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 "export.h"
  55 #include "compression.h"
  56 #include "rcu-string.h"
  57 #include "dev-replace.h"
  58 #include "free-space-cache.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/btrfs.h>
  62
  63 static const struct super_operations btrfs_super_ops;
  64 static struct file_system_type btrfs_fs_type;
  65
  66 static const char *btrfs_decode_error(int errno)
  67 {
  68         char *errstr = "unknown";
  69
  70         switch (errno) {
  71         case -EIO:
  72                 errstr = "IO failure";
  73                 break;
  74         case -ENOMEM:
  75                 errstr = "Out of memory";
  76                 break;
  77         case -EROFS:
  78                 errstr = "Readonly filesystem";
  79                 break;
  80         case -EEXIST:
  81                 errstr = "Object already exists";
  82                 break;
  83         case -ENOSPC:
  84                 errstr = "No space left";
  85                 break;
  86         case -ENOENT:
  87                 errstr = "No such entry";
  88                 break;
  89         }
  90
  91         return errstr;
  92 }
  93
  94 static void save_error_info(struct btrfs_fs_info *fs_info)
  95 {
  96         /*
  97          * today we only save the error info into ram.  Long term we'll
  98          * also send it down to the disk
  99          */
 100         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 101 }
 102
 103 /* btrfs handle error by forcing the filesystem readonly */
 104 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 105 {
 106         struct super_block *sb = fs_info->sb;
 107
 108         if (sb->s_flags & MS_RDONLY)
 109                 return;
 110
 111         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 112                 sb->s_flags |= MS_RDONLY;
 113                 btrfs_info(fs_info, "forced readonly");
 114                 /*
 115                  * Note that a running device replace operation is not
 116                  * canceled here although there is no way to update
 117                  * the progress. It would add the risk of a deadlock,
 118                  * therefore the canceling is ommited. The only penalty
 119                  * is that some I/O remains active until the procedure
 120                  * completes. The next time when the filesystem is
 121                  * mounted writeable again, the device replace
 122                  * operation continues.
 123                  */
 124         }
 125 }
 126
 127 #ifdef CONFIG_PRINTK
 128 /*
 129  * __btrfs_std_error decodes expected errors from the caller and
 130  * invokes the approciate error response.
 131  */
 132 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 133                        unsigned int line, int errno, const char *fmt, ...)
 134 {
 135         struct super_block *sb = fs_info->sb;
 136         const char *errstr;
 137
 138         /*
 139          * Special case: if the error is EROFS, and we're already
 140          * under MS_RDONLY, then it is safe here.
 141          */
 142         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 143                 return;
 144
 145         errstr = btrfs_decode_error(errno);
 146         if (fmt) {
 147                 struct va_format vaf;
 148                 va_list args;
 149
 150                 va_start(args, fmt);
 151                 vaf.fmt = fmt;
 152                 vaf.va = &args;
 153
 154                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 155                         sb->s_id, function, line, errno, errstr, &vaf);
 156                 va_end(args);
 157         } else {
 158                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: errno=%d %s\n",
 159                         sb->s_id, function, line, errno, errstr);
 160         }
 161
 162         /* Don't go through full error handling during mount */
 163         save_error_info(fs_info);
 164         if (sb->s_flags & MS_BORN)
 165                 btrfs_handle_error(fs_info);
 166 }
 167
 168 static const char * const logtypes[] = {
 169         "emergency",
 170         "alert",
 171         "critical",
 172         "error",
 173         "warning",
 174         "notice",
 175         "info",
 176         "debug",
 177 };
 178
 179 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 180 {
 181         struct super_block *sb = fs_info->sb;
 182         char lvl[4];
 183         struct va_format vaf;
 184         va_list args;
 185         const char *type = logtypes[4];
 186         int kern_level;
 187
 188         va_start(args, fmt);
 189
 190         kern_level = printk_get_level(fmt);
 191         if (kern_level) {
 192                 size_t size = printk_skip_level(fmt) - fmt;
 193                 memcpy(lvl, fmt,  size);
 194                 lvl[size] = '\0';
 195                 fmt += size;
 196                 type = logtypes[kern_level - '0'];
 197         } else
 198                 *lvl = '\0';
 199
 200         vaf.fmt = fmt;
 201         vaf.va = &args;
 202
 203         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 204
 205         va_end(args);
 206 }
 207
 208 #else
 209
 210 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 211                        unsigned int line, int errno, const char *fmt, ...)
 212 {
 213         struct super_block *sb = fs_info->sb;
 214
 215         /*
 216          * Special case: if the error is EROFS, and we're already
 217          * under MS_RDONLY, then it is safe here.
 218          */
 219         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 220                 return;
 221
 222         /* Don't go through full error handling during mount */
 223         if (sb->s_flags & MS_BORN) {
 224                 save_error_info(fs_info);
 225                 btrfs_handle_error(fs_info);
 226         }
 227 }
 228 #endif
 229
 230 /*
 231  * We only mark the transaction aborted and then set the file system read-only.
 232  * This will prevent new transactions from starting or trying to join this
 233  * one.
 234  *
 235  * This means that error recovery at the call site is limited to freeing
 236  * any local memory allocations and passing the error code up without
 237  * further cleanup. The transaction should complete as it normally would
 238  * in the call path but will return -EIO.
 239  *
 240  * We'll complete the cleanup in btrfs_end_transaction and
 241  * btrfs_commit_transaction.
 242  */
 243 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 244                                struct btrfs_root *root, const char *function,
 245                                unsigned int line, int errno)
 246 {
 247         /*
 248          * Report first abort since mount
 249          */
 250         if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,
 251                                 &root->fs_info->fs_state)) {
 252                 WARN(1, KERN_DEBUG "btrfs: Transaction aborted (error %d)\n",
 253                                 errno);
 254         }
 255         trans->aborted = errno;
 256         /* Nothing used. The other threads that have joined this
 257          * transaction may be able to continue. */
 258         if (!trans->blocks_used) {
 259                 const char *errstr;
 260
 261                 errstr = btrfs_decode_error(errno);
 262                 btrfs_warn(root->fs_info,
 263                            "%s:%d: Aborting unused transaction(%s).",
 264                            function, line, errstr);
 265                 return;
 266         }
 267         ACCESS_ONCE(trans->transaction->aborted) = errno;
 268         /* Wake up anybody who may be waiting on this transaction */
 269         wake_up(&root->fs_info->transaction_wait);
 270         wake_up(&root->fs_info->transaction_blocked_wait);
 271         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 272 }
 273 /*
 274  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 275  * issues an alert, and either panics or BUGs, depending on mount options.
 276  */
 277 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 278                    unsigned int line, int errno, const char *fmt, ...)
