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");
 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                         printk(KERN_INFO "btrfs: setting nodatacow\n");
 417                         btrfs_set_opt(info->mount_opt, NODATACOW);
 418                         btrfs_set_opt(info->mount_opt, NODATASUM);
 419                         break;
 420                 case Opt_compress_force:
 421                 case Opt_compress_force_type:
 422                         compress_force = true;
 423                 case Opt_compress:
 424                 case Opt_compress_type:
 425                         if (token == Opt_compress ||
 426                             token == Opt_compress_force ||
 427                             strcmp(args[0].from, "zlib") == 0) {
 428                                 compress_type = "zlib";
 429                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 430                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 431                         } else if (strcmp(args[0].from, "lzo") == 0) {
 432                                 compress_type = "lzo";
 433                                 info->compress_type = BTRFS_COMPRESS_LZO;
 434                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 435                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 436                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 437                                 compress_type = "no";
 438                                 info->compress_type = BTRFS_COMPRESS_NONE;
 439                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 440                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 441                                 compress_force = false;
 442                         } else {
 443                                 ret = -EINVAL;
 444                                 goto out;
 445                         }
 446
 447                         if (compress_force) {
 448                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 449                                 pr_info("btrfs: force %s compression\n",
 450                                         compress_type);
 451                         } else
 452                                 pr_info("btrfs: use %s compression\n",
 453                                         compress_type);
 454                         break;
 455                 case Opt_ssd:
 456                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 457                         btrfs_set_opt(info->mount_opt, SSD);
 458                         break;
 459                 case Opt_ssd_spread:
 460                         printk(KERN_INFO "btrfs: use spread ssd "
 461                                "allocation scheme\n");
 462                         btrfs_set_opt(info->mount_opt, SSD);
 463                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 464                         break;
 465                 case Opt_nossd:
 466                         printk(KERN_INFO "btrfs: not using ssd allocation "
 467                                "scheme\n");
 468                         btrfs_set_opt(info->mount_opt, NOSSD);
 469                         btrfs_clear_opt(info->mount_opt, SSD);
 470                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 471                         break;
 472                 case Opt_nobarrier:
 473                         printk(KERN_INFO "btrfs: turning off barriers\n");
 474                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 475                         break;
 476                 case Opt_thread_pool:
 477                         intarg = 0;
 478                         match_int(&args[0], &intarg);
 479                         if (intarg)
 480                                 info->thread_pool_size = intarg;
 481                         break;
 482                 case Opt_max_inline:
 483                         num = match_strdup(&args[0]);
 484                         if (num) {
 485                                 info->max_inline = memparse(num, NULL);
 486                                 kfree(num);
 487
 488                                 if (info->max_inline) {
 489                                         info->max_inline = max_t(u64,
 490                                                 info->max_inline,
 491                                                 root->sectorsize);
 492                                 }
 493                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 494                                         (unsigned long long)info->max_inline);
 495                         }
 496                         break;
 497                 case Opt_alloc_start:
 498                         num = match_strdup(&args[0]);
 499                         if (num) {
 500                                 info->alloc_start = memparse(num, NULL);
 501                                 kfree(num);
 502                                 printk(KERN_INFO
 503                                         "btrfs: allocations start at %llu\n",
 504                                         (unsigned long long)info->alloc_start);
 505                         }
 506                         break;
 507                 case Opt_noacl:
 508                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 509                         break;
 510                 case Opt_notreelog:
 511                         printk(KERN_INFO "btrfs: disabling tree log\n");
 512                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 513                         break;
 514                 case Opt_flushoncommit:
 515                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 516                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 517                         break;
 518                 case Opt_ratio:
 519                         intarg = 0;
 520                         match_int(&args[0], &intarg);
 521                         if (intarg) {
 522                                 info->metadata_ratio = intarg;
 523                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 524                                        info->metadata_ratio);
 525                         }
 526                         break;
 527                 case Opt_discard:
 528                         btrfs_set_opt(info->mount_opt, DISCARD);
 529                         break;
 530                 case Opt_space_cache:
 531                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 532                         break;
 533                 case Opt_no_space_cache:
 534                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 535                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 536                         break;
 537                 case Opt_inode_cache:
 538                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 539                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 540                         break;
 541                 case Opt_clear_cache:
 542                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 543                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 544                         break;
 545                 case Opt_user_subvol_rm_allowed:
 546                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 547                         break;
 548                 case Opt_enospc_debug:
 549                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 550                         break;
 551                 case Opt_defrag:
 552                         printk(KERN_INFO "btrfs: enabling auto defrag");
 553                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 554                         break;
