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 "compat.h"
  45 #include "delayed-inode.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "ioctl.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
  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(struct btrfs_fs_info *fs_info, int errno,
  67                                       char nbuf[16])
  68 {
  69         char *errstr = NULL;
  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         default:
  85                 if (nbuf) {
  86                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  87                                 errstr = nbuf;
  88                 }
  89                 break;
  90         }
  91
  92         return errstr;
  93 }
  94
  95 static void __save_error_info(struct btrfs_fs_info *fs_info)
  96 {
  97         /*
  98          * today we only save the error info into ram.  Long term we'll
  99          * also send it down to the disk
 100          */
 101         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
 102 }
 103
 104 static void save_error_info(struct btrfs_fs_info *fs_info)
 105 {
 106         __save_error_info(fs_info);
 107 }
 108
 109 /* btrfs handle error by forcing the filesystem readonly */
 110 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 111 {
 112         struct super_block *sb = fs_info->sb;
 113
 114         if (sb->s_flags & MS_RDONLY)
 115                 return;
 116
 117         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 118                 sb->s_flags |= MS_RDONLY;
 119                 printk(KERN_INFO "btrfs is forced readonly\n");
 120                 /*
 121                  * Note that a running device replace operation is not
 122                  * canceled here although there is no way to update
 123                  * the progress. It would add the risk of a deadlock,
 124                  * therefore the canceling is ommited. The only penalty
 125                  * is that some I/O remains active until the procedure
 126                  * completes. The next time when the filesystem is
 127                  * mounted writeable again, the device replace
 128                  * operation continues.
 129                  */
 130 //              WARN_ON(1);
 131         }
 132 }
 133
 134 #ifdef CONFIG_PRINTK
 135 /*
 136  * __btrfs_std_error decodes expected errors from the caller and
 137  * invokes the approciate error response.
 138  */
 139 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 140                        unsigned int line, int errno, const char *fmt, ...)
 141 {
 142         struct super_block *sb = fs_info->sb;
 143         char nbuf[16];
 144         const char *errstr;
 145         va_list args;
 146         va_start(args, fmt);
 147
 148         /*
 149          * Special case: if the error is EROFS, and we're already
 150          * under MS_RDONLY, then it is safe here.
 151          */
 152         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 153                 return;
 154
 155         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 156         if (fmt) {
 157                 struct va_format vaf = {
 158                         .fmt = fmt,
 159                         .va = &args,
 160                 };
 161
 162                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s (%pV)\n",
 163                         sb->s_id, function, line, errstr, &vaf);
 164         } else {
 165                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 166                         sb->s_id, function, line, errstr);
 167         }
 168
 169         /* Don't go through full error handling during mount */
 170         if (sb->s_flags & MS_BORN) {
 171                 save_error_info(fs_info);
 172                 btrfs_handle_error(fs_info);
 173         }
 174         va_end(args);
 175 }
 176
 177 static const char * const logtypes[] = {
 178         "emergency",
 179         "alert",
 180         "critical",
 181         "error",
 182         "warning",
 183         "notice",
 184         "info",
 185         "debug",
 186 };
 187
 188 void btrfs_printk(struct btrfs_fs_info *fs_info, const char *fmt, ...)
 189 {
 190         struct super_block *sb = fs_info->sb;
 191         char lvl[4];
 192         struct va_format vaf;
 193         va_list args;
 194         const char *type = logtypes[4];
 195         int kern_level;
 196
 197         va_start(args, fmt);
 198
 199         kern_level = printk_get_level(fmt);
 200         if (kern_level) {
 201                 size_t size = printk_skip_level(fmt) - fmt;
 202                 memcpy(lvl, fmt,  size);
 203                 lvl[size] = '\0';
 204                 fmt += size;
 205                 type = logtypes[kern_level - '0'];
 206         } else
 207                 *lvl = '\0';
 208
 209         vaf.fmt = fmt;
 210         vaf.va = &args;
 211
 212         printk("%sBTRFS %s (device %s): %pV", lvl, type, sb->s_id, &vaf);
 213
 214         va_end(args);
 215 }
 216
 217 #else
 218
 219 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 220                        unsigned int line, int errno, const char *fmt, ...)
 221 {
 222         struct super_block *sb = fs_info->sb;
 223
 224         /*
 225          * Special case: if the error is EROFS, and we're already
 226          * under MS_RDONLY, then it is safe here.
 227          */
 228         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 229                 return;
 230
 231         /* Don't go through full error handling during mount */
 232         if (sb->s_flags & MS_BORN) {
 233                 save_error_info(fs_info);
 234                 btrfs_handle_error(fs_info);
 235         }
 236 }
 237 #endif
 238
 239 /*
 240  * We only mark the transaction aborted and then set the file system read-only.
 241  * This will prevent new transactions from starting or trying to join this
 242  * one.
 243  *
 244  * This means that error recovery at the call site is limited to freeing
 245  * any local memory allocations and passing the error code up without
 246  * further cleanup. The transaction should complete as it normally would
 247  * in the call path but will return -EIO.
 248  *
 249  * We'll complete the cleanup in btrfs_end_transaction and
 250  * btrfs_commit_transaction.
 251  */
 252 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 253                                struct btrfs_root *root, const char *function,
 254                                unsigned int line, int errno)
 255 {
 256         WARN_ONCE(1, KERN_DEBUG "btrfs: Transaction aborted\n");
 257         trans->aborted = errno;
 258         /* Nothing used. The other threads that have joined this
 259          * transaction may be able to continue. */
 260         if (!trans->blocks_used) {
 261                 char nbuf[16];
 262                 const char *errstr;
 263
 264                 errstr = btrfs_decode_error(root->fs_info, errno, nbuf);
 265                 btrfs_printk(root->fs_info,
 266                              "%s:%d: Aborting unused transaction(%s).\n",
 267                              function, line, errstr);
 268                 return;
 269         }
 270         ACCESS_ONCE(trans->transaction->aborted) = errno;
 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 nbuf[16];
 281         char *s_id = "<unknown>";
 282         const char *errstr;
 283         struct va_format vaf = { .fmt = fmt };
 284         va_list args;
 285
 286         if (fs_info)
 287                 s_id = fs_info->sb->s_id;
 288
 289         va_start(args, fmt);
 290         vaf.va = &args;
 291
 292         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 293         if (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR)
 294                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
 295                         s_id, function, line, &vaf, errstr);
 296
 297         printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
 298                s_id, function, line, &vaf, errstr);
 299         va_end(args);
 300         /* Caller calls BUG() */
 301 }
 302
 303 static void btrfs_put_super(struct super_block *sb)
 304 {
 305         (void)close_ctree(btrfs_sb(sb)->tree_root);
 306         /* FIXME: need to fix VFS to return error? */
 307         /* AV: return it _where_?  ->put_super() can be triggered by any number
 308          * of async events, up to and including delivery of SIGKILL to the
 309          * last process that kept it busy.  Or segfault in the aforementioned
 310          * process...  Whom would you report that to?