 279 {
 280         char *s_id = "<unknown>";
 281         const char *errstr;
 282         struct va_format vaf = { .fmt = fmt };
 283         va_list args;
 284
 285         if (fs_info)
 286                 s_id = fs_info->sb->s_id;
 287
 288         va_start(args, fmt);
 289         vaf.va = &args;
 290
 291         errstr = btrfs_decode_error(errno);
 292         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 293                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 294                         s_id, function, line, &vaf, errno, errstr);
 295
 296         printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 297                s_id, function, line, &vaf, errno, errstr);
 298         va_end(args);
 299         /* Caller calls BUG() */
 300 }
 301
 302 static void btrfs_put_super(struct super_block *sb)
 303 {
 304         (void)close_ctree(btrfs_sb(sb)->tree_root);
 305         /* FIXME: need to fix VFS to return error? */
 306         /* AV: return it _where_?  ->put_super() can be triggered by any number
 307          * of async events, up to and including delivery of SIGKILL to the
 308          * last process that kept it busy.  Or segfault in the aforementioned
 309          * process...  Whom would you report that to?
 310          */
 311 }
 312
 313 enum {
 314         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 315         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 316         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 317         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 318         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 319         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 320         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 321         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 322         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 323         Opt_check_integrity_print_mask, Opt_fatal_errors,
 324         Opt_err,
 325 };
 326
 327 static match_table_t tokens = {
 328         {Opt_degraded, "degraded"},
 329         {Opt_subvol, "subvol=%s"},
 330         {Opt_subvolid, "subvolid=%d"},
 331         {Opt_device, "device=%s"},
 332         {Opt_nodatasum, "nodatasum"},
 333         {Opt_nodatacow, "nodatacow"},
 334         {Opt_nobarrier, "nobarrier"},
 335         {Opt_max_inline, "max_inline=%s"},
 336         {Opt_alloc_start, "alloc_start=%s"},
 337         {Opt_thread_pool, "thread_pool=%d"},
 338         {Opt_compress, "compress"},
 339         {Opt_compress_type, "compress=%s"},
 340         {Opt_compress_force, "compress-force"},
 341         {Opt_compress_force_type, "compress-force=%s"},
 342         {Opt_ssd, "ssd"},
 343         {Opt_ssd_spread, "ssd_spread"},
 344         {Opt_nossd, "nossd"},
 345         {Opt_noacl, "noacl"},
 346         {Opt_notreelog, "notreelog"},
 347         {Opt_flushoncommit, "flushoncommit"},
 348         {Opt_ratio, "metadata_ratio=%d"},
 349         {Opt_discard, "discard"},
 350         {Opt_space_cache, "space_cache"},
 351         {Opt_clear_cache, "clear_cache"},
 352         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 353         {Opt_enospc_debug, "enospc_debug"},
 354         {Opt_subvolrootid, "subvolrootid=%d"},
 355         {Opt_defrag, "autodefrag"},
 356         {Opt_inode_cache, "inode_cache"},
 357         {Opt_no_space_cache, "nospace_cache"},
 358         {Opt_recovery, "recovery"},
 359         {Opt_skip_balance, "skip_balance"},
 360         {Opt_check_integrity, "check_int"},
 361         {Opt_check_integrity_including_extent_data, "check_int_data"},
 362         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 363         {Opt_fatal_errors, "fatal_errors=%s"},
 364         {Opt_err, NULL},
 365 };
 366
 367 /*
 368  * Regular mount options parser.  Everything that is needed only when
 369  * reading in a new superblock is parsed here.
 370  * XXX JDM: This needs to be cleaned up for remount.
 371  */
 372 int btrfs_parse_options(struct btrfs_root *root, char *options)
 373 {
 374         struct btrfs_fs_info *info = root->fs_info;
 375         substring_t args[MAX_OPT_ARGS];
 376         char *p, *num, *orig = NULL;
 377         u64 cache_gen;
 378         int intarg;
 379         int ret = 0;
 380         char *compress_type;
 381         bool compress_force = false;
 382
 383         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 384         if (cache_gen)
 385                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 386
 387         if (!options)
 388                 goto out;
 389
 390         /*
 391          * strsep changes the string, duplicate it because parse_options
 392          * gets called twice
 393          */
 394         options = kstrdup(options, GFP_NOFS);
 395         if (!options)
 396                 return -ENOMEM;
 397
 398         orig = options;
 399
 400         while ((p = strsep(&options, ",")) != NULL) {
 401                 int token;
 402                 if (!*p)
 403                         continue;
 404
 405                 token = match_token(p, tokens, args);
 406                 switch (token) {
 407                 case Opt_degraded:
 408                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 409                         btrfs_set_opt(info->mount_opt, DEGRADED);
 410                         break;
 411                 case Opt_subvol:
 412                 case Opt_subvolid:
 413                 case Opt_subvolrootid:
 414                 case Opt_device:
 415                         /*
 416                          * These are parsed by btrfs_parse_early_options
 417                          * and can be happily ignored here.
 418                          */
 419                         break;
 420                 case Opt_nodatasum:
 421                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 422                         btrfs_set_opt(info->mount_opt, NODATASUM);
 423                         break;
 424                 case Opt_nodatacow:
 425                         if (!btrfs_test_opt(root, COMPRESS) ||
 426                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
 427                                         printk(KERN_INFO "btrfs: setting nodatacow, compression disabled\n");
 428                         } else {
 429                                 printk(KERN_INFO "btrfs: setting nodatacow\n");
 430                         }
 431                         info->compress_type = BTRFS_COMPRESS_NONE;
 432                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 433                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 434                         btrfs_set_opt(info->mount_opt, NODATACOW);
 435                         btrfs_set_opt(info->mount_opt, NODATASUM);
 436                         break;
 437                 case Opt_compress_force:
 438                 case Opt_compress_force_type:
 439                         compress_force = true;
 440                         /* Fallthrough */
 441                 case Opt_compress:
 442                 case Opt_compress_type:
 443                         if (token == Opt_compress ||
 444                             token == Opt_compress_force ||
 445                             strcmp(args[0].from, "zlib") == 0) {
 446                                 compress_type = "zlib";
 447                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 448                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 449                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 450                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 451                         } else if (strcmp(args[0].from, "lzo") == 0) {
 452                                 compress_type = "lzo";
 453                                 info->compress_type = BTRFS_COMPRESS_LZO;
 454                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 455                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 456                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 457                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 458                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 459                                 compress_type = "no";
 460                                 info->compress_type = BTRFS_COMPRESS_NONE;
 461                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 462                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 463                                 compress_force = false;
 464                         } else {
 465                                 ret = -EINVAL;
 466                                 goto out;
 467                         }
 468
 469                         if (compress_force) {
 470                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 471                                 pr_info("btrfs: force %s compression\n",
 472                                         compress_type);
 473                         } else
 474                                 pr_info("btrfs: use %s compression\n",
 475                                         compress_type);
 476                         break;
 477                 case Opt_ssd:
 478                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 479                         btrfs_set_opt(info->mount_opt, SSD);
 480                         break;
 481                 case Opt_ssd_spread:
 482                         printk(KERN_INFO "btrfs: use spread ssd "
 483                                "allocation scheme\n");
 484                         btrfs_set_opt(info->mount_opt, SSD);
 485                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 486                         break;
 487                 case Opt_nossd:
 488                         printk(KERN_INFO "btrfs: not using ssd allocation "
 489                                "scheme\n");
 490                         btrfs_set_opt(info->mount_opt, NOSSD);
 491                         btrfs_clear_opt(info->mount_opt, SSD);
 492                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 493                         break;
 494                 case Opt_nobarrier:
 495                         printk(KERN_INFO "btrfs: turning off barriers\n");
 496                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 497                         break;
 498                 case Opt_thread_pool:
 499                         intarg = 0;
 500                         match_int(&args[0], &intarg);
 501                         if (intarg)
 502                                 info->thread_pool_size = intarg;
 503                         break;
 504                 case Opt_max_inline:
 505                         num = match_strdup(&args[0]);
 506                         if (num) {
 507                                 info->max_inline = memparse(num, NULL);
