
1594                 goto error_fs_info;
1595
1596         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1597                 error = -EACCES;
1598                 goto error_close_devices;
1599         }
1600
1601         bdev = fs_devices->latest_bdev;
1602         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1603                  fs_info);
1604         if (IS_ERR(s)) {
1605                 error = PTR_ERR(s);
1606                 goto error_close_devices;
1607         }
1608
1609         if (s->s_root) {
1610                 btrfs_close_devices(fs_devices);
1611                 free_fs_info(fs_info);
1612                 if ((flags ^ s->s_flags) & MS_RDONLY)
1613                         error = -EBUSY;
1614         } else {
1615                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1616                 btrfs_sb(s)->bdev_holder = fs_type;
1617                 error = btrfs_fill_super(s, fs_devices, data);
1618         }
1619         if (error) {
1620                 deactivate_locked_super(s);
1621                 goto error_sec_opts;
1622         }
1623
1624         fs_info = btrfs_sb(s);
1625         error = setup_security_options(fs_info, s, &new_sec_opts);
1626         if (error) {
1627                 deactivate_locked_super(s);
1628                 goto error_sec_opts;
1629         }
1630
1631         return dget(s->s_root);
1632
1633 error_close_devices:
1634         btrfs_close_devices(fs_devices);
1635 error_fs_info:
1636         free_fs_info(fs_info);
1637 error_sec_opts:
1638         security_free_mnt_opts(&new_sec_opts);
1639         return ERR_PTR(error);
1640 }
1641
1642 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1643                                      int new_pool_size, int old_pool_size)
1644 {
1645         if (new_pool_size == old_pool_size)
1646                 return;
1647
1648         fs_info->thread_pool_size = new_pool_size;
1649
1650         btrfs_info(fs_info, "resize thread pool %d -> %d",
1651                old_pool_size, new_pool_size);
1652
1653         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1654         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1655         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1656         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1657         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1658         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1659         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1660                                 new_pool_size);
1661         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1662         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1663         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1664         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1665         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1666                                 new_pool_size);
1667 }
1668
1669 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1670 {
1671         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1672 }
1673
1674 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1675                                        unsigned long old_opts, int flags)
1676 {
1677         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1678             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1679              (flags & MS_RDONLY))) {
1680                 /* wait for any defraggers to finish */
1681                 wait_event(fs_info->transaction_wait,
1682                            (atomic_read(&fs_info->defrag_running) == 0));
1683                 if (flags & MS_RDONLY)
1684                         sync_filesystem(fs_info->sb);
1685         }
1686 }
1687
1688 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1689                                          unsigned long old_opts)
1690 {
1691         /*
1692          * We need to cleanup all defragable inodes if the autodefragment is
1693          * close or the filesystem is read only.
1694          */
1695         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1696             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1697              (fs_info->sb->s_flags & MS_RDONLY))) {
1698                 btrfs_cleanup_defrag_inodes(fs_info);
1699         }
1700
1701         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1702 }
1703
1704 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1705 {
1706         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1707         struct btrfs_root *root = fs_info->tree_root;
1708         unsigned old_flags = sb->s_flags;
1709         unsigned long old_opts = fs_info->mount_opt;
1710         unsigned long old_compress_type = fs_info->compress_type;
1711         u64 old_max_inline = fs_info->max_inline;
1712         u64 old_alloc_start = fs_info->alloc_start;
1713         int old_thread_pool_size = fs_info->thread_pool_size;
1714         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1715         int ret;
1716
1717         sync_filesystem(sb);
1718         btrfs_remount_prepare(fs_info);
1719
1720         if (data) {
1721                 struct security_mnt_opts new_sec_opts;
1722
1723                 security_init_mnt_opts(&new_sec_opts);
1724                 ret = parse_security_options(data, &new_sec_opts);
1725                 if (ret)
1726                         goto restore;
1727                 ret = setup_security_options(fs_info, sb,
1728                                              &new_sec_opts);
1729                 if (ret) {
1730                         security_free_mnt_opts(&new_sec_opts);
1731                         goto restore;
1732                 }
1733         }
1734
1735         ret = btrfs_parse_options(fs_info, data, *flags);
1736         if (ret) {
1737                 ret = -EINVAL;
1738                 goto restore;
1739         }
1740
1741         btrfs_remount_begin(fs_info, old_opts, *flags);
1742         btrfs_resize_thread_pool(fs_info,
1743                 fs_info->thread_pool_size, old_thread_pool_size);
1744
1745         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1746                 goto out;
1747
1748         if (*flags & MS_RDONLY) {
1749                 /*
1750                  * this also happens on 'umount -rf' or on shutdown, when
1751                  * the filesystem is busy.