 555                 case Opt_recovery:
 556                         printk(KERN_INFO "btrfs: enabling auto recovery");
 557                         btrfs_set_opt(info->mount_opt, RECOVERY);
 558                         break;
 559                 case Opt_skip_balance:
 560                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 561                         break;
 562 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 563                 case Opt_check_integrity_including_extent_data:
 564                         printk(KERN_INFO "btrfs: enabling check integrity"
 565                                " including extent data\n");
 566                         btrfs_set_opt(info->mount_opt,
 567                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 568                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 569                         break;
 570                 case Opt_check_integrity:
 571                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 572                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 573                         break;
 574                 case Opt_check_integrity_print_mask:
 575                         intarg = 0;
 576                         match_int(&args[0], &intarg);
 577                         if (intarg) {
 578                                 info->check_integrity_print_mask = intarg;
 579                                 printk(KERN_INFO "btrfs:"
 580                                        " check_integrity_print_mask 0x%x\n",
 581                                        info->check_integrity_print_mask);
 582                         }
 583                         break;
 584 #else
 585                 case Opt_check_integrity_including_extent_data:
 586                 case Opt_check_integrity:
 587                 case Opt_check_integrity_print_mask:
 588                         printk(KERN_ERR "btrfs: support for check_integrity*"
 589                                " not compiled in!\n");
 590                         ret = -EINVAL;
 591                         goto out;
 592 #endif
 593                 case Opt_fatal_errors:
 594                         if (strcmp(args[0].from, "panic") == 0)
 595                                 btrfs_set_opt(info->mount_opt,
 596                                               PANIC_ON_FATAL_ERROR);
 597                         else if (strcmp(args[0].from, "bug") == 0)
 598                                 btrfs_clear_opt(info->mount_opt,
 599                                               PANIC_ON_FATAL_ERROR);
 600                         else {
 601                                 ret = -EINVAL;
 602                                 goto out;
 603                         }
 604                         break;
 605                 case Opt_err:
 606                         printk(KERN_INFO "btrfs: unrecognized mount option "
 607                                "'%s'\n", p);
 608                         ret = -EINVAL;
 609                         goto out;
 610                 default:
 611                         break;
 612                 }
 613         }
 614 out:
 615         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 616                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 617         kfree(orig);
 618         return ret;
 619 }
 620
 621 /*
 622  * Parse mount options that are required early in the mount process.
 623  *
 624  * All other options will be parsed on much later in the mount process and
 625  * only when we need to allocate a new super block.
 626  */
 627 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 628                 void *holder, char **subvol_name, u64 *subvol_objectid,
 629                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
 630 {
 631         substring_t args[MAX_OPT_ARGS];
 632         char *device_name, *opts, *orig, *p;
 633         int error = 0;
 634         int intarg;
 635
 636         if (!options)
 637                 return 0;
 638
 639         /*
 640          * strsep changes the string, duplicate it because parse_options
 641          * gets called twice
 642          */
 643         opts = kstrdup(options, GFP_KERNEL);
 644         if (!opts)
 645                 return -ENOMEM;
 646         orig = opts;
 647
 648         while ((p = strsep(&opts, ",")) != NULL) {
 649                 int token;
 650                 if (!*p)
 651                         continue;
 652
 653                 token = match_token(p, tokens, args);
 654                 switch (token) {
 655                 case Opt_subvol:
 656                         kfree(*subvol_name);
 657                         *subvol_name = match_strdup(&args[0]);
 658                         break;
 659                 case Opt_subvolid:
 660                         intarg = 0;
 661                         error = match_int(&args[0], &intarg);
 662                         if (!error) {
 663                                 /* we want the original fs_tree */
 664                                 if (!intarg)
 665                                         *subvol_objectid =
 666                                                 BTRFS_FS_TREE_OBJECTID;
 667                                 else
 668                                         *subvol_objectid = intarg;
 669                         }
 670                         break;
 671                 case Opt_subvolrootid:
 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_rootid =
 678                                                 BTRFS_FS_TREE_OBJECTID;
 679                                 else
 680                                         *subvol_rootid = intarg;
 681                         }
 682                         break;
 683                 case Opt_device:
 684                         device_name = match_strdup(&args[0]);
 685                         if (!device_name) {
 686                                 error = -ENOMEM;
 687                                 goto out;
 688                         }
 689                         error = btrfs_scan_one_device(device_name,
 690                                         flags, holder, fs_devices);
 691                         kfree(device_name);
 692                         if (error)
 693                                 goto out;
 694                         break;
 695                 default:
 696                         break;
 697                 }
 698         }
 699
 700 out:
 701         kfree(orig);
 702         return error;
 703 }
 704
 705 static struct dentry *get_default_root(struct super_block *sb,
 706                                        u64 subvol_objectid)
 707 {
 708         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 709         struct btrfs_root *root = fs_info->tree_root;
 710         struct btrfs_root *new_root;
 711         struct btrfs_dir_item *di;
 712         struct btrfs_path *path;
 713         struct btrfs_key location;
 714         struct inode *inode;
 715         u64 dir_id;
 716         int new = 0;
 717
 718         /*
 719          * We have a specific subvol we want to mount, just setup location and
 720          * go look up the root.
 721          */
 722         if (subvol_objectid) {
 723                 location.objectid = subvol_objectid;
 724                 location.type = BTRFS_ROOT_ITEM_KEY;
 725                 location.offset = (u64)-1;
 726                 goto find_root;
 727         }
 728
 729         path = btrfs_alloc_path();
 730         if (!path)
 731                 return ERR_PTR(-ENOMEM);
 732         path->leave_spinning = 1;
 733
 734         /*
 735          * Find the "default" dir item which points to the root item that we
 736          * will mount by default if we haven't been given a specific subvolume
 737          * to mount.