 508                                 kfree(num);
 509
 510                                 if (info->max_inline) {
 511                                         info->max_inline = max_t(u64,
 512                                                 info->max_inline,
 513                                                 root->sectorsize);
 514                                 }
 515                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 516                                         (unsigned long long)info->max_inline);
 517                         }
 518                         break;
 519                 case Opt_alloc_start:
 520                         num = match_strdup(&args[0]);
 521                         if (num) {
 522                                 mutex_lock(&info->chunk_mutex);
 523                                 info->alloc_start = memparse(num, NULL);
 524                                 mutex_unlock(&info->chunk_mutex);
 525                                 kfree(num);
 526                                 printk(KERN_INFO
 527                                         "btrfs: allocations start at %llu\n",
 528                                         (unsigned long long)info->alloc_start);
 529                         }
 530                         break;
 531                 case Opt_noacl:
 532                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 533                         break;
 534                 case Opt_notreelog:
 535                         printk(KERN_INFO "btrfs: disabling tree log\n");
 536                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 537                         break;
 538                 case Opt_flushoncommit:
 539                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 540                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 541                         break;
 542                 case Opt_ratio:
 543                         intarg = 0;
 544                         match_int(&args[0], &intarg);
 545                         if (intarg) {
 546                                 info->metadata_ratio = intarg;
 547                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 548                                        info->metadata_ratio);
 549                         }
 550                         break;
 551                 case Opt_discard:
 552                         btrfs_set_opt(info->mount_opt, DISCARD);
 553                         break;
 554                 case Opt_space_cache:
 555                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 556                         break;
 557                 case Opt_no_space_cache:
 558                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 559                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 560                         break;
 561                 case Opt_inode_cache:
 562                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 563                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 564                         break;
 565                 case Opt_clear_cache:
 566                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 567                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 568                         break;
 569                 case Opt_user_subvol_rm_allowed:
 570                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 571                         break;
 572                 case Opt_enospc_debug:
 573                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 574                         break;
 575                 case Opt_defrag:
 576                         printk(KERN_INFO "btrfs: enabling auto defrag\n");
 577                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 578                         break;
 579                 case Opt_recovery:
 580                         printk(KERN_INFO "btrfs: enabling auto recovery\n");
 581                         btrfs_set_opt(info->mount_opt, RECOVERY);
 582                         break;
 583                 case Opt_skip_balance:
 584                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 585                         break;
 586 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 587                 case Opt_check_integrity_including_extent_data:
 588                         printk(KERN_INFO "btrfs: enabling check integrity"
 589                                " including extent data\n");
 590                         btrfs_set_opt(info->mount_opt,
 591                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 592                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 593                         break;
 594                 case Opt_check_integrity:
 595                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 596                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 597                         break;
 598                 case Opt_check_integrity_print_mask:
 599                         intarg = 0;
 600                         match_int(&args[0], &intarg);
 601                         if (intarg) {
 602                                 info->check_integrity_print_mask = intarg;
 603                                 printk(KERN_INFO "btrfs:"
 604                                        " check_integrity_print_mask 0x%x\n",
 605                                        info->check_integrity_print_mask);
 606                         }
 607                         break;
 608 #else
 609                 case Opt_check_integrity_including_extent_data:
 610                 case Opt_check_integrity:
 611                 case Opt_check_integrity_print_mask:
 612                         printk(KERN_ERR "btrfs: support for check_integrity*"
 613                                " not compiled in!\n");
 614                         ret = -EINVAL;
 615                         goto out;
 616 #endif
 617                 case Opt_fatal_errors:
 618                         if (strcmp(args[0].from, "panic") == 0)
 619                                 btrfs_set_opt(info->mount_opt,
 620                                               PANIC_ON_FATAL_ERROR);
 621                         else if (strcmp(args[0].from, "bug") == 0)
 622                                 btrfs_clear_opt(info->mount_opt,
 623                                               PANIC_ON_FATAL_ERROR);
 624                         else {
 625                                 ret = -EINVAL;
 626                                 goto out;
 627                         }
 628                         break;
 629                 case Opt_err:
 630                         printk(KERN_INFO "btrfs: unrecognized mount option "
 631                                "'%s'\n", p);
 632                         ret = -EINVAL;
 633                         goto out;
 634                 default:
 635                         break;
 636                 }
 637         }
 638 out:
 639         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 640                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 641         kfree(orig);
 642         return ret;
 643 }
 644
 645 /*
 646  * Parse mount options that are required early in the mount process.
 647  *
 648  * All other options will be parsed on much later in the mount process and
 649  * only when we need to allocate a new super block.
 650  */
 651 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 652                 void *holder, char **subvol_name, u64 *subvol_objectid,
 653                 struct btrfs_fs_devices **fs_devices)
 654 {
 655         substring_t args[MAX_OPT_ARGS];
 656         char *device_name, *opts, *orig, *p;
 657         int error = 0;
 658         int intarg;
 659
 660         if (!options)
 661                 return 0;
 662
 663         /*
 664          * strsep changes the string, duplicate it because parse_options
 665          * gets called twice
 666          */
 667         opts = kstrdup(options, GFP_KERNEL);
 668         if (!opts)
 669                 return -ENOMEM;
 670         orig = opts;
 671
 672         while ((p = strsep(&opts, ",")) != NULL) {
 673                 int token;
 674                 if (!*p)
 675                         continue;
 676
 677                 token = match_token(p, tokens, args);
 678                 switch (token) {
 679                 case Opt_subvol:
 680                         kfree(*subvol_name);
 681                         *subvol_name = match_strdup(&args[0]);
 682                         break;
 683                 case Opt_subvolid:
 684                         intarg = 0;
 685                         error = match_int(&args[0], &intarg);
 686                         if (!error) {
 687                                 /* we want the original fs_tree */
 688                                 if (!intarg)
 689                                         *subvol_objectid =
 690                                                 BTRFS_FS_TREE_OBJECTID;
 691                                 else
 692                                         *subvol_objectid = intarg;
 693                         }
 694                         break;
 695                 case Opt_subvolrootid:
 696                         printk(KERN_WARNING
 697                                 "btrfs: 'subvolrootid' mount option is deprecated and has no effect\n");
 698                         break;
 699                 case Opt_device:
 700                         device_name = match_strdup(&args[0]);
 701                         if (!device_name) {
 702                                 error = -ENOMEM;
 703                                 goto out;
 704                         }
 705                         error = btrfs_scan_one_device(device_name,
 706                                         flags, holder, fs_devices);
 707                         kfree(device_name);
 708                         if (error)
 709                                 goto out;
 710                         break;
 711                 default:
 712                         break;
 713                 }
 714         }
 715
 716 out:
 717         kfree(orig);
 718         return error;
 719 }
 720
 721 static struct dentry *get_default_root(struct super_block *sb,
 722                                        u64 subvol_objectid)
 723 {
 724         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 725         struct btrfs_root *root = fs_info->tree_root;
 726         struct btrfs_root *new_root;
 727         struct btrfs_dir_item *di;
 728         struct btrfs_path *path;
 729         struct btrfs_key location;
 730         struct inode *inode;
 731         u64 dir_id;
 732         int new = 0;
 733
 734         /*
 735          * We have a specific subvol we want to mount, just setup location and
 736          * go look up the root.