1752                  */
1753                 cancel_work_sync(&fs_info->async_reclaim_work);
1754
1755                 /* wait for the uuid_scan task to finish */
1756                 down(&fs_info->uuid_tree_rescan_sem);
1757                 /* avoid complains from lockdep et al. */
1758                 up(&fs_info->uuid_tree_rescan_sem);
1759
1760                 sb->s_flags |= MS_RDONLY;
1761
1762                 /*
1763                  * Setting MS_RDONLY will put the cleaner thread to
1764                  * sleep at the next loop if it's already active.
1765                  * If it's already asleep, we'll leave unused block
1766                  * groups on disk until we're mounted read-write again
1767                  * unless we clean them up here.
1768                  */
1769                 btrfs_delete_unused_bgs(fs_info);
1770
1771                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1772                 btrfs_scrub_cancel(fs_info);
1773                 btrfs_pause_balance(fs_info);
1774
1775                 ret = btrfs_commit_super(fs_info);
1776                 if (ret)
1777                         goto restore;
1778         } else {
1779                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1780                         btrfs_err(fs_info,
1781                                 "Remounting read-write after error is not allowed");
1782                         ret = -EINVAL;
1783                         goto restore;
1784                 }
1785                 if (fs_info->fs_devices->rw_devices == 0) {
1786                         ret = -EACCES;
1787                         goto restore;
1788                 }
1789
1790                 if (fs_info->fs_devices->missing_devices >
1791                      fs_info->num_tolerated_disk_barrier_failures &&
1792                     !(*flags & MS_RDONLY)) {
1793                         btrfs_warn(fs_info,
1794                                 "too many missing devices, writeable remount is not allowed");
1795                         ret = -EACCES;
1796                         goto restore;
1797                 }
1798
1799                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1800                         ret = -EINVAL;
1801                         goto restore;
1802                 }
1803
1804                 ret = btrfs_cleanup_fs_roots(fs_info);
1805                 if (ret)
1806                         goto restore;
1807
1808                 /* recover relocation */
1809                 mutex_lock(&fs_info->cleaner_mutex);
1810                 ret = btrfs_recover_relocation(root);
1811                 mutex_unlock(&fs_info->cleaner_mutex);
1812                 if (ret)
1813                         goto restore;
1814
1815                 ret = btrfs_resume_balance_async(fs_info);
1816                 if (ret)
1817                         goto restore;
1818
1819                 ret = btrfs_resume_dev_replace_async(fs_info);
1820                 if (ret) {
1821                         btrfs_warn(fs_info, "failed to resume dev_replace");
1822                         goto restore;
1823                 }
1824
1825                 if (!fs_info->uuid_root) {
1826                         btrfs_info(fs_info, "creating UUID tree");
1827                         ret = btrfs_create_uuid_tree(fs_info);
1828                         if (ret) {
1829                                 btrfs_warn(fs_info,
1830                                            "failed to create the UUID tree %d",
1831                                            ret);
1832                                 goto restore;
1833                         }
1834                 }
1835                 sb->s_flags &= ~MS_RDONLY;
1836
1837                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1838         }
1839 out:
1840         wake_up_process(fs_info->transaction_kthread);
1841         btrfs_remount_cleanup(fs_info, old_opts);
1842         return 0;
1843
1844 restore:
1845         /* We've hit an error - don't reset MS_RDONLY */
1846         if (sb->s_flags & MS_RDONLY)
1847                 old_flags |= MS_RDONLY;
1848         sb->s_flags = old_flags;
1849         fs_info->mount_opt = old_opts;
1850         fs_info->compress_type = old_compress_type;
1851         fs_info->max_inline = old_max_inline;
1852         mutex_lock(&fs_info->chunk_mutex);
1853         fs_info->alloc_start = old_alloc_start;
1854         mutex_unlock(&fs_info->chunk_mutex);
