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
  59 #define CREATE_TRACE_POINTS
  60 #include <trace/events/btrfs.h>
  61
  62 static const struct super_operations btrfs_super_ops;
  63 static struct file_system_type btrfs_fs_type;
  64
  65 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
  66                                       char nbuf[16])
  67 {
  68         char *errstr = NULL;
  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         default:
  84                 if (nbuf) {
  85                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  86                                 errstr = nbuf;
  87                 }
  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         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
 101 }
 102
 103 static void save_error_info(struct btrfs_fs_info *fs_info)
 104 {
 105         __save_error_info(fs_info);
 106 }
 107
 108 /* btrfs handle error by forcing the filesystem readonly */
 109 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 110 {
 111         struct super_block *sb = fs_info->sb;
 112
 113         if (sb->s_flags & MS_RDONLY)
 114                 return;
 115
 116         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 117                 sb->s_flags |= MS_RDONLY;
 118                 printk(KERN_INFO "btrfs is forced readonly\n");
 119                 __btrfs_scrub_cancel(fs_info);
 120 //              WARN_ON(1);
 121         }
 122 }
 123
 124 #ifdef CONFIG_PRINTK
 125 /*
 126  * __btrfs_std_error decodes expected errors from the caller and
 127  * invokes the approciate error response.
 128  */
 129 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 130                        unsigned int line, int errno, const char *fmt, ...)
 131 {
 132         struct super_block *sb = fs_info->sb;
 133         char nbuf[16];
 134         const char *errstr;
 135         va_list args;
 136         va_start(args, fmt);
 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(fs_info, errno, nbuf);
 146         if (fmt) {
 147                 struct va_format vaf = {
 148                         .fmt = fmt,
 149                         .va = &args,
 150                 };
 151
 152                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s (%pV)\n",
 153                         sb->s_id, function, line, errstr, &vaf);
 154         } else {
 155                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 156                         sb->s_id, function, line, errstr);
 157         }
 158
 159         /* Don't go through full error handling during mount */
 160         if (sb->s_flags & MS_BORN) {
 161                 save_error_info(fs_info);
 162                 btrfs_handle_error(fs_info);
 163         }
 164         va_end(args);
 165 }
 166
 167 static const char * const logtypes[] = {
 168         "emergency",
 169         "alert",
 170         "critical",
 171         "error",
 172         "warning",
 173         "notice",
 174         "info",
 175         "debug",
 176 };
 177
 178 void btrfs_printk(struct btrfs_fs_info *fs_info, const char *fmt, ...)
 179 {
 180         struct super_block *sb = fs_info->sb;
 181         char lvl[4];
 182         struct va_format vaf;
 183         va_list args;
 184         const char *type = logtypes[4];
 185         int kern_level;
 186
 187         va_start(args, fmt);
 188
 189         kern_level = printk_get_level(fmt);
 190         if (kern_level) {
 191                 size_t size = printk_skip_level(fmt) - fmt;
 192                 memcpy(lvl, fmt,  size);
 193                 lvl[size] = '\0';
 194                 fmt += size;
 195                 type = logtypes[kern_level - '0'];
 196         } else
 197                 *lvl = '\0';
 198
 199         vaf.fmt = fmt;
 200         vaf.va = &args;
 201
 202         printk("%sBTRFS %s (device %s): %pV", lvl, type, sb->s_id, &vaf);
 203
 204         va_end(args);
 205 }
 206
 207 #else
 208
 209 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 210                        unsigned int line, int errno, const char *fmt, ...)
 211 {
 212         struct super_block *sb = fs_info->sb;
 213
 214         /*
 215          * Special case: if the error is EROFS, and we're already
 216          * under MS_RDONLY, then it is safe here.
 217          */
 218         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 219                 return;
 220
 221         /* Don't go through full error handling during mount */
 222         if (sb->s_flags & MS_BORN) {
 223                 save_error_info(fs_info);
 224                 btrfs_handle_error(fs_info);
 225         }
 226 }
 227 #endif
 228
 229 /*
 230  * We only mark the transaction aborted and then set the file system read-only.
 231  * This will prevent new transactions from starting or trying to join this
 232  * one.
 233  *
 234  * This means that error recovery at the call site is limited to freeing
 235  * any local memory allocations and passing the error code up without
 236  * further cleanup. The transaction should complete as it normally would
 237  * in the call path but will return -EIO.
 238  *
 239  * We'll complete the cleanup in btrfs_end_transaction and
 240  * btrfs_commit_transaction.
 241  */
 242 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 243                                struct btrfs_root *root, const char *function,
 244                                unsigned int line, int errno)
 245 {
 246         WARN_ONCE(1, KERN_DEBUG "btrfs: Transaction aborted\n");
 247         trans->aborted = errno;
 248         /* Nothing used. The other threads that have joined this
 249          * transaction may be able to continue. */
 250         if (!trans->blocks_used) {
 251                 char nbuf[16];
 252                 const char *errstr;
 253
 254                 errstr = btrfs_decode_error(root->fs_info, errno, nbuf);
 255                 btrfs_printk(root->fs_info,
 256                              "%s:%d: Aborting unused transaction(%s).\n",
 257                              function, line, errstr);
 258                 return;
 259         }
 260         trans->transaction->aborted = errno;
 261         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 262 }
 263 /*
 264  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 265  * issues an alert, and either panics or BUGs, depending on mount options.
 266  */
 267 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 268                    unsigned int line, int errno, const char *fmt, ...)
 269 {
 270         char nbuf[16];
 271         char *s_id = "<unknown>";
 272         const char *errstr;
 273         struct va_format vaf = { .fmt = fmt };
 274         va_list args;
 275
 276         if (fs_info)
 277                 s_id = fs_info->sb->s_id;
 278
 279         va_start(args, fmt);
 280         vaf.va = &args;
 281
 282         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 283         if (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR)
 284                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
 285                         s_id, function, line, &vaf, errstr);
 286
 287         printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
 288                s_id, function, line, &vaf, errstr);
 289         va_end(args);
 290         /* Caller calls BUG() */
 291 }
 292
 293 static void btrfs_put_super(struct super_block *sb)
 294 {
 295         (void)close_ctree(btrfs_sb(sb)->tree_root);
 296         /* FIXME: need to fix VFS to return error? */
 297         /* AV: return it _where_?  ->put_super() can be triggered by any number
 298          * of async events, up to and including delivery of SIGKILL to the
 299          * last process that kept it busy.  Or segfault in the aforementioned
 300          * process...  Whom would you report that to?
