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 "compat.h"
  46 #include "delayed-inode.h"
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "xattr.h"
  53 #include "volumes.h"
  54 #include "version.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
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/btrfs.h>
  63
  64 static const struct super_operations btrfs_super_ops;
  65 static struct file_system_type btrfs_fs_type;
  66
  67 static const char *btrfs_decode_error(int errno)
  68 {
  69         char *errstr = "unknown";
  70
  71         switch (errno) {
  72         case -EIO:
  73                 errstr = "IO failure";
  74                 break;
  75         case -ENOMEM:
  76                 errstr = "Out of memory";
  77                 break;
  78         case -EROFS:
  79                 errstr = "Readonly filesystem";
  80                 break;
  81         case -EEXIST:
  82                 errstr = "Object already exists";
  83                 break;
  84         case -ENOSPC:
  85                 errstr = "No space left";
  86                 break;
  87         case -ENOENT:
  88                 errstr = "No such entry";
  89                 break;
  90         }
  91
  92         return errstr;
  93 }
  94
  95 static void save_error_info(struct btrfs_fs_info *fs_info)
  96 {
  97         /*
  98          * today we only save the error info into ram.  Long term we'll
  99          * also send it down to the disk
 100          */
 101         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 102 }
 103
 104 /* btrfs handle error by forcing the filesystem readonly */
 105 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 106 {
 107         struct super_block *sb = fs_info->sb;
 108
 109         if (sb->s_flags & MS_RDONLY)
 110                 return;
 111
 112         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 113                 sb->s_flags |= MS_RDONLY;
 114                 btrfs_info(fs_info, "forced readonly");
 115                 /*
 116                  * Note that a running device replace operation is not
 117                  * canceled here although there is no way to update
 118                  * the progress. It would add the risk of a deadlock,
 119                  * therefore the canceling is ommited. The only penalty
 120                  * is that some I/O remains active until the procedure
 121                  * completes. The next time when the filesystem is
 122                  * mounted writeable again, the device replace
 123                  * operation continues.
 124                  */
 125         }
 126 }
 127
 128 #ifdef CONFIG_PRINTK
 129 /*
 130  * __btrfs_std_error decodes expected errors from the caller and
 131  * invokes the approciate error response.
 132  */
 133 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 134                        unsigned int line, int errno, const char *fmt, ...)
 135 {
 136         struct super_block *sb = fs_info->sb;
 137         const char *errstr;
 138
 139         /*
 140          * Special case: if the error is EROFS, and we're already
 141          * under MS_RDONLY, then it is safe here.
 142          */
 143         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 144                 return;
 145
 146         errstr = btrfs_decode_error(errno);
 147         if (fmt) {
 148                 struct va_format vaf;
 149                 va_list args;
 150
 151                 va_start(args, fmt);
 152                 vaf.fmt = fmt;
 153                 vaf.va = &args;
 154
 155                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 156                         sb->s_id, function, line, errno, errstr, &vaf);
 157                 va_end(args);
 158         } else {
 159                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: errno=%d %s\n",
 160                         sb->s_id, function, line, errno, errstr);
 161         }
 162
 163         /* Don't go through full error handling during mount */
 164         save_error_info(fs_info);
 165         if (sb->s_flags & MS_BORN)
 166                 btrfs_handle_error(fs_info);
 167 }
 168
 169 static const char * const logtypes[] = {
 170         "emergency",
 171         "alert",
 172         "critical",
 173         "error",
 174         "warning",
 175         "notice",
 176         "info",
 177         "debug",
 178 };
 179
 180 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 181 {
 182         struct super_block *sb = fs_info->sb;
 183         char lvl[4];
 184         struct va_format vaf;
 185         va_list args;
 186         const char *type = logtypes[4];
 187         int kern_level;
 188
 189         va_start(args, fmt);
 190
 191         kern_level = printk_get_level(fmt);
 192         if (kern_level) {
 193                 size_t size = printk_skip_level(fmt) - fmt;
 194                 memcpy(lvl, fmt,  size);
 195                 lvl[size] = '\0';
 196                 fmt += size;
 197                 type = logtypes[kern_level - '0'];
 198         } else
 199                 *lvl = '\0';
 200
 201         vaf.fmt = fmt;
 202         vaf.va = &args;
 203
 204         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 205
 206         va_end(args);
 207 }
 208
 209 #else
 210
 211 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 212                        unsigned int line, int errno, const char *fmt, ...)
 213 {
 214         struct super_block *sb = fs_info->sb;
 215
 216         /*
 217          * Special case: if the error is EROFS, and we're already
 218          * under MS_RDONLY, then it is safe here.
 219          */
 220         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 221                 return;
 222
 223         /* Don't go through full error handling during mount */
 224         if (sb->s_flags & MS_BORN) {
 225                 save_error_info(fs_info);
 226                 btrfs_handle_error(fs_info);
 227         }
 228 }
 229 #endif
 230
 231 /*
 232  * We only mark the transaction aborted and then set the file system read-only.
 233  * This will prevent new transactions from starting or trying to join this
 234  * one.
 235  *
 236  * This means that error recovery at the call site is limited to freeing
 237  * any local memory allocations and passing the error code up without
 238  * further cleanup. The transaction should complete as it normally would
 239  * in the call path but will return -EIO.
 240  *
 241  * We'll complete the cleanup in btrfs_end_transaction and
 242  * btrfs_commit_transaction.
 243  */
 244 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 245                                struct btrfs_root *root, const char *function,
 246                                unsigned int line, int errno)
 247 {
 248         /*
 249          * Report first abort since mount
 250          */
 251         if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,
 252                                 &root->fs_info->fs_state)) {
 253                 WARN(1, KERN_DEBUG "btrfs: Transaction aborted (error %d)\n",
 254                                 errno);
 255         }
 256         trans->aborted = errno;
 257         /* Nothing used. The other threads that have joined this
 258          * transaction may be able to continue. */
 259         if (!trans->blocks_used) {
 260                 const char *errstr;
 261
 262                 errstr = btrfs_decode_error(errno);
 263                 btrfs_warn(root->fs_info,
 264                            "%s:%d: Aborting unused transaction(%s).",
 265                            function, line, errstr);
 266                 return;
 267         }
 268         ACCESS_ONCE(trans->transaction->aborted) = errno;
 269         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 270 }
 271 /*
 272  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 273  * issues an alert, and either panics or BUGs, depending on mount options.
 274  */
 275 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 276                    unsigned int line, int errno, const char *fmt, ...)