1855         btrfs_resize_thread_pool(fs_info,
1856                 old_thread_pool_size, fs_info->thread_pool_size);
1857         fs_info->metadata_ratio = old_metadata_ratio;
1858         btrfs_remount_cleanup(fs_info, old_opts);
1859         return ret;
1860 }
1861
1862 /* Used to sort the devices by max_avail(descending sort) */
1863 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1864                                        const void *dev_info2)
1865 {
1866         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1867             ((struct btrfs_device_info *)dev_info2)->max_avail)
1868                 return -1;
1869         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1870                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1871                 return 1;
1872         else
1873         return 0;
1874 }
1875
1876 /*
1877  * sort the devices by max_avail, in which max free extent size of each device
1878  * is stored.(Descending Sort)
1879  */
1880 static inline void btrfs_descending_sort_devices(
1881                                         struct btrfs_device_info *devices,
1882                                         size_t nr_devices)
1883 {
1884         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1885              btrfs_cmp_device_free_bytes, NULL);
1886 }
1887
1888 /*
1889  * The helper to calc the free space on the devices that can be used to store
1890  * file data.
1891  */
1892 static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1893                                        u64 *free_bytes)
1894 {
1895         struct btrfs_root *root = fs_info->tree_root;
1896         struct btrfs_device_info *devices_info;
1897         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1898         struct btrfs_device *device;
1899         u64 skip_space;
1900         u64 type;
1901         u64 avail_space;
1902         u64 used_space;
1903         u64 min_stripe_size;
1904         int min_stripes = 1, num_stripes = 1;
1905         int i = 0, nr_devices;
1906         int ret;
1907
1908         /*
1909          * We aren't under the device list lock, so this is racy-ish, but good
1910          * enough for our purposes.
1911          */
1912         nr_devices = fs_info->fs_devices->open_devices;
1913         if (!nr_devices) {
1914                 smp_mb();
1915                 nr_devices = fs_info->fs_devices->open_devices;
1916                 ASSERT(nr_devices);
1917                 if (!nr_devices) {
1918                         *free_bytes = 0;
1919                         return 0;
1920                 }
1921         }
1922
1923         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1924                                GFP_NOFS);
1925         if (!devices_info)
1926                 return -ENOMEM;
1927
1928         /* calc min stripe number for data space allocation */
1929         type = btrfs_get_alloc_profile(root, 1);
1930         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1931                 min_stripes = 2;
1932                 num_stripes = nr_devices;
1933         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1934                 min_stripes = 2;
1935                 num_stripes = 2;
1936         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1937                 min_stripes = 4;
1938                 num_stripes = 4;
1939         }
1940
1941         if (type & BTRFS_BLOCK_GROUP_DUP)
1942                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1943         else
1944                 min_stripe_size = BTRFS_STRIPE_LEN;
1945
1946         if (fs_info->alloc_start)
1947                 mutex_lock(&fs_devices->device_list_mutex);
1948         rcu_read_lock();
1949         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1950                 if (!device->in_fs_metadata || !device->bdev ||
1951                     device->is_tgtdev_for_dev_replace)
1952                         continue;
1953
1954                 if (i >= nr_devices)
1955                         break;
1956
1957                 avail_space = device->total_bytes - device->bytes_used;
1958
1959                 /* align with stripe_len */
1960                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1961                 avail_space *= BTRFS_STRIPE_LEN;
1962
1963                 /*
1964                  * In order to avoid overwriting the superblock on the drive,
1965                  * btrfs starts at an offset of at least 1MB when doing chunk
1966                  * allocation.