 301          */
 302 }
 303
 304 enum {
 305         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 306         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 307         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 308         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 309         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 310         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 311         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 312         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 313         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 314         Opt_check_integrity_print_mask, Opt_fatal_errors,
 315         Opt_err,
 316 };
 317
 318 static match_table_t tokens = {
 319         {Opt_degraded, "degraded"},
 320         {Opt_subvol, "subvol=%s"},
 321         {Opt_subvolid, "subvolid=%d"},
 322         {Opt_device, "device=%s"},
 323         {Opt_nodatasum, "nodatasum"},
 324         {Opt_nodatacow, "nodatacow"},
 325         {Opt_nobarrier, "nobarrier"},
 326         {Opt_max_inline, "max_inline=%s"},
 327         {Opt_alloc_start, "alloc_start=%s"},
 328         {Opt_thread_pool, "thread_pool=%d"},
 329         {Opt_compress, "compress"},
 330         {Opt_compress_type, "compress=%s"},
 331         {Opt_compress_force, "compress-force"},
 332         {Opt_compress_force_type, "compress-force=%s"},
 333         {Opt_ssd, "ssd"},
 334         {Opt_ssd_spread, "ssd_spread"},
 335         {Opt_nossd, "nossd"},
 336         {Opt_noacl, "noacl"},
 337         {Opt_notreelog, "notreelog"},
 338         {Opt_flushoncommit, "flushoncommit"},
 339         {Opt_ratio, "metadata_ratio=%d"},
 340         {Opt_discard, "discard"},
 341         {Opt_space_cache, "space_cache"},
 342         {Opt_clear_cache, "clear_cache"},
 343         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 344         {Opt_enospc_debug, "enospc_debug"},
 345         {Opt_subvolrootid, "subvolrootid=%d"},
 346         {Opt_defrag, "autodefrag"},
 347         {Opt_inode_cache, "inode_cache"},
 348         {Opt_no_space_cache, "nospace_cache"},
 349         {Opt_recovery, "recovery"},
 350         {Opt_skip_balance, "skip_balance"},
 351         {Opt_check_integrity, "check_int"},
 352         {Opt_check_integrity_including_extent_data, "check_int_data"},
 353         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 354         {Opt_fatal_errors, "fatal_errors=%s"},
 355         {Opt_err, NULL},
 356 };
 357
 358 /*
 359  * Regular mount options parser.  Everything that is needed only when
 360  * reading in a new superblock is parsed here.
 361  * XXX JDM: This needs to be cleaned up for remount.
 362  */
 363 int btrfs_parse_options(struct btrfs_root *root, char *options)
 364 {
 365         struct btrfs_fs_info *info = root->fs_info;
 366         substring_t args[MAX_OPT_ARGS];
 367         char *p, *num, *orig = NULL;
 368         u64 cache_gen;
 369         int intarg;
 370         int ret = 0;
 371         char *compress_type;
 372         bool compress_force = false;
 373
 374         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 375         if (cache_gen)
 376                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 377
 378         if (!options)
 379                 goto out;
 380
 381         /*
 382          * strsep changes the string, duplicate it because parse_options
 383          * gets called twice
 384          */
 385         options = kstrdup(options, GFP_NOFS);
 386         if (!options)
 387                 return -ENOMEM;
 388
 389         orig = options;
 390
 391         while ((p = strsep(&options, ",")) != NULL) {
 392                 int token;
 393                 if (!*p)
 394                         continue;
 395
 396                 token = match_token(p, tokens, args);
 397                 switch (token) {
 398                 case Opt_degraded:
 399                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 400                         btrfs_set_opt(info->mount_opt, DEGRADED);
 401                         break;
 402                 case Opt_subvol:
 403                 case Opt_subvolid:
 404                 case Opt_subvolrootid:
 405                 case Opt_device:
 406                         /*
 407                          * These are parsed by btrfs_parse_early_options
 408                          * and can be happily ignored here.
