fs/btrfs/super.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/time.h>
  26 #include <linux/init.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mount.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/parser.h>
  37 #include <linux/ctype.h>
  38 #include <linux/namei.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/magic.h>
  41 #include <linux/slab.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/ratelimit.h>
  44 #include <linux/btrfs.h>
  45 #include "compat.h"
  46 #include "delayed-inode.h"
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "xattr.h"
  53 #include "volumes.h"
  54 #include "version.h"
  55 #include "export.h"
  56 #include "compression.h"
  57 #include "rcu-string.h"
  58 #include "dev-replace.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/btrfs.h>
  62
  63 static const struct super_operations btrfs_super_ops;
  64 static struct file_system_type btrfs_fs_type;
  65
  66 static const char *btrfs_decode_error(int errno, 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         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 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 (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 117                 sb->s_flags |= MS_RDONLY;
 118                 printk(KERN_INFO "btrfs is forced readonly\n");
 119                 /*
 120                  * Note that a running device replace operation is not
 121                  * canceled here although there is no way to update
 122                  * the progress. It would add the risk of a deadlock,
 123                  * therefore the canceling is ommited. The only penalty
 124                  * is that some I/O remains active until the procedure
 125                  * completes. The next time when the filesystem is
 126                  * mounted writeable again, the device replace
 127                  * operation continues.
 128                  */
 129 //              WARN_ON(1);
 130         }
 131 }
 132
 133 #ifdef CONFIG_PRINTK
 134 /*
 135  * __btrfs_std_error decodes expected errors from the caller and
 136  * invokes the approciate error response.
 137  */
 138 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 139                        unsigned int line, int errno, const char *fmt, ...)
 140 {
 141         struct super_block *sb = fs_info->sb;
 142         char nbuf[16];
 143         const char *errstr;
 144
 145         /*
 146          * Special case: if the error is EROFS, and we're already
 147          * under MS_RDONLY, then it is safe here.
 148          */
 149         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 150                 return;
 151
 152         errstr = btrfs_decode_error(errno, nbuf);
 153         if (fmt) {
 154                 struct va_format vaf;
 155                 va_list args;
 156
 157                 va_start(args, fmt);
 158                 vaf.fmt = fmt;
 159                 vaf.va = &args;
 160
 161                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s (%pV)\n",
 162                         sb->s_id, function, line, errstr, &vaf);
 163                 va_end(args);
 164         } else {
 165                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 166                         sb->s_id, function, line, errstr);
 167         }
 168
 169         /* Don't go through full error handling during mount */
 170         if (sb->s_flags & MS_BORN) {
 171                 save_error_info(fs_info);
 172                 btrfs_handle_error(fs_info);
 173         }
 174 }
 175
 176 static const char * const logtypes[] = {
 177         "emergency",
 178         "alert",
 179         "critical",
 180         "error",
 181         "warning",
 182         "notice",
 183         "info",
 184         "debug",
 185 };
 186
 187 void btrfs_printk(struct btrfs_fs_info *fs_info, const char *fmt, ...)
 188 {
 189         struct super_block *sb = fs_info->sb;
 190         char lvl[4];
 191         struct va_format vaf;
 192         va_list args;
 193         const char *type = logtypes[4];
 194         int kern_level;
 195
 196         va_start(args, fmt);
 197
 198         kern_level = printk_get_level(fmt);
 199         if (kern_level) {
 200                 size_t size = printk_skip_level(fmt) - fmt;
 201                 memcpy(lvl, fmt,  size);
 202                 lvl[size] = '\0';
 203                 fmt += size;
 204                 type = logtypes[kern_level - '0'];
 205         } else
 206                 *lvl = '\0';
 207
 208         vaf.fmt = fmt;
 209         vaf.va = &args;
 210
 211         printk("%sBTRFS %s (device %s): %pV", lvl, type, sb->s_id, &vaf);
 212
 213         va_end(args);
 214 }
 215
 216 #else
 217
 218 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 219                        unsigned int line, int errno, const char *fmt, ...)
 220 {
 221         struct super_block *sb = fs_info->sb;
 222
 223         /*
 224          * Special case: if the error is EROFS, and we're already
 225          * under MS_RDONLY, then it is safe here.
 226          */
 227         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 228                 return;
 229
 230         /* Don't go through full error handling during mount */
 231         if (sb->s_flags & MS_BORN) {
 232                 save_error_info(fs_info);
 233                 btrfs_handle_error(fs_info);
 234         }
 235 }
 236 #endif
 237
 238 /*
 239  * We only mark the transaction aborted and then set the file system read-only.
 240  * This will prevent new transactions from starting or trying to join this
 241  * one.
 242  *
 243  * This means that error recovery at the call site is limited to freeing
 244  * any local memory allocations and passing the error code up without
 245  * further cleanup. The transaction should complete as it normally would
 246  * in the call path but will return -EIO.
 247  *
 248  * We'll complete the cleanup in btrfs_end_transaction and
 249  * btrfs_commit_transaction.
 250  */
 251 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 252                                struct btrfs_root *root, const char *function,
 253                                unsigned int line, int errno)
 254 {
 255         WARN_ONCE(1, KERN_DEBUG "btrfs: Transaction aborted\n");
 256         trans->aborted = errno;
 257         /* Nothing used. The other threads that have joined this
 258          * transaction may be able to continue. */
 259         if (!trans->blocks_used) {
 260                 char nbuf[16];
 261                 const char *errstr;
 262
 263                 errstr = btrfs_decode_error(errno, nbuf);
 264                 btrfs_printk(root->fs_info,
 265                              "%s:%d: Aborting unused transaction(%s).\n",
 266                              function, line, errstr);
 267                 return;
 268         }
 269         ACCESS_ONCE(trans->transaction->aborted) = errno;
 270         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 271 }
 272 /*
 273  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 274  * issues an alert, and either panics or BUGs, depending on mount options.