 738          */
 739         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 740         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 741         if (IS_ERR(di)) {
 742                 btrfs_free_path(path);
 743                 return ERR_CAST(di);
 744         }
 745         if (!di) {
 746                 /*
 747                  * Ok the default dir item isn't there.  This is weird since
 748                  * it's always been there, but don't freak out, just try and
 749                  * mount to root most subvolume.
 750                  */
 751                 btrfs_free_path(path);
 752                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 753                 new_root = fs_info->fs_root;
 754                 goto setup_root;
 755         }
 756
 757         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 758         btrfs_free_path(path);
 759
 760 find_root:
 761         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 762         if (IS_ERR(new_root))
 763                 return ERR_CAST(new_root);
 764
 765         if (btrfs_root_refs(&new_root->root_item) == 0)
 766                 return ERR_PTR(-ENOENT);
 767
 768         dir_id = btrfs_root_dirid(&new_root->root_item);
 769 setup_root:
 770         location.objectid = dir_id;
 771         location.type = BTRFS_INODE_ITEM_KEY;
 772         location.offset = 0;
 773
 774         inode = btrfs_iget(sb, &location, new_root, &new);
 775         if (IS_ERR(inode))
 776                 return ERR_CAST(inode);
 777
 778         /*
 779          * If we're just mounting the root most subvol put the inode and return
 780          * a reference to the dentry.  We will have already gotten a reference
 781          * to the inode in btrfs_fill_super so we're good to go.
 782          */
 783         if (!new && sb->s_root->d_inode == inode) {
 784                 iput(inode);
 785                 return dget(sb->s_root);
 786         }
 787
 788         return d_obtain_alias(inode);
 789 }
 790
 791 static int btrfs_fill_super(struct super_block *sb,
 792                             struct btrfs_fs_devices *fs_devices,
 793                             void *data, int silent)
 794 {
 795         struct inode *inode;
 796         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 797         struct btrfs_key key;
 798         int err;
 799
 800         sb->s_maxbytes = MAX_LFS_FILESIZE;
 801         sb->s_magic = BTRFS_SUPER_MAGIC;
 802         sb->s_op = &btrfs_super_ops;
 803         sb->s_d_op = &btrfs_dentry_operations;
 804         sb->s_export_op = &btrfs_export_ops;
 805         sb->s_xattr = btrfs_xattr_handlers;
 806         sb->s_time_gran = 1;
 807 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 808         sb->s_flags |= MS_POSIXACL;
 809 #endif
 810         sb->s_flags |= MS_I_VERSION;
 811         err = open_ctree(sb, fs_devices, (char *)data);
 812         if (err) {
 813                 printk("btrfs: open_ctree failed\n");
 814                 return err;
 815         }
 816
 817         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 818         key.type = BTRFS_INODE_ITEM_KEY;
 819         key.offset = 0;
 820         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 821         if (IS_ERR(inode)) {
 822                 err = PTR_ERR(inode);
 823                 goto fail_close;
 824         }
 825
 826         sb->s_root = d_make_root(inode);
 827         if (!sb->s_root) {
 828                 err = -ENOMEM;
 829                 goto fail_close;
 830         }
 831
 832         save_mount_options(sb, data);
 833         cleancache_init_fs(sb);
 834         sb->s_flags |= MS_ACTIVE;
 835         return 0;
 836
 837 fail_close:
 838         close_ctree(fs_info->tree_root);
 839         return err;
 840 }
 841
 842 int btrfs_sync_fs(struct super_block *sb, int wait)
 843 {
 844         struct btrfs_trans_handle *trans;
 845         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 846         struct btrfs_root *root = fs_info->tree_root;
 847
 848         trace_btrfs_sync_fs(wait);
 849
 850         if (!wait) {
 851                 filemap_flush(fs_info->btree_inode->i_mapping);
 852                 return 0;
 853         }
 854
 855         btrfs_wait_ordered_extents(root, 0);
 856
 857         trans = btrfs_join_transaction_freeze(root);
 858         if (IS_ERR(trans)) {
 859                 /* Frozen, don't bother */
 860                 if (PTR_ERR(trans) == -EPERM)
 861                         return 0;
 862                 return PTR_ERR(trans);
 863         }
 864         return btrfs_commit_transaction(trans, root);
 865 }
 866
 867 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 868 {
 869         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 870         struct btrfs_root *root = info->tree_root;
 871         char *compress_type;
 872
 873         if (btrfs_test_opt(root, DEGRADED))
 874                 seq_puts(seq, ",degraded");
 875         if (btrfs_test_opt(root, NODATASUM))
 876                 seq_puts(seq, ",nodatasum");
 877         if (btrfs_test_opt(root, NODATACOW))
 878                 seq_puts(seq, ",nodatacow");
 879         if (btrfs_test_opt(root, NOBARRIER))
 880                 seq_puts(seq, ",nobarrier");
 881         if (info->max_inline != 8192 * 1024)
 882                 seq_printf(seq, ",max_inline=%llu",
 883                            (unsigned long long)info->max_inline);
 884         if (info->alloc_start != 0)
 885                 seq_printf(seq, ",alloc_start=%llu",
 886                            (unsigned long long)info->alloc_start);
 887         if (info->thread_pool_size !=  min_t(unsigned long,
 888                                              num_online_cpus() + 2, 8))