 277 {
 278         char *s_id = "<unknown>";
 279         const char *errstr;
 280         struct va_format vaf = { .fmt = fmt };
 281         va_list args;
 282
 283         if (fs_info)
 284                 s_id = fs_info->sb->s_id;
 285
 286         va_start(args, fmt);
 287         vaf.va = &args;
 288
 289         errstr = btrfs_decode_error(errno);
 290         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 291                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 292                         s_id, function, line, &vaf, errno, errstr);
 293
 294         printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 295                s_id, function, line, &vaf, errno, errstr);
 296         va_end(args);
 297         /* Caller calls BUG() */
 298 }
 299
 300 static void btrfs_put_super(struct super_block *sb)
 301 {
 302         (void)close_ctree(btrfs_sb(sb)->tree_root);
 303         /* FIXME: need to fix VFS to return error? */
 304         /* AV: return it _where_?  ->put_super() can be triggered by any number
 305          * of async events, up to and including delivery of SIGKILL to the
 306          * last process that kept it busy.  Or segfault in the aforementioned
 307          * process...  Whom would you report that to?
 308          */
 309 }
 310
 311 enum {
 312         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 313         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 314         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 315         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 316         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 317         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 318         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 319         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 320         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 321         Opt_check_integrity_print_mask, Opt_fatal_errors,
 322         Opt_err,
 323 };
 324
 325 static match_table_t tokens = {
 326         {Opt_degraded, "degraded"},
 327         {Opt_subvol, "subvol=%s"},
 328         {Opt_subvolid, "subvolid=%d"},
 329         {Opt_device, "device=%s"},
 330         {Opt_nodatasum, "nodatasum"},
 331         {Opt_nodatacow, "nodatacow"},
 332         {Opt_nobarrier, "nobarrier"},
 333         {Opt_max_inline, "max_inline=%s"},
 334         {Opt_alloc_start, "alloc_start=%s"},
 335         {Opt_thread_pool, "thread_pool=%d"},
 336         {Opt_compress, "compress"},
 337         {Opt_compress_type, "compress=%s"},
 338         {Opt_compress_force, "compress-force"},
 339         {Opt_compress_force_type, "compress-force=%s"},
 340         {Opt_ssd, "ssd"},
 341         {Opt_ssd_spread, "ssd_spread"},
 342         {Opt_nossd, "nossd"},
 343         {Opt_noacl, "noacl"},
 344         {Opt_notreelog, "notreelog"},
 345         {Opt_flushoncommit, "flushoncommit"},
 346         {Opt_ratio, "metadata_ratio=%d"},
 347         {Opt_discard, "discard"},
 348         {Opt_space_cache, "space_cache"},
 349         {Opt_clear_cache, "clear_cache"},
 350         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 351         {Opt_enospc_debug, "enospc_debug"},
 352         {Opt_subvolrootid, "subvolrootid=%d"},
 353         {Opt_defrag, "autodefrag"},
 354         {Opt_inode_cache, "inode_cache"},
 355         {Opt_no_space_cache, "nospace_cache"},
 356         {Opt_recovery, "recovery"},
 357         {Opt_skip_balance, "skip_balance"},
 358         {Opt_check_integrity, "check_int"},
 359         {Opt_check_integrity_including_extent_data, "check_int_data"},
 360         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 361         {Opt_fatal_errors, "fatal_errors=%s"},
 362         {Opt_err, NULL},
 363 };
 364
 365 /*
 366  * Regular mount options parser.  Everything that is needed only when
 367  * reading in a new superblock is parsed here.
 368  * XXX JDM: This needs to be cleaned up for remount.
 369  */
 370 int btrfs_parse_options(struct btrfs_root *root, char *options)
 371 {
 372         struct btrfs_fs_info *info = root->fs_info;
 373         substring_t args[MAX_OPT_ARGS];
 374         char *p, *num, *orig = NULL;
 375         u64 cache_gen;
 376         int intarg;
 377         int ret = 0;
 378         char *compress_type;
 379         bool compress_force = false;
 380
 381         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 382         if (cache_gen)
 383                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 384
 385         if (!options)
 386                 goto out;
 387
 388         /*
 389          * strsep changes the string, duplicate it because parse_options
 390          * gets called twice
 391          */
 392         options = kstrdup(options, GFP_NOFS);
 393         if (!options)
 394                 return -ENOMEM;
 395
 396         orig = options;
 397
 398         while ((p = strsep(&options, ",")) != NULL) {
 399                 int token;
 400                 if (!*p)
 401                         continue;
 402
 403                 token = match_token(p, tokens, args);
 404                 switch (token) {
 405                 case Opt_degraded:
 406                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 407                         btrfs_set_opt(info->mount_opt, DEGRADED);
 408                         break;
 409                 case Opt_subvol:
 410                 case Opt_subvolid:
 411                 case Opt_subvolrootid:
 412                 case Opt_device:
 413                         /*
 414                          * These are parsed by btrfs_parse_early_options
 415                          * and can be happily ignored here.
 416                          */
 417                         break;
 418                 case Opt_nodatasum:
 419                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 420                         btrfs_set_opt(info->mount_opt, NODATASUM);
 421                         break;
 422                 case Opt_nodatacow:
 423                         if (!btrfs_test_opt(root, COMPRESS) ||
 424                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
 425                                         printk(KERN_INFO "btrfs: setting nodatacow, compression disabled\n");
 426                         } else {
 427                                 printk(KERN_INFO "btrfs: setting nodatacow\n");
 428                         }
 429                         info->compress_type = BTRFS_COMPRESS_NONE;
 430                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 431                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 432                         btrfs_set_opt(info->mount_opt, NODATACOW);
 433                         btrfs_set_opt(info->mount_opt, NODATASUM);
 434                         break;
 435                 case Opt_compress_force:
 436                 case Opt_compress_force_type:
 437                         compress_force = true;
 438                         /* Fallthrough */
 439                 case Opt_compress:
 440                 case Opt_compress_type:
 441                         if (token == Opt_compress ||
 442                             token == Opt_compress_force ||
 443                             strcmp(args[0].from, "zlib") == 0) {
 444                                 compress_type = "zlib";
 445                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 446                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 447                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 448                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 449                         } else if (strcmp(args[0].from, "lzo") == 0) {
 450                                 compress_type = "lzo";
 451                                 info->compress_type = BTRFS_COMPRESS_LZO;
 452                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 453                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 454                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 455                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 456                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 457                                 compress_type = "no";
 458                                 info->compress_type = BTRFS_COMPRESS_NONE;
 459                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 460                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 461                                 compress_force = false;
 462                         } else {
 463                                 ret = -EINVAL;
 464                                 goto out;
 465                         }
 466
 467                         if (compress_force) {
 468                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 469                                 pr_info("btrfs: force %s compression\n",
 470                                         compress_type);
 471                         } else
 472                                 pr_info("btrfs: use %s compression\n",
 473                                         compress_type);
 474                         break;
 475                 case Opt_ssd:
 476                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 477                         btrfs_set_opt(info->mount_opt, SSD);
 478                         break;
 479                 case Opt_ssd_spread:
 480                         printk(KERN_INFO "btrfs: use spread ssd "
 481                                "allocation scheme\n");
 482                         btrfs_set_opt(info->mount_opt, SSD);
 483                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 484                         break;
 485                 case Opt_nossd:
 486                         printk(KERN_INFO "btrfs: not using ssd allocation "
 487                                "scheme\n");
 488                         btrfs_set_opt(info->mount_opt, NOSSD);
 489                         btrfs_clear_opt(info->mount_opt, SSD);
 490                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 491                         break;
 492                 case Opt_nobarrier:
 493                         printk(KERN_INFO "btrfs: turning off barriers\n");
 494                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 495                         break;
 496                 case Opt_thread_pool:
 497                         intarg = 0;
 498                         match_int(&args[0], &intarg);
 499                         if (intarg)
 500                                 info->thread_pool_size = intarg;
 501                         break;
 502                 case Opt_max_inline:
 503                         num = match_strdup(&args[0]);
 504                         if (num) {
 505                                 info->max_inline = memparse(num, NULL);