 311          */
 312 }
 313
 314 enum {
 315         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 316         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 317         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 318         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 319         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 320         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 321         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 322         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 323         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 324         Opt_check_integrity_print_mask, Opt_fatal_errors,
 325         Opt_err,
 326 };
 327
 328 static match_table_t tokens = {
 329         {Opt_degraded, "degraded"},
 330         {Opt_subvol, "subvol=%s"},
 331         {Opt_subvolid, "subvolid=%d"},
 332         {Opt_device, "device=%s"},
 333         {Opt_nodatasum, "nodatasum"},
 334         {Opt_nodatacow, "nodatacow"},
 335         {Opt_nobarrier, "nobarrier"},
 336         {Opt_max_inline, "max_inline=%s"},
 337         {Opt_alloc_start, "alloc_start=%s"},
 338         {Opt_thread_pool, "thread_pool=%d"},
 339         {Opt_compress, "compress"},
 340         {Opt_compress_type, "compress=%s"},
 341         {Opt_compress_force, "compress-force"},
 342         {Opt_compress_force_type, "compress-force=%s"},
 343         {Opt_ssd, "ssd"},
 344         {Opt_ssd_spread, "ssd_spread"},
 345         {Opt_nossd, "nossd"},
 346         {Opt_noacl, "noacl"},
 347         {Opt_notreelog, "notreelog"},
 348         {Opt_flushoncommit, "flushoncommit"},
 349         {Opt_ratio, "metadata_ratio=%d"},
 350         {Opt_discard, "discard"},
 351         {Opt_space_cache, "space_cache"},
 352         {Opt_clear_cache, "clear_cache"},
 353         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 354         {Opt_enospc_debug, "enospc_debug"},
 355         {Opt_subvolrootid, "subvolrootid=%d"},
 356         {Opt_defrag, "autodefrag"},
 357         {Opt_inode_cache, "inode_cache"},
 358         {Opt_no_space_cache, "nospace_cache"},
 359         {Opt_recovery, "recovery"},
 360         {Opt_skip_balance, "skip_balance"},
 361         {Opt_check_integrity, "check_int"},
 362         {Opt_check_integrity_including_extent_data, "check_int_data"},
 363         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 364         {Opt_fatal_errors, "fatal_errors=%s"},
 365         {Opt_err, NULL},
 366 };
 367
 368 /*
 369  * Regular mount options parser.  Everything that is needed only when
 370  * reading in a new superblock is parsed here.
 371  * XXX JDM: This needs to be cleaned up for remount.
 372  */
 373 int btrfs_parse_options(struct btrfs_root *root, char *options)
 374 {
 375         struct btrfs_fs_info *info = root->fs_info;
 376         substring_t args[MAX_OPT_ARGS];
 377         char *p, *num, *orig = NULL;
 378         u64 cache_gen;
 379         int intarg;
 380         int ret = 0;
 381         char *compress_type;
 382         bool compress_force = false;
 383
 384         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 385         if (cache_gen)
 386                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 387
 388         if (!options)
 389                 goto out;
 390
 391         /*
 392          * strsep changes the string, duplicate it because parse_options
 393          * gets called twice
 394          */
 395         options = kstrdup(options, GFP_NOFS);
 396         if (!options)
 397                 return -ENOMEM;
 398
 399         orig = options;
 400
 401         while ((p = strsep(&options, ",")) != NULL) {
 402                 int token;
 403                 if (!*p)
 404                         continue;
 405
 406                 token = match_token(p, tokens, args);
 407                 switch (token) {
 408                 case Opt_degraded:
 409                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 410                         btrfs_set_opt(info->mount_opt, DEGRADED);
 411                         break;
 412                 case Opt_subvol:
 413                 case Opt_subvolid:
 414                 case Opt_subvolrootid:
 415                 case Opt_device:
 416                         /*
 417                          * These are parsed by btrfs_parse_early_options
 418                          * and can be happily ignored here.