 889                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 890         if (btrfs_test_opt(root, COMPRESS)) {
 891                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 892                         compress_type = "zlib";
 893                 else
 894                         compress_type = "lzo";
 895                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 896                         seq_printf(seq, ",compress-force=%s", compress_type);
 897                 else
 898                         seq_printf(seq, ",compress=%s", compress_type);
 899         }
 900         if (btrfs_test_opt(root, NOSSD))
 901                 seq_puts(seq, ",nossd");
 902         if (btrfs_test_opt(root, SSD_SPREAD))
 903                 seq_puts(seq, ",ssd_spread");
 904         else if (btrfs_test_opt(root, SSD))
 905                 seq_puts(seq, ",ssd");
 906         if (btrfs_test_opt(root, NOTREELOG))
 907                 seq_puts(seq, ",notreelog");
 908         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 909                 seq_puts(seq, ",flushoncommit");
 910         if (btrfs_test_opt(root, DISCARD))
 911                 seq_puts(seq, ",discard");
 912         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 913                 seq_puts(seq, ",noacl");
 914         if (btrfs_test_opt(root, SPACE_CACHE))
 915                 seq_puts(seq, ",space_cache");
 916         else
 917                 seq_puts(seq, ",nospace_cache");
 918         if (btrfs_test_opt(root, CLEAR_CACHE))
 919                 seq_puts(seq, ",clear_cache");
 920         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 921                 seq_puts(seq, ",user_subvol_rm_allowed");
 922         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 923                 seq_puts(seq, ",enospc_debug");
 924         if (btrfs_test_opt(root, AUTO_DEFRAG))
 925                 seq_puts(seq, ",autodefrag");
 926         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 927                 seq_puts(seq, ",inode_cache");
 928         if (btrfs_test_opt(root, SKIP_BALANCE))
 929                 seq_puts(seq, ",skip_balance");
 930         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 931                 seq_puts(seq, ",fatal_errors=panic");
 932         return 0;
 933 }
 934
 935 static int btrfs_test_super(struct super_block *s, void *data)
 936 {
 937         struct btrfs_fs_info *p = data;
 938         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 939
 940         return fs_info->fs_devices == p->fs_devices;
 941 }
 942
 943 static int btrfs_set_super(struct super_block *s, void *data)
 944 {
 945         int err = set_anon_super(s, data);
 946         if (!err)
 947                 s->s_fs_info = data;
 948         return err;
 949 }
 950
 951 /*
 952  * subvolumes are identified by ino 256
 953  */
 954 static inline int is_subvolume_inode(struct inode *inode)
 955 {
 956         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 957                 return 1;
 958         return 0;
 959 }
 960
 961 /*
 962  * This will strip out the subvol=%s argument for an argument string and add
 963  * subvolid=0 to make sure we get the actual tree root for path walking to the
 964  * subvol we want.
 965  */
 966 static char *setup_root_args(char *args)
 967 {
 968         unsigned len = strlen(args) + 2 + 1;
 969         char *src, *dst, *buf;
 970
 971         /*
 972          * We need the same args as before, but with this substitution:
 973          * s!subvol=[^,]+!subvolid=0!
 974          *
 975          * Since the replacement string is up to 2 bytes longer than the
 976          * original, allocate strlen(args) + 2 + 1 bytes.
 977          */
 978
 979         src = strstr(args, "subvol=");
 980         /* This shouldn't happen, but just in case.. */
 981         if (!src)
 982                 return NULL;
 983
 984         buf = dst = kmalloc(len, GFP_NOFS);
 985         if (!buf)
 986                 return NULL;
 987
 988         /*
 989          * If the subvol= arg is not at the start of the string,
 990          * copy whatever precedes it into buf.
 991          */
 992         if (src != args) {
 993                 *src++ = '\0';
 994                 strcpy(buf, args);
 995                 dst += strlen(args);
 996         }
 997
 998         strcpy(dst, "subvolid=0");
 999         dst += strlen("subvolid=0");
1000
1001         /*
1002          * If there is a "," after the original subvol=... string,
1003          * copy that suffix into our buffer.  Otherwise, we're done.
1004          */
1005         src = strchr(src, ',');
1006         if (src)
1007                 strcpy(dst, src);
1008
1009         return buf;
1010 }
1011
1012 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1013                                    const char *device_name, char *data)
1014 {
1015         struct dentry *root;
1016         struct vfsmount *mnt;
1017         char *newargs;
1018
1019         newargs = setup_root_args(data);
1020         if (!newargs)
1021                 return ERR_PTR(-ENOMEM);
1022         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1023                              newargs);
1024         kfree(newargs);
1025         if (IS_ERR(mnt))
1026                 return ERR_CAST(mnt);
1027
1028         root = mount_subtree(mnt, subvol_name);
1029
1030         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1031                 struct super_block *s = root->d_sb;
1032                 dput(root);
1033                 root = ERR_PTR(-EINVAL);
1034                 deactivate_locked_super(s);
1035                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1036                                 subvol_name);
1037         }
1038
1039         return root;
1040 }
1041
1042 /*
1043  * Find a superblock for the given device / mount point.