 506                                 kfree(num);
 507
 508                                 if (info->max_inline) {
 509                                         info->max_inline = max_t(u64,
 510                                                 info->max_inline,
 511                                                 root->sectorsize);
 512                                 }
 513                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 514                                         (unsigned long long)info->max_inline);
 515                         }
 516                         break;
 517                 case Opt_alloc_start:
 518                         num = match_strdup(&args[0]);
 519                         if (num) {
 520                                 mutex_lock(&info->chunk_mutex);
 521                                 info->alloc_start = memparse(num, NULL);
 522                                 mutex_unlock(&info->chunk_mutex);
 523                                 kfree(num);
 524                                 printk(KERN_INFO
 525                                         "btrfs: allocations start at %llu\n",
 526                                         (unsigned long long)info->alloc_start);
 527                         }
 528                         break;
 529                 case Opt_noacl:
 530                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 531                         break;
 532                 case Opt_notreelog:
 533                         printk(KERN_INFO "btrfs: disabling tree log\n");
 534                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 535                         break;
 536                 case Opt_flushoncommit:
 537                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 538                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 539                         break;
 540                 case Opt_ratio:
 541                         intarg = 0;
 542                         match_int(&args[0], &intarg);
 543                         if (intarg) {
 544                                 info->metadata_ratio = intarg;
 545                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 546                                        info->metadata_ratio);
 547                         }
 548                         break;
 549                 case Opt_discard:
 550                         btrfs_set_opt(info->mount_opt, DISCARD);
 551                         break;
 552                 case Opt_space_cache:
 553                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 554                         break;
 555                 case Opt_no_space_cache:
 556                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 557                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 558                         break;
 559                 case Opt_inode_cache:
 560                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 561                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 562                         break;
 563                 case Opt_clear_cache:
 564                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 565                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 566                         break;
 567                 case Opt_user_subvol_rm_allowed:
 568                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 569                         break;
 570                 case Opt_enospc_debug:
 571                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 572                         break;
 573                 case Opt_defrag:
 574                         printk(KERN_INFO "btrfs: enabling auto defrag\n");
 575                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 576                         break;
 577                 case Opt_recovery:
 578                         printk(KERN_INFO "btrfs: enabling auto recovery\n");
 579                         btrfs_set_opt(info->mount_opt, RECOVERY);
 580                         break;
 581                 case Opt_skip_balance:
 582                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 583                         break;
 584 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 585                 case Opt_check_integrity_including_extent_data:
 586                         printk(KERN_INFO "btrfs: enabling check integrity"
 587                                " including extent data\n");
 588                         btrfs_set_opt(info->mount_opt,
 589                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 590                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 591                         break;
 592                 case Opt_check_integrity:
 593                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 594                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 595                         break;
 596                 case Opt_check_integrity_print_mask:
 597                         intarg = 0;
 598                         match_int(&args[0], &intarg);
 599                         if (intarg) {
 600                                 info->check_integrity_print_mask = intarg;
 601                                 printk(KERN_INFO "btrfs:"
 602                                        " check_integrity_print_mask 0x%x\n",
 603                                        info->check_integrity_print_mask);
 604                         }
 605                         break;
 606 #else
 607                 case Opt_check_integrity_including_extent_data:
 608                 case Opt_check_integrity:
 609                 case Opt_check_integrity_print_mask:
 610                         printk(KERN_ERR "btrfs: support for check_integrity*"
 611                                " not compiled in!\n");
 612                         ret = -EINVAL;
 613                         goto out;
 614 #endif
 615                 case Opt_fatal_errors:
 616                         if (strcmp(args[0].from, "panic") == 0)
 617                                 btrfs_set_opt(info->mount_opt,
 618                                               PANIC_ON_FATAL_ERROR);
 619                         else if (strcmp(args[0].from, "bug") == 0)
 620                                 btrfs_clear_opt(info->mount_opt,
 621                                               PANIC_ON_FATAL_ERROR);
 622                         else {
 623                                 ret = -EINVAL;
 624                                 goto out;
 625                         }
 626                         break;
 627                 case Opt_err:
 628                         printk(KERN_INFO "btrfs: unrecognized mount option "
 629                                "'%s'\n", p);
 630                         ret = -EINVAL;
 631                         goto out;
 632                 default:
 633                         break;
 634                 }
 635         }
 636 out:
 637         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 638                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 639         kfree(orig);
 640         return ret;
 641 }
 642
 643 /*
 644  * Parse mount options that are required early in the mount process.
 645  *
 646  * All other options will be parsed on much later in the mount process and
 647  * only when we need to allocate a new super block.
 648  */
 649 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 650                 void *holder, char **subvol_name, u64 *subvol_objectid,
 651                 struct btrfs_fs_devices **fs_devices)
 652 {
 653         substring_t args[MAX_OPT_ARGS];
 654         char *device_name, *opts, *orig, *p;
 655         int error = 0;
 656         int intarg;
 657
 658         if (!options)
 659                 return 0;
 660
 661         /*
 662          * strsep changes the string, duplicate it because parse_options
 663          * gets called twice
 664          */
 665         opts = kstrdup(options, GFP_KERNEL);
 666         if (!opts)
 667                 return -ENOMEM;
 668         orig = opts;
 669
 670         while ((p = strsep(&opts, ",")) != NULL) {
 671                 int token;
 672                 if (!*p)
 673                         continue;
 674
 675                 token = match_token(p, tokens, args);
 676                 switch (token) {
 677                 case Opt_subvol:
 678                         kfree(*subvol_name);
 679                         *subvol_name = match_strdup(&args[0]);
 680                         break;
 681                 case Opt_subvolid:
 682                         intarg = 0;
 683                         error = match_int(&args[0], &intarg);
 684                         if (!error) {
 685                                 /* we want the original fs_tree */
 686                                 if (!intarg)
 687                                         *subvol_objectid =
 688                                                 BTRFS_FS_TREE_OBJECTID;
 689                                 else
 690                                         *subvol_objectid = intarg;
 691                         }
 692                         break;
 693                 case Opt_subvolrootid:
 694                         printk(KERN_WARNING
 695                                 "btrfs: 'subvolrootid' mount option is deprecated and has no effect\n");
 696                         break;
 697                 case Opt_device:
 698                         device_name = match_strdup(&args[0]);
 699                         if (!device_name) {
 700                                 error = -ENOMEM;
 701                                 goto out;
 702                         }
 703                         error = btrfs_scan_one_device(device_name,
 704                                         flags, holder, fs_devices);
 705                         kfree(device_name);
 706                         if (error)
 707                                 goto out;
 708                         break;
 709                 default:
 710                         break;
 711                 }
 712         }
 713
 714 out:
 715         kfree(orig);
 716         return error;
 717 }
 718
 719 static struct dentry *get_default_root(struct super_block *sb,
 720                                        u64 subvol_objectid)
 721 {
 722         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 723         struct btrfs_root *root = fs_info->tree_root;
 724         struct btrfs_root *new_root;
 725         struct btrfs_dir_item *di;
 726         struct btrfs_path *path;
 727         struct btrfs_key location;
 728         struct inode *inode;
 729         u64 dir_id;
 730         int new = 0;
 731
 732         /*
 733          * We have a specific subvol we want to mount, just setup location and
 734          * go look up the root.