 419                          */
 420                         break;
 421                 case Opt_nodatasum:
 422                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 423                         btrfs_set_opt(info->mount_opt, NODATASUM);
 424                         break;
 425                 case Opt_nodatacow:
 426                         if (!btrfs_test_opt(root, COMPRESS) ||
 427                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
 428                                         printk(KERN_INFO "btrfs: setting nodatacow, compression disabled\n");
 429                         } else {
 430                                 printk(KERN_INFO "btrfs: setting nodatacow\n");
 431                         }
 432                         info->compress_type = BTRFS_COMPRESS_NONE;
 433                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 434                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 435                         btrfs_set_opt(info->mount_opt, NODATACOW);
 436                         btrfs_set_opt(info->mount_opt, NODATASUM);
 437                         break;
 438                 case Opt_compress_force:
 439                 case Opt_compress_force_type:
 440                         compress_force = true;
 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                                 info->alloc_start = memparse(num, NULL);
 523                                 kfree(num);
 524                                 printk(KERN_INFO
 525                                         "btrfs: allocations start at %llu\n",
 526                                         (unsigned long long)info->alloc_start);
 527                         }
 528                         break;
 529                 case Opt_noacl:
 530                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 531                         break;
 532                 case Opt_notreelog:
 533                         printk(KERN_INFO "btrfs: disabling tree log\n");
 534                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 535                         break;
 536                 case Opt_flushoncommit:
 537                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 538                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 539                         break;
 540                 case Opt_ratio:
 541                         intarg = 0;
 542                         match_int(&args[0], &intarg);
 543                         if (intarg) {
 544                                 info->metadata_ratio = intarg;
 545                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 546                                        info->metadata_ratio);
 547                         }
 548                         break;
 549                 case Opt_discard:
 550                         btrfs_set_opt(info->mount_opt, DISCARD);
 551                         break;
 552                 case Opt_space_cache:
 553                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 554                         break;
 555                 case Opt_no_space_cache:
 556                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 557                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 558                         break;
 559                 case Opt_inode_cache:
 560                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 561                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 562                         break;
 563                 case Opt_clear_cache:
 564                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 565                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 566                         break;
 567                 case Opt_user_subvol_rm_allowed:
 568                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 569                         break;
 570                 case Opt_enospc_debug:
 571                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 572                         break;
 573                 case Opt_defrag:
 574                         printk(KERN_INFO "btrfs: enabling auto defrag\n");
 575                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 576                         break;
 577                 case Opt_recovery:
 578                         printk(KERN_INFO "btrfs: enabling auto recovery\n");
 579                         btrfs_set_opt(info->mount_opt, RECOVERY);
 580                         break;
 581                 case Opt_skip_balance:
 582                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 583                         break;
 584 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 585                 case Opt_check_integrity_including_extent_data:
 586                         printk(KERN_INFO "btrfs: enabling check integrity"
 587                                " including extent data\n");
 588                         btrfs_set_opt(info->mount_opt,
 589                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 590                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 591                         break;
 592                 case Opt_check_integrity:
 593                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 594                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 595                         break;
 596                 case Opt_check_integrity_print_mask:
 597                         intarg = 0;
 598                         match_int(&args[0], &intarg);
 599                         if (intarg) {
 600                                 info->check_integrity_print_mask = intarg;
 601                                 printk(KERN_INFO "btrfs:"
 602                                        " check_integrity_print_mask 0x%x\n",
 603                                        info->check_integrity_print_mask);
 604                         }
 605                         break;
 606 #else
 607                 case Opt_check_integrity_including_extent_data:
 608                 case Opt_check_integrity:
 609                 case Opt_check_integrity_print_mask:
 610                         printk(KERN_ERR "btrfs: support for check_integrity*"
 611                                " not compiled in!\n");
 612                         ret = -EINVAL;
 613                         goto out;
 614 #endif
 615                 case Opt_fatal_errors:
 616                         if (strcmp(args[0].from, "panic") == 0)
 617                                 btrfs_set_opt(info->mount_opt,
 618                                               PANIC_ON_FATAL_ERROR);
 619                         else if (strcmp(args[0].from, "bug") == 0)
 620                                 btrfs_clear_opt(info->mount_opt,
 621                                               PANIC_ON_FATAL_ERROR);
 622                         else {
 623                                 ret = -EINVAL;
 624                                 goto out;
 625                         }
 626                         break;
 627                 case Opt_err:
 628                         printk(KERN_INFO "btrfs: unrecognized mount option "
 629                                "'%s'\n", p);
 630                         ret = -EINVAL;
 631                         goto out;
 632                 default:
 633                         break;
 634                 }
 635         }
 636 out:
 637         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 638                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 639         kfree(orig);
 640         return ret;
 641 }
 642
 643 /*
 644  * Parse mount options that are required early in the mount process.
 645  *
 646  * All other options will be parsed on much later in the mount process and
 647  * only when we need to allocate a new super block.
 648  */
 649 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 650                 void *holder, char **subvol_name, u64 *subvol_objectid,
 651                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
 652 {
 653         substring_t args[MAX_OPT_ARGS];
 654         char *device_name, *opts, *orig, *p;
 655         int error = 0;
 656         int intarg;
 657
 658         if (!options)
 659                 return 0;
 660
 661         /*
 662          * strsep changes the string, duplicate it because parse_options
 663          * gets called twice
 664          */
 665         opts = kstrdup(options, GFP_KERNEL);
 666         if (!opts)
 667                 return -ENOMEM;
 668         orig = opts;
 669
 670         while ((p = strsep(&opts, ",")) != NULL) {
 671                 int token;
 672                 if (!*p)
 673                         continue;
 674
 675                 token = match_token(p, tokens, args);
 676                 switch (token) {
 677                 case Opt_subvol:
 678                         kfree(*subvol_name);
 679                         *subvol_name = match_strdup(&args[0]);
 680                         break;
 681                 case Opt_subvolid:
 682                         intarg = 0;
 683                         error = match_int(&args[0], &intarg);
 684                         if (!error) {
 685                                 /* we want the original fs_tree */
 686                                 if (!intarg)
 687                                         *subvol_objectid =
 688                                                 BTRFS_FS_TREE_OBJECTID;
 689                                 else
 690                                         *subvol_objectid = intarg;
 691                         }
 692                         break;
 693                 case Opt_subvolrootid:
 694                         intarg = 0;
 695                         error = match_int(&args[0], &intarg);
 696                         if (!error) {
 697                                 /* we want the original fs_tree */
 698                                 if (!intarg)
 699                                         *subvol_rootid =
 700                                                 BTRFS_FS_TREE_OBJECTID;
 701                                 else
 702                                         *subvol_rootid = intarg;
 703                         }
 704                         break;
 705                 case Opt_device:
 706                         device_name = match_strdup(&args[0]);
 707                         if (!device_name) {
 708                                 error = -ENOMEM;
 709                                 goto out;
 710                         }
 711                         error = btrfs_scan_one_device(device_name,
 712                                         flags, holder, fs_devices);
 713                         kfree(device_name);
 714                         if (error)
 715                                 goto out;
 716                         break;
 717                 default:
 718                         break;
 719                 }
 720         }
 721
 722 out:
 723         kfree(orig);
 724         return error;
 725 }
 726
 727 static struct dentry *get_default_root(struct super_block *sb,
 728                                        u64 subvol_objectid)
 729 {
 730         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 731         struct btrfs_root *root = fs_info->tree_root;
 732         struct btrfs_root *new_root;
 733         struct btrfs_dir_item *di;
 734         struct btrfs_path *path;
 735         struct btrfs_key location;
 736         struct inode *inode;
 737         u64 dir_id;
 738         int new = 0;
 739
 740         /*
 741          * We have a specific subvol we want to mount, just setup location and
 742          * go look up the root.
 743          */
 744         if (subvol_objectid) {
 745                 location.objectid = subvol_objectid;
 746                 location.type = BTRFS_ROOT_ITEM_KEY;
 747                 location.offset = (u64)-1;
 748                 goto find_root;
 749         }
 750
 751         path = btrfs_alloc_path();
 752         if (!path)
 753                 return ERR_PTR(-ENOMEM);
 754         path->leave_spinning = 1;
 755
 756         /*
 757          * Find the "default" dir item which points to the root item that we
 758          * will mount by default if we haven't been given a specific subvolume
 759          * to mount.