1044  *
1045  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1046  *        for multiple device setup.  Make sure to keep it in sync.
1047  */
1048 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1049                 const char *device_name, void *data)
1050 {
1051         struct block_device *bdev = NULL;
1052         struct super_block *s;
1053         struct dentry *root;
1054         struct btrfs_fs_devices *fs_devices = NULL;
1055         struct btrfs_fs_info *fs_info = NULL;
1056         fmode_t mode = FMODE_READ;
1057         char *subvol_name = NULL;
1058         u64 subvol_objectid = 0;
1059         u64 subvol_rootid = 0;
1060         int error = 0;
1061
1062         if (!(flags & MS_RDONLY))
1063                 mode |= FMODE_WRITE;
1064
1065         error = btrfs_parse_early_options(data, mode, fs_type,
1066                                           &subvol_name, &subvol_objectid,
1067                                           &subvol_rootid, &fs_devices);
1068         if (error) {
1069                 kfree(subvol_name);
1070                 return ERR_PTR(error);
1071         }
1072
1073         if (subvol_name) {
1074                 root = mount_subvol(subvol_name, flags, device_name, data);
1075                 kfree(subvol_name);
1076                 return root;
1077         }
1078
1079         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1080         if (error)
1081                 return ERR_PTR(error);
1082
1083         /*
1084          * Setup a dummy root and fs_info for test/set super.  This is because
1085          * we don't actually fill this stuff out until open_ctree, but we need
1086          * it for searching for existing supers, so this lets us do that and
1087          * then open_ctree will properly initialize everything later.
1088          */
1089         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1090         if (!fs_info)
1091                 return ERR_PTR(-ENOMEM);
1092
1093         fs_info->fs_devices = fs_devices;
1094
1095         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1096         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1097         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1098                 error = -ENOMEM;
1099                 goto error_fs_info;
1100         }
1101
1102         error = btrfs_open_devices(fs_devices, mode, fs_type);
1103         if (error)
1104                 goto error_fs_info;
1105
1106         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1107                 error = -EACCES;
1108                 goto error_close_devices;
1109         }
1110
1111         bdev = fs_devices->latest_bdev;
1112         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1113                  fs_info);
1114         if (IS_ERR(s)) {
1115                 error = PTR_ERR(s);
1116                 goto error_close_devices;
1117         }
1118
1119         if (s->s_root) {
1120                 btrfs_close_devices(fs_devices);
1121                 free_fs_info(fs_info);
1122                 if ((flags ^ s->s_flags) & MS_RDONLY)
1123                         error = -EBUSY;
1124         } else {
1125                 char b[BDEVNAME_SIZE];
1126
1127                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1128                 btrfs_sb(s)->bdev_holder = fs_type;
1129                 error = btrfs_fill_super(s, fs_devices, data,
1130                                          flags & MS_SILENT ? 1 : 0);
1131         }
1132
1133         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1134         if (IS_ERR(root))
1135                 deactivate_locked_super(s);
1136
1137         return root;
1138
1139 error_close_devices:
1140         btrfs_close_devices(fs_devices);
1141 error_fs_info:
1142         free_fs_info(fs_info);
1143         return ERR_PTR(error);
1144 }
1145
1146 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1147 {
1148         spin_lock_irq(&workers->lock);
1149         workers->max_workers = new_limit;
1150         spin_unlock_irq(&workers->lock);
1151 }
1152
1153 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1154                                      int new_pool_size, int old_pool_size)
1155 {
1156         if (new_pool_size == old_pool_size)
1157                 return;
1158
1159         fs_info->thread_pool_size = new_pool_size;
1160
1161         printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1162                old_pool_size, new_pool_size);
1163
1164         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1165         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1166         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1167         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1168         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1169         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1170         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1171         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1172         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1173         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1174         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1175         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1176         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1177         btrfs_set_max_workers(&fs_info->scrub_workers, new_pool_size);
1178 }
1179
1180 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1181 {
1182         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1183         struct btrfs_root *root = fs_info->tree_root;
1184         unsigned old_flags = sb->s_flags;
1185         unsigned long old_opts = fs_info->mount_opt;
1186         unsigned long old_compress_type = fs_info->compress_type;
1187         u64 old_max_inline = fs_info->max_inline;
1188         u64 old_alloc_start = fs_info->alloc_start;
1189         int old_thread_pool_size = fs_info->thread_pool_size;
1190         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1191         int ret;
1192
1193         ret = btrfs_parse_options(root, data);
1194         if (ret) {
1195                 ret = -EINVAL;
1196                 goto restore;
1197         }
1198
1199         btrfs_resize_thread_pool(fs_info,
1200                 fs_info->thread_pool_size, old_thread_pool_size);
1201
1202         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1203                 return 0;