 409                          */
 410                         break;
 411                 case Opt_nodatasum:
 412                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 413                         btrfs_set_opt(info->mount_opt, NODATASUM);
 414                         break;
 415                 case Opt_nodatacow:
 416                         if (!btrfs_test_opt(root, COMPRESS) ||
 417                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
 418                                         printk(KERN_INFO "btrfs: setting nodatacow, compression disabled\n");
 419                         } else {
 420                                 printk(KERN_INFO "btrfs: setting nodatacow\n");
 421                         }
 422                         info->compress_type = BTRFS_COMPRESS_NONE;
 423                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 424                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 425                         btrfs_set_opt(info->mount_opt, NODATACOW);
 426                         btrfs_set_opt(info->mount_opt, NODATASUM);
 427                         break;
 428                 case Opt_compress_force:
 429                 case Opt_compress_force_type:
 430                         compress_force = true;
 431                 case Opt_compress:
 432                 case Opt_compress_type:
 433                         if (token == Opt_compress ||
 434                             token == Opt_compress_force ||
 435                             strcmp(args[0].from, "zlib") == 0) {
 436                                 compress_type = "zlib";
 437                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 438                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 439                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 440                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 441                         } else if (strcmp(args[0].from, "lzo") == 0) {
 442                                 compress_type = "lzo";
 443                                 info->compress_type = BTRFS_COMPRESS_LZO;
 444                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 445                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 446                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 447                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 448                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 449                                 compress_type = "no";
 450                                 info->compress_type = BTRFS_COMPRESS_NONE;
 451                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 452                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 453                                 compress_force = false;
 454                         } else {
 455                                 ret = -EINVAL;
 456                                 goto out;
 457                         }
 458
 459                         if (compress_force) {
 460                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 461                                 pr_info("btrfs: force %s compression\n",
 462                                         compress_type);
 463                         } else
 464                                 pr_info("btrfs: use %s compression\n",
 465                                         compress_type);
 466                         break;
 467                 case Opt_ssd:
 468                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 469                         btrfs_set_opt(info->mount_opt, SSD);
 470                         break;
 471                 case Opt_ssd_spread:
 472                         printk(KERN_INFO "btrfs: use spread ssd "
 473                                "allocation scheme\n");
 474                         btrfs_set_opt(info->mount_opt, SSD);
 475                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 476                         break;
 477                 case Opt_nossd:
 478                         printk(KERN_INFO "btrfs: not using ssd allocation "
 479                                "scheme\n");
 480                         btrfs_set_opt(info->mount_opt, NOSSD);
 481                         btrfs_clear_opt(info->mount_opt, SSD);
 482                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 483                         break;
 484                 case Opt_nobarrier:
 485                         printk(KERN_INFO "btrfs: turning off barriers\n");
 486                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 487                         break;
 488                 case Opt_thread_pool:
 489                         intarg = 0;
 490                         match_int(&args[0], &intarg);
 491                         if (intarg)
 492                                 info->thread_pool_size = intarg;
 493                         break;
 494                 case Opt_max_inline:
 495                         num = match_strdup(&args[0]);
 496                         if (num) {
 497                                 info->max_inline = memparse(num, NULL);
 498                                 kfree(num);
 499
 500                                 if (info->max_inline) {
 501                                         info->max_inline = max_t(u64,
 502                                                 info->max_inline,
 503                                                 root->sectorsize);
 504                                 }
 505                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 506                                         (unsigned long long)info->max_inline);
 507                         }
 508                         break;
 509                 case Opt_alloc_start:
 510                         num = match_strdup(&args[0]);
 511                         if (num) {
 512                                 info->alloc_start = memparse(num, NULL);
 513                                 kfree(num);
 514                                 printk(KERN_INFO
 515                                         "btrfs: allocations start at %llu\n",
 516                                         (unsigned long long)info->alloc_start);
 517                         }
 518                         break;
 519                 case Opt_noacl:
 520                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 521                         break;
 522                 case Opt_notreelog:
 523                         printk(KERN_INFO "btrfs: disabling tree log\n");
 524                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 525                         break;
 526                 case Opt_flushoncommit:
 527                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 528                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 529                         break;
 530                 case Opt_ratio:
 531                         intarg = 0;
 532                         match_int(&args[0], &intarg);
 533                         if (intarg) {
 534                                 info->metadata_ratio = intarg;
 535                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 536                                        info->metadata_ratio);
 537                         }
 538                         break;
 539                 case Opt_discard:
 540                         btrfs_set_opt(info->mount_opt, DISCARD);
 541                         break;
 542                 case Opt_space_cache:
 543                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 544                         break;
 545                 case Opt_no_space_cache:
 546                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 547                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 548                         break;
 549                 case Opt_inode_cache:
 550                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 551                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 552                         break;
 553                 case Opt_clear_cache:
 554                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 555                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 556                         break;
 557                 case Opt_user_subvol_rm_allowed:
 558                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 559                         break;
 560                 case Opt_enospc_debug:
 561                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 562                         break;
 563                 case Opt_defrag:
 564                         printk(KERN_INFO "btrfs: enabling auto defrag\n");
 565                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 566                         break;
 567                 case Opt_recovery:
 568                         printk(KERN_INFO "btrfs: enabling auto recovery\n");
 569                         btrfs_set_opt(info->mount_opt, RECOVERY);
 570                         break;
 571                 case Opt_skip_balance:
 572                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 573                         break;
 574 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 575                 case Opt_check_integrity_including_extent_data:
 576                         printk(KERN_INFO "btrfs: enabling check integrity"
 577                                " including extent data\n");
 578                         btrfs_set_opt(info->mount_opt,
 579                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 580                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 581                         break;
 582                 case Opt_check_integrity:
 583                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 584                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 585                         break;
 586                 case Opt_check_integrity_print_mask:
 587                         intarg = 0;
 588                         match_int(&args[0], &intarg);
 589                         if (intarg) {
 590                                 info->check_integrity_print_mask = intarg;
 591                                 printk(KERN_INFO "btrfs:"
 592                                        " check_integrity_print_mask 0x%x\n",
 593                                        info->check_integrity_print_mask);
 594                         }
 595                         break;
 596 #else
 597                 case Opt_check_integrity_including_extent_data:
 598                 case Opt_check_integrity:
 599                 case Opt_check_integrity_print_mask:
 600                         printk(KERN_ERR "btrfs: support for check_integrity*"
 601                                " not compiled in!\n");
 602                         ret = -EINVAL;
 603                         goto out;
 604 #endif
 605                 case Opt_fatal_errors:
 606                         if (strcmp(args[0].from, "panic") == 0)
 607                                 btrfs_set_opt(info->mount_opt,
 608                                               PANIC_ON_FATAL_ERROR);
 609                         else if (strcmp(args[0].from, "bug") == 0)
 610                                 btrfs_clear_opt(info->mount_opt,
 611                                               PANIC_ON_FATAL_ERROR);
 612                         else {
 613                                 ret = -EINVAL;
 614                                 goto out;
 615                         }
 616                         break;
 617                 case Opt_err:
 618                         printk(KERN_INFO "btrfs: unrecognized mount option "
 619                                "'%s'\n", p);
 620                         ret = -EINVAL;
 621                         goto out;
 622                 default:
 623                         break;
 624                 }
 625         }
 626 out:
 627         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 628                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 629         kfree(orig);
 630         return ret;
 631 }
 632
 633 /*
 634  * Parse mount options that are required early in the mount process.
 635  *
 636  * All other options will be parsed on much later in the mount process and
 637  * only when we need to allocate a new super block.