 737          */
 738         if (subvol_objectid) {
 739                 location.objectid = subvol_objectid;
 740                 location.type = BTRFS_ROOT_ITEM_KEY;
 741                 location.offset = (u64)-1;
 742                 goto find_root;
 743         }
 744
 745         path = btrfs_alloc_path();
 746         if (!path)
 747                 return ERR_PTR(-ENOMEM);
 748         path->leave_spinning = 1;
 749
 750         /*
 751          * Find the "default" dir item which points to the root item that we
 752          * will mount by default if we haven't been given a specific subvolume
 753          * to mount.
 754          */
 755         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 756         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 757         if (IS_ERR(di)) {
 758                 btrfs_free_path(path);
 759                 return ERR_CAST(di);
 760         }
 761         if (!di) {
 762                 /*
 763                  * Ok the default dir item isn't there.  This is weird since
 764                  * it's always been there, but don't freak out, just try and
 765                  * mount to root most subvolume.
 766                  */
 767                 btrfs_free_path(path);
 768                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 769                 new_root = fs_info->fs_root;
 770                 goto setup_root;
 771         }
 772
 773         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 774         btrfs_free_path(path);
 775
 776 find_root:
 777         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 778         if (IS_ERR(new_root))
 779                 return ERR_CAST(new_root);
 780
 781         dir_id = btrfs_root_dirid(&new_root->root_item);
 782 setup_root:
 783         location.objectid = dir_id;
 784         location.type = BTRFS_INODE_ITEM_KEY;
 785         location.offset = 0;
 786
 787         inode = btrfs_iget(sb, &location, new_root, &new);
 788         if (IS_ERR(inode))
 789                 return ERR_CAST(inode);
 790
 791         /*
 792          * If we're just mounting the root most subvol put the inode and return
 793          * a reference to the dentry.  We will have already gotten a reference
 794          * to the inode in btrfs_fill_super so we're good to go.
 795          */
 796         if (!new && sb->s_root->d_inode == inode) {
 797                 iput(inode);
 798                 return dget(sb->s_root);
 799         }
 800
 801         return d_obtain_alias(inode);
 802 }
 803
 804 static int btrfs_fill_super(struct super_block *sb,
 805                             struct btrfs_fs_devices *fs_devices,
 806                             void *data, int silent)
 807 {
 808         struct inode *inode;
 809         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 810         struct btrfs_key key;
 811         int err;
 812
 813         sb->s_maxbytes = MAX_LFS_FILESIZE;
 814         sb->s_magic = BTRFS_SUPER_MAGIC;
 815         sb->s_op = &btrfs_super_ops;
 816         sb->s_d_op = &btrfs_dentry_operations;
 817         sb->s_export_op = &btrfs_export_ops;
 818         sb->s_xattr = btrfs_xattr_handlers;
 819         sb->s_time_gran = 1;
 820 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 821         sb->s_flags |= MS_POSIXACL;
 822 #endif
 823         sb->s_flags |= MS_I_VERSION;
 824         err = open_ctree(sb, fs_devices, (char *)data);
 825         if (err) {
 826                 printk("btrfs: open_ctree failed\n");
 827                 return err;
 828         }
 829
 830         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 831         key.type = BTRFS_INODE_ITEM_KEY;
 832         key.offset = 0;
 833         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 834         if (IS_ERR(inode)) {
 835                 err = PTR_ERR(inode);
 836                 goto fail_close;
 837         }
 838
 839         sb->s_root = d_make_root(inode);
 840         if (!sb->s_root) {
 841                 err = -ENOMEM;
 842                 goto fail_close;
 843         }
 844
 845         save_mount_options(sb, data);
 846         cleancache_init_fs(sb);
 847         sb->s_flags |= MS_ACTIVE;
 848         return 0;
 849
 850 fail_close:
 851         close_ctree(fs_info->tree_root);
 852         return err;
 853 }
 854
 855 int btrfs_sync_fs(struct super_block *sb, int wait)
 856 {
 857         struct btrfs_trans_handle *trans;
 858         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 859         struct btrfs_root *root = fs_info->tree_root;
 860
 861         trace_btrfs_sync_fs(wait);
 862
 863         if (!wait) {
 864                 filemap_flush(fs_info->btree_inode->i_mapping);
 865                 return 0;
 866         }
 867
 868         btrfs_wait_all_ordered_extents(fs_info, 1);
 869
 870         trans = btrfs_attach_transaction_barrier(root);
 871         if (IS_ERR(trans)) {
 872                 /* no transaction, don't bother */
 873                 if (PTR_ERR(trans) == -ENOENT)
 874                         return 0;
 875                 return PTR_ERR(trans);
 876         }
 877         return btrfs_commit_transaction(trans, root);
 878 }
 879
 880 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 881 {
 882         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 883         struct btrfs_root *root = info->tree_root;
 884         char *compress_type;
 885
 886         if (btrfs_test_opt(root, DEGRADED))
 887                 seq_puts(seq, ",degraded");
 888         if (btrfs_test_opt(root, NODATASUM))
 889                 seq_puts(seq, ",nodatasum");
 890         if (btrfs_test_opt(root, NODATACOW))
 891                 seq_puts(seq, ",nodatacow");
 892         if (btrfs_test_opt(root, NOBARRIER))
 893                 seq_puts(seq, ",nobarrier");
 894         if (info->max_inline != 8192 * 1024)
 895                 seq_printf(seq, ",max_inline=%llu",
 896                            (unsigned long long)info->max_inline);
 897         if (info->alloc_start != 0)
 898                 seq_printf(seq, ",alloc_start=%llu",
 899                            (unsigned long long)info->alloc_start);
 900         if (info->thread_pool_size !=  min_t(unsigned long,
 901                                              num_online_cpus() + 2, 8))
 902                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 903         if (btrfs_test_opt(root, COMPRESS)) {
 904                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 905                         compress_type = "zlib";
 906                 else
 907                         compress_type = "lzo";
 908                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 909                         seq_printf(seq, ",compress-force=%s", compress_type);
 910                 else
 911                         seq_printf(seq, ",compress=%s", compress_type);
 912         }
 913         if (btrfs_test_opt(root, NOSSD))