 760          */
 761         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 762         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 763         if (IS_ERR(di)) {
 764                 btrfs_free_path(path);
 765                 return ERR_CAST(di);
 766         }
 767         if (!di) {
 768                 /*
 769                  * Ok the default dir item isn't there.  This is weird since
 770                  * it's always been there, but don't freak out, just try and
 771                  * mount to root most subvolume.
 772                  */
 773                 btrfs_free_path(path);
 774                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 775                 new_root = fs_info->fs_root;
 776                 goto setup_root;
 777         }
 778
 779         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 780         btrfs_free_path(path);
 781
 782 find_root:
 783         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 784         if (IS_ERR(new_root))
 785                 return ERR_CAST(new_root);
 786
 787         if (btrfs_root_refs(&new_root->root_item) == 0)
 788                 return ERR_PTR(-ENOENT);
 789
 790         dir_id = btrfs_root_dirid(&new_root->root_item);
 791 setup_root:
 792         location.objectid = dir_id;
 793         location.type = BTRFS_INODE_ITEM_KEY;
 794         location.offset = 0;
 795
 796         inode = btrfs_iget(sb, &location, new_root, &new);
 797         if (IS_ERR(inode))
 798                 return ERR_CAST(inode);
 799
 800         /*
 801          * If we're just mounting the root most subvol put the inode and return
 802          * a reference to the dentry.  We will have already gotten a reference
 803          * to the inode in btrfs_fill_super so we're good to go.
 804          */
 805         if (!new && sb->s_root->d_inode == inode) {
 806                 iput(inode);
 807                 return dget(sb->s_root);
 808         }
 809
 810         return d_obtain_alias(inode);
 811 }
 812
 813 static int btrfs_fill_super(struct super_block *sb,
 814                             struct btrfs_fs_devices *fs_devices,
 815                             void *data, int silent)
 816 {
 817         struct inode *inode;
 818         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 819         struct btrfs_key key;
 820         int err;
 821
 822         sb->s_maxbytes = MAX_LFS_FILESIZE;
 823         sb->s_magic = BTRFS_SUPER_MAGIC;
 824         sb->s_op = &btrfs_super_ops;
 825         sb->s_d_op = &btrfs_dentry_operations;
 826         sb->s_export_op = &btrfs_export_ops;
 827         sb->s_xattr = btrfs_xattr_handlers;
 828         sb->s_time_gran = 1;
 829 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 830         sb->s_flags |= MS_POSIXACL;
 831 #endif
 832         sb->s_flags |= MS_I_VERSION;
 833         err = open_ctree(sb, fs_devices, (char *)data);
 834         if (err) {
 835                 printk("btrfs: open_ctree failed\n");
 836                 return err;
 837         }
 838
 839         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 840         key.type = BTRFS_INODE_ITEM_KEY;
 841         key.offset = 0;
 842         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 843         if (IS_ERR(inode)) {
 844                 err = PTR_ERR(inode);
 845                 goto fail_close;
 846         }
 847
 848         sb->s_root = d_make_root(inode);
 849         if (!sb->s_root) {
 850                 err = -ENOMEM;
 851                 goto fail_close;
 852         }
 853
 854         save_mount_options(sb, data);
 855         cleancache_init_fs(sb);
 856         sb->s_flags |= MS_ACTIVE;
 857         return 0;
 858
 859 fail_close:
 860         close_ctree(fs_info->tree_root);
 861         return err;
 862 }
 863
 864 int btrfs_sync_fs(struct super_block *sb, int wait)
 865 {
 866         struct btrfs_trans_handle *trans;
 867         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 868         struct btrfs_root *root = fs_info->tree_root;
 869
 870         trace_btrfs_sync_fs(wait);
 871
 872         if (!wait) {
 873                 filemap_flush(fs_info->btree_inode->i_mapping);
 874                 return 0;
 875         }
 876
 877         btrfs_wait_ordered_extents(root, 0);
 878
 879         trans = btrfs_attach_transaction(root);
 880         if (IS_ERR(trans)) {
 881                 /* no transaction, don't bother */
 882                 if (PTR_ERR(trans) == -ENOENT)
 883                         return 0;
 884                 return PTR_ERR(trans);
 885         }
 886         return btrfs_commit_transaction(trans, root);
 887 }
 888
 889 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 890 {
 891         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 892         struct btrfs_root *root = info->tree_root;
 893         char *compress_type;
 894
 895         if (btrfs_test_opt(root, DEGRADED))
 896                 seq_puts(seq, ",degraded");
 897         if (btrfs_test_opt(root, NODATASUM))
 898                 seq_puts(seq, ",nodatasum");
 899         if (btrfs_test_opt(root, NODATACOW))
 900                 seq_puts(seq, ",nodatacow");
 901         if (btrfs_test_opt(root, NOBARRIER))
 902                 seq_puts(seq, ",nobarrier");
 903         if (info->max_inline != 8192 * 1024)
 904                 seq_printf(seq, ",max_inline=%llu",
 905                            (unsigned long long)info->max_inline);
 906         if (info->alloc_start != 0)
 907                 seq_printf(seq, ",alloc_start=%llu",
 908                            (unsigned long long)info->alloc_start);
 909         if (info->thread_pool_size !=  min_t(unsigned long,
 910                                              num_online_cpus() + 2, 8))
 911                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 912         if (btrfs_test_opt(root, COMPRESS)) {
 913                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 914                         compress_type = "zlib";
 915                 else
 916                         compress_type = "lzo";
 917                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 918                         seq_printf(seq, ",compress-force=%s", compress_type);
 919                 else
 920                         seq_printf(seq, ",compress=%s", compress_type);
 921         }
 922         if (btrfs_test_opt(root, NOSSD))
 923                 seq_puts(seq, ",nossd");
 924         if (btrfs_test_opt(root, SSD_SPREAD))
 925                 seq_puts(seq, ",ssd_spread");
 926         else if (btrfs_test_opt(root, SSD))
 927                 seq_puts(seq, ",ssd");
 928         if (btrfs_test_opt(root, NOTREELOG))
 929                 seq_puts(seq, ",notreelog");