 638  */
 639 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 640                 void *holder, char **subvol_name, u64 *subvol_objectid,
 641                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
 642 {
 643         substring_t args[MAX_OPT_ARGS];
 644         char *device_name, *opts, *orig, *p;
 645         int error = 0;
 646         int intarg;
 647
 648         if (!options)
 649                 return 0;
 650
 651         /*
 652          * strsep changes the string, duplicate it because parse_options
 653          * gets called twice
 654          */
 655         opts = kstrdup(options, GFP_KERNEL);
 656         if (!opts)
 657                 return -ENOMEM;
 658         orig = opts;
 659
 660         while ((p = strsep(&opts, ",")) != NULL) {
 661                 int token;
 662                 if (!*p)
 663                         continue;
 664
 665                 token = match_token(p, tokens, args);
 666                 switch (token) {
 667                 case Opt_subvol:
 668                         kfree(*subvol_name);
 669                         *subvol_name = match_strdup(&args[0]);
 670                         break;
 671                 case Opt_subvolid:
 672                         intarg = 0;
 673                         error = match_int(&args[0], &intarg);
 674                         if (!error) {
 675                                 /* we want the original fs_tree */
 676                                 if (!intarg)
 677                                         *subvol_objectid =
 678                                                 BTRFS_FS_TREE_OBJECTID;
 679                                 else
 680                                         *subvol_objectid = intarg;
 681                         }
 682                         break;
 683                 case Opt_subvolrootid:
 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_rootid =
 690                                                 BTRFS_FS_TREE_OBJECTID;
 691                                 else
 692                                         *subvol_rootid = intarg;
 693                         }
 694                         break;
 695                 case Opt_device:
 696                         device_name = match_strdup(&args[0]);
 697                         if (!device_name) {
 698                                 error = -ENOMEM;
 699                                 goto out;
 700                         }
 701                         error = btrfs_scan_one_device(device_name,
 702                                         flags, holder, fs_devices);
 703                         kfree(device_name);
 704                         if (error)
 705                                 goto out;
 706                         break;
 707                 default:
 708                         break;
 709                 }
 710         }
 711
 712 out:
 713         kfree(orig);
 714         return error;
 715 }
 716
 717 static struct dentry *get_default_root(struct super_block *sb,
 718                                        u64 subvol_objectid)
 719 {
 720         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 721         struct btrfs_root *root = fs_info->tree_root;
 722         struct btrfs_root *new_root;
 723         struct btrfs_dir_item *di;
 724         struct btrfs_path *path;
 725         struct btrfs_key location;
 726         struct inode *inode;
 727         u64 dir_id;
 728         int new = 0;
 729
 730         /*
 731          * We have a specific subvol we want to mount, just setup location and
 732          * go look up the root.
 733          */
 734         if (subvol_objectid) {
 735                 location.objectid = subvol_objectid;
 736                 location.type = BTRFS_ROOT_ITEM_KEY;
 737                 location.offset = (u64)-1;
 738                 goto find_root;
 739         }
 740
 741         path = btrfs_alloc_path();
 742         if (!path)
 743                 return ERR_PTR(-ENOMEM);
 744         path->leave_spinning = 1;
 745
 746         /*
 747          * Find the "default" dir item which points to the root item that we
 748          * will mount by default if we haven't been given a specific subvolume
 749          * to mount.
 750          */
 751         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 752         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 753         if (IS_ERR(di)) {
 754                 btrfs_free_path(path);
 755                 return ERR_CAST(di);
 756         }
 757         if (!di) {
 758                 /*
 759                  * Ok the default dir item isn't there.  This is weird since
 760                  * it's always been there, but don't freak out, just try and
 761                  * mount to root most subvolume.
 762                  */
 763                 btrfs_free_path(path);
 764                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 765                 new_root = fs_info->fs_root;
 766                 goto setup_root;
 767         }
 768
 769         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 770         btrfs_free_path(path);
 771
 772 find_root:
 773         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 774         if (IS_ERR(new_root))
 775                 return ERR_CAST(new_root);
 776
 777         if (btrfs_root_refs(&new_root->root_item) == 0)
 778                 return ERR_PTR(-ENOENT);
 779
 780         dir_id = btrfs_root_dirid(&new_root->root_item);
 781 setup_root:
 782         location.objectid = dir_id;
 783         location.type = BTRFS_INODE_ITEM_KEY;
 784         location.offset = 0;
 785
 786         inode = btrfs_iget(sb, &location, new_root, &new);
 787         if (IS_ERR(inode))
 788                 return ERR_CAST(inode);
 789
 790         /*
 791          * If we're just mounting the root most subvol put the inode and return
 792          * a reference to the dentry.  We will have already gotten a reference
 793          * to the inode in btrfs_fill_super so we're good to go.