 914                 seq_puts(seq, ",nossd");
 915         if (btrfs_test_opt(root, SSD_SPREAD))
 916                 seq_puts(seq, ",ssd_spread");
 917         else if (btrfs_test_opt(root, SSD))
 918                 seq_puts(seq, ",ssd");
 919         if (btrfs_test_opt(root, NOTREELOG))
 920                 seq_puts(seq, ",notreelog");
 921         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 922                 seq_puts(seq, ",flushoncommit");
 923         if (btrfs_test_opt(root, DISCARD))
 924                 seq_puts(seq, ",discard");
 925         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 926                 seq_puts(seq, ",noacl");
 927         if (btrfs_test_opt(root, SPACE_CACHE))
 928                 seq_puts(seq, ",space_cache");
 929         else
 930                 seq_puts(seq, ",nospace_cache");
 931         if (btrfs_test_opt(root, CLEAR_CACHE))
 932                 seq_puts(seq, ",clear_cache");
 933         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 934                 seq_puts(seq, ",user_subvol_rm_allowed");
 935         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 936                 seq_puts(seq, ",enospc_debug");
 937         if (btrfs_test_opt(root, AUTO_DEFRAG))
 938                 seq_puts(seq, ",autodefrag");
 939         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 940                 seq_puts(seq, ",inode_cache");
 941         if (btrfs_test_opt(root, SKIP_BALANCE))
 942                 seq_puts(seq, ",skip_balance");
 943         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 944                 seq_puts(seq, ",fatal_errors=panic");
 945         return 0;
 946 }
 947
 948 static int btrfs_test_super(struct super_block *s, void *data)
 949 {
 950         struct btrfs_fs_info *p = data;
 951         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 952
 953         return fs_info->fs_devices == p->fs_devices;
 954 }
 955
 956 static int btrfs_set_super(struct super_block *s, void *data)
 957 {
 958         int err = set_anon_super(s, data);
 959         if (!err)
 960                 s->s_fs_info = data;
 961         return err;
 962 }
 963
 964 /*
 965  * subvolumes are identified by ino 256
 966  */
 967 static inline int is_subvolume_inode(struct inode *inode)
 968 {
 969         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 970                 return 1;
 971         return 0;
 972 }
 973
 974 /*
 975  * This will strip out the subvol=%s argument for an argument string and add
 976  * subvolid=0 to make sure we get the actual tree root for path walking to the
 977  * subvol we want.
 978  */
 979 static char *setup_root_args(char *args)
 980 {
 981         unsigned len = strlen(args) + 2 + 1;
 982         char *src, *dst, *buf;
 983
 984         /*
 985          * We need the same args as before, but with this substitution:
 986          * s!subvol=[^,]+!subvolid=0!
 987          *
 988          * Since the replacement string is up to 2 bytes longer than the
 989          * original, allocate strlen(args) + 2 + 1 bytes.
 990          */
 991
 992         src = strstr(args, "subvol=");
 993         /* This shouldn't happen, but just in case.. */
 994         if (!src)
 995                 return NULL;
 996
 997         buf = dst = kmalloc(len, GFP_NOFS);
 998         if (!buf)
 999                 return NULL;
1000
1001         /*
1002          * If the subvol= arg is not at the start of the string,
1003          * copy whatever precedes it into buf.
1004          */
1005         if (src != args) {
1006                 *src++ = '\0';
1007                 strcpy(buf, args);
1008                 dst += strlen(args);
1009         }
1010
1011         strcpy(dst, "subvolid=0");
1012         dst += strlen("subvolid=0");
1013
1014         /*
1015          * If there is a "," after the original subvol=... string,
1016          * copy that suffix into our buffer.  Otherwise, we're done.
1017          */
1018         src = strchr(src, ',');
1019         if (src)
1020                 strcpy(dst, src);
1021
1022         return buf;
1023 }
1024
1025 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1026                                    const char *device_name, char *data)
1027 {
1028         struct dentry *root;
1029         struct vfsmount *mnt;
1030         char *newargs;
1031
1032         newargs = setup_root_args(data);
1033         if (!newargs)
1034                 return ERR_PTR(-ENOMEM);
1035         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1036                              newargs);
1037         kfree(newargs);
1038         if (IS_ERR(mnt))
1039                 return ERR_CAST(mnt);
1040
1041         root = mount_subtree(mnt, subvol_name);
1042
1043         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1044                 struct super_block *s = root->d_sb;
1045                 dput(root);
1046                 root = ERR_PTR(-EINVAL);
1047                 deactivate_locked_super(s);
1048                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1049                                 subvol_name);
1050         }
1051
1052         return root;
1053 }
1054
1055 /*
1056  * Find a superblock for the given device / mount point.
1057  *
1058  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1059  *        for multiple device setup.  Make sure to keep it in sync.
1060  */
1061 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1062                 const char *device_name, void *data)
1063 {
1064         struct block_device *bdev = NULL;
1065         struct super_block *s;
1066         struct dentry *root;
1067         struct btrfs_fs_devices *fs_devices = NULL;
1068         struct btrfs_fs_info *fs_info = NULL;
1069         fmode_t mode = FMODE_READ;
1070         char *subvol_name = NULL;
1071         u64 subvol_objectid = 0;
1072         int error = 0;
1073
1074         if (!(flags & MS_RDONLY))
1075                 mode |= FMODE_WRITE;
1076
1077         error = btrfs_parse_early_options(data, mode, fs_type,
1078                                           &subvol_name, &subvol_objectid,
1079                                           &fs_devices);
1080         if (error) {
1081                 kfree(subvol_name);
1082                 return ERR_PTR(error);
1083         }
1084
1085         if (subvol_name) {
1086                 root = mount_subvol(subvol_name, flags, device_name, data);
1087                 kfree(subvol_name);
1088                 return root;
1089         }
1090
1091         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1092         if (error)
1093                 return ERR_PTR(error);
1094
1095         /*
1096          * Setup a dummy root and fs_info for test/set super.  This is because
1097          * we don't actually fill this stuff out until open_ctree, but we need
1098          * it for searching for existing supers, so this lets us do that and
1099          * then open_ctree will properly initialize everything later.