 275  */
 276 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 277                    unsigned int line, int errno, const char *fmt, ...)
 278 {
 279         char nbuf[16];
 280         char *s_id = "<unknown>";
 281         const char *errstr;
 282         struct va_format vaf = { .fmt = fmt };
 283         va_list args;
 284
 285         if (fs_info)
 286                 s_id = fs_info->sb->s_id;
 287
 288         va_start(args, fmt);
 289         vaf.va = &args;
 290
 291         errstr = btrfs_decode_error(errno, nbuf);
 292         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 293                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
 294                         s_id, function, line, &vaf, errstr);
 295
 296         printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (%s)\n",
 297                s_id, function, line, &vaf, errstr);
 298         va_end(args);
 299         /* Caller calls BUG() */
 300 }
 301
 302 static void btrfs_put_super(struct super_block *sb)
 303 {
 304         (void)close_ctree(btrfs_sb(sb)->tree_root);
 305         /* FIXME: need to fix VFS to return error? */
 306         /* AV: return it _where_?  ->put_super() can be triggered by any number
 307          * of async events, up to and including delivery of SIGKILL to the
 308          * last process that kept it busy.  Or segfault in the aforementioned
 309          * process...  Whom would you report that to?
 310          */
 311 }
 312
 313 enum {
 314         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 315         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 316         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 317         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 318         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 319         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 320         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 321         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 322         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 323         Opt_check_integrity_print_mask, Opt_fatal_errors,
 324         Opt_err,
 325 };
 326
 327 static match_table_t tokens = {
 328         {Opt_degraded, "degraded"},
 329         {Opt_subvol, "subvol=%s"},
 330         {Opt_subvolid, "subvolid=%d"},
 331         {Opt_device, "device=%s"},
 332         {Opt_nodatasum, "nodatasum"},
 333         {Opt_nodatacow, "nodatacow"},
 334         {Opt_nobarrier, "nobarrier"},
 335         {Opt_max_inline, "max_inline=%s"},
 336         {Opt_alloc_start, "alloc_start=%s"},
 337         {Opt_thread_pool, "thread_pool=%d"},
 338         {Opt_compress, "compress"},
 339         {Opt_compress_type, "compress=%s"},
 340         {Opt_compress_force, "compress-force"},
 341         {Opt_compress_force_type, "compress-force=%s"},
 342         {Opt_ssd, "ssd"},
 343         {Opt_ssd_spread, "ssd_spread"},
 344         {Opt_nossd, "nossd"},
 345         {Opt_noacl, "noacl"},
 346         {Opt_notreelog, "notreelog"},
 347         {Opt_flushoncommit, "flushoncommit"},
 348         {Opt_ratio, "metadata_ratio=%d"},
 349         {Opt_discard, "discard"},
 350         {Opt_space_cache, "space_cache"},
 351         {Opt_clear_cache, "clear_cache"},
 352         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 353         {Opt_enospc_debug, "enospc_debug"},
 354         {Opt_subvolrootid, "subvolrootid=%d"},
 355         {Opt_defrag, "autodefrag"},
 356         {Opt_inode_cache, "inode_cache"},
 357         {Opt_no_space_cache, "nospace_cache"},
 358         {Opt_recovery, "recovery"},
 359         {Opt_skip_balance, "skip_balance"},
 360         {Opt_check_integrity, "check_int"},
 361         {Opt_check_integrity_including_extent_data, "check_int_data"},
 362         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 363         {Opt_fatal_errors, "fatal_errors=%s"},
 364         {Opt_err, NULL},
 365 };
 366
 367 /*
 368  * Regular mount options parser.  Everything that is needed only when
 369  * reading in a new superblock is parsed here.
 370  * XXX JDM: This needs to be cleaned up for remount.
 371  */
 372 int btrfs_parse_options(struct btrfs_root *root, char *options)
 373 {
 374         struct btrfs_fs_info *info = root->fs_info;
 375         substring_t args[MAX_OPT_ARGS];
 376         char *p, *num, *orig = NULL;
 377         u64 cache_gen;
 378         int intarg;
 379         int ret = 0;
 380         char *compress_type;
 381         bool compress_force = false;
 382
 383         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 384         if (cache_gen)
 385                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 386
 387         if (!options)
 388                 goto out;
 389
 390         /*
 391          * strsep changes the string, duplicate it because parse_options
 392          * gets called twice
 393          */
 394         options = kstrdup(options, GFP_NOFS);
 395         if (!options)
 396                 return -ENOMEM;
 397
 398         orig = options;
 399
 400         while ((p = strsep(&options, ",")) != NULL) {
 401                 int token;
 402                 if (!*p)
 403                         continue;
 404
 405                 token = match_token(p, tokens, args);
 406                 switch (token) {
 407                 case Opt_degraded:
 408                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 409                         btrfs_set_opt(info->mount_opt, DEGRADED);
 410                         break;
 411                 case Opt_subvol:
 412                 case Opt_subvolid:
 413                 case Opt_subvolrootid:
 414                 case Opt_device:
 415                         /*
 416                          * These are parsed by btrfs_parse_early_options
 417                          * and can be happily ignored here.