 930         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 931                 seq_puts(seq, ",flushoncommit");
 932         if (btrfs_test_opt(root, DISCARD))
 933                 seq_puts(seq, ",discard");
 934         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 935                 seq_puts(seq, ",noacl");
 936         if (btrfs_test_opt(root, SPACE_CACHE))
 937                 seq_puts(seq, ",space_cache");
 938         else
 939                 seq_puts(seq, ",nospace_cache");
 940         if (btrfs_test_opt(root, CLEAR_CACHE))
 941                 seq_puts(seq, ",clear_cache");
 942         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 943                 seq_puts(seq, ",user_subvol_rm_allowed");
 944         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 945                 seq_puts(seq, ",enospc_debug");
 946         if (btrfs_test_opt(root, AUTO_DEFRAG))
 947                 seq_puts(seq, ",autodefrag");
 948         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 949                 seq_puts(seq, ",inode_cache");
 950         if (btrfs_test_opt(root, SKIP_BALANCE))
 951                 seq_puts(seq, ",skip_balance");
 952         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 953                 seq_puts(seq, ",fatal_errors=panic");
 954         return 0;
 955 }
 956
 957 static int btrfs_test_super(struct super_block *s, void *data)
 958 {
 959         struct btrfs_fs_info *p = data;
 960         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 961
 962         return fs_info->fs_devices == p->fs_devices;
 963 }
 964
 965 static int btrfs_set_super(struct super_block *s, void *data)
 966 {
 967         int err = set_anon_super(s, data);
 968         if (!err)
 969                 s->s_fs_info = data;
 970         return err;
 971 }
 972
 973 /*
 974  * subvolumes are identified by ino 256
 975  */
 976 static inline int is_subvolume_inode(struct inode *inode)
 977 {
 978         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 979                 return 1;
 980         return 0;
 981 }
 982
 983 /*
 984  * This will strip out the subvol=%s argument for an argument string and add
 985  * subvolid=0 to make sure we get the actual tree root for path walking to the
 986  * subvol we want.
 987  */
 988 static char *setup_root_args(char *args)
 989 {
 990         unsigned len = strlen(args) + 2 + 1;
 991         char *src, *dst, *buf;
 992
 993         /*
 994          * We need the same args as before, but with this substitution:
 995          * s!subvol=[^,]+!subvolid=0!
 996          *
 997          * Since the replacement string is up to 2 bytes longer than the
 998          * original, allocate strlen(args) + 2 + 1 bytes.
 999          */
1000
1001         src = strstr(args, "subvol=");
1002         /* This shouldn't happen, but just in case.. */
1003         if (!src)
1004                 return NULL;
1005
1006         buf = dst = kmalloc(len, GFP_NOFS);
1007         if (!buf)
1008                 return NULL;
1009
1010         /*
1011          * If the subvol= arg is not at the start of the string,
1012          * copy whatever precedes it into buf.
1013          */
1014         if (src != args) {
1015                 *src++ = '\0';
1016                 strcpy(buf, args);
1017                 dst += strlen(args);
1018         }
1019
1020         strcpy(dst, "subvolid=0");
1021         dst += strlen("subvolid=0");
1022
1023         /*
1024          * If there is a "," after the original subvol=... string,
1025          * copy that suffix into our buffer.  Otherwise, we're done.
1026          */
1027         src = strchr(src, ',');
1028         if (src)
1029                 strcpy(dst, src);
1030
1031         return buf;
1032 }
1033
1034 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1035                                    const char *device_name, char *data)
1036 {
1037         struct dentry *root;
1038         struct vfsmount *mnt;
1039         char *newargs;
1040
1041         newargs = setup_root_args(data);
1042         if (!newargs)
1043                 return ERR_PTR(-ENOMEM);
1044         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1045                              newargs);
1046         kfree(newargs);
1047         if (IS_ERR(mnt))
1048                 return ERR_CAST(mnt);
1049
1050         root = mount_subtree(mnt, subvol_name);
1051
1052         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1053                 struct super_block *s = root->d_sb;
1054                 dput(root);
1055                 root = ERR_PTR(-EINVAL);
1056                 deactivate_locked_super(s);
1057                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1058                                 subvol_name);
1059         }
1060
1061         return root;
1062 }
1063
1064 /*
1065  * Find a superblock for the given device / mount point.
1066  *
1067  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1068  *        for multiple device setup.  Make sure to keep it in sync.
1069  */
1070 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1071                 const char *device_name, void *data)
1072 {
1073         struct block_device *bdev = NULL;
1074         struct super_block *s;
1075         struct dentry *root;
1076         struct btrfs_fs_devices *fs_devices = NULL;
1077         struct btrfs_fs_info *fs_info = NULL;
1078         fmode_t mode = FMODE_READ;
1079         char *subvol_name = NULL;
1080         u64 subvol_objectid = 0;
1081         u64 subvol_rootid = 0;
1082         int error = 0;
1083
1084         if (!(flags & MS_RDONLY))
1085                 mode |= FMODE_WRITE;
1086
1087         error = btrfs_parse_early_options(data, mode, fs_type,
1088                                           &subvol_name, &subvol_objectid,
1089                                           &subvol_rootid, &fs_devices);
1090         if (error) {
1091                 kfree(subvol_name);
1092                 return ERR_PTR(error);
1093         }
1094
1095         if (subvol_name) {
1096                 root = mount_subvol(subvol_name, flags, device_name, data);
1097                 kfree(subvol_name);
1098                 return root;
1099         }
1100
1101         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1102         if (error)
1103                 return ERR_PTR(error);
1104
1105         /*
1106          * Setup a dummy root and fs_info for test/set super.  This is because
1107          * we don't actually fill this stuff out until open_ctree, but we need
1108          * it for searching for existing supers, so this lets us do that and
1109          * then open_ctree will properly initialize everything later.