1100          */
1101         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1102         if (!fs_info)
1103                 return ERR_PTR(-ENOMEM);
1104
1105         fs_info->fs_devices = fs_devices;
1106
1107         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1108         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1109         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1110                 error = -ENOMEM;
1111                 goto error_fs_info;
1112         }
1113
1114         error = btrfs_open_devices(fs_devices, mode, fs_type);
1115         if (error)
1116                 goto error_fs_info;
1117
1118         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1119                 error = -EACCES;
1120                 goto error_close_devices;
1121         }
1122
1123         bdev = fs_devices->latest_bdev;
1124         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1125                  fs_info);
1126         if (IS_ERR(s)) {
1127                 error = PTR_ERR(s);
1128                 goto error_close_devices;
1129         }
1130
1131         if (s->s_root) {
1132                 btrfs_close_devices(fs_devices);
1133                 free_fs_info(fs_info);
1134                 if ((flags ^ s->s_flags) & MS_RDONLY)
1135                         error = -EBUSY;
1136         } else {
1137                 char b[BDEVNAME_SIZE];
1138
1139                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1140                 btrfs_sb(s)->bdev_holder = fs_type;
1141                 error = btrfs_fill_super(s, fs_devices, data,
1142                                          flags & MS_SILENT ? 1 : 0);
1143         }
1144
1145         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1146         if (IS_ERR(root))
1147                 deactivate_locked_super(s);
1148
1149         return root;
1150
1151 error_close_devices:
1152         btrfs_close_devices(fs_devices);
1153 error_fs_info:
1154         free_fs_info(fs_info);
1155         return ERR_PTR(error);
1156 }
1157
1158 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1159 {
1160         spin_lock_irq(&workers->lock);
1161         workers->max_workers = new_limit;
1162         spin_unlock_irq(&workers->lock);
1163 }
1164
1165 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1166                                      int new_pool_size, int old_pool_size)
1167 {
1168         if (new_pool_size == old_pool_size)
1169                 return;
1170
1171         fs_info->thread_pool_size = new_pool_size;
1172
1173         printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1174                old_pool_size, new_pool_size);
1175
1176         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1177         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1178         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1179         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1180         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1181         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1182         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1183         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1184         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1185         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1186         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1187         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1188         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1189         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1190                               new_pool_size);
1191 }
1192
1193 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1194 {
1195         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1196 }
1197
1198 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1199                                        unsigned long old_opts, int flags)
1200 {
1201         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1202             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1203              (flags & MS_RDONLY))) {
1204                 /* wait for any defraggers to finish */
1205                 wait_event(fs_info->transaction_wait,
1206                            (atomic_read(&fs_info->defrag_running) == 0));
1207                 if (flags & MS_RDONLY)
1208                         sync_filesystem(fs_info->sb);
1209         }
1210 }
1211
1212 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1213                                          unsigned long old_opts)
1214 {
1215         /*
1216          * We need cleanup all defragable inodes if the autodefragment is
1217          * close or the fs is R/O.
1218          */
1219         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1220             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1221              (fs_info->sb->s_flags & MS_RDONLY))) {
1222                 btrfs_cleanup_defrag_inodes(fs_info);
1223         }
1224
1225         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1226 }
1227
1228 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1229 {
1230         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1231         struct btrfs_root *root = fs_info->tree_root;
1232         unsigned old_flags = sb->s_flags;
1233         unsigned long old_opts = fs_info->mount_opt;
1234         unsigned long old_compress_type = fs_info->compress_type;
1235         u64 old_max_inline = fs_info->max_inline;
1236         u64 old_alloc_start = fs_info->alloc_start;
1237         int old_thread_pool_size = fs_info->thread_pool_size;
1238         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1239         int ret;
1240
1241         btrfs_remount_prepare(fs_info);
1242
1243         ret = btrfs_parse_options(root, data);
1244         if (ret) {
1245                 ret = -EINVAL;
1246                 goto restore;
1247         }
1248
1249         btrfs_remount_begin(fs_info, old_opts, *flags);
1250         btrfs_resize_thread_pool(fs_info,
1251                 fs_info->thread_pool_size, old_thread_pool_size);
1252
1253         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1254                 goto out;
1255
1256         if (*flags & MS_RDONLY) {
1257                 /*
1258                  * this also happens on 'umount -rf' or on shutdown, when
1259                  * the filesystem is busy.
1260                  */
1261                 sb->s_flags |= MS_RDONLY;
1262
1263                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1264                 btrfs_scrub_cancel(fs_info);
1265                 btrfs_pause_balance(fs_info);
1266
1267                 ret = btrfs_commit_super(root);
1268                 if (ret)
1269                         goto restore;
1270         } else {
1271                 if (fs_info->fs_devices->rw_devices == 0) {
1272                         ret = -EACCES;
1273                         goto restore;
1274                 }
1275
1276                 if (fs_info->fs_devices->missing_devices >
1277                      fs_info->num_tolerated_disk_barrier_failures &&
1278                     !(*flags & MS_RDONLY)) {
1279                         printk(KERN_WARNING
1280                                "Btrfs: too many missing devices, writeable remount is not allowed\n");
1281                         ret = -EACCES;
1282                         goto restore;
1283                 }
1284
1285                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1286                         ret = -EINVAL;
1287                         goto restore;
1288                 }
1289
1290                 ret = btrfs_cleanup_fs_roots(fs_info);
1291                 if (ret)
1292                         goto restore;
1293
1294                 /* recover relocation */
1295                 ret = btrfs_recover_relocation(root);
1296                 if (ret)
1297                         goto restore;
1298
1299                 ret = btrfs_resume_balance_async(fs_info);
1300                 if (ret)
1301                         goto restore;
1302
1303                 ret = btrfs_resume_dev_replace_async(fs_info);
1304                 if (ret) {
1305                         pr_warn("btrfs: failed to resume dev_replace\n");
1306                         goto restore;
1307                 }
1308                 sb->s_flags &= ~MS_RDONLY;
1309         }
1310 out:
1311         btrfs_remount_cleanup(fs_info, old_opts);
1312         return 0;
1313
1314 restore:
1315         /* We've hit an error - don't reset MS_RDONLY */
1316         if (sb->s_flags & MS_RDONLY)
1317                 old_flags |= MS_RDONLY;
1318         sb->s_flags = old_flags;
1319         fs_info->mount_opt = old_opts;
1320         fs_info->compress_type = old_compress_type;
1321         fs_info->max_inline = old_max_inline;
1322         mutex_lock(&fs_info->chunk_mutex);
1323         fs_info->alloc_start = old_alloc_start;
1324         mutex_unlock(&fs_info->chunk_mutex);
1325         btrfs_resize_thread_pool(fs_info,
1326                 old_thread_pool_size, fs_info->thread_pool_size);
1327         fs_info->metadata_ratio = old_metadata_ratio;
1328         btrfs_remount_cleanup(fs_info, old_opts);
1329         return ret;
1330 }
1331
1332 /* Used to sort the devices by max_avail(descending sort) */
1333 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1334                                        const void *dev_info2)
1335 {
1336         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1337             ((struct btrfs_device_info *)dev_info2)->max_avail)
1338                 return -1;
1339         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1340                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1341                 return 1;
1342         else
1343         return 0;
1344 }
1345
1346 /*
1347  * sort the devices by max_avail, in which max free extent size of each device
1348  * is stored.(Descending Sort)
1349  */
1350 static inline void btrfs_descending_sort_devices(
1351                                         struct btrfs_device_info *devices,
1352                                         size_t nr_devices)
1353 {
1354         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1355              btrfs_cmp_device_free_bytes, NULL);
1356 }
1357
1358 /*
1359  * The helper to calc the free space on the devices that can be used to store
1360  * file data.