 794          */
 795         if (!new && sb->s_root->d_inode == inode) {
 796                 iput(inode);
 797                 return dget(sb->s_root);
 798         }
 799
 800         return d_obtain_alias(inode);
 801 }
 802
 803 static int btrfs_fill_super(struct super_block *sb,
 804                             struct btrfs_fs_devices *fs_devices,
 805                             void *data, int silent)
 806 {
 807         struct inode *inode;
 808         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 809         struct btrfs_key key;
 810         int err;
 811
 812         sb->s_maxbytes = MAX_LFS_FILESIZE;
 813         sb->s_magic = BTRFS_SUPER_MAGIC;
 814         sb->s_op = &btrfs_super_ops;
 815         sb->s_d_op = &btrfs_dentry_operations;
 816         sb->s_export_op = &btrfs_export_ops;
 817         sb->s_xattr = btrfs_xattr_handlers;
 818         sb->s_time_gran = 1;
 819 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 820         sb->s_flags |= MS_POSIXACL;
 821 #endif
 822         sb->s_flags |= MS_I_VERSION;
 823         err = open_ctree(sb, fs_devices, (char *)data);
 824         if (err) {
 825                 printk("btrfs: open_ctree failed\n");
 826                 return err;
 827         }
 828
 829         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 830         key.type = BTRFS_INODE_ITEM_KEY;
 831         key.offset = 0;
 832         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 833         if (IS_ERR(inode)) {
 834                 err = PTR_ERR(inode);
 835                 goto fail_close;
 836         }
 837
 838         sb->s_root = d_make_root(inode);
 839         if (!sb->s_root) {
 840                 err = -ENOMEM;
 841                 goto fail_close;
 842         }
 843
 844         save_mount_options(sb, data);
 845         cleancache_init_fs(sb);
 846         sb->s_flags |= MS_ACTIVE;
 847         return 0;
 848
 849 fail_close:
 850         close_ctree(fs_info->tree_root);
 851         return err;
 852 }
 853
 854 int btrfs_sync_fs(struct super_block *sb, int wait)
 855 {
 856         struct btrfs_trans_handle *trans;
 857         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 858         struct btrfs_root *root = fs_info->tree_root;
 859
 860         trace_btrfs_sync_fs(wait);
 861
 862         if (!wait) {
 863                 filemap_flush(fs_info->btree_inode->i_mapping);
 864                 return 0;
 865         }
 866
 867         btrfs_wait_ordered_extents(root, 0);
 868
 869         trans = btrfs_attach_transaction(root);
 870         if (IS_ERR(trans)) {
 871                 /* no transaction, don't bother */
 872                 if (PTR_ERR(trans) == -ENOENT)
 873                         return 0;
 874                 return PTR_ERR(trans);
 875         }
 876         return btrfs_commit_transaction(trans, root);
 877 }
 878
 879 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 880 {
 881         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 882         struct btrfs_root *root = info->tree_root;
 883         char *compress_type;
 884
 885         if (btrfs_test_opt(root, DEGRADED))
 886                 seq_puts(seq, ",degraded");
 887         if (btrfs_test_opt(root, NODATASUM))
 888                 seq_puts(seq, ",nodatasum");
 889         if (btrfs_test_opt(root, NODATACOW))
 890                 seq_puts(seq, ",nodatacow");
 891         if (btrfs_test_opt(root, NOBARRIER))
 892                 seq_puts(seq, ",nobarrier");
 893         if (info->max_inline != 8192 * 1024)
 894                 seq_printf(seq, ",max_inline=%llu",
 895                            (unsigned long long)info->max_inline);
 896         if (info->alloc_start != 0)
 897                 seq_printf(seq, ",alloc_start=%llu",
 898                            (unsigned long long)info->alloc_start);
 899         if (info->thread_pool_size !=  min_t(unsigned long,
 900                                              num_online_cpus() + 2, 8))
 901                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 902         if (btrfs_test_opt(root, COMPRESS)) {
 903                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 904                         compress_type = "zlib";
 905                 else
 906                         compress_type = "lzo";
 907                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 908                         seq_printf(seq, ",compress-force=%s", compress_type);
 909                 else
 910                         seq_printf(seq, ",compress=%s", compress_type);
 911         }
 912         if (btrfs_test_opt(root, NOSSD))
 913                 seq_puts(seq, ",nossd");
 914         if (btrfs_test_opt(root, SSD_SPREAD))
 915                 seq_puts(seq, ",ssd_spread");
 916         else if (btrfs_test_opt(root, SSD))
 917                 seq_puts(seq, ",ssd");
 918         if (btrfs_test_opt(root, NOTREELOG))
 919                 seq_puts(seq, ",notreelog");
 920         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 921                 seq_puts(seq, ",flushoncommit");
 922         if (btrfs_test_opt(root, DISCARD))
 923                 seq_puts(seq, ",discard");
 924         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 925                 seq_puts(seq, ",noacl");
 926         if (btrfs_test_opt(root, SPACE_CACHE))
 927                 seq_puts(seq, ",space_cache");
 928         else
 929                 seq_puts(seq, ",nospace_cache");
 930         if (btrfs_test_opt(root, CLEAR_CACHE))
 931                 seq_puts(seq, ",clear_cache");
 932         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 933                 seq_puts(seq, ",user_subvol_rm_allowed");
 934         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 935                 seq_puts(seq, ",enospc_debug");
 936         if (btrfs_test_opt(root, AUTO_DEFRAG))
 937                 seq_puts(seq, ",autodefrag");
 938         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 939                 seq_puts(seq, ",inode_cache");
 940         if (btrfs_test_opt(root, SKIP_BALANCE))
 941                 seq_puts(seq, ",skip_balance");
 942         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 943                 seq_puts(seq, ",fatal_errors=panic");
 944         return 0;
 945 }
 946
 947 static int btrfs_test_super(struct super_block *s, void *data)
 948 {
 949         struct btrfs_fs_info *p = data;
 950         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 951
 952         return fs_info->fs_devices == p->fs_devices;
 953 }
 954
 955 static int btrfs_set_super(struct super_block *s, void *data)
 956 {
 957         int err = set_anon_super(s, data);
 958         if (!err)
 959                 s->s_fs_info = data;
 960         return err;
 961 }
 962
 963 /*
 964  * subvolumes are identified by ino 256
 965  */
 966 static inline int is_subvolume_inode(struct inode *inode)
 967 {
 968         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 969                 return 1;
 970         return 0;
 971 }
 972
 973 /*
 974  * This will strip out the subvol=%s argument for an argument string and add
 975  * subvolid=0 to make sure we get the actual tree root for path walking to the
 976  * subvol we want.
 977  */
 978 static char *setup_root_args(char *args)
 979 {
 980         unsigned len = strlen(args) + 2 + 1;
 981         char *src, *dst, *buf;
 982
 983         /*
 984          * We need the same args as before, but with this substitution:
 985          * s!subvol=[^,]+!subvolid=0!
 986          *
 987          * Since the replacement string is up to 2 bytes longer than the
 988          * original, allocate strlen(args) + 2 + 1 bytes.
 989          */
 990
 991         src = strstr(args, "subvol=");
 992         /* This shouldn't happen, but just in case.. */
 993         if (!src)
 994                 return NULL;
 995
 996         buf = dst = kmalloc(len, GFP_NOFS);
 997         if (!buf)
 998                 return NULL;
 999
1000         /*
1001          * If the subvol= arg is not at the start of the string,
1002          * copy whatever precedes it into buf.