 735          */
 736         if (subvol_objectid) {
 737                 location.objectid = subvol_objectid;
 738                 location.type = BTRFS_ROOT_ITEM_KEY;
 739                 location.offset = (u64)-1;
 740                 goto find_root;
 741         }
 742
 743         path = btrfs_alloc_path();
 744         if (!path)
 745                 return ERR_PTR(-ENOMEM);
 746         path->leave_spinning = 1;
 747
 748         /*
 749          * Find the "default" dir item which points to the root item that we
 750          * will mount by default if we haven't been given a specific subvolume
 751          * to mount.
 752          */
 753         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 754         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 755         if (IS_ERR(di)) {
 756                 btrfs_free_path(path);
 757                 return ERR_CAST(di);
 758         }
 759         if (!di) {
 760                 /*
 761                  * Ok the default dir item isn't there.  This is weird since
 762                  * it's always been there, but don't freak out, just try and
 763                  * mount to root most subvolume.
 764                  */
 765                 btrfs_free_path(path);
 766                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 767                 new_root = fs_info->fs_root;
 768                 goto setup_root;
 769         }
 770
 771         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 772         btrfs_free_path(path);
 773
 774 find_root:
 775         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 776         if (IS_ERR(new_root))
 777                 return ERR_CAST(new_root);
 778
 779         if (btrfs_root_refs(&new_root->root_item) == 0)
 780                 return ERR_PTR(-ENOENT);
 781
 782         dir_id = btrfs_root_dirid(&new_root->root_item);
 783 setup_root:
 784         location.objectid = dir_id;
 785         location.type = BTRFS_INODE_ITEM_KEY;
 786         location.offset = 0;
 787
 788         inode = btrfs_iget(sb, &location, new_root, &new);
 789         if (IS_ERR(inode))
 790                 return ERR_CAST(inode);
 791
 792         /*
 793          * If we're just mounting the root most subvol put the inode and return
 794          * a reference to the dentry.  We will have already gotten a reference
 795          * to the inode in btrfs_fill_super so we're good to go.
 796          */
 797         if (!new && sb->s_root->d_inode == inode) {
 798                 iput(inode);
 799                 return dget(sb->s_root);
 800         }
 801
 802         return d_obtain_alias(inode);
 803 }
 804
 805 static int btrfs_fill_super(struct super_block *sb,
 806                             struct btrfs_fs_devices *fs_devices,
 807                             void *data, int silent)
 808 {
 809         struct inode *inode;
 810         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 811         struct btrfs_key key;
 812         int err;
 813
 814         sb->s_maxbytes = MAX_LFS_FILESIZE;
 815         sb->s_magic = BTRFS_SUPER_MAGIC;
 816         sb->s_op = &btrfs_super_ops;
 817         sb->s_d_op = &btrfs_dentry_operations;
 818         sb->s_export_op = &btrfs_export_ops;
 819         sb->s_xattr = btrfs_xattr_handlers;
 820         sb->s_time_gran = 1;
 821 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 822         sb->s_flags |= MS_POSIXACL;
 823 #endif
 824         sb->s_flags |= MS_I_VERSION;
 825         err = open_ctree(sb, fs_devices, (char *)data);
 826         if (err) {
 827                 printk("btrfs: open_ctree failed\n");
 828                 return err;
 829         }
 830
 831         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 832         key.type = BTRFS_INODE_ITEM_KEY;
 833         key.offset = 0;
 834         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 835         if (IS_ERR(inode)) {
 836                 err = PTR_ERR(inode);
 837                 goto fail_close;
 838         }
 839
 840         sb->s_root = d_make_root(inode);
 841         if (!sb->s_root) {
 842                 err = -ENOMEM;
 843                 goto fail_close;
 844         }
 845
 846         save_mount_options(sb, data);
 847         cleancache_init_fs(sb);
 848         sb->s_flags |= MS_ACTIVE;
 849         return 0;
 850
 851 fail_close:
 852         close_ctree(fs_info->tree_root);
 853         return err;
 854 }
 855
 856 int btrfs_sync_fs(struct super_block *sb, int wait)
 857 {
 858         struct btrfs_trans_handle *trans;
 859         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 860         struct btrfs_root *root = fs_info->tree_root;
 861
 862         trace_btrfs_sync_fs(wait);
 863
 864         if (!wait) {
 865                 filemap_flush(fs_info->btree_inode->i_mapping);
 866                 return 0;
 867         }
 868
 869         btrfs_wait_ordered_extents(root, 1);
 870
 871         trans = btrfs_attach_transaction_barrier(root);
 872         if (IS_ERR(trans)) {
 873                 /* no transaction, don't bother */
 874                 if (PTR_ERR(trans) == -ENOENT)
 875                         return 0;
 876                 return PTR_ERR(trans);
 877         }
 878         return btrfs_commit_transaction(trans, root);
 879 }
 880
 881 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 882 {
 883         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 884         struct btrfs_root *root = info->tree_root;
 885         char *compress_type;
 886
 887         if (btrfs_test_opt(root, DEGRADED))
 888                 seq_puts(seq, ",degraded");
 889         if (btrfs_test_opt(root, NODATASUM))
 890                 seq_puts(seq, ",nodatasum");
 891         if (btrfs_test_opt(root, NODATACOW))
 892                 seq_puts(seq, ",nodatacow");
 893         if (btrfs_test_opt(root, NOBARRIER))
 894                 seq_puts(seq, ",nobarrier");
 895         if (info->max_inline != 8192 * 1024)
 896                 seq_printf(seq, ",max_inline=%llu",
 897                            (unsigned long long)info->max_inline);
 898         if (info->alloc_start != 0)
 899                 seq_printf(seq, ",alloc_start=%llu",
 900                            (unsigned long long)info->alloc_start);
 901         if (info->thread_pool_size !=  min_t(unsigned long,
 902                                              num_online_cpus() + 2, 8))
 903                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 904         if (btrfs_test_opt(root, COMPRESS)) {
 905                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 906                         compress_type = "zlib";
 907                 else
 908                         compress_type = "lzo";
 909                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 910                         seq_printf(seq, ",compress-force=%s", compress_type);
 911                 else
 912                         seq_printf(seq, ",compress=%s", compress_type);
 913         }