 418                          */
 419                         break;
 420                 case Opt_nodatasum:
 421                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 422                         btrfs_set_opt(info->mount_opt, NODATASUM);
 423                         break;
 424                 case Opt_nodatacow:
 425                         if (!btrfs_test_opt(root, COMPRESS) ||
 426                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
 427                                         printk(KERN_INFO "btrfs: setting nodatacow, compression disabled\n");
 428                         } else {
 429                                 printk(KERN_INFO "btrfs: setting nodatacow\n");
 430                         }
 431                         info->compress_type = BTRFS_COMPRESS_NONE;
 432                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 433                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 434                         btrfs_set_opt(info->mount_opt, NODATACOW);
 435                         btrfs_set_opt(info->mount_opt, NODATASUM);
 436                         break;
 437                 case Opt_compress_force:
 438                 case Opt_compress_force_type:
 439                         compress_force = true;
 440                         /* Fallthrough */
 441                 case Opt_compress:
 442                 case Opt_compress_type:
 443                         if (token == Opt_compress ||
 444                             token == Opt_compress_force ||
 445                             strcmp(args[0].from, "zlib") == 0) {
 446                                 compress_type = "zlib";
 447                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 448                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 449                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 450                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 451                         } else if (strcmp(args[0].from, "lzo") == 0) {
 452                                 compress_type = "lzo";
 453                                 info->compress_type = BTRFS_COMPRESS_LZO;
 454                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 455                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 456                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 457                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 458                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 459                                 compress_type = "no";
 460                                 info->compress_type = BTRFS_COMPRESS_NONE;
 461                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 462                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 463                                 compress_force = false;
 464                         } else {
 465                                 ret = -EINVAL;
 466                                 goto out;
 467                         }
 468
 469                         if (compress_force) {
 470                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 471                                 pr_info("btrfs: force %s compression\n",
 472                                         compress_type);
 473                         } else
 474                                 pr_info("btrfs: use %s compression\n",
 475                                         compress_type);
 476                         break;
 477                 case Opt_ssd:
 478                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 479                         btrfs_set_opt(info->mount_opt, SSD);
 480                         break;
 481                 case Opt_ssd_spread:
 482                         printk(KERN_INFO "btrfs: use spread ssd "
 483                                "allocation scheme\n");
 484                         btrfs_set_opt(info->mount_opt, SSD);
 485                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 486                         break;
 487                 case Opt_nossd:
 488                         printk(KERN_INFO "btrfs: not using ssd allocation "
 489                                "scheme\n");
 490                         btrfs_set_opt(info->mount_opt, NOSSD);
 491                         btrfs_clear_opt(info->mount_opt, SSD);
 492                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 493                         break;
 494                 case Opt_nobarrier:
 495                         printk(KERN_INFO "btrfs: turning off barriers\n");
 496                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 497                         break;
 498                 case Opt_thread_pool:
 499                         intarg = 0;
 500                         match_int(&args[0], &intarg);
 501                         if (intarg)
 502                                 info->thread_pool_size = intarg;
 503                         break;
 504                 case Opt_max_inline:
 505                         num = match_strdup(&args[0]);
 506                         if (num) {
 507                                 info->max_inline = memparse(num, NULL);
 508                                 kfree(num);
 509
 510                                 if (info->max_inline) {
 511                                         info->max_inline = max_t(u64,
 512                                                 info->max_inline,
 513                                                 root->sectorsize);
 514                                 }
 515                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 516                                         (unsigned long long)info->max_inline);
 517                         }
 518                         break;
 519                 case Opt_alloc_start:
 520                         num = match_strdup(&args[0]);
 521                         if (num) {
 522                                 mutex_lock(&info->chunk_mutex);
 523                                 info->alloc_start = memparse(num, NULL);
 524                                 mutex_unlock(&info->chunk_mutex);
 525                                 kfree(num);
 526                                 printk(KERN_INFO
 527                                         "btrfs: allocations start at %llu\n",
 528                                         (unsigned long long)info->alloc_start);
 529                         }
 530                         break;
 531                 case Opt_noacl:
 532                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 533                         break;
 534                 case Opt_notreelog:
 535                         printk(KERN_INFO "btrfs: disabling tree log\n");
 536                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 537                         break;
 538                 case Opt_flushoncommit:
 539                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 540                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 541                         break;
 542                 case Opt_ratio:
 543                         intarg = 0;
 544                         match_int(&args[0], &intarg);
 545                         if (intarg) {
 546                                 info->metadata_ratio = intarg;
 547                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 548                                        info->metadata_ratio);
 549                         }
 550                         break;
 551                 case Opt_discard:
 552                         btrfs_set_opt(info->mount_opt, DISCARD);
 553                         break;
 554                 case Opt_space_cache:
 555                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 556                         break;
 557                 case Opt_no_space_cache:
 558                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 559                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 560                         break;
 561                 case Opt_inode_cache:
 562                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 563                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 564                         break;
 565                 case Opt_clear_cache:
 566                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 567                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 568                         break;
 569                 case Opt_user_subvol_rm_allowed:
 570                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 571                         break;
 572                 case Opt_enospc_debug:
 573                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 574                         break;
 575                 case Opt_defrag:
 576                         printk(KERN_INFO "btrfs: enabling auto defrag\n");
 577                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 578                         break;
 579                 case Opt_recovery:
 580                         printk(KERN_INFO "btrfs: enabling auto recovery\n");
 581                         btrfs_set_opt(info->mount_opt, RECOVERY);
 582                         break;
 583                 case Opt_skip_balance:
 584                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 585                         break;
 586 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 587                 case Opt_check_integrity_including_extent_data:
 588                         printk(KERN_INFO "btrfs: enabling check integrity"
 589                                " including extent data\n");
 590                         btrfs_set_opt(info->mount_opt,
 591                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 592                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 593                         break;
 594                 case Opt_check_integrity:
 595                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 596                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 597                         break;
 598                 case Opt_check_integrity_print_mask:
 599                         intarg = 0;
 600                         match_int(&args[0], &intarg);
 601                         if (intarg) {
 602                                 info->check_integrity_print_mask = intarg;
 603                                 printk(KERN_INFO "btrfs:"
 604                                        " check_integrity_print_mask 0x%x\n",
 605                                        info->check_integrity_print_mask);
 606                         }
 607                         break;
 608 #else
 609                 case Opt_check_integrity_including_extent_data:
 610                 case Opt_check_integrity:
 611                 case Opt_check_integrity_print_mask:
 612                         printk(KERN_ERR "btrfs: support for check_integrity*"
 613                                " not compiled in!\n");
 614                         ret = -EINVAL;
 615                         goto out;
 616 #endif
 617                 case Opt_fatal_errors:
 618                         if (strcmp(args[0].from, "panic") == 0)
 619                                 btrfs_set_opt(info->mount_opt,
 620                                               PANIC_ON_FATAL_ERROR);
 621                         else if (strcmp(args[0].from, "bug") == 0)
 622                                 btrfs_clear_opt(info->mount_opt,
 623                                               PANIC_ON_FATAL_ERROR);
 624                         else {
 625                                 ret = -EINVAL;
 626                                 goto out;
 627                         }
 628                         break;
 629                 case Opt_err:
 630                         printk(KERN_INFO "btrfs: unrecognized mount option "
 631                                "'%s'\n", p);
 632                         ret = -EINVAL;
 633                         goto out;
 634                 default:
 635                         break;
 636                 }
 637         }
 638 out:
 639         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 640                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 641         kfree(orig);
 642         return ret;
 643 }
 644
 645 /*
 646  * Parse mount options that are required early in the mount process.