1003          */
1004         if (src != args) {
1005                 *src++ = '\0';
1006                 strcpy(buf, args);
1007                 dst += strlen(args);
1008         }
1009
1010         strcpy(dst, "subvolid=0");
1011         dst += strlen("subvolid=0");
1012
1013         /*
1014          * If there is a "," after the original subvol=... string,
1015          * copy that suffix into our buffer.  Otherwise, we're done.
1016          */
1017         src = strchr(src, ',');
1018         if (src)
1019                 strcpy(dst, src);
1020
1021         return buf;
1022 }
1023
1024 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1025                                    const char *device_name, char *data)
1026 {
1027         struct dentry *root;
1028         struct vfsmount *mnt;
1029         char *newargs;
1030
1031         newargs = setup_root_args(data);
1032         if (!newargs)
1033                 return ERR_PTR(-ENOMEM);
1034         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1035                              newargs);
1036         kfree(newargs);
1037         if (IS_ERR(mnt))
1038                 return ERR_CAST(mnt);
1039
1040         root = mount_subtree(mnt, subvol_name);
1041
1042         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1043                 struct super_block *s = root->d_sb;
1044                 dput(root);
1045                 root = ERR_PTR(-EINVAL);
1046                 deactivate_locked_super(s);
1047                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1048                                 subvol_name);
1049         }
1050
1051         return root;
1052 }
1053
1054 /*
1055  * Find a superblock for the given device / mount point.
1056  *
1057  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1058  *        for multiple device setup.  Make sure to keep it in sync.
1059  */
1060 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1061                 const char *device_name, void *data)
1062 {
1063         struct block_device *bdev = NULL;
1064         struct super_block *s;
1065         struct dentry *root;
1066         struct btrfs_fs_devices *fs_devices = NULL;
1067         struct btrfs_fs_info *fs_info = NULL;
1068         fmode_t mode = FMODE_READ;
1069         char *subvol_name = NULL;
1070         u64 subvol_objectid = 0;
1071         u64 subvol_rootid = 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                                           &subvol_rootid, &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_workers, new_pool_size);
1190 }
1191
1192 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1193 {
1194         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1195         struct btrfs_root *root = fs_info->tree_root;
1196         unsigned old_flags = sb->s_flags;
1197         unsigned long old_opts = fs_info->mount_opt;
1198         unsigned long old_compress_type = fs_info->compress_type;
1199         u64 old_max_inline = fs_info->max_inline;
1200         u64 old_alloc_start = fs_info->alloc_start;
1201         int old_thread_pool_size = fs_info->thread_pool_size;
1202         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1203         int ret;
1204
1205         ret = btrfs_parse_options(root, data);
1206         if (ret) {
1207                 ret = -EINVAL;
1208                 goto restore;
1209         }
1210
1211         btrfs_resize_thread_pool(fs_info,
1212                 fs_info->thread_pool_size, old_thread_pool_size);
1213
1214         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1215                 return 0;
1216
1217         if (*flags & MS_RDONLY) {
1218                 sb->s_flags |= MS_RDONLY;
1219
1220                 ret = btrfs_commit_super(root);
1221                 if (ret)
1222                         goto restore;
1223         } else {
1224                 if (fs_info->fs_devices->rw_devices == 0) {
1225                         ret = -EACCES;
1226                         goto restore;
1227                 }
1228
1229                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1230                         ret = -EINVAL;
1231                         goto restore;
1232                 }
1233
1234                 ret = btrfs_cleanup_fs_roots(fs_info);
1235                 if (ret)
1236                         goto restore;
1237
1238                 /* recover relocation */
1239                 ret = btrfs_recover_relocation(root);
1240                 if (ret)
1241                         goto restore;
1242
1243                 ret = btrfs_resume_balance_async(fs_info);
1244                 if (ret)
1245                         goto restore;
1246
1247                 sb->s_flags &= ~MS_RDONLY;
1248         }
1249
1250         return 0;
1251
1252 restore:
1253         /* We've hit an error - don't reset MS_RDONLY */
1254         if (sb->s_flags & MS_RDONLY)
1255                 old_flags |= MS_RDONLY;
1256         sb->s_flags = old_flags;
1257         fs_info->mount_opt = old_opts;
1258         fs_info->compress_type = old_compress_type;
1259         fs_info->max_inline = old_max_inline;
1260         fs_info->alloc_start = old_alloc_start;
1261         btrfs_resize_thread_pool(fs_info,
1262                 old_thread_pool_size, fs_info->thread_pool_size);
1263         fs_info->metadata_ratio = old_metadata_ratio;
1264         return ret;
1265 }
1266
1267 /* Used to sort the devices by max_avail(descending sort) */
1268 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1269                                        const void *dev_info2)
1270 {
1271         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1272             ((struct btrfs_device_info *)dev_info2)->max_avail)
1273                 return -1;
1274         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1275                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1276                 return 1;
1277         else
1278         return 0;
1279 }
1280
1281 /*
1282  * sort the devices by max_avail, in which max free extent size of each device
1283  * is stored.(Descending Sort)
1284  */
1285 static inline void btrfs_descending_sort_devices(
1286                                         struct btrfs_device_info *devices,
1287                                         size_t nr_devices)
1288 {
1289         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1290              btrfs_cmp_device_free_bytes, NULL);
1291 }
1292
1293 /*
1294  * The helper to calc the free space on the devices that can be used to store
1295  * file data.