 914         if (btrfs_test_opt(root, NOSSD))
 915                 seq_puts(seq, ",nossd");
 916         if (btrfs_test_opt(root, SSD_SPREAD))
 917                 seq_puts(seq, ",ssd_spread");
 918         else if (btrfs_test_opt(root, SSD))
 919                 seq_puts(seq, ",ssd");
 920         if (btrfs_test_opt(root, NOTREELOG))
 921                 seq_puts(seq, ",notreelog");
 922         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 923                 seq_puts(seq, ",flushoncommit");
 924         if (btrfs_test_opt(root, DISCARD))
 925                 seq_puts(seq, ",discard");
 926         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 927                 seq_puts(seq, ",noacl");
 928         if (btrfs_test_opt(root, SPACE_CACHE))
 929                 seq_puts(seq, ",space_cache");
 930         else
 931                 seq_puts(seq, ",nospace_cache");
 932         if (btrfs_test_opt(root, CLEAR_CACHE))
 933                 seq_puts(seq, ",clear_cache");
 934         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 935                 seq_puts(seq, ",user_subvol_rm_allowed");
 936         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 937                 seq_puts(seq, ",enospc_debug");
 938         if (btrfs_test_opt(root, AUTO_DEFRAG))
 939                 seq_puts(seq, ",autodefrag");
 940         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 941                 seq_puts(seq, ",inode_cache");
 942         if (btrfs_test_opt(root, SKIP_BALANCE))
 943                 seq_puts(seq, ",skip_balance");
 944         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 945                 seq_puts(seq, ",fatal_errors=panic");
 946         return 0;
 947 }
 948
 949 static int btrfs_test_super(struct super_block *s, void *data)
 950 {
 951         struct btrfs_fs_info *p = data;
 952         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 953
 954         return fs_info->fs_devices == p->fs_devices;
 955 }
 956
 957 static int btrfs_set_super(struct super_block *s, void *data)
 958 {
 959         int err = set_anon_super(s, data);
 960         if (!err)
 961                 s->s_fs_info = data;
 962         return err;
 963 }
 964
 965 /*
 966  * subvolumes are identified by ino 256
 967  */
 968 static inline int is_subvolume_inode(struct inode *inode)
 969 {
 970         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 971                 return 1;
 972         return 0;
 973 }
 974
 975 /*
 976  * This will strip out the subvol=%s argument for an argument string and add
 977  * subvolid=0 to make sure we get the actual tree root for path walking to the
 978  * subvol we want.
 979  */
 980 static char *setup_root_args(char *args)
 981 {
 982         unsigned len = strlen(args) + 2 + 1;
 983         char *src, *dst, *buf;
 984
 985         /*
 986          * We need the same args as before, but with this substitution:
 987          * s!subvol=[^,]+!subvolid=0!
 988          *
 989          * Since the replacement string is up to 2 bytes longer than the
 990          * original, allocate strlen(args) + 2 + 1 bytes.
 991          */
 992
 993         src = strstr(args, "subvol=");
 994         /* This shouldn't happen, but just in case.. */
 995         if (!src)
 996                 return NULL;
 997
 998         buf = dst = kmalloc(len, GFP_NOFS);
 999         if (!buf)
1000                 return NULL;
1001
1002         /*
1003          * If the subvol= arg is not at the start of the string,
1004          * copy whatever precedes it into buf.
1005          */
1006         if (src != args) {
1007                 *src++ = '\0';
1008                 strcpy(buf, args);
1009                 dst += strlen(args);
1010         }
1011
1012         strcpy(dst, "subvolid=0");
1013         dst += strlen("subvolid=0");
1014
1015         /*
1016          * If there is a "," after the original subvol=... string,
1017          * copy that suffix into our buffer.  Otherwise, we're done.
1018          */
1019         src = strchr(src, ',');
1020         if (src)
1021                 strcpy(dst, src);
1022
1023         return buf;
1024 }
1025
1026 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1027                                    const char *device_name, char *data)
1028 {
1029         struct dentry *root;
1030         struct vfsmount *mnt;
1031         char *newargs;
1032
1033         newargs = setup_root_args(data);
1034         if (!newargs)
1035                 return ERR_PTR(-ENOMEM);
1036         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1037                              newargs);
1038         kfree(newargs);
1039         if (IS_ERR(mnt))
1040                 return ERR_CAST(mnt);
1041
1042         root = mount_subtree(mnt, subvol_name);
1043
1044         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1045                 struct super_block *s = root->d_sb;
1046                 dput(root);
1047                 root = ERR_PTR(-EINVAL);
1048                 deactivate_locked_super(s);
1049                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1050                                 subvol_name);
1051         }
1052
1053         return root;
1054 }
1055
1056 /*
1057  * Find a superblock for the given device / mount point.
1058  *
1059  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1060  *        for multiple device setup.  Make sure to keep it in sync.
1061  */
1062 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1063                 const char *device_name, void *data)
1064 {
1065         struct block_device *bdev = NULL;
1066         struct super_block *s;
1067         struct dentry *root;
1068         struct btrfs_fs_devices *fs_devices = NULL;
1069         struct btrfs_fs_info *fs_info = NULL;
1070         fmode_t mode = FMODE_READ;
1071         char *subvol_name = NULL;
1072         u64 subvol_objectid = 0;
1073         int error = 0;
1074
1075         if (!(flags & MS_RDONLY))
1076                 mode |= FMODE_WRITE;
1077
1078         error = btrfs_parse_early_options(data, mode, fs_type,
1079                                           &subvol_name, &subvol_objectid,
1080                                           &fs_devices);
1081         if (error) {
1082                 kfree(subvol_name);
1083                 return ERR_PTR(error);
1084         }
1085
1086         if (subvol_name) {
1087                 root = mount_subvol(subvol_name, flags, device_name, data);
1088                 kfree(subvol_name);
1089                 return root;
1090         }
1091
1092         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1093         if (error)
1094                 return ERR_PTR(error);
1095
1096         /*
1097          * Setup a dummy root and fs_info for test/set super.  This is because
1098          * we don't actually fill this stuff out until open_ctree, but we need
1099          * it for searching for existing supers, so this lets us do that and
1100          * then open_ctree will properly initialize everything later.