1110          */
1111         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1112         if (!fs_info)
1113                 return ERR_PTR(-ENOMEM);
1114
1115         fs_info->fs_devices = fs_devices;
1116
1117         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1118         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1119         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1120                 error = -ENOMEM;
1121                 goto error_fs_info;
1122         }
1123
1124         error = btrfs_open_devices(fs_devices, mode, fs_type);
1125         if (error)
1126                 goto error_fs_info;
1127
1128         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1129                 error = -EACCES;
1130                 goto error_close_devices;
1131         }
1132
1133         bdev = fs_devices->latest_bdev;
1134         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1135                  fs_info);
1136         if (IS_ERR(s)) {
1137                 error = PTR_ERR(s);
1138                 goto error_close_devices;
1139         }
1140
1141         if (s->s_root) {
1142                 btrfs_close_devices(fs_devices);
1143                 free_fs_info(fs_info);
1144                 if ((flags ^ s->s_flags) & MS_RDONLY)
1145                         error = -EBUSY;
1146         } else {
1147                 char b[BDEVNAME_SIZE];
1148
1149                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1150                 btrfs_sb(s)->bdev_holder = fs_type;
1151                 error = btrfs_fill_super(s, fs_devices, data,
1152                                          flags & MS_SILENT ? 1 : 0);
1153         }
1154
1155         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1156         if (IS_ERR(root))
1157                 deactivate_locked_super(s);
1158
1159         return root;
1160
1161 error_close_devices:
1162         btrfs_close_devices(fs_devices);
1163 error_fs_info:
1164         free_fs_info(fs_info);
1165         return ERR_PTR(error);
1166 }
1167
1168 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1169 {
1170         spin_lock_irq(&workers->lock);
1171         workers->max_workers = new_limit;
1172         spin_unlock_irq(&workers->lock);
1173 }
1174
1175 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1176                                      int new_pool_size, int old_pool_size)
1177 {
1178         if (new_pool_size == old_pool_size)
1179                 return;
1180
1181         fs_info->thread_pool_size = new_pool_size;
1182
1183         printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1184                old_pool_size, new_pool_size);
1185
1186         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1187         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1188         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1189         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1190         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1191         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1192         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1193         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1194         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1195         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1196         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1197         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1198         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1199         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1200                               new_pool_size);
1201 }
1202
1203 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1204 {
1205         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1206         struct btrfs_root *root = fs_info->tree_root;
1207         unsigned old_flags = sb->s_flags;
1208         unsigned long old_opts = fs_info->mount_opt;
1209         unsigned long old_compress_type = fs_info->compress_type;
1210         u64 old_max_inline = fs_info->max_inline;
1211         u64 old_alloc_start = fs_info->alloc_start;
1212         int old_thread_pool_size = fs_info->thread_pool_size;
1213         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1214         int ret;
1215
1216         ret = btrfs_parse_options(root, data);
1217         if (ret) {
1218                 ret = -EINVAL;
1219                 goto restore;
1220         }
1221
1222         btrfs_resize_thread_pool(fs_info,
1223                 fs_info->thread_pool_size, old_thread_pool_size);
1224
1225         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1226                 return 0;
1227
1228         if (*flags & MS_RDONLY) {
1229                 /*
1230                  * this also happens on 'umount -rf' or on shutdown, when
1231                  * the filesystem is busy.
1232                  */
1233                 sb->s_flags |= MS_RDONLY;
1234
1235                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1236                 btrfs_scrub_cancel(fs_info);
1237
1238                 ret = btrfs_commit_super(root);
1239                 if (ret)
1240                         goto restore;
1241         } else {
1242                 if (fs_info->fs_devices->rw_devices == 0) {
1243                         ret = -EACCES;
1244                         goto restore;
1245                 }
1246
1247                 if (fs_info->fs_devices->missing_devices >
1248                      fs_info->num_tolerated_disk_barrier_failures &&
1249                     !(*flags & MS_RDONLY)) {
1250                         printk(KERN_WARNING
1251                                "Btrfs: too many missing devices, writeable remount is not allowed\n");
1252                         ret = -EACCES;
1253                         goto restore;
1254                 }
1255
1256                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1257                         ret = -EINVAL;
1258                         goto restore;
1259                 }
1260
1261                 ret = btrfs_cleanup_fs_roots(fs_info);
1262                 if (ret)
1263                         goto restore;
1264
1265                 /* recover relocation */
1266                 ret = btrfs_recover_relocation(root);
1267                 if (ret)
1268                         goto restore;
1269
1270                 ret = btrfs_resume_balance_async(fs_info);
1271                 if (ret)
1272                         goto restore;
1273
1274                 ret = btrfs_resume_dev_replace_async(fs_info);
1275                 if (ret) {
1276                         pr_warn("btrfs: failed to resume dev_replace\n");
1277                         goto restore;
1278                 }
1279                 sb->s_flags &= ~MS_RDONLY;
1280         }
1281
1282         return 0;
1283
1284 restore:
1285         /* We've hit an error - don't reset MS_RDONLY */
1286         if (sb->s_flags & MS_RDONLY)
1287                 old_flags |= MS_RDONLY;
1288         sb->s_flags = old_flags;
1289         fs_info->mount_opt = old_opts;
1290         fs_info->compress_type = old_compress_type;
1291         fs_info->max_inline = old_max_inline;
1292         fs_info->alloc_start = old_alloc_start;
1293         btrfs_resize_thread_pool(fs_info,
1294                 old_thread_pool_size, fs_info->thread_pool_size);
1295         fs_info->metadata_ratio = old_metadata_ratio;
1296         return ret;
1297 }
1298
1299 /* Used to sort the devices by max_avail(descending sort) */
1300 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1301                                        const void *dev_info2)
1302 {
1303         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1304             ((struct btrfs_device_info *)dev_info2)->max_avail)
1305                 return -1;
1306         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1307                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1308                 return 1;
1309         else
1310         return 0;
1311 }
1312
1313 /*
1314  * sort the devices by max_avail, in which max free extent size of each device
1315  * is stored.(Descending Sort)
1316  */
1317 static inline void btrfs_descending_sort_devices(
1318                                         struct btrfs_device_info *devices,
1319                                         size_t nr_devices)
1320 {
1321         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1322              btrfs_cmp_device_free_bytes, NULL);
1323 }
1324
1325 /*
1326  * The helper to calc the free space on the devices that can be used to store
1327  * file data.