1204
1205         if (*flags & MS_RDONLY) {
1206                 sb->s_flags |= MS_RDONLY;
1207
1208                 ret = btrfs_commit_super(root);
1209                 if (ret)
1210                         goto restore;
1211         } else {
1212                 if (fs_info->fs_devices->rw_devices == 0) {
1213                         ret = -EACCES;
1214                         goto restore;
1215                 }
1216
1217                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1218                         ret = -EINVAL;
1219                         goto restore;
1220                 }
1221
1222                 ret = btrfs_cleanup_fs_roots(fs_info);
1223                 if (ret)
1224                         goto restore;
1225
1226                 /* recover relocation */
1227                 ret = btrfs_recover_relocation(root);
1228                 if (ret)
1229                         goto restore;
1230
1231                 ret = btrfs_resume_balance_async(fs_info);
1232                 if (ret)
1233                         goto restore;
1234
1235                 sb->s_flags &= ~MS_RDONLY;
1236         }
1237
1238         return 0;
1239
1240 restore:
1241         /* We've hit an error - don't reset MS_RDONLY */
1242         if (sb->s_flags & MS_RDONLY)
1243                 old_flags |= MS_RDONLY;
1244         sb->s_flags = old_flags;
1245         fs_info->mount_opt = old_opts;
1246         fs_info->compress_type = old_compress_type;
1247         fs_info->max_inline = old_max_inline;
1248         fs_info->alloc_start = old_alloc_start;
1249         btrfs_resize_thread_pool(fs_info,
1250                 old_thread_pool_size, fs_info->thread_pool_size);
1251         fs_info->metadata_ratio = old_metadata_ratio;
1252         return ret;
1253 }
1254
1255 /* Used to sort the devices by max_avail(descending sort) */
1256 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1257                                        const void *dev_info2)
1258 {
1259         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1260             ((struct btrfs_device_info *)dev_info2)->max_avail)
1261                 return -1;
1262         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1263                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1264                 return 1;
1265         else
1266         return 0;
1267 }
1268
1269 /*
1270  * sort the devices by max_avail, in which max free extent size of each device
1271  * is stored.(Descending Sort)
1272  */
1273 static inline void btrfs_descending_sort_devices(
1274                                         struct btrfs_device_info *devices,
1275                                         size_t nr_devices)
1276 {
1277         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1278              btrfs_cmp_device_free_bytes, NULL);
1279 }
1280
1281 /*
1282  * The helper to calc the free space on the devices that can be used to store
1283  * file data.
1284  */
1285 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1286 {
1287         struct btrfs_fs_info *fs_info = root->fs_info;
1288         struct btrfs_device_info *devices_info;
1289         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1290         struct btrfs_device *device;
1291         u64 skip_space;
1292         u64 type;
1293         u64 avail_space;
1294         u64 used_space;
1295         u64 min_stripe_size;
1296         int min_stripes = 1, num_stripes = 1;
1297         int i = 0, nr_devices;
1298         int ret;
1299
1300         nr_devices = fs_info->fs_devices->open_devices;
1301         BUG_ON(!nr_devices);
1302
1303         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1304                                GFP_NOFS);
1305         if (!devices_info)
1306                 return -ENOMEM;
1307
1308         /* calc min stripe number for data space alloction */
1309         type = btrfs_get_alloc_profile(root, 1);
1310         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1311                 min_stripes = 2;
1312                 num_stripes = nr_devices;
1313         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1314                 min_stripes = 2;
1315                 num_stripes = 2;
1316         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1317                 min_stripes = 4;
1318                 num_stripes = 4;
1319         }
1320
1321         if (type & BTRFS_BLOCK_GROUP_DUP)
1322                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1323         else
1324                 min_stripe_size = BTRFS_STRIPE_LEN;
1325
1326         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1327                 if (!device->in_fs_metadata || !device->bdev)
1328                         continue;
1329
1330                 avail_space = device->total_bytes - device->bytes_used;
1331
1332                 /* align with stripe_len */
1333                 do_div(avail_space, BTRFS_STRIPE_LEN);
1334                 avail_space *= BTRFS_STRIPE_LEN;
1335
1336                 /*
1337                  * In order to avoid overwritting the superblock on the drive,
1338                  * btrfs starts at an offset of at least 1MB when doing chunk
1339                  * allocation.
1340                  */
1341                 skip_space = 1024 * 1024;
1342
1343                 /* user can set the offset in fs_info->alloc_start. */
1344                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1345                     device->total_bytes)
1346                         skip_space = max(fs_info->alloc_start, skip_space);
1347
1348                 /*
1349                  * btrfs can not use the free space in [0, skip_space - 1],
1350                  * we must subtract it from the total. In order to implement
1351                  * it, we account the used space in this range first.
1352                  */
1353                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1354                                                      &used_space);
1355                 if (ret) {
1356                         kfree(devices_info);
1357                         return ret;
1358                 }
1359
1360                 /* calc the free space in [0, skip_space - 1] */
1361                 skip_space -= used_space;
1362
1363                 /*
1364                  * we can use the free space in [0, skip_space - 1], subtract
1365                  * it from the total.