1967                  */
1968                 skip_space = SZ_1M;
1969
1970                 /* user can set the offset in fs_info->alloc_start. */
1971                 if (fs_info->alloc_start &&
1972                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1973                     device->total_bytes) {
1974                         rcu_read_unlock();
1975                         skip_space = max(fs_info->alloc_start, skip_space);
1976
1977                         /*
1978                          * btrfs can not use the free space in
1979                          * [0, skip_space - 1], we must subtract it from the
1980                          * total. In order to implement it, we account the used
1981                          * space in this range first.
1982                          */
1983                         ret = btrfs_account_dev_extents_size(device, 0,
1984                                                              skip_space - 1,
1985                                                              &used_space);
1986                         if (ret) {
1987                                 kfree(devices_info);
1988                                 mutex_unlock(&fs_devices->device_list_mutex);
1989                                 return ret;
1990                         }
1991
1992                         rcu_read_lock();
1993
1994                         /* calc the free space in [0, skip_space - 1] */
1995                         skip_space -= used_space;
1996                 }
1997
1998                 /*
1999                  * we can use the free space in [0, skip_space - 1], subtract
2000                  * it from the total.
2001                  */
2002                 if (avail_space && avail_space >= skip_space)
2003                         avail_space -= skip_space;
2004                 else
2005                         avail_space = 0;
2006
2007                 if (avail_space < min_stripe_size)
2008                         continue;
2009
2010                 devices_info[i].dev = device;
2011                 devices_info[i].max_avail = avail_space;
2012
2013                 i++;
2014         }
2015         rcu_read_unlock();
2016         if (fs_info->alloc_start)
2017                 mutex_unlock(&fs_devices->device_list_mutex);
2018
2019         nr_devices = i;
2020
2021         btrfs_descending_sort_devices(devices_info, nr_devices);
2022
2023         i = nr_devices - 1;
2024         avail_space = 0;
2025         while (nr_devices >= min_stripes) {
2026                 if (num_stripes > nr_devices)
2027                         num_stripes = nr_devices;
2028
2029                 if (devices_info[i].max_avail >= min_stripe_size) {
2030                         int j;
2031                         u64 alloc_size;
2032
2033                         avail_space += devices_info[i].max_avail * num_stripes;
2034                         alloc_size = devices_info[i].max_avail;
2035                         for (j = i + 1 - num_stripes; j <= i; j++)
2036                                 devices_info[j].max_avail -= alloc_size;
2037                 }
2038                 i--;
2039                 nr_devices--;
2040         }
2041
2042         kfree(devices_info);
2043         *free_bytes = avail_space;
2044         return 0;
2045 }
2046
2047 /*
2048  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2049  *
2050  * If there's a redundant raid level at DATA block groups, use the respective
2051  * multiplier to scale the sizes.
2052  *
2053  * Unused device space usage is based on simulating the chunk allocator
2054  * algorithm that respects the device sizes, order of allocations and the
2055  * 'alloc_start' value, this is a close approximation of the actual use but
2056  * there are other factors that may change the result (like a new metadata
2057  * chunk).
2058  *
2059  * If metadata is exhausted, f_bavail will be 0.