1328  */
1329 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1330 {
1331         struct btrfs_fs_info *fs_info = root->fs_info;
1332         struct btrfs_device_info *devices_info;
1333         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1334         struct btrfs_device *device;
1335         u64 skip_space;
1336         u64 type;
1337         u64 avail_space;
1338         u64 used_space;
1339         u64 min_stripe_size;
1340         int min_stripes = 1, num_stripes = 1;
1341         int i = 0, nr_devices;
1342         int ret;
1343
1344         nr_devices = fs_info->fs_devices->open_devices;
1345         BUG_ON(!nr_devices);
1346
1347         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1348                                GFP_NOFS);
1349         if (!devices_info)
1350                 return -ENOMEM;
1351
1352         /* calc min stripe number for data space alloction */
1353         type = btrfs_get_alloc_profile(root, 1);
1354         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1355                 min_stripes = 2;
1356                 num_stripes = nr_devices;
1357         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1358                 min_stripes = 2;
1359                 num_stripes = 2;
1360         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1361                 min_stripes = 4;
1362                 num_stripes = 4;
1363         }
1364
1365         if (type & BTRFS_BLOCK_GROUP_DUP)
1366                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1367         else
1368                 min_stripe_size = BTRFS_STRIPE_LEN;
1369
1370         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1371                 if (!device->in_fs_metadata || !device->bdev ||
1372                     device->is_tgtdev_for_dev_replace)
1373                         continue;
1374
1375                 avail_space = device->total_bytes - device->bytes_used;
1376
1377                 /* align with stripe_len */
1378                 do_div(avail_space, BTRFS_STRIPE_LEN);
1379                 avail_space *= BTRFS_STRIPE_LEN;
1380
1381                 /*
1382                  * In order to avoid overwritting the superblock on the drive,
1383                  * btrfs starts at an offset of at least 1MB when doing chunk
1384                  * allocation.
1385                  */
1386                 skip_space = 1024 * 1024;
1387
1388                 /* user can set the offset in fs_info->alloc_start. */
1389                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1390                     device->total_bytes)
1391                         skip_space = max(fs_info->alloc_start, skip_space);
1392
1393                 /*
1394                  * btrfs can not use the free space in [0, skip_space - 1],
1395                  * we must subtract it from the total. In order to implement
1396                  * it, we account the used space in this range first.
1397                  */
1398                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1399                                                      &used_space);
1400                 if (ret) {
1401                         kfree(devices_info);
1402                         return ret;
1403                 }
1404
1405                 /* calc the free space in [0, skip_space - 1] */
1406                 skip_space -= used_space;
1407
1408                 /*
1409                  * we can use the free space in [0, skip_space - 1], subtract
1410                  * it from the total.
1411                  */
1412                 if (avail_space && avail_space >= skip_space)
1413                         avail_space -= skip_space;
1414                 else
1415                         avail_space = 0;
1416
1417                 if (avail_space < min_stripe_size)
1418                         continue;
1419
1420                 devices_info[i].dev = device;
1421                 devices_info[i].max_avail = avail_space;
1422
1423                 i++;
1424         }
1425
1426         nr_devices = i;
1427
1428         btrfs_descending_sort_devices(devices_info, nr_devices);
1429
1430         i = nr_devices - 1;
1431         avail_space = 0;
1432         while (nr_devices >= min_stripes) {
1433                 if (num_stripes > nr_devices)
1434                         num_stripes = nr_devices;
1435
1436                 if (devices_info[i].max_avail >= min_stripe_size) {
1437                         int j;
1438                         u64 alloc_size;
1439
1440                         avail_space += devices_info[i].max_avail * num_stripes;
1441                         alloc_size = devices_info[i].max_avail;
1442                         for (j = i + 1 - num_stripes; j <= i; j++)
1443                                 devices_info[j].max_avail -= alloc_size;
1444                 }
1445                 i--;
1446                 nr_devices--;
1447         }
1448
1449         kfree(devices_info);
1450         *free_bytes = avail_space;
1451         return 0;
1452 }
1453
1454 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1455 {
1456         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1457         struct btrfs_super_block *disk_super = fs_info->super_copy;
1458         struct list_head *head = &fs_info->space_info;
1459         struct btrfs_space_info *found;
1460         u64 total_used = 0;
1461         u64 total_free_data = 0;
1462         int bits = dentry->d_sb->s_blocksize_bits;
1463         __be32 *fsid = (__be32 *)fs_info->fsid;
1464         int ret;
1465
1466         /* holding chunk_muext to avoid allocating new chunks */
1467         mutex_lock(&fs_info->chunk_mutex);
1468         rcu_read_lock();
1469         list_for_each_entry_rcu(found, head, list) {
1470                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1471                         total_free_data += found->disk_total - found->disk_used;
1472                         total_free_data -=
1473                                 btrfs_account_ro_block_groups_free_space(found);
1474                 }
1475
1476                 total_used += found->disk_used;
1477         }
1478         rcu_read_unlock();
1479
1480         buf->f_namelen = BTRFS_NAME_LEN;
1481         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1482         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1483         buf->f_bsize = dentry->d_sb->s_blocksize;
1484         buf->f_type = BTRFS_SUPER_MAGIC;
1485         buf->f_bavail = total_free_data;
1486         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1487         if (ret) {
1488                 mutex_unlock(&fs_info->chunk_mutex);
1489                 return ret;
1490         }
1491         buf->f_bavail += total_free_data;
1492         buf->f_bavail = buf->f_bavail >> bits;
1493         mutex_unlock(&fs_info->chunk_mutex);
1494
1495         /* We treat it as constant endianness (it doesn't matter _which_)
1496            because we want the fsid to come out the same whether mounted
1497            on a big-endian or little-endian host */
1498         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1499         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1500         /* Mask in the root object ID too, to disambiguate subvols */
1501         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1502         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1503
1504         return 0;
1505 }
1506
1507 static void btrfs_kill_super(struct super_block *sb)
1508 {
