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