 647  *
 648  * All other options will be parsed on much later in the mount process and
 649  * only when we need to allocate a new super block.
 650  */
 651 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 652                 void *holder, char **subvol_name, u64 *subvol_objectid,
 653                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
 654 {
 655         substring_t args[MAX_OPT_ARGS];
 656         char *device_name, *opts, *orig, *p;
 657         int error = 0;
 658         int intarg;
 659
 660         if (!options)
 661                 return 0;
 662
 663         /*
 664          * strsep changes the string, duplicate it because parse_options
 665          * gets called twice
 666          */
 667         opts = kstrdup(options, GFP_KERNEL);
 668         if (!opts)
 669                 return -ENOMEM;
 670         orig = opts;
 671
 672         while ((p = strsep(&opts, ",")) != NULL) {
 673                 int token;
 674                 if (!*p)
 675                         continue;
 676
 677                 token = match_token(p, tokens, args);
 678                 switch (token) {
 679                 case Opt_subvol:
 680                         kfree(*subvol_name);
 681                         *subvol_name = match_strdup(&args[0]);
 682                         break;
 683                 case Opt_subvolid:
 684                         intarg = 0;
 685                         error = match_int(&args[0], &intarg);
 686                         if (!error) {
 687                                 /* we want the original fs_tree */
 688                                 if (!intarg)
 689                                         *subvol_objectid =
 690                                                 BTRFS_FS_TREE_OBJECTID;
 691                                 else
 692                                         *subvol_objectid = intarg;
 693                         }
 694                         break;
 695                 case Opt_subvolrootid:
 696                         intarg = 0;
 697                         error = match_int(&args[0], &intarg);
 698                         if (!error) {
 699                                 /* we want the original fs_tree */
 700                                 if (!intarg)
 701                                         *subvol_rootid =
 702                                                 BTRFS_FS_TREE_OBJECTID;
 703                                 else
 704                                         *subvol_rootid = intarg;
 705                         }
 706                         break;
 707                 case Opt_device:
 708                         device_name = match_strdup(&args[0]);
 709                         if (!device_name) {
 710                                 error = -ENOMEM;
 711                                 goto out;
 712                         }
 713                         error = btrfs_scan_one_device(device_name,
 714                                         flags, holder, fs_devices);
 715                         kfree(device_name);
 716                         if (error)
 717                                 goto out;
 718                         break;
 719                 default:
 720                         break;
 721                 }
 722         }
 723
 724 out:
 725         kfree(orig);
 726         return error;
 727 }
 728
 729 static struct dentry *get_default_root(struct super_block *sb,
 730                                        u64 subvol_objectid)
 731 {
 732         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 733         struct btrfs_root *root = fs_info->tree_root;
 734         struct btrfs_root *new_root;
 735         struct btrfs_dir_item *di;
 736         struct btrfs_path *path;
 737         struct btrfs_key location;
 738         struct inode *inode;
 739         u64 dir_id;
 740         int new = 0;
 741
 742         /*
 743          * We have a specific subvol we want to mount, just setup location and
 744          * go look up the root.
 745          */
 746         if (subvol_objectid) {
 747                 location.objectid = subvol_objectid;
 748                 location.type = BTRFS_ROOT_ITEM_KEY;
 749                 location.offset = (u64)-1;
 750                 goto find_root;
 751         }
 752
 753         path = btrfs_alloc_path();
 754         if (!path)
 755                 return ERR_PTR(-ENOMEM);
 756         path->leave_spinning = 1;
 757
 758         /*
 759          * Find the "default" dir item which points to the root item that we
 760          * will mount by default if we haven't been given a specific subvolume
 761          * to mount.
 762          */
 763         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 764         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 765         if (IS_ERR(di)) {
 766                 btrfs_free_path(path);
 767                 return ERR_CAST(di);
 768         }
 769         if (!di) {
 770                 /*
 771                  * Ok the default dir item isn't there.  This is weird since
 772                  * it's always been there, but don't freak out, just try and
 773                  * mount to root most subvolume.
 774                  */
 775                 btrfs_free_path(path);
 776                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 777                 new_root = fs_info->fs_root;
 778                 goto setup_root;
 779         }
 780
 781         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 782         btrfs_free_path(path);
 783
 784 find_root:
 785         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 786         if (IS_ERR(new_root))
 787                 return ERR_CAST(new_root);
 788
 789         if (btrfs_root_refs(&new_root->root_item) == 0)
 790                 return ERR_PTR(-ENOENT);
 791
 792         dir_id = btrfs_root_dirid(&new_root->root_item);
 793 setup_root:
 794         location.objectid = dir_id;
 795         location.type = BTRFS_INODE_ITEM_KEY;
 796         location.offset = 0;
 797
 798         inode = btrfs_iget(sb, &location, new_root, &new);
 799         if (IS_ERR(inode))
 800                 return ERR_CAST(inode);
 801
 802         /*
 803          * If we're just mounting the root most subvol put the inode and return
 804          * a reference to the dentry.  We will have already gotten a reference
 805          * to the inode in btrfs_fill_super so we're good to go.