1101          */
1102         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1103         if (!fs_info)
1104                 return ERR_PTR(-ENOMEM);
1105
1106         fs_info->fs_devices = fs_devices;
1107
1108         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1109         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1110         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1111                 error = -ENOMEM;
1112                 goto error_fs_info;
1113         }
1114
1115         error = btrfs_open_devices(fs_devices, mode, fs_type);
1116         if (error)
1117                 goto error_fs_info;
1118
1119         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1120                 error = -EACCES;
1121                 goto error_close_devices;
1122         }
1123
1124         bdev = fs_devices->latest_bdev;
1125         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1126                  fs_info);
1127         if (IS_ERR(s)) {
1128                 error = PTR_ERR(s);
1129                 goto error_close_devices;
1130         }
1131
1132         if (s->s_root) {
1133                 btrfs_close_devices(fs_devices);
1134                 free_fs_info(fs_info);
1135                 if ((flags ^ s->s_flags) & MS_RDONLY)
1136                         error = -EBUSY;
1137         } else {
1138                 char b[BDEVNAME_SIZE];
1139
1140                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1141                 btrfs_sb(s)->bdev_holder = fs_type;
1142                 error = btrfs_fill_super(s, fs_devices, data,
1143                                          flags & MS_SILENT ? 1 : 0);
1144         }
1145
1146         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1147         if (IS_ERR(root))
1148                 deactivate_locked_super(s);
1149
1150         return root;
1151
1152 error_close_devices:
1153         btrfs_close_devices(fs_devices);
1154 error_fs_info:
1155         free_fs_info(fs_info);
1156         return ERR_PTR(error);
1157 }
1158
1159 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1160 {
1161         spin_lock_irq(&workers->lock);
1162         workers->max_workers = new_limit;
1163         spin_unlock_irq(&workers->lock);
1164 }
1165
1166 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1167                                      int new_pool_size, int old_pool_size)
1168 {
1169         if (new_pool_size == old_pool_size)
1170                 return;
1171
1172         fs_info->thread_pool_size = new_pool_size;
1173
1174         printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1175                old_pool_size, new_pool_size);
1176
1177         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1178         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1179         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1180         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1181         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1182         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1183         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1184         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1185         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1186         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1187         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1188         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1189         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1190         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1191                               new_pool_size);
1192 }
1193
1194 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1195 {
1196         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1197 }
1198
1199 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1200                                        unsigned long old_opts, int flags)
1201 {
1202         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1203             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1204              (flags & MS_RDONLY))) {
1205                 /* wait for any defraggers to finish */
1206                 wait_event(fs_info->transaction_wait,
1207                            (atomic_read(&fs_info->defrag_running) == 0));
1208                 if (flags & MS_RDONLY)
1209                         sync_filesystem(fs_info->sb);
1210         }
1211 }
1212
1213 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1214                                          unsigned long old_opts)
1215 {
1216         /*
1217          * We need cleanup all defragable inodes if the autodefragment is
1218          * close or the fs is R/O.
1219          */
1220         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1221             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1222              (fs_info->sb->s_flags & MS_RDONLY))) {
1223                 btrfs_cleanup_defrag_inodes(fs_info);
1224         }
1225
1226         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1227 }
1228
1229 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1230 {
1231         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1232         struct btrfs_root *root = fs_info->tree_root;
1233         unsigned old_flags = sb->s_flags;
1234         unsigned long old_opts = fs_info->mount_opt;
1235         unsigned long old_compress_type = fs_info->compress_type;
1236         u64 old_max_inline = fs_info->max_inline;
1237         u64 old_alloc_start = fs_info->alloc_start;
1238         int old_thread_pool_size = fs_info->thread_pool_size;
1239         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1240         int ret;
1241
1242         sync_filesystem(sb);
1243         btrfs_remount_prepare(fs_info);
1244
1245         ret = btrfs_parse_options(root, data);
1246         if (ret) {
1247                 ret = -EINVAL;
1248                 goto restore;
1249         }
1250
1251         btrfs_remount_begin(fs_info, old_opts, *flags);
1252         btrfs_resize_thread_pool(fs_info,
1253                 fs_info->thread_pool_size, old_thread_pool_size);
1254
1255         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1256                 goto out;
1257
1258         if (*flags & MS_RDONLY) {
1259                 /*
1260                  * this also happens on 'umount -rf' or on shutdown, when
1261                  * the filesystem is busy.
1262                  */
1263                 sb->s_flags |= MS_RDONLY;
1264
1265                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1266                 btrfs_scrub_cancel(fs_info);
1267                 btrfs_pause_balance(fs_info);
1268
1269                 ret = btrfs_commit_super(root);
1270                 if (ret)
1271                         goto restore;
1272         } else {
1273                 if (fs_info->fs_devices->rw_devices == 0) {
1274                         ret = -EACCES;
1275                         goto restore;
1276                 }
1277
1278                 if (fs_info->fs_devices->missing_devices >
1279                      fs_info->num_tolerated_disk_barrier_failures &&
1280                     !(*flags & MS_RDONLY)) {
1281                         printk(KERN_WARNING
1282                                "Btrfs: too many missing devices, writeable remount is not allowed\n");
1283                         ret = -EACCES;
1284                         goto restore;
1285                 }
1286
1287                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1288                         ret = -EINVAL;
1289                         goto restore;
1290                 }
1291
1292                 ret = btrfs_cleanup_fs_roots(fs_info);
1293                 if (ret)
1294                         goto restore;
1295
1296                 /* recover relocation */
1297                 ret = btrfs_recover_relocation(root);
1298                 if (ret)
1299                         goto restore;
1300
1301                 ret = btrfs_resume_balance_async(fs_info);
1302                 if (ret)
1303                         goto restore;
1304
1305                 ret = btrfs_resume_dev_replace_async(fs_info);
1306                 if (ret) {
1307                         pr_warn("btrfs: failed to resume dev_replace\n");
1308                         goto restore;
1309                 }
1310                 sb->s_flags &= ~MS_RDONLY;
1311         }
1312 out:
1313         btrfs_remount_cleanup(fs_info, old_opts);
1314         return 0;
1315
1316 restore:
1317         /* We've hit an error - don't reset MS_RDONLY */
1318         if (sb->s_flags & MS_RDONLY)
1319                 old_flags |= MS_RDONLY;
1320         sb->s_flags = old_flags;
1321         fs_info->mount_opt = old_opts;
1322         fs_info->compress_type = old_compress_type;
1323         fs_info->max_inline = old_max_inline;
1324         mutex_lock(&fs_info->chunk_mutex);
1325         fs_info->alloc_start = old_alloc_start;
1326         mutex_unlock(&fs_info->chunk_mutex);
1327         btrfs_resize_thread_pool(fs_info,
1328                 old_thread_pool_size, fs_info->thread_pool_size);
1329         fs_info->metadata_ratio = old_metadata_ratio;
1330         btrfs_remount_cleanup(fs_info, old_opts);
1331         return ret;
1332 }
1333
1334 /* Used to sort the devices by max_avail(descending sort) */
1335 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1336                                        const void *dev_info2)
1337 {
1338         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1339             ((struct btrfs_device_info *)dev_info2)->max_avail)
1340                 return -1;
1341         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1342                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1343                 return 1;
1344         else
1345         return 0;
1346 }
1347
1348 /*
1349  * sort the devices by max_avail, in which max free extent size of each device
1350  * is stored.(Descending Sort)
1351  */
1352 static inline void btrfs_descending_sort_devices(
1353                                         struct btrfs_device_info *devices,
1354                                         size_t nr_devices)
1355 {
1356         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1357              btrfs_cmp_device_free_bytes, NULL);
1358 }
1359
1360 /*
1361  * The helper to calc the free space on the devices that can be used to store
1362  * file data.