2060  */
2061 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2062 {
2063         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2064         struct btrfs_super_block *disk_super = fs_info->super_copy;
2065         struct list_head *head = &fs_info->space_info;
2066         struct btrfs_space_info *found;
2067         u64 total_used = 0;
2068         u64 total_free_data = 0;
2069         u64 total_free_meta = 0;
2070         int bits = dentry->d_sb->s_blocksize_bits;
2071         __be32 *fsid = (__be32 *)fs_info->fsid;
2072         unsigned factor = 1;
2073         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2074         int ret;
2075         u64 thresh = 0;
2076         int mixed = 0;
2077
2078         rcu_read_lock();
2079         list_for_each_entry_rcu(found, head, list) {
2080                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2081                         int i;
2082
2083                         total_free_data += found->disk_total - found->disk_used;
2084                         total_free_data -=
2085                                 btrfs_account_ro_block_groups_free_space(found);
2086
2087                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2088                                 if (!list_empty(&found->block_groups[i])) {
2089                                         switch (i) {
2090                                         case BTRFS_RAID_DUP:
2091                                         case BTRFS_RAID_RAID1:
2092                                         case BTRFS_RAID_RAID10:
2093                                                 factor = 2;
2094                                         }
2095                                 }
2096                         }
2097                 }
2098
2099                 /*
2100                  * Metadata in mixed block goup profiles are accounted in data
2101                  */
2102                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2103                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2104                                 mixed = 1;
2105                         else
2106                                 total_free_meta += found->disk_total -
2107                                         found->disk_used;
2108                 }
2109
2110                 total_used += found->disk_used;
2111         }
2112
2113         rcu_read_unlock();
2114
2115         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2116         buf->f_blocks >>= bits;
2117         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2118
2119         /* Account global block reserve as used, it's in logical size already */
2120         spin_lock(&block_rsv->lock);
2121         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2122         if (buf->f_bfree >= block_rsv->size >> bits)
2123                 buf->f_bfree -= block_rsv->size >> bits;
2124         else
2125                 buf->f_bfree = 0;
2126         spin_unlock(&block_rsv->lock);
2127
2128         buf->f_bavail = div_u64(total_free_data, factor);
2129         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2130         if (ret)
2131                 return ret;
2132         buf->f_bavail += div_u64(total_free_data, factor);
2133         buf->f_bavail = buf->f_bavail >> bits;
2134
2135         /*
2136          * We calculate the remaining metadata space minus global reserve. If
2137          * this is (supposedly) smaller than zero, there's no space. But this
2138          * does not hold in practice, the exhausted state happens where's still
2139          * some positive delta. So we apply some guesswork and compare the
2140          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2141          *
2142          * We probably cannot calculate the exact threshold value because this
2143          * depends on the internal reservations requested by various
2144          * operations, so some operations that consume a few metadata will
2145          * succeed even if the Avail is zero. But this is better than the other
2146          * way around.
2147          */
2148         thresh = 4 * 1024 * 1024;
2149
2150         if (!mixed && total_free_meta - thresh < block_rsv->size)
2151                 buf->f_bavail = 0;
2152
2153         buf->f_type = BTRFS_SUPER_MAGIC;
2154         buf->f_bsize = dentry->d_sb->s_blocksize;
2155         buf->f_namelen = BTRFS_NAME_LEN;
2156
2157         /* We treat it as constant endianness (it doesn't matter _which_)
2158            because we want the fsid to come out the same whether mounted
2159            on a big-endian or little-endian host */
2160         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2161         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2162         /* Mask in the root object ID too, to disambiguate subvols */
2163         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2164         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2165
2166         return 0;
2167 }
2168
2169 static void btrfs_kill_super(struct super_block *sb)
2170 {
2171         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2172         kill_anon_super(sb);
2173         free_fs_info(fs_info);
2174 }
2175
2176 static struct file_system_type btrfs_fs_type = {
2177         .owner          = THIS_MODULE,
2178         .name           = "btrfs",
2179         .mount          = btrfs_mount,
2180         .kill_sb        = btrfs_kill_super,
2181         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2182 };
2183 MODULE_ALIAS_FS("btrfs");
2184
2185 static int btrfs_control_open(struct inode *inode, struct file *file)
2186 {
2187         /*
2188          * The control file's private_data is used to hold the
2189          * transaction when it is started and is used to keep
2190          * track of whether a transaction is already in progress.
2191          */
2192         file->private_data = NULL;
2193         return 0;
2194 }
2195
2196 /*
2197  * used by btrfsctl to scan devices when no FS is mounted
2198  */
2199 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2200                                 unsigned long arg)
2201 {
2202         struct btrfs_ioctl_vol_args *vol;
2203         struct btrfs_fs_devices *fs_devices;
2204         int ret = -ENOTTY;
2205
2206         if (!capable(CAP_SYS_ADMIN))
2207                 return -EPERM;
2208
2209         vol = memdup_user((void __user *)arg, sizeof(*vol));
2210         if (IS_ERR(vol))
2211                 return PTR_ERR(vol);
2212
2213         switch (cmd) {
2214         case BTRFS_IOC_SCAN_DEV:
2215                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2216                                             &btrfs_fs_type, &fs_devices);
2217                 break;
2218         case BTRFS_IOC_DEVICES_READY:
2219                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2220                                             &btrfs_fs_type, &fs_devices);
2221                 if (ret)
2222                         break;
2223                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2224                 break;
2225         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2226                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2227                 break;
2228         }
2229
2230         kfree(vol);
2231         return ret;
2232 }
2233
2234 static int btrfs_freeze(struct super_block *sb)
2235 {
2236         struct btrfs_trans_handle *trans;
2237         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2238         struct btrfs_root *root = fs_info->tree_root;
2239
2240         fs_info->fs_frozen = 1;
2241         /*
2242          * We don't need a barrier here, we'll wait for any transaction that
2243          * could be in progress on other threads (and do delayed iputs that
2244          * we want to avoid on a frozen filesystem), or do the commit
2245          * ourselves.