 806          */
 807         if (!new && sb->s_root->d_inode == inode) {
 808                 iput(inode);
 809                 return dget(sb->s_root);
 810         }
 811
 812         return d_obtain_alias(inode);
 813 }
 814
 815 static int btrfs_fill_super(struct super_block *sb,
 816                             struct btrfs_fs_devices *fs_devices,
 817                             void *data, int silent)
 818 {
 819         struct inode *inode;
 820         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 821         struct btrfs_key key;
 822         int err;
 823
 824         sb->s_maxbytes = MAX_LFS_FILESIZE;
 825         sb->s_magic = BTRFS_SUPER_MAGIC;
 826         sb->s_op = &btrfs_super_ops;
 827         sb->s_d_op = &btrfs_dentry_operations;
 828         sb->s_export_op = &btrfs_export_ops;
 829         sb->s_xattr = btrfs_xattr_handlers;
 830         sb->s_time_gran = 1;
 831 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 832         sb->s_flags |= MS_POSIXACL;
 833 #endif
 834         sb->s_flags |= MS_I_VERSION;
 835         err = open_ctree(sb, fs_devices, (char *)data);
 836         if (err) {
 837                 printk("btrfs: open_ctree failed\n");
 838                 return err;
 839         }
 840
 841         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 842         key.type = BTRFS_INODE_ITEM_KEY;
 843         key.offset = 0;
 844         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 845         if (IS_ERR(inode)) {
 846                 err = PTR_ERR(inode);
 847                 goto fail_close;
 848         }
 849
 850         sb->s_root = d_make_root(inode);
 851         if (!sb->s_root) {
 852                 err = -ENOMEM;
 853                 goto fail_close;
 854         }
 855
 856         save_mount_options(sb, data);
 857         cleancache_init_fs(sb);
 858         sb->s_flags |= MS_ACTIVE;
 859         return 0;
 860
 861 fail_close:
 862         close_ctree(fs_info->tree_root);
 863         return err;
 864 }
 865
 866 int btrfs_sync_fs(struct super_block *sb, int wait)
 867 {
 868         struct btrfs_trans_handle *trans;
 869         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 870         struct btrfs_root *root = fs_info->tree_root;
 871
 872         trace_btrfs_sync_fs(wait);
 873
 874         if (!wait) {
 875                 filemap_flush(fs_info->btree_inode->i_mapping);
 876                 return 0;
 877         }
 878
 879         btrfs_wait_ordered_extents(root, 0);
 880
 881         trans = btrfs_attach_transaction_barrier(root);
 882         if (IS_ERR(trans)) {
 883                 /* no transaction, don't bother */
 884                 if (PTR_ERR(trans) == -ENOENT)
 885                         return 0;
 886                 return PTR_ERR(trans);
 887         }
 888         return btrfs_commit_transaction(trans, root);
 889 }
 890
 891 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 892 {
 893         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 894         struct btrfs_root *root = info->tree_root;
 895         char *compress_type;
 896
 897         if (btrfs_test_opt(root, DEGRADED))
 898                 seq_puts(seq, ",degraded");
 899         if (btrfs_test_opt(root, NODATASUM))
 900                 seq_puts(seq, ",nodatasum");
 901         if (btrfs_test_opt(root, NODATACOW))
 902                 seq_puts(seq, ",nodatacow");
 903         if (btrfs_test_opt(root, NOBARRIER))
 904                 seq_puts(seq, ",nobarrier");
 905         if (info->max_inline != 8192 * 1024)
 906                 seq_printf(seq, ",max_inline=%llu",
 907                            (unsigned long long)info->max_inline);
 908         if (info->alloc_start != 0)
 909                 seq_printf(seq, ",alloc_start=%llu",
 910                            (unsigned long long)info->alloc_start);
 911         if (info->thread_pool_size !=  min_t(unsigned long,
 912                                              num_online_cpus() + 2, 8))
 913                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 914         if (btrfs_test_opt(root, COMPRESS)) {
 915                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 916                         compress_type = "zlib";
 917                 else
 918                         compress_type = "lzo";
 919                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 920                         seq_printf(seq, ",compress-force=%s", compress_type);
 921                 else
 922                         seq_printf(seq, ",compress=%s", compress_type);
 923         }
 924         if (btrfs_test_opt(root, NOSSD))
 925                 seq_puts(seq, ",nossd");
 926         if (btrfs_test_opt(root, SSD_SPREAD))
 927                 seq_puts(seq, ",ssd_spread");
 928         else if (btrfs_test_opt(root, SSD))
 929                 seq_puts(seq, ",ssd");
 930         if (btrfs_test_opt(root, NOTREELOG))
 931                 seq_puts(seq, ",notreelog");
 932         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 933                 seq_puts(seq, ",flushoncommit");
 934         if (btrfs_test_opt(root, DISCARD))
 935                 seq_puts(seq, ",discard");
 936         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 937                 seq_puts(seq, ",noacl");
 938         if (btrfs_test_opt(root, SPACE_CACHE))
 939                 seq_puts(seq, ",space_cache");
 940         else
 941                 seq_puts(seq, ",nospace_cache");
 942         if (btrfs_test_opt(root, CLEAR_CACHE))
 943                 seq_puts(seq, ",clear_cache");
 944         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 945                 seq_puts(seq, ",user_subvol_rm_allowed");
 946         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 947                 seq_puts(seq, ",enospc_debug");
 948         if (btrfs_test_opt(root, AUTO_DEFRAG))
 949                 seq_puts(seq, ",autodefrag");
 950         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 951                 seq_puts(seq, ",inode_cache");
 952         if (btrfs_test_opt(root, SKIP_BALANCE))
 953                 seq_puts(seq, ",skip_balance");
 954         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 955                 seq_puts(seq, ",fatal_errors=panic");
 956         return 0;
 957 }
 958
 959 static int btrfs_test_super(struct super_block *s, void *data)
 960 {
 961         struct btrfs_fs_info *p = data;
 962         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 963
 964         return fs_info->fs_devices == p->fs_devices;
 965 }
 966
 967 static int btrfs_set_super(struct super_block *s, void *data)
 968 {
 969         int err = set_anon_super(s, data);
 970         if (!err)
 971                 s->s_fs_info = data;
 972         return err;
 973 }
 974
 975 /*
 976  * subvolumes are identified by ino 256
 977  */
 978 static inline int is_subvolume_inode(struct inode *inode)
 979 {
 980         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 981                 return 1;
 982         return 0;
 983 }
 984
 985 /*
 986  * This will strip out the subvol=%s argument for an argument string and add
 987  * subvolid=0 to make sure we get the actual tree root for path walking to the
 988  * subvol we want.