1366                  */
1367                 if (avail_space && avail_space >= skip_space)
1368                         avail_space -= skip_space;
1369                 else
1370                         avail_space = 0;
1371
1372                 if (avail_space < min_stripe_size)
1373                         continue;
1374
1375                 devices_info[i].dev = device;
1376                 devices_info[i].max_avail = avail_space;
1377
1378                 i++;
1379         }
1380
1381         nr_devices = i;
1382
1383         btrfs_descending_sort_devices(devices_info, nr_devices);
1384
1385         i = nr_devices - 1;
1386         avail_space = 0;
1387         while (nr_devices >= min_stripes) {
1388                 if (num_stripes > nr_devices)
1389                         num_stripes = nr_devices;
1390
1391                 if (devices_info[i].max_avail >= min_stripe_size) {
1392                         int j;
1393                         u64 alloc_size;
1394
1395                         avail_space += devices_info[i].max_avail * num_stripes;
1396                         alloc_size = devices_info[i].max_avail;
1397                         for (j = i + 1 - num_stripes; j <= i; j++)
1398                                 devices_info[j].max_avail -= alloc_size;
1399                 }
1400                 i--;
1401                 nr_devices--;
1402         }
1403
1404         kfree(devices_info);
1405         *free_bytes = avail_space;
1406         return 0;
1407 }
1408
1409 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1410 {
1411         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1412         struct btrfs_super_block *disk_super = fs_info->super_copy;
1413         struct list_head *head = &fs_info->space_info;
1414         struct btrfs_space_info *found;
1415         u64 total_used = 0;
1416         u64 total_free_data = 0;
1417         int bits = dentry->d_sb->s_blocksize_bits;
1418         __be32 *fsid = (__be32 *)fs_info->fsid;
1419         int ret;
1420
1421         /* holding chunk_muext to avoid allocating new chunks */
1422         mutex_lock(&fs_info->chunk_mutex);
1423         rcu_read_lock();
1424         list_for_each_entry_rcu(found, head, list) {
1425                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1426                         total_free_data += found->disk_total - found->disk_used;
1427                         total_free_data -=
1428                                 btrfs_account_ro_block_groups_free_space(found);
1429                 }
1430
1431                 total_used += found->disk_used;
1432         }
1433         rcu_read_unlock();
1434
1435         buf->f_namelen = BTRFS_NAME_LEN;
1436         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1437         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1438         buf->f_bsize = dentry->d_sb->s_blocksize;
1439         buf->f_type = BTRFS_SUPER_MAGIC;
1440         buf->f_bavail = total_free_data;
1441         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1442         if (ret) {
1443                 mutex_unlock(&fs_info->chunk_mutex);
1444                 return ret;
1445         }
1446         buf->f_bavail += total_free_data;
1447         buf->f_bavail = buf->f_bavail >> bits;
1448         mutex_unlock(&fs_info->chunk_mutex);
1449
1450         /* We treat it as constant endianness (it doesn't matter _which_)
1451            because we want the fsid to come out the same whether mounted
1452            on a big-endian or little-endian host */
1453         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1454         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1455         /* Mask in the root object ID too, to disambiguate subvols */
1456         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1457         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1458
1459         return 0;
1460 }
1461
1462 static void btrfs_kill_super(struct super_block *sb)
1463 {
1464         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1465         kill_anon_super(sb);
1466         free_fs_info(fs_info);
1467 }
1468
1469 static struct file_system_type btrfs_fs_type = {
1470         .owner          = THIS_MODULE,
1471         .name           = "btrfs",
1472         .mount          = btrfs_mount,
1473         .kill_sb        = btrfs_kill_super,
1474         .fs_flags       = FS_REQUIRES_DEV,
1475 };
1476
1477 /*
1478  * used by btrfsctl to scan devices when no FS is mounted
1479  */
1480 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1481                                 unsigned long arg)
1482 {
1483         struct btrfs_ioctl_vol_args *vol;
1484         struct btrfs_fs_devices *fs_devices;
1485         int ret = -ENOTTY;
1486
1487         if (!capable(CAP_SYS_ADMIN))
1488                 return -EPERM;
1489
1490         vol = memdup_user((void __user *)arg, sizeof(*vol));
1491         if (IS_ERR(vol))
1492                 return PTR_ERR(vol);
1493
1494         switch (cmd) {
1495         case BTRFS_IOC_SCAN_DEV:
1496                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1497                                             &btrfs_fs_type, &fs_devices);
1498                 break;
1499         case BTRFS_IOC_DEVICES_READY:
1500                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1501                                             &btrfs_fs_type, &fs_devices);
1502                 if (ret)
1503                         break;
1504                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1505                 break;
1506         }
1507