1361  */
1362 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1363 {
1364         struct btrfs_fs_info *fs_info = root->fs_info;
1365         struct btrfs_device_info *devices_info;
1366         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1367         struct btrfs_device *device;
1368         u64 skip_space;
1369         u64 type;
1370         u64 avail_space;
1371         u64 used_space;
1372         u64 min_stripe_size;
1373         int min_stripes = 1, num_stripes = 1;
1374         int i = 0, nr_devices;
1375         int ret;
1376
1377         nr_devices = fs_info->fs_devices->open_devices;
1378         BUG_ON(!nr_devices);
1379
1380         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1381                                GFP_NOFS);
1382         if (!devices_info)
1383                 return -ENOMEM;
1384
1385         /* calc min stripe number for data space alloction */
1386         type = btrfs_get_alloc_profile(root, 1);
1387         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1388                 min_stripes = 2;
1389                 num_stripes = nr_devices;
1390         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1391                 min_stripes = 2;
1392                 num_stripes = 2;
1393         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1394                 min_stripes = 4;
1395                 num_stripes = 4;
1396         }
1397
1398         if (type & BTRFS_BLOCK_GROUP_DUP)
1399                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1400         else
1401                 min_stripe_size = BTRFS_STRIPE_LEN;
1402
1403         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1404                 if (!device->in_fs_metadata || !device->bdev ||
1405                     device->is_tgtdev_for_dev_replace)
1406                         continue;
1407
1408                 avail_space = device->total_bytes - device->bytes_used;
1409
1410                 /* align with stripe_len */
1411                 do_div(avail_space, BTRFS_STRIPE_LEN);
1412                 avail_space *= BTRFS_STRIPE_LEN;
1413
1414                 /*
1415                  * In order to avoid overwritting the superblock on the drive,
1416                  * btrfs starts at an offset of at least 1MB when doing chunk
1417                  * allocation.
1418                  */
1419                 skip_space = 1024 * 1024;
1420
1421                 /* user can set the offset in fs_info->alloc_start. */
1422                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1423                     device->total_bytes)
1424                         skip_space = max(fs_info->alloc_start, skip_space);
1425
1426                 /*
1427                  * btrfs can not use the free space in [0, skip_space - 1],
1428                  * we must subtract it from the total. In order to implement
1429                  * it, we account the used space in this range first.
1430                  */
1431                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1432                                                      &used_space);
1433                 if (ret) {
1434                         kfree(devices_info);
1435                         return ret;
1436                 }
1437
1438                 /* calc the free space in [0, skip_space - 1] */
1439                 skip_space -= used_space;
1440
1441                 /*
1442                  * we can use the free space in [0, skip_space - 1], subtract
1443                  * it from the total.
1444                  */
1445                 if (avail_space && avail_space >= skip_space)
1446                         avail_space -= skip_space;
1447                 else
1448                         avail_space = 0;
1449
1450                 if (avail_space < min_stripe_size)
1451                         continue;
1452
1453                 devices_info[i].dev = device;
1454                 devices_info[i].max_avail = avail_space;
1455
1456                 i++;
1457         }
1458
1459         nr_devices = i;
1460
1461         btrfs_descending_sort_devices(devices_info, nr_devices);
1462
1463         i = nr_devices - 1;
1464         avail_space = 0;
1465         while (nr_devices >= min_stripes) {
1466                 if (num_stripes > nr_devices)
1467                         num_stripes = nr_devices;
1468
1469                 if (devices_info[i].max_avail >= min_stripe_size) {
1470                         int j;
1471                         u64 alloc_size;
1472
1473                         avail_space += devices_info[i].max_avail * num_stripes;
1474                         alloc_size = devices_info[i].max_avail;
1475                         for (j = i + 1 - num_stripes; j <= i; j++)
1476                                 devices_info[j].max_avail -= alloc_size;
1477                 }
1478                 i--;
1479                 nr_devices--;
1480         }
1481
1482         kfree(devices_info);
1483         *free_bytes = avail_space;
1484         return 0;
1485 }
1486
1487 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1488 {
1489         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1490         struct btrfs_super_block *disk_super = fs_info->super_copy;
1491         struct list_head *head = &fs_info->space_info;
1492         struct btrfs_space_info *found;
1493         u64 total_used = 0;
1494         u64 total_free_data = 0;
1495         int bits = dentry->d_sb->s_blocksize_bits;
1496         __be32 *fsid = (__be32 *)fs_info->fsid;
1497         int ret;
1498
1499         /* holding chunk_muext to avoid allocating new chunks */
1500         mutex_lock(&fs_info->chunk_mutex);
1501         rcu_read_lock();
1502         list_for_each_entry_rcu(found, head, list) {
1503                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1504                         total_free_data += found->disk_total - found->disk_used;
1505                         total_free_data -=
1506                                 btrfs_account_ro_block_groups_free_space(found);
1507                 }
1508
1509                 total_used += found->disk_used;
1510         }
1511         rcu_read_unlock();
1512
1513         buf->f_namelen = BTRFS_NAME_LEN;
1514         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1515         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1516         buf->f_bsize = dentry->d_sb->s_blocksize;
1517         buf->f_type = BTRFS_SUPER_MAGIC;
1518         buf->f_bavail = total_free_data;
1519         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1520         if (ret) {
1521                 mutex_unlock(&fs_info->chunk_mutex);
1522                 return ret;
1523         }
1524         buf->f_bavail += total_free_data;
1525         buf->f_bavail = buf->f_bavail >> bits;
1526         mutex_unlock(&fs_info->chunk_mutex);
1527
1528         /* We treat it as constant endianness (it doesn't matter _which_)
1529            because we want the fsid to come out the same whether mounted
1530            on a big-endian or little-endian host */
1531         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1532         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1533         /* Mask in the root object ID too, to disambiguate subvols */
1534         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1535         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1536
1537         return 0;
1538 }
1539
1540 static void btrfs_kill_super(struct super_block *sb)
1541 {
1542         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1543         kill_anon_super(sb);
1544         free_fs_info(fs_info);
1545 }
1546
1547 static struct file_system_type btrfs_fs_type = {
1548         .owner          = THIS_MODULE,
1549         .name           = "btrfs",
1550         .mount          = btrfs_mount,