 989  */
 990 static char *setup_root_args(char *args)
 991 {
 992         unsigned len = strlen(args) + 2 + 1;
 993         char *src, *dst, *buf;
 994
 995         /*
 996          * We need the same args as before, but with this substitution:
 997          * s!subvol=[^,]+!subvolid=0!
 998          *
 999          * Since the replacement string is up to 2 bytes longer than the
1000          * original, allocate strlen(args) + 2 + 1 bytes.
1001          */
1002
1003         src = strstr(args, "subvol=");
1004         /* This shouldn't happen, but just in case.. */
1005         if (!src)
1006                 return NULL;
1007
1008         buf = dst = kmalloc(len, GFP_NOFS);
1009         if (!buf)
1010                 return NULL;
1011
1012         /*
1013          * If the subvol= arg is not at the start of the string,
1014          * copy whatever precedes it into buf.
1015          */
1016         if (src != args) {
1017                 *src++ = '\0';
1018                 strcpy(buf, args);
1019                 dst += strlen(args);
1020         }
1021
1022         strcpy(dst, "subvolid=0");
1023         dst += strlen("subvolid=0");
1024
1025         /*
1026          * If there is a "," after the original subvol=... string,
1027          * copy that suffix into our buffer.  Otherwise, we're done.
1028          */
1029         src = strchr(src, ',');
1030         if (src)
1031                 strcpy(dst, src);
1032
1033         return buf;
1034 }
1035
1036 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1037                                    const char *device_name, char *data)
1038 {
1039         struct dentry *root;
1040         struct vfsmount *mnt;
1041         char *newargs;
1042
1043         newargs = setup_root_args(data);
1044         if (!newargs)
1045                 return ERR_PTR(-ENOMEM);
1046         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1047                              newargs);
1048         kfree(newargs);
1049         if (IS_ERR(mnt))
1050                 return ERR_CAST(mnt);
1051
1052         root = mount_subtree(mnt, subvol_name);
1053
1054         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1055                 struct super_block *s = root->d_sb;
1056                 dput(root);
1057                 root = ERR_PTR(-EINVAL);
1058                 deactivate_locked_super(s);
1059                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1060                                 subvol_name);
1061         }
1062
1063         return root;
1064 }
1065
1066 /*
1067  * Find a superblock for the given device / mount point.
1068  *
1069  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1070  *        for multiple device setup.  Make sure to keep it in sync.
1071  */
1072 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1073                 const char *device_name, void *data)
1074 {
1075         struct block_device *bdev = NULL;
1076         struct super_block *s;
1077         struct dentry *root;
1078         struct btrfs_fs_devices *fs_devices = NULL;
1079         struct btrfs_fs_info *fs_info = NULL;
1080         fmode_t mode = FMODE_READ;
1081         char *subvol_name = NULL;
1082         u64 subvol_objectid = 0;
1083         u64 subvol_rootid = 0;
1084         int error = 0;
1085
1086         if (!(flags & MS_RDONLY))
1087                 mode |= FMODE_WRITE;
1088
1089         error = btrfs_parse_early_options(data, mode, fs_type,
1090                                           &subvol_name, &subvol_objectid,
1091                                           &subvol_rootid, &fs_devices);
1092         if (error) {
1093                 kfree(subvol_name);
1094                 return ERR_PTR(error);
1095         }
1096
1097         if (subvol_name) {
1098                 root = mount_subvol(subvol_name, flags, device_name, data);
1099                 kfree(subvol_name);
1100                 return root;
1101         }
1102
1103         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1104         if (error)
1105                 return ERR_PTR(error);
1106
1107         /*
1108          * Setup a dummy root and fs_info for test/set super.  This is because
1109          * we don't actually fill this stuff out until open_ctree, but we need
1110          * it for searching for existing supers, so this lets us do that and
1111          * then open_ctree will properly initialize everything later.