2246          */
2247         trans = btrfs_attach_transaction_barrier(root);
2248         if (IS_ERR(trans)) {
2249                 /* no transaction, don't bother */
2250                 if (PTR_ERR(trans) == -ENOENT)
2251                         return 0;
2252                 return PTR_ERR(trans);
2253         }
2254         return btrfs_commit_transaction(trans);
2255 }
2256
2257 static int btrfs_unfreeze(struct super_block *sb)
2258 {
2259         btrfs_sb(sb)->fs_frozen = 0;
2260         return 0;
2261 }
2262
2263 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2264 {
2265         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2266         struct btrfs_fs_devices *cur_devices;
2267         struct btrfs_device *dev, *first_dev = NULL;
2268         struct list_head *head;
2269         struct rcu_string *name;
2270
2271         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2272         cur_devices = fs_info->fs_devices;
2273         while (cur_devices) {
2274                 head = &cur_devices->devices;
2275                 list_for_each_entry(dev, head, dev_list) {
2276                         if (dev->missing)
2277                                 continue;
2278                         if (!dev->name)
2279                                 continue;
2280                         if (!first_dev || dev->devid < first_dev->devid)
2281                                 first_dev = dev;
2282                 }
2283                 cur_devices = cur_devices->seed;
2284         }
2285
2286         if (first_dev) {
2287                 rcu_read_lock();
2288                 name = rcu_dereference(first_dev->name);
2289                 seq_escape(m, name->str, " \t\n\\");
2290                 rcu_read_unlock();
2291         } else {
2292                 WARN_ON(1);
2293         }
2294         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2295         return 0;
2296 }
2297
2298 static const struct super_operations btrfs_super_ops = {
2299         .drop_inode     = btrfs_drop_inode,
2300         .evict_inode    = btrfs_evict_inode,
2301         .put_super      = btrfs_put_super,
2302         .sync_fs        = btrfs_sync_fs,
2303         .show_options   = btrfs_show_options,
2304         .show_devname   = btrfs_show_devname,
2305         .write_inode    = btrfs_write_inode,
2306         .alloc_inode    = btrfs_alloc_inode,
2307         .destroy_inode  = btrfs_destroy_inode,
2308         .statfs         = btrfs_statfs,
2309         .remount_fs     = btrfs_remount,
2310         .freeze_fs      = btrfs_freeze,
2311         .unfreeze_fs    = btrfs_unfreeze,
2312 };
2313
2314 static const struct file_operations btrfs_ctl_fops = {
2315         .open = btrfs_control_open,
2316         .unlocked_ioctl  = btrfs_control_ioctl,
2317         .compat_ioctl = btrfs_control_ioctl,
2318         .owner   = THIS_MODULE,
2319         .llseek = noop_llseek,
2320 };
2321
2322 static struct miscdevice btrfs_misc = {
2323         .minor          = BTRFS_MINOR,
2324         .name           = "btrfs-control",
2325         .fops           = &btrfs_ctl_fops
2326 };
2327
2328 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2329 MODULE_ALIAS("devname:btrfs-control");
2330
2331 static int btrfs_interface_init(void)
2332 {
2333         return misc_register(&btrfs_misc);
2334 }
2335
2336 static void btrfs_interface_exit(void)
2337 {
2338         misc_deregister(&btrfs_misc);
2339 }
2340
2341 static void btrfs_print_mod_info(void)
2342 {
2343         pr_info("Btrfs loaded, crc32c=%s"
2344 #ifdef CONFIG_BTRFS_DEBUG
2345                         ", debug=on"
2346 #endif