1296  */
1297 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1298 {
1299         struct btrfs_fs_info *fs_info = root->fs_info;
1300         struct btrfs_device_info *devices_info;
1301         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1302         struct btrfs_device *device;
1303         u64 skip_space;
1304         u64 type;
1305         u64 avail_space;
1306         u64 used_space;
1307         u64 min_stripe_size;
1308         int min_stripes = 1, num_stripes = 1;
1309         int i = 0, nr_devices;
1310         int ret;
1311
1312         nr_devices = fs_info->fs_devices->open_devices;
1313         BUG_ON(!nr_devices);
1314
1315         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1316                                GFP_NOFS);
1317         if (!devices_info)
1318                 return -ENOMEM;
1319
1320         /* calc min stripe number for data space alloction */
1321         type = btrfs_get_alloc_profile(root, 1);
1322         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1323                 min_stripes = 2;
1324                 num_stripes = nr_devices;
1325         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1326                 min_stripes = 2;
1327                 num_stripes = 2;
1328         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1329                 min_stripes = 4;
1330                 num_stripes = 4;
1331         }
1332
1333         if (type & BTRFS_BLOCK_GROUP_DUP)
1334                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1335         else
1336                 min_stripe_size = BTRFS_STRIPE_LEN;
1337
1338         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1339                 if (!device->in_fs_metadata || !device->bdev)
1340                         continue;
1341
1342                 avail_space = device->total_bytes - device->bytes_used;
1343
1344                 /* align with stripe_len */
1345                 do_div(avail_space, BTRFS_STRIPE_LEN);
1346                 avail_space *= BTRFS_STRIPE_LEN;
1347
1348                 /*
1349                  * In order to avoid overwritting the superblock on the drive,
1350                  * btrfs starts at an offset of at least 1MB when doing chunk
1351                  * allocation.
1352                  */
1353                 skip_space = 1024 * 1024;
1354
1355                 /* user can set the offset in fs_info->alloc_start. */
1356                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1357                     device->total_bytes)
1358                         skip_space = max(fs_info->alloc_start, skip_space);
1359
1360                 /*
1361                  * btrfs can not use the free space in [0, skip_space - 1],
1362                  * we must subtract it from the total. In order to implement
1363                  * it, we account the used space in this range first.
1364                  */
1365                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1366                                                      &used_space);
1367                 if (ret) {
1368                         kfree(devices_info);
1369                         return ret;
1370                 }
1371
1372                 /* calc the free space in [0, skip_space - 1] */
1373                 skip_space -= used_space;
1374
1375                 /*
1376                  * we can use the free space in [0, skip_space - 1], subtract
1377                  * it from the total.
1378                  */
1379                 if (avail_space && avail_space >= skip_space)
1380                         avail_space -= skip_space;
1381                 else
1382                         avail_space = 0;
1383
1384                 if (avail_space < min_stripe_size)
1385                         continue;
1386
1387                 devices_info[i].dev = device;
1388                 devices_info[i].max_avail = avail_space;
1389
1390                 i++;
1391         }
1392
1393         nr_devices = i;
1394
1395         btrfs_descending_sort_devices(devices_info, nr_devices);
1396
1397         i = nr_devices - 1;
1398         avail_space = 0;
1399         while (nr_devices >= min_stripes) {
1400                 if (num_stripes > nr_devices)
1401                         num_stripes = nr_devices;
1402
1403                 if (devices_info[i].max_avail >= min_stripe_size) {
1404                         int j;
1405                         u64 alloc_size;
1406
1407                         avail_space += devices_info[i].max_avail * num_stripes;
1408                         alloc_size = devices_info[i].max_avail;
1409                         for (j = i + 1 - num_stripes; j <= i; j++)
1410                                 devices_info[j].max_avail -= alloc_size;
1411                 }
1412                 i--;
1413                 nr_devices--;
1414         }
1415
1416         kfree(devices_info);
1417         *free_bytes = avail_space;
1418         return 0;
1419 }
1420
1421 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1422 {
1423         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1424         struct btrfs_super_block *disk_super = fs_info->super_copy;
1425         struct list_head *head = &fs_info->space_info;
1426         struct btrfs_space_info *found;
1427         u64 total_used = 0;
1428         u64 total_free_data = 0;
1429         int bits = dentry->d_sb->s_blocksize_bits;
1430         __be32 *fsid = (__be32 *)fs_info->fsid;
1431         int ret;
1432
1433         /* holding chunk_muext to avoid allocating new chunks */
1434         mutex_lock(&fs_info->chunk_mutex);
1435         rcu_read_lock();
1436         list_for_each_entry_rcu(found, head, list) {
1437                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1438                         total_free_data += found->disk_total - found->disk_used;
1439                         total_free_data -=
1440                                 btrfs_account_ro_block_groups_free_space(found);
1441                 }
1442
1443                 total_used += found->disk_used;
1444         }
1445         rcu_read_unlock();
1446
1447         buf->f_namelen = BTRFS_NAME_LEN;
1448         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1449         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1450         buf->f_bsize = dentry->d_sb->s_blocksize;
1451         buf->f_type = BTRFS_SUPER_MAGIC;
1452         buf->f_bavail = total_free_data;
1453         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1454         if (ret) {
1455                 mutex_unlock(&fs_info->chunk_mutex);
1456                 return ret;
1457         }
1458         buf->f_bavail += total_free_data;
1459         buf->f_bavail = buf->f_bavail >> bits;
1460         mutex_unlock(&fs_info->chunk_mutex);
1461
1462         /* We treat it as constant endianness (it doesn't matter _which_)
1463            because we want the fsid to come out the same whether mounted
1464            on a big-endian or little-endian host */
1465         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1466         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1467         /* Mask in the root object ID too, to disambiguate subvols */
1468         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1469         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1470
1471         return 0;
1472 }
1473