1112          */
1113         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1114         if (!fs_info)
1115                 return ERR_PTR(-ENOMEM);
1116
1117         fs_info->fs_devices = fs_devices;
1118
1119         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1120         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1121         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1122                 error = -ENOMEM;
1123                 goto error_fs_info;
1124         }
1125
1126         error = btrfs_open_devices(fs_devices, mode, fs_type);
1127         if (error)
1128                 goto error_fs_info;
1129
1130         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1131                 error = -EACCES;
1132                 goto error_close_devices;
1133         }
1134
1135         bdev = fs_devices->latest_bdev;
1136         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1137                  fs_info);
1138         if (IS_ERR(s)) {
1139                 error = PTR_ERR(s);
1140                 goto error_close_devices;
1141         }
1142
1143         if (s->s_root) {
1144                 btrfs_close_devices(fs_devices);
1145                 free_fs_info(fs_info);
1146                 if ((flags ^ s->s_flags) & MS_RDONLY)
1147                         error = -EBUSY;
1148         } else {
1149                 char b[BDEVNAME_SIZE];
1150
1151                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1152                 btrfs_sb(s)->bdev_holder = fs_type;
1153                 error = btrfs_fill_super(s, fs_devices, data,
1154                                          flags & MS_SILENT ? 1 : 0);
1155         }
1156
1157         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1158         if (IS_ERR(root))
1159                 deactivate_locked_super(s);
1160
1161         return root;
1162
1163 error_close_devices:
1164         btrfs_close_devices(fs_devices);
1165 error_fs_info:
1166         free_fs_info(fs_info);
1167         return ERR_PTR(error);
1168 }
1169
1170 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1171 {
1172         spin_lock_irq(&workers->lock);
1173         workers->max_workers = new_limit;
1174         spin_unlock_irq(&workers->lock);
1175 }
1176
1177 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1178                                      int new_pool_size, int old_pool_size)
1179 {
1180         if (new_pool_size == old_pool_size)
1181                 return;
1182
1183         fs_info->thread_pool_size = new_pool_size;
1184
1185         printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1186                old_pool_size, new_pool_size);
1187
1188         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1189         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1190         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1191         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1192         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1193         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1194         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1195         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1196         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1197         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1198         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1199         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1200         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1201         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1202                               new_pool_size);
1203 }
1204
1205 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info,
1206                                          unsigned long old_opts, int flags)
1207 {
1208         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1209
1210         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1211             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1212              (flags & MS_RDONLY))) {
1213                 /* wait for any defraggers to finish */
1214                 wait_event(fs_info->transaction_wait,
1215                            (atomic_read(&fs_info->defrag_running) == 0));
1216                 if (flags & MS_RDONLY)
1217                         sync_filesystem(fs_info->sb);
1218         }
1219 }
1220
1221 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1222                                          unsigned long old_opts)
1223 {
1224         /*
1225          * We need cleanup all defragable inodes if the autodefragment is
1226          * close or the fs is R/O.
1227          */
1228         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1229             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1230              (fs_info->sb->s_flags & MS_RDONLY))) {
1231                 btrfs_cleanup_defrag_inodes(fs_info);
1232         }
1233
1234         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1235 }
1236
1237 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1238 {
1239         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1240         struct btrfs_root *root = fs_info->tree_root;
1241         unsigned old_flags = sb->s_flags;
1242         unsigned long old_opts = fs_info->mount_opt;
1243         unsigned long old_compress_type = fs_info->compress_type;
1244         u64 old_max_inline = fs_info->max_inline;
1245         u64 old_alloc_start = fs_info->alloc_start;
1246         int old_thread_pool_size = fs_info->thread_pool_size;
1247         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1248         int ret;
1249
1250         btrfs_remount_prepare(fs_info, old_opts, *flags);
1251
1252         ret = btrfs_parse_options(root, data);
1253         if (ret) {
1254                 ret = -EINVAL;
1255                 goto restore;
1256         }
1257
1258         btrfs_resize_thread_pool(fs_info,
1259                 fs_info->thread_pool_size, old_thread_pool_size);
1260
1261         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1262                 goto out;
1263
1264         if (*flags & MS_RDONLY) {
1265                 /*
1266                  * this also happens on 'umount -rf' or on shutdown, when
1267                  * the filesystem is busy.
1268                  */
1269                 sb->s_flags |= MS_RDONLY;
1270
1271                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1272                 btrfs_scrub_cancel(fs_info);
1273
1274                 ret = btrfs_commit_super(root);
1275                 if (ret)
1276                         goto restore;
1277         } else {
1278                 if (fs_info->fs_devices->rw_devices == 0) {
1279                         ret = -EACCES;
1280                         goto restore;
1281                 }
1282
1283                 if (fs_info->fs_devices->missing_devices >
1284                      fs_info->num_tolerated_disk_barrier_failures &&
1285                     !(*flags & MS_RDONLY)) {
1286                         printk(KERN_WARNING
1287                                "Btrfs: too many missing devices, writeable remount is not allowed\n");
1288                         ret = -EACCES;
1289                         goto restore;
1290                 }
1291
1292                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1293                         ret = -EINVAL;
1294                         goto restore;
1295                 }
1296
1297                 ret = btrfs_cleanup_fs_roots(fs_info);
1298                 if (ret)
1299                         goto restore;
1300
1301                 /* recover relocation */
1302                 ret = btrfs_recover_relocation(root);
1303                 if (ret)
1304                         goto restore;
1305
1306                 ret = btrfs_resume_balance_async(fs_info);
1307                 if (ret)
1308                         goto restore;
1309
1310                 ret = btrfs_resume_dev_replace_async(fs_info);
1311                 if (ret) {
1312                         pr_warn("btrfs: failed to resume dev_replace\n");
1313                         goto restore;
1314                 }
1315                 sb->s_flags &= ~MS_RDONLY;
1316         }
1317 out:
1318         btrfs_remount_cleanup(fs_info, old_opts);
1319         return 0;
1320
1321 restore:
1322         /* We've hit an error - don't reset MS_RDONLY */
1323         if (sb->s_flags & MS_RDONLY)
1324                 old_flags |= MS_RDONLY;
1325         sb->s_flags = old_flags;
1326         fs_info->mount_opt = old_opts;
1327         fs_info->compress_type = old_compress_type;
1328         fs_info->max_inline = old_max_inline;
1329         mutex_lock(&fs_info->chunk_mutex);
1330         fs_info->alloc_start = old_alloc_start;
1331         mutex_unlock(&fs_info->chunk_mutex);
1332         btrfs_resize_thread_pool(fs_info,
1333                 old_thread_pool_size, fs_info->thread_pool_size);
1334         fs_info->metadata_ratio = old_metadata_ratio;
1335         btrfs_remount_cleanup(fs_info, old_opts);
1336         return ret;
1337 }
1338
1339 /* Used to sort the devices by max_avail(descending sort) */
1340 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1341                                        const void *dev_info2)
1342 {
1343         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1344             ((struct btrfs_device_info *)dev_info2)->max_avail)
1345                 return -1;
1346         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1347                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1348                 return 1;
1349         else
1350         return 0;
1351 }
1352
1353 /*
1354  * sort the devices by max_avail, in which max free extent size of each device
1355  * is stored.(Descending Sort)
1356  */
1357 static inline void btrfs_descending_sort_devices(
1358                                         struct btrfs_device_info *devices,
1359                                         size_t nr_devices)
1360 {
1361         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1362              btrfs_cmp_device_free_bytes, NULL);
1363 }
1364
1365 /*
1366  * The helper to calc the free space on the devices that can be used to store
1367  * file data.