2347 #ifdef CONFIG_BTRFS_ASSERT
2348                         ", assert=on"
2349 #endif
2350 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2351                         ", integrity-checker=on"
2352 #endif
2353                         "\n",
2354                         btrfs_crc32c_impl());
2355 }
2356
2357 static int __init init_btrfs_fs(void)
2358 {
2359         int err;
2360
2361         err = btrfs_hash_init();
2362         if (err)
2363                 return err;
2364
2365         btrfs_props_init();
2366
2367         err = btrfs_init_sysfs();
2368         if (err)
2369                 goto free_hash;
2370
2371         btrfs_init_compress();
2372
2373         err = btrfs_init_cachep();
2374         if (err)
2375                 goto free_compress;
2376
2377         err = extent_io_init();
2378         if (err)
2379                 goto free_cachep;
2380
2381         err = extent_map_init();
2382         if (err)
2383                 goto free_extent_io;
2384
2385         err = ordered_data_init();
2386         if (err)
2387                 goto free_extent_map;
2388
2389         err = btrfs_delayed_inode_init();
2390         if (err)
2391                 goto free_ordered_data;
2392
2393         err = btrfs_auto_defrag_init();
2394         if (err)
2395                 goto free_delayed_inode;
2396
2397         err = btrfs_delayed_ref_init();
2398         if (err)
2399                 goto free_auto_defrag;
2400
2401         err = btrfs_prelim_ref_init();
2402         if (err)
2403                 goto free_delayed_ref;
2404
2405         err = btrfs_end_io_wq_init();
2406         if (err)
2407                 goto free_prelim_ref;
2408
2409         err = btrfs_interface_init();
2410         if (err)
2411                 goto free_end_io_wq;
2412
2413         btrfs_init_lockdep();
2414
2415         btrfs_print_mod_info();
2416
2417         err = btrfs_run_sanity_tests();
2418         if (err)
2419                 goto unregister_ioctl;
2420
2421         err = register_filesystem(&btrfs_fs_type);
2422         if (err)
2423                 goto unregister_ioctl;
2424
2425         return 0;
2426
2427 unregister_ioctl:
2428         btrfs_interface_exit();
2429 free_end_io_wq:
2430         btrfs_end_io_wq_exit();
2431 free_prelim_ref:
2432         btrfs_prelim_ref_exit();
2433 free_delayed_ref:
2434         btrfs_delayed_ref_exit();
2435 free_auto_defrag:
2436         btrfs_auto_defrag_exit();
2437 free_delayed_inode:
2438         btrfs_delayed_inode_exit();
2439 free_ordered_data:
2440         ordered_data_exit();
2441 free_extent_map:
2442         extent_map_exit();
2443 free_extent_io:
2444         extent_io_exit();
2445 free_cachep:
2446         btrfs_destroy_cachep();
2447 free_compress:
2448         btrfs_exit_compress();
2449         btrfs_exit_sysfs();
2450 free_hash:
2451         btrfs_hash_exit();
2452         return err;
2453 }
2454
2455 static void __exit exit_btrfs_fs(void)
2456 {
2457         btrfs_destroy_cachep();
2458         btrfs_delayed_ref_exit();
2459         btrfs_auto_defrag_exit();
2460         btrfs_delayed_inode_exit();
2461         btrfs_prelim_ref_exit();
2462         ordered_data_exit();
2463         extent_map_exit();
2464         extent_io_exit();
2465         btrfs_interface_exit();
2466         btrfs_end_io_wq_exit();
2467         unregister_filesystem(&btrfs_fs_type);
2468         btrfs_exit_sysfs();
2469         btrfs_cleanup_fs_uuids();
2470         btrfs_exit_compress();
2471         btrfs_hash_exit();
2472 }
2473
2474 late_initcall(init_btrfs_fs);
2475 module_exit(exit_btrfs_fs)
2476
2477 MODULE_LICENSE("GPL");