1363  */
1364 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1365 {
1366         struct btrfs_fs_info *fs_info = root->fs_info;
1367         struct btrfs_device_info *devices_info;
1368         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1369         struct btrfs_device *device;
1370         u64 skip_space;
1371         u64 type;
1372         u64 avail_space;
1373         u64 used_space;
1374         u64 min_stripe_size;
1375         int min_stripes = 1, num_stripes = 1;
1376         int i = 0, nr_devices;
1377         int ret;
1378
1379         nr_devices = fs_info->fs_devices->open_devices;
1380         BUG_ON(!nr_devices);
1381
1382         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1383                                GFP_NOFS);
1384         if (!devices_info)
1385                 return -ENOMEM;
1386
1387         /* calc min stripe number for data space alloction */
1388         type = btrfs_get_alloc_profile(root, 1);
1389         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1390                 min_stripes = 2;
1391                 num_stripes = nr_devices;
1392         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1393                 min_stripes = 2;
1394                 num_stripes = 2;
1395         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1396                 min_stripes = 4;
1397                 num_stripes = 4;
1398         }
1399
1400         if (type & BTRFS_BLOCK_GROUP_DUP)
1401                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1402         else
1403                 min_stripe_size = BTRFS_STRIPE_LEN;
1404
1405         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1406                 if (!device->in_fs_metadata || !device->bdev ||
1407                     device->is_tgtdev_for_dev_replace)
1408                         continue;
1409
1410                 avail_space = device->total_bytes - device->bytes_used;
1411
1412                 /* align with stripe_len */
1413                 do_div(avail_space, BTRFS_STRIPE_LEN);
1414                 avail_space *= BTRFS_STRIPE_LEN;
1415
1416                 /*
1417                  * In order to avoid overwritting the superblock on the drive,
1418                  * btrfs starts at an offset of at least 1MB when doing chunk
1419                  * allocation.
1420                  */
1421                 skip_space = 1024 * 1024;
1422
1423                 /* user can set the offset in fs_info->alloc_start. */
1424                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1425                     device->total_bytes)
1426                         skip_space = max(fs_info->alloc_start, skip_space);
1427
1428                 /*
1429                  * btrfs can not use the free space in [0, skip_space - 1],
1430                  * we must subtract it from the total. In order to implement
1431                  * it, we account the used space in this range first.
1432                  */
1433                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1434                                                      &used_space);
1435                 if (ret) {
1436                         kfree(devices_info);
1437                         return ret;
1438                 }
1439
1440                 /* calc the free space in [0, skip_space - 1] */
1441                 skip_space -= used_space;
1442
1443                 /*
1444                  * we can use the free space in [0, skip_space - 1], subtract
1445                  * it from the total.
1446                  */
1447                 if (avail_space && avail_space >= skip_space)
1448                         avail_space -= skip_space;
1449                 else
1450                         avail_space = 0;
1451
1452                 if (avail_space < min_stripe_size)
1453                         continue;
1454
1455                 devices_info[i].dev = device;
1456                 devices_info[i].max_avail = avail_space;
1457
1458                 i++;
1459         }
1460
1461         nr_devices = i;
1462
1463         btrfs_descending_sort_devices(devices_info, nr_devices);
1464
1465         i = nr_devices - 1;
1466         avail_space = 0;
1467         while (nr_devices >= min_stripes) {
1468                 if (num_stripes > nr_devices)
1469                         num_stripes = nr_devices;
1470
1471                 if (devices_info[i].max_avail >= min_stripe_size) {
1472                         int j;
1473                         u64 alloc_size;
1474
1475                         avail_space += devices_info[i].max_avail * num_stripes;
1476                         alloc_size = devices_info[i].max_avail;
1477                         for (j = i + 1 - num_stripes; j <= i; j++)
1478                                 devices_info[j].max_avail -= alloc_size;
1479                 }
1480                 i--;
1481                 nr_devices--;
1482         }
1483
1484         kfree(devices_info);
1485         *free_bytes = avail_space;
1486         return 0;
1487 }
1488
1489 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1490 {
1491         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1492         struct btrfs_super_block *disk_super = fs_info->super_copy;
1493         struct list_head *head = &fs_info->space_info;
1494         struct btrfs_space_info *found;
1495         u64 total_used = 0;
1496         u64 total_free_data = 0;
1497         int bits = dentry->d_sb->s_blocksize_bits;
1498         __be32 *fsid = (__be32 *)fs_info->fsid;
1499         int ret;
1500
1501         /* holding chunk_muext to avoid allocating new chunks */
1502         mutex_lock(&fs_info->chunk_mutex);
1503         rcu_read_lock();
1504         list_for_each_entry_rcu(found, head, list) {
1505                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1506                         total_free_data += found->disk_total - found->disk_used;
1507                         total_free_data -=
1508                                 btrfs_account_ro_block_groups_free_space(found);
1509                 }
1510
1511                 total_used += found->disk_used;
1512         }
1513         rcu_read_unlock();
1514
1515         buf->f_namelen = BTRFS_NAME_LEN;
1516         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1517         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1518         buf->f_bsize = dentry->d_sb->s_blocksize;
1519         buf->f_type = BTRFS_SUPER_MAGIC;
1520         buf->f_bavail = total_free_data;
1521         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1522         if (ret) {
1523                 mutex_unlock(&fs_info->chunk_mutex);
1524                 return ret;
1525         }
1526         buf->f_bavail += total_free_data;
1527         buf->f_bavail = buf->f_bavail >> bits;
1528         mutex_unlock(&fs_info->chunk_mutex);
1529
1530         /* We treat it as constant endianness (it doesn't matter _which_)
1531            because we want the fsid to come out the same whether mounted
1532            on a big-endian or little-endian host */
1533         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1534         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1535         /* Mask in the root object ID too, to disambiguate subvols */
1536         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1537         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1538
1539         return 0;
1540 }
1541
1542 static void btrfs_kill_super(struct super_block *sb)
1543 {
1544         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1545         kill_anon_super(sb);
1546         free_fs_info(fs_info);
1547 }
1548
1549 static struct file_system_type btrfs_fs_type = {