1474 static void btrfs_kill_super(struct super_block *sb)
1475 {
1476         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1477         kill_anon_super(sb);
1478         free_fs_info(fs_info);
1479 }
1480
1481 static struct file_system_type btrfs_fs_type = {
1482         .owner          = THIS_MODULE,
1483         .name           = "btrfs",
1484         .mount          = btrfs_mount,
1485         .kill_sb        = btrfs_kill_super,
1486         .fs_flags       = FS_REQUIRES_DEV,
1487 };
1488
1489 /*
1490  * used by btrfsctl to scan devices when no FS is mounted
1491  */
1492 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1493                                 unsigned long arg)
1494 {
1495         struct btrfs_ioctl_vol_args *vol;
1496         struct btrfs_fs_devices *fs_devices;
1497         int ret = -ENOTTY;
1498
1499         if (!capable(CAP_SYS_ADMIN))
1500                 return -EPERM;
1501
1502         vol = memdup_user((void __user *)arg, sizeof(*vol));
1503         if (IS_ERR(vol))
1504                 return PTR_ERR(vol);
1505
1506         switch (cmd) {
1507         case BTRFS_IOC_SCAN_DEV:
1508                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1509                                             &btrfs_fs_type, &fs_devices);
1510                 break;
1511         case BTRFS_IOC_DEVICES_READY:
1512                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1513                                             &btrfs_fs_type, &fs_devices);
1514                 if (ret)
1515                         break;
1516                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1517                 break;
1518         }
1519
1520         kfree(vol);
1521         return ret;
1522 }
1523
1524 static int btrfs_freeze(struct super_block *sb)
1525 {
1526         struct btrfs_trans_handle *trans;
1527         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1528
1529         trans = btrfs_attach_transaction(root);
1530         if (IS_ERR(trans)) {
1531                 /* no transaction, don't bother */
1532                 if (PTR_ERR(trans) == -ENOENT)
1533                         return 0;
1534                 return PTR_ERR(trans);
1535         }
1536         return btrfs_commit_transaction(trans, root);
1537 }
1538
1539 static int btrfs_unfreeze(struct super_block *sb)
1540 {
1541         return 0;
1542 }
1543
1544 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1545 {
1546         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1547         struct btrfs_fs_devices *cur_devices;
1548         struct btrfs_device *dev, *first_dev = NULL;
1549         struct list_head *head;
1550         struct rcu_string *name;
1551
1552         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1553         cur_devices = fs_info->fs_devices;
1554         while (cur_devices) {
1555                 head = &cur_devices->devices;
1556                 list_for_each_entry(dev, head, dev_list) {
1557                         if (dev->missing)
1558                                 continue;
1559                         if (!first_dev || dev->devid < first_dev->devid)
1560                                 first_dev = dev;
1561                 }
1562                 cur_devices = cur_devices->seed;
1563         }
1564
1565         if (first_dev) {
1566                 rcu_read_lock();
1567                 name = rcu_dereference(first_dev->name);
1568                 seq_escape(m, name->str, " \t\n\\");
1569                 rcu_read_unlock();
1570         } else {
1571                 WARN_ON(1);
1572         }
1573         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1574         return 0;
1575 }
1576
1577 static const struct super_operations btrfs_super_ops = {
1578         .drop_inode     = btrfs_drop_inode,
1579         .evict_inode    = btrfs_evict_inode,
1580         .put_super      = btrfs_put_super,
1581         .sync_fs        = btrfs_sync_fs,
1582         .show_options   = btrfs_show_options,
1583         .show_devname   = btrfs_show_devname,
1584         .write_inode    = btrfs_write_inode,
1585         .alloc_inode    = btrfs_alloc_inode,
1586         .destroy_inode  = btrfs_destroy_inode,
1587         .statfs         = btrfs_statfs,
1588         .remount_fs     = btrfs_remount,
1589         .freeze_fs      = btrfs_freeze,
1590         .unfreeze_fs    = btrfs_unfreeze,
1591 };
1592
1593 static const struct file_operations btrfs_ctl_fops = {
1594         .unlocked_ioctl  = btrfs_control_ioctl,
1595         .compat_ioctl = btrfs_control_ioctl,
1596         .owner   = THIS_MODULE,
1597         .llseek = noop_llseek,
1598 };
1599
1600 static struct miscdevice btrfs_misc = {
1601         .minor          = BTRFS_MINOR,
1602         .name           = "btrfs-control",
1603         .fops           = &btrfs_ctl_fops
1604 };
1605
1606 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1607 MODULE_ALIAS("devname:btrfs-control");
1608
1609 static int btrfs_interface_init(void)
1610 {
1611         return misc_register(&btrfs_misc);
1612 }
1613
1614 static void btrfs_interface_exit(void)
1615 {
1616         if (misc_deregister(&btrfs_misc) < 0)
1617                 printk(KERN_INFO "btrfs: misc_deregister failed for control device\n");
1618 }
1619
1620 static int __init init_btrfs_fs(void)
1621 {
1622         int err;
1623
1624         err = btrfs_init_sysfs();
1625         if (err)
1626                 return err;
1627
1628         btrfs_init_compress();
1629
1630         err = btrfs_init_cachep();
1631         if (err)
1632                 goto free_compress;
1633
1634         err = extent_io_init();
1635         if (err)
1636                 goto free_cachep;
1637
1638         err = extent_map_init();
1639         if (err)
1640                 goto free_extent_io;
1641
1642         err = ordered_data_init();
1643         if (err)
1644                 goto free_extent_map;
1645
1646         err = btrfs_delayed_inode_init();
1647         if (err)
1648                 goto free_ordered_data;
1649
1650         err = btrfs_interface_init();
1651         if (err)
1652                 goto free_delayed_inode;
1653
1654         err = register_filesystem(&btrfs_fs_type);
1655         if (err)
1656                 goto unregister_ioctl;
1657
1658         btrfs_init_lockdep();
1659
1660         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1661         return 0;
1662
1663 unregister_ioctl:
1664         btrfs_interface_exit();
1665 free_delayed_inode:
1666         btrfs_delayed_inode_exit();
1667 free_ordered_data:
1668         ordered_data_exit();
1669 free_extent_map:
1670         extent_map_exit();
1671 free_extent_io:
1672         extent_io_exit();
1673 free_cachep:
1674         btrfs_destroy_cachep();
1675 free_compress:
1676         btrfs_exit_compress();
1677         btrfs_exit_sysfs();
1678         return err;
1679 }
1680
1681 static void __exit exit_btrfs_fs(void)
1682 {
1683         btrfs_destroy_cachep();
1684         btrfs_delayed_inode_exit();
1685         ordered_data_exit();
1686         extent_map_exit();
1687         extent_io_exit();
1688         btrfs_interface_exit();
1689         unregister_filesystem(&btrfs_fs_type);
1690         btrfs_exit_sysfs();
1691         btrfs_cleanup_fs_uuids();
1692         btrfs_exit_compress();
1693 }
1694
1695 module_init(init_btrfs_fs)
1696 module_exit(exit_btrfs_fs)
1697
1698 MODULE_LICENSE("GPL");