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