1509         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1510         kill_anon_super(sb);
1511         free_fs_info(fs_info);
1512 }
1513
1514 static struct file_system_type btrfs_fs_type = {
1515         .owner          = THIS_MODULE,
1516         .name           = "btrfs",
1517         .mount          = btrfs_mount,
1518         .kill_sb        = btrfs_kill_super,
1519         .fs_flags       = FS_REQUIRES_DEV,
1520 };
1521
1522 /*
1523  * used by btrfsctl to scan devices when no FS is mounted
1524  */
1525 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1526                                 unsigned long arg)
1527 {
1528         struct btrfs_ioctl_vol_args *vol;
1529         struct btrfs_fs_devices *fs_devices;
1530         int ret = -ENOTTY;
1531
1532         if (!capable(CAP_SYS_ADMIN))
1533                 return -EPERM;
1534
1535         vol = memdup_user((void __user *)arg, sizeof(*vol));
1536         if (IS_ERR(vol))
1537                 return PTR_ERR(vol);
1538
1539         switch (cmd) {
1540         case BTRFS_IOC_SCAN_DEV:
1541                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1542                                             &btrfs_fs_type, &fs_devices);
1543                 break;
1544         case BTRFS_IOC_DEVICES_READY:
1545                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1546                                             &btrfs_fs_type, &fs_devices);
1547                 if (ret)
1548                         break;
1549                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1550                 break;
1551         }
1552
1553         kfree(vol);
1554         return ret;
1555 }
1556
1557 static int btrfs_freeze(struct super_block *sb)
1558 {
1559         struct btrfs_trans_handle *trans;
1560         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1561
1562         trans = btrfs_attach_transaction(root);
1563         if (IS_ERR(trans)) {
1564                 /* no transaction, don't bother */
1565                 if (PTR_ERR(trans) == -ENOENT)
1566                         return 0;
1567                 return PTR_ERR(trans);
1568         }
1569         return btrfs_commit_transaction(trans, root);
1570 }
1571
1572 static int btrfs_unfreeze(struct super_block *sb)
1573 {
1574         return 0;
1575 }
1576
1577 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1578 {
1579         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1580         struct btrfs_fs_devices *cur_devices;
1581         struct btrfs_device *dev, *first_dev = NULL;
1582         struct list_head *head;
1583         struct rcu_string *name;
1584
1585         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1586         cur_devices = fs_info->fs_devices;
1587         while (cur_devices) {
1588                 head = &cur_devices->devices;
1589                 list_for_each_entry(dev, head, dev_list) {
1590                         if (dev->missing)
1591                                 continue;
1592                         if (!first_dev || dev->devid < first_dev->devid)
1593                                 first_dev = dev;
1594                 }
1595                 cur_devices = cur_devices->seed;
1596         }
1597
1598         if (first_dev) {
1599                 rcu_read_lock();
1600                 name = rcu_dereference(first_dev->name);
1601                 seq_escape(m, name->str, " \t\n\\");
1602                 rcu_read_unlock();
1603         } else {
1604                 WARN_ON(1);
1605         }
1606         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1607         return 0;
1608 }
1609
1610 static const struct super_operations btrfs_super_ops = {
1611         .drop_inode     = btrfs_drop_inode,
1612         .evict_inode    = btrfs_evict_inode,
1613         .put_super      = btrfs_put_super,
1614         .sync_fs        = btrfs_sync_fs,
1615         .show_options   = btrfs_show_options,
1616         .show_devname   = btrfs_show_devname,
1617         .write_inode    = btrfs_write_inode,
1618         .alloc_inode    = btrfs_alloc_inode,
1619         .destroy_inode  = btrfs_destroy_inode,
1620         .statfs         = btrfs_statfs,
1621         .remount_fs     = btrfs_remount,
1622         .freeze_fs      = btrfs_freeze,
1623         .unfreeze_fs    = btrfs_unfreeze,
1624 };
1625
1626 static const struct file_operations btrfs_ctl_fops = {
1627         .unlocked_ioctl  = btrfs_control_ioctl,
1628         .compat_ioctl = btrfs_control_ioctl,
1629         .owner   = THIS_MODULE,
1630         .llseek = noop_llseek,
1631 };
1632
1633 static struct miscdevice btrfs_misc = {
1634         .minor          = BTRFS_MINOR,
1635         .name           = "btrfs-control",
1636         .fops           = &btrfs_ctl_fops
1637 };
1638
1639 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1640 MODULE_ALIAS("devname:btrfs-control");
1641
1642 static int btrfs_interface_init(void)
1643 {
1644         return misc_register(&btrfs_misc);
1645 }
1646
1647 static void btrfs_interface_exit(void)
1648 {
1649         if (misc_deregister(&btrfs_misc) < 0)
1650                 printk(KERN_INFO "btrfs: misc_deregister failed for control device\n");
1651 }
1652
1653 static int __init init_btrfs_fs(void)
1654 {
1655         int err;
1656
1657         err = btrfs_init_sysfs();
1658         if (err)
1659                 return err;
1660
1661         btrfs_init_compress();
1662
1663         err = btrfs_init_cachep();
1664         if (err)
1665                 goto free_compress;
1666
1667         err = extent_io_init();
1668         if (err)
1669                 goto free_cachep;
1670
1671         err = extent_map_init();
1672         if (err)
1673                 goto free_extent_io;
1674
1675         err = ordered_data_init();
1676         if (err)
1677                 goto free_extent_map;
1678
1679         err = btrfs_delayed_inode_init();
1680         if (err)
1681                 goto free_ordered_data;
1682
1683         err = btrfs_auto_defrag_init();
1684         if (err)
1685                 goto free_delayed_inode;
1686
1687         err = btrfs_interface_init();
1688         if (err)
1689                 goto free_auto_defrag;
1690
1691         err = register_filesystem(&btrfs_fs_type);
1692         if (err)
1693                 goto unregister_ioctl;
1694
1695         btrfs_init_lockdep();
1696
1697         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1698         return 0;
1699
1700 unregister_ioctl:
1701         btrfs_interface_exit();
1702 free_auto_defrag:
1703         btrfs_auto_defrag_exit();
1704 free_delayed_inode:
1705         btrfs_delayed_inode_exit();
1706 free_ordered_data:
1707         ordered_data_exit();
1708 free_extent_map:
1709         extent_map_exit();
1710 free_extent_io:
1711         extent_io_exit();
1712 free_cachep:
1713         btrfs_destroy_cachep();
1714 free_compress:
1715         btrfs_exit_compress();
1716         btrfs_exit_sysfs();
1717         return err;
1718 }
1719
1720 static void __exit exit_btrfs_fs(void)
1721 {
1722         btrfs_destroy_cachep();
1723         btrfs_auto_defrag_exit();
1724         btrfs_delayed_inode_exit();
1725         ordered_data_exit();
1726         extent_map_exit();
1727         extent_io_exit();
1728         btrfs_interface_exit();
1729         unregister_filesystem(&btrfs_fs_type);
1730         btrfs_exit_sysfs();
1731         btrfs_cleanup_fs_uuids();
1732         btrfs_exit_compress();
1733 }
1734
1735 module_init(init_btrfs_fs)
1736 module_exit(exit_btrfs_fs)
1737
1738 MODULE_LICENSE("GPL");