1550         .owner          = THIS_MODULE,
1551         .name           = "btrfs",
1552         .mount          = btrfs_mount,
1553         .kill_sb        = btrfs_kill_super,
1554         .fs_flags       = FS_REQUIRES_DEV,
1555 };
1556 MODULE_ALIAS_FS("btrfs");
1557
1558 /*
1559  * used by btrfsctl to scan devices when no FS is mounted
1560  */
1561 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1562                                 unsigned long arg)
1563 {
1564         struct btrfs_ioctl_vol_args *vol;
1565         struct btrfs_fs_devices *fs_devices;
1566         int ret = -ENOTTY;
1567
1568         if (!capable(CAP_SYS_ADMIN))
1569                 return -EPERM;
1570
1571         vol = memdup_user((void __user *)arg, sizeof(*vol));
1572         if (IS_ERR(vol))
1573                 return PTR_ERR(vol);
1574
1575         switch (cmd) {
1576         case BTRFS_IOC_SCAN_DEV:
1577                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1578                                             &btrfs_fs_type, &fs_devices);
1579                 break;
1580         case BTRFS_IOC_DEVICES_READY:
1581                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1582                                             &btrfs_fs_type, &fs_devices);
1583                 if (ret)
1584                         break;
1585                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1586                 break;
1587         }
1588
1589         kfree(vol);
1590         return ret;
1591 }
1592
1593 static int btrfs_freeze(struct super_block *sb)
1594 {
1595         struct btrfs_trans_handle *trans;
1596         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1597
1598         trans = btrfs_attach_transaction_barrier(root);
1599         if (IS_ERR(trans)) {
1600                 /* no transaction, don't bother */
1601                 if (PTR_ERR(trans) == -ENOENT)
1602                         return 0;
1603                 return PTR_ERR(trans);
1604         }
1605         return btrfs_commit_transaction(trans, root);
1606 }
1607
1608 static int btrfs_unfreeze(struct super_block *sb)
1609 {
1610         return 0;
1611 }
1612
1613 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1614 {
1615         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1616         struct btrfs_fs_devices *cur_devices;
1617         struct btrfs_device *dev, *first_dev = NULL;
1618         struct list_head *head;
1619         struct rcu_string *name;
1620
1621         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1622         cur_devices = fs_info->fs_devices;
1623         while (cur_devices) {
1624                 head = &cur_devices->devices;
1625                 list_for_each_entry(dev, head, dev_list) {
1626                         if (dev->missing)
1627                                 continue;
1628                         if (!first_dev || dev->devid < first_dev->devid)
1629                                 first_dev = dev;
1630                 }
1631                 cur_devices = cur_devices->seed;
1632         }
1633
1634         if (first_dev) {
1635                 rcu_read_lock();
1636                 name = rcu_dereference(first_dev->name);
1637                 seq_escape(m, name->str, " \t\n\\");
1638                 rcu_read_unlock();
1639         } else {
1640                 WARN_ON(1);
1641         }
1642         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1643         return 0;
1644 }
1645
1646 static const struct super_operations btrfs_super_ops = {
1647         .drop_inode     = btrfs_drop_inode,
1648         .evict_inode    = btrfs_evict_inode,
1649         .put_super      = btrfs_put_super,
1650         .sync_fs        = btrfs_sync_fs,
1651         .show_options   = btrfs_show_options,
1652         .show_devname   = btrfs_show_devname,
1653         .write_inode    = btrfs_write_inode,
1654         .alloc_inode    = btrfs_alloc_inode,
1655         .destroy_inode  = btrfs_destroy_inode,
1656         .statfs         = btrfs_statfs,
1657         .remount_fs     = btrfs_remount,
1658         .freeze_fs      = btrfs_freeze,
1659         .unfreeze_fs    = btrfs_unfreeze,
1660 };
1661
1662 static const struct file_operations btrfs_ctl_fops = {
1663         .unlocked_ioctl  = btrfs_control_ioctl,
1664         .compat_ioctl = btrfs_control_ioctl,
1665         .owner   = THIS_MODULE,
1666         .llseek = noop_llseek,
1667 };
1668
1669 static struct miscdevice btrfs_misc = {
1670         .minor          = BTRFS_MINOR,
1671         .name           = "btrfs-control",
1672         .fops           = &btrfs_ctl_fops
1673 };
1674
1675 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1676 MODULE_ALIAS("devname:btrfs-control");
1677
1678 static int btrfs_interface_init(void)
1679 {
1680         return misc_register(&btrfs_misc);
1681 }
1682
1683 static void btrfs_interface_exit(void)
1684 {
1685         if (misc_deregister(&btrfs_misc) < 0)
1686                 printk(KERN_INFO "btrfs: misc_deregister failed for control device\n");
1687 }
1688
1689 static int __init init_btrfs_fs(void)
1690 {
1691         int err;
1692
1693         err = btrfs_init_sysfs();
1694         if (err)
1695                 return err;
1696
1697         btrfs_init_compress();
1698
1699         err = btrfs_init_cachep();
1700         if (err)
1701                 goto free_compress;
1702
1703         err = extent_io_init();
1704         if (err)
1705                 goto free_cachep;
1706
1707         err = extent_map_init();
1708         if (err)
1709                 goto free_extent_io;
1710
1711         err = ordered_data_init();
1712         if (err)
1713                 goto free_extent_map;
1714
1715         err = btrfs_delayed_inode_init();
1716         if (err)
1717                 goto free_ordered_data;
1718
1719         err = btrfs_auto_defrag_init();
1720         if (err)
1721                 goto free_delayed_inode;
1722
1723         err = btrfs_delayed_ref_init();
1724         if (err)
1725                 goto free_auto_defrag;
1726
1727         err = btrfs_interface_init();
1728         if (err)
1729                 goto free_delayed_ref;
1730
1731         err = register_filesystem(&btrfs_fs_type);
1732         if (err)
1733                 goto unregister_ioctl;
1734
1735         btrfs_init_lockdep();
1736
1737 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1738         btrfs_test_free_space_cache();
1739 #endif
1740
1741         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1742         return 0;
1743
1744 unregister_ioctl:
1745         btrfs_interface_exit();
1746 free_delayed_ref:
1747         btrfs_delayed_ref_exit();
1748 free_auto_defrag:
1749         btrfs_auto_defrag_exit();
1750 free_delayed_inode:
1751         btrfs_delayed_inode_exit();
1752 free_ordered_data:
1753         ordered_data_exit();
1754 free_extent_map:
1755         extent_map_exit();
1756 free_extent_io:
1757         extent_io_exit();
1758 free_cachep:
1759         btrfs_destroy_cachep();
1760 free_compress:
1761         btrfs_exit_compress();
1762         btrfs_exit_sysfs();
1763         return err;
1764 }
1765
1766 static void __exit exit_btrfs_fs(void)
1767 {
1768         btrfs_destroy_cachep();
1769         btrfs_delayed_ref_exit();
1770         btrfs_auto_defrag_exit();
1771         btrfs_delayed_inode_exit();
1772         ordered_data_exit();
1773         extent_map_exit();
1774         extent_io_exit();
1775         btrfs_interface_exit();
1776         unregister_filesystem(&btrfs_fs_type);
1777         btrfs_exit_sysfs();
1778         btrfs_cleanup_fs_uuids();
1779         btrfs_exit_compress();
1780 }
1781
1782 module_init(init_btrfs_fs)
1783 module_exit(exit_btrfs_fs)
1784
1785 MODULE_LICENSE("GPL");