1551         .kill_sb        = btrfs_kill_super,
1552         .fs_flags       = FS_REQUIRES_DEV,
1553 };
1554 MODULE_ALIAS_FS("btrfs");
1555
1556 /*
1557  * used by btrfsctl to scan devices when no FS is mounted
1558  */
1559 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1560                                 unsigned long arg)
1561 {
1562         struct btrfs_ioctl_vol_args *vol;
1563         struct btrfs_fs_devices *fs_devices;
1564         int ret = -ENOTTY;
1565
1566         if (!capable(CAP_SYS_ADMIN))
1567                 return -EPERM;
1568
1569         vol = memdup_user((void __user *)arg, sizeof(*vol));
1570         if (IS_ERR(vol))
1571                 return PTR_ERR(vol);
1572
1573         switch (cmd) {
1574         case BTRFS_IOC_SCAN_DEV:
1575                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1576                                             &btrfs_fs_type, &fs_devices);
1577                 break;
1578         case BTRFS_IOC_DEVICES_READY:
1579                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1580                                             &btrfs_fs_type, &fs_devices);
1581                 if (ret)
1582                         break;
1583                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1584                 break;
1585         }
1586
1587         kfree(vol);
1588         return ret;
1589 }
1590
1591 static int btrfs_freeze(struct super_block *sb)
1592 {
1593         struct btrfs_trans_handle *trans;
1594         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1595
1596         trans = btrfs_attach_transaction_barrier(root);
1597         if (IS_ERR(trans)) {
1598                 /* no transaction, don't bother */
1599                 if (PTR_ERR(trans) == -ENOENT)
1600                         return 0;
1601                 return PTR_ERR(trans);
1602         }
1603         return btrfs_commit_transaction(trans, root);
1604 }
1605
1606 static int btrfs_unfreeze(struct super_block *sb)
1607 {
1608         return 0;
1609 }
1610
1611 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1612 {
1613         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1614         struct btrfs_fs_devices *cur_devices;
1615         struct btrfs_device *dev, *first_dev = NULL;
1616         struct list_head *head;
1617         struct rcu_string *name;
1618
1619         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1620         cur_devices = fs_info->fs_devices;
1621         while (cur_devices) {
1622                 head = &cur_devices->devices;
1623                 list_for_each_entry(dev, head, dev_list) {
1624                         if (dev->missing)
1625                                 continue;
1626                         if (!first_dev || dev->devid < first_dev->devid)
1627                                 first_dev = dev;
1628                 }
1629                 cur_devices = cur_devices->seed;
1630         }
1631
1632         if (first_dev) {
1633                 rcu_read_lock();
1634                 name = rcu_dereference(first_dev->name);
1635                 seq_escape(m, name->str, " \t\n\\");
1636                 rcu_read_unlock();
1637         } else {
1638                 WARN_ON(1);
1639         }
1640         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1641         return 0;
1642 }
1643
1644 static const struct super_operations btrfs_super_ops = {
1645         .drop_inode     = btrfs_drop_inode,
1646         .evict_inode    = btrfs_evict_inode,
1647         .put_super      = btrfs_put_super,
1648         .sync_fs        = btrfs_sync_fs,
1649         .show_options   = btrfs_show_options,
1650         .show_devname   = btrfs_show_devname,
1651         .write_inode    = btrfs_write_inode,
1652         .alloc_inode    = btrfs_alloc_inode,
1653         .destroy_inode  = btrfs_destroy_inode,
1654         .statfs         = btrfs_statfs,
1655         .remount_fs     = btrfs_remount,
1656         .freeze_fs      = btrfs_freeze,
1657         .unfreeze_fs    = btrfs_unfreeze,
1658 };
1659
1660 static const struct file_operations btrfs_ctl_fops = {
1661         .unlocked_ioctl  = btrfs_control_ioctl,
1662         .compat_ioctl = btrfs_control_ioctl,
1663         .owner   = THIS_MODULE,
1664         .llseek = noop_llseek,
1665 };
1666
1667 static struct miscdevice btrfs_misc = {
1668         .minor          = BTRFS_MINOR,
1669         .name           = "btrfs-control",
1670         .fops           = &btrfs_ctl_fops
1671 };
1672
1673 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1674 MODULE_ALIAS("devname:btrfs-control");
1675
1676 static int btrfs_interface_init(void)
1677 {
1678         return misc_register(&btrfs_misc);
1679 }
1680
1681 static void btrfs_interface_exit(void)
1682 {
1683         if (misc_deregister(&btrfs_misc) < 0)
1684                 printk(KERN_INFO "btrfs: misc_deregister failed for control device\n");
1685 }
1686
1687 static void btrfs_print_info(void)
1688 {
1689         printk(KERN_INFO "Btrfs loaded"
1690 #ifdef CONFIG_BTRFS_DEBUG
1691                         ", debug=on"
1692 #endif
1693 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1694                         ", integrity-checker=on"
1695 #endif
1696                         "\n");
1697 }
1698
1699 static int __init init_btrfs_fs(void)
1700 {
1701         int err;
1702
1703         err = btrfs_init_sysfs();
1704         if (err)
1705                 return err;
1706
1707         btrfs_init_compress();
1708
1709         err = btrfs_init_cachep();
1710         if (err)
1711                 goto free_compress;
1712
1713         err = extent_io_init();
1714         if (err)
1715                 goto free_cachep;
1716
1717         err = extent_map_init();
1718         if (err)
1719                 goto free_extent_io;
1720
1721         err = ordered_data_init();
1722         if (err)
1723                 goto free_extent_map;
1724
1725         err = btrfs_delayed_inode_init();
1726         if (err)
1727                 goto free_ordered_data;
1728
1729         err = btrfs_auto_defrag_init();
1730         if (err)
1731                 goto free_delayed_inode;
1732
1733         err = btrfs_delayed_ref_init();
1734         if (err)
1735                 goto free_auto_defrag;
1736
1737         err = btrfs_interface_init();
1738         if (err)
1739                 goto free_delayed_ref;
1740
1741         err = register_filesystem(&btrfs_fs_type);
1742         if (err)
1743                 goto unregister_ioctl;
1744
1745         btrfs_init_lockdep();
1746
1747         btrfs_print_info();
1748         btrfs_test_free_space_cache();
1749
1750         return 0;
1751
1752 unregister_ioctl:
1753         btrfs_interface_exit();
1754 free_delayed_ref:
1755         btrfs_delayed_ref_exit();
1756 free_auto_defrag:
1757         btrfs_auto_defrag_exit();
1758 free_delayed_inode:
1759         btrfs_delayed_inode_exit();
1760 free_ordered_data:
1761         ordered_data_exit();
1762 free_extent_map:
1763         extent_map_exit();
1764 free_extent_io:
1765         extent_io_exit();
1766 free_cachep:
1767         btrfs_destroy_cachep();
1768 free_compress:
1769         btrfs_exit_compress();
1770         btrfs_exit_sysfs();
1771         return err;
1772 }
1773
1774 static void __exit exit_btrfs_fs(void)
1775 {
1776         btrfs_destroy_cachep();
1777         btrfs_delayed_ref_exit();
1778         btrfs_auto_defrag_exit();
1779         btrfs_delayed_inode_exit();
1780         ordered_data_exit();
1781         extent_map_exit();
1782         extent_io_exit();
1783         btrfs_interface_exit();
1784         unregister_filesystem(&btrfs_fs_type);
1785         btrfs_exit_sysfs();
1786         btrfs_cleanup_fs_uuids();
1787         btrfs_exit_compress();
1788 }
1789
1790 module_init(init_btrfs_fs)
1791 module_exit(exit_btrfs_fs)
1792
1793 MODULE_LICENSE("GPL");