1508         kfree(vol);
1509         return ret;
1510 }
1511
1512 static int btrfs_freeze(struct super_block *sb)
1513 {
1514         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1515         mutex_lock(&fs_info->transaction_kthread_mutex);
1516         mutex_lock(&fs_info->cleaner_mutex);
1517         return 0;
1518 }
1519
1520 static int btrfs_unfreeze(struct super_block *sb)
1521 {
1522         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1523         mutex_unlock(&fs_info->cleaner_mutex);
1524         mutex_unlock(&fs_info->transaction_kthread_mutex);
1525         return 0;
1526 }
1527
1528 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1529 {
1530         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1531         struct btrfs_fs_devices *cur_devices;
1532         struct btrfs_device *dev, *first_dev = NULL;
1533         struct list_head *head;
1534         struct rcu_string *name;
1535
1536         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1537         cur_devices = fs_info->fs_devices;
1538         while (cur_devices) {
1539                 head = &cur_devices->devices;
1540                 list_for_each_entry(dev, head, dev_list) {
1541                         if (dev->missing)
1542                                 continue;
1543                         if (!first_dev || dev->devid < first_dev->devid)
1544                                 first_dev = dev;
1545                 }
1546                 cur_devices = cur_devices->seed;
1547         }
1548
1549         if (first_dev) {
1550                 rcu_read_lock();
1551                 name = rcu_dereference(first_dev->name);
1552                 seq_escape(m, name->str, " \t\n\\");
1553                 rcu_read_unlock();
1554         } else {
1555                 WARN_ON(1);
1556         }
1557         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1558         return 0;
1559 }
1560
1561 static const struct super_operations btrfs_super_ops = {
1562         .drop_inode     = btrfs_drop_inode,
1563         .evict_inode    = btrfs_evict_inode,
1564         .put_super      = btrfs_put_super,
1565         .sync_fs        = btrfs_sync_fs,
1566         .show_options   = btrfs_show_options,
1567         .show_devname   = btrfs_show_devname,
1568         .write_inode    = btrfs_write_inode,
1569         .alloc_inode    = btrfs_alloc_inode,
1570         .destroy_inode  = btrfs_destroy_inode,
1571         .statfs         = btrfs_statfs,
1572         .remount_fs     = btrfs_remount,
1573         .freeze_fs      = btrfs_freeze,
1574         .unfreeze_fs    = btrfs_unfreeze,
1575 };
1576
1577 static const struct file_operations btrfs_ctl_fops = {
1578         .unlocked_ioctl  = btrfs_control_ioctl,
1579         .compat_ioctl = btrfs_control_ioctl,
1580         .owner   = THIS_MODULE,
1581         .llseek = noop_llseek,
1582 };
1583
1584 static struct miscdevice btrfs_misc = {
1585         .minor          = BTRFS_MINOR,
1586         .name           = "btrfs-control",
1587         .fops           = &btrfs_ctl_fops
1588 };
1589
1590 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1591 MODULE_ALIAS("devname:btrfs-control");
1592
1593 static int btrfs_interface_init(void)
1594 {
1595         return misc_register(&btrfs_misc);
1596 }
1597
1598 static void btrfs_interface_exit(void)
1599 {
1600         if (misc_deregister(&btrfs_misc) < 0)
1601                 printk(KERN_INFO "misc_deregister failed for control device");
1602 }
1603
1604 static int __init init_btrfs_fs(void)
1605 {
1606         int err;
1607
1608         err = btrfs_init_sysfs();
1609         if (err)
1610                 return err;
1611
1612         btrfs_init_compress();
1613
1614         err = btrfs_init_cachep();
1615         if (err)
1616                 goto free_compress;
1617
1618         err = extent_io_init();
1619         if (err)
1620                 goto free_cachep;
1621
1622         err = extent_map_init();
1623         if (err)
1624                 goto free_extent_io;
1625
1626         err = ordered_data_init();
1627         if (err)
1628                 goto free_extent_map;
1629
1630         err = btrfs_delayed_inode_init();
1631         if (err)
1632                 goto free_ordered_data;
1633
1634         err = btrfs_interface_init();
1635         if (err)
1636                 goto free_delayed_inode;
1637
1638         err = register_filesystem(&btrfs_fs_type);
1639         if (err)
1640                 goto unregister_ioctl;
1641
1642         btrfs_init_lockdep();
1643
1644         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1645         return 0;
1646
1647 unregister_ioctl:
1648         btrfs_interface_exit();
1649 free_delayed_inode:
1650         btrfs_delayed_inode_exit();
1651 free_ordered_data:
1652         ordered_data_exit();
1653 free_extent_map:
1654         extent_map_exit();
1655 free_extent_io:
1656         extent_io_exit();
1657 free_cachep:
1658         btrfs_destroy_cachep();
1659 free_compress:
1660         btrfs_exit_compress();
1661         btrfs_exit_sysfs();
1662         return err;
1663 }
1664
1665 static void __exit exit_btrfs_fs(void)
1666 {
1667         btrfs_destroy_cachep();
1668         btrfs_delayed_inode_exit();
1669         ordered_data_exit();
1670         extent_map_exit();
1671         extent_io_exit();
1672         btrfs_interface_exit();
1673         unregister_filesystem(&btrfs_fs_type);
1674         btrfs_exit_sysfs();
1675         btrfs_cleanup_fs_uuids();
1676         btrfs_exit_compress();
1677 }
1678
1679 module_init(init_btrfs_fs)
1680 module_exit(exit_btrfs_fs)
1681
1682 MODULE_LICENSE("GPL");