1368  */
1369 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1370 {
1371         struct btrfs_fs_info *fs_info = root->fs_info;
1372         struct btrfs_device_info *devices_info;
1373         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1374         struct btrfs_device *device;
1375         u64 skip_space;
1376         u64 type;
1377         u64 avail_space;
1378         u64 used_space;
1379         u64 min_stripe_size;
1380         int min_stripes = 1, num_stripes = 1;
1381         int i = 0, nr_devices;
1382         int ret;
1383
1384         nr_devices = fs_info->fs_devices->open_devices;
1385         BUG_ON(!nr_devices);
1386
1387         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1388                                GFP_NOFS);
1389         if (!devices_info)
1390                 return -ENOMEM;
1391
1392         /* calc min stripe number for data space alloction */
1393         type = btrfs_get_alloc_profile(root, 1);
1394         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1395                 min_stripes = 2;
1396                 num_stripes = nr_devices;
1397         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1398                 min_stripes = 2;
1399                 num_stripes = 2;
1400         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1401                 min_stripes = 4;
1402                 num_stripes = 4;
1403         }
1404
1405         if (type & BTRFS_BLOCK_GROUP_DUP)
1406                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1407         else
1408                 min_stripe_size = BTRFS_STRIPE_LEN;
1409
1410         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1411                 if (!device->in_fs_metadata || !device->bdev ||
1412                     device->is_tgtdev_for_dev_replace)
1413                         continue;
1414
1415                 avail_space = device->total_bytes - device->bytes_used;
1416
1417                 /* align with stripe_len */
1418                 do_div(avail_space, BTRFS_STRIPE_LEN);
1419                 avail_space *= BTRFS_STRIPE_LEN;
1420
1421                 /*
1422                  * In order to avoid overwritting the superblock on the drive,
1423                  * btrfs starts at an offset of at least 1MB when doing chunk
1424                  * allocation.
1425                  */
1426                 skip_space = 1024 * 1024;
1427
1428                 /* user can set the offset in fs_info->alloc_start. */
1429                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1430                     device->total_bytes)
1431                         skip_space = max(fs_info->alloc_start, skip_space);
1432
1433                 /*
1434                  * btrfs can not use the free space in [0, skip_space - 1],
1435                  * we must subtract it from the total. In order to implement
1436                  * it, we account the used space in this range first.
1437                  */
1438                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1439                                                      &used_space);
1440                 if (ret) {
1441                         kfree(devices_info);
1442                         return ret;
1443                 }
1444
1445                 /* calc the free space in [0, skip_space - 1] */
1446                 skip_space -= used_space;
1447
1448                 /*
1449                  * we can use the free space in [0, skip_space - 1], subtract
1450                  * it from the total.
1451                  */
1452                 if (avail_space && avail_space >= skip_space)
1453                         avail_space -= skip_space;
1454                 else
1455                         avail_space = 0;
1456
1457                 if (avail_space < min_stripe_size)
1458                         continue;
1459
1460                 devices_info[i].dev = device;
1461                 devices_info[i].max_avail = avail_space;
1462
1463                 i++;
1464         }
1465
1466         nr_devices = i;
1467
1468         btrfs_descending_sort_devices(devices_info, nr_devices);
1469
1470         i = nr_devices - 1;
1471         avail_space = 0;
1472         while (nr_devices >= min_stripes) {
1473                 if (num_stripes > nr_devices)
1474                         num_stripes = nr_devices;
1475
1476                 if (devices_info[i].max_avail >= min_stripe_size) {
1477                         int j;
1478                         u64 alloc_size;
1479
1480                         avail_space += devices_info[i].max_avail * num_stripes;
1481                         alloc_size = devices_info[i].max_avail;
1482                         for (j = i + 1 - num_stripes; j <= i; j++)
1483                                 devices_info[j].max_avail -= alloc_size;
1484                 }
1485                 i--;
1486                 nr_devices--;
1487         }
1488
1489         kfree(devices_info);
1490         *free_bytes = avail_space;
1491         return 0;
1492 }
1493
1494 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1495 {
1496         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1497         struct btrfs_super_block *disk_super = fs_info->super_copy;
1498         struct list_head *head = &fs_info->space_info;
1499         struct btrfs_space_info *found;
1500         u64 total_used = 0;
1501         u64 total_free_data = 0;
1502         int bits = dentry->d_sb->s_blocksize_bits;
1503         __be32 *fsid = (__be32 *)fs_info->fsid;
1504         int ret;
1505
1506         /* holding chunk_muext to avoid allocating new chunks */
1507         mutex_lock(&fs_info->chunk_mutex);
1508         rcu_read_lock();
1509         list_for_each_entry_rcu(found, head, list) {
1510                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1511                         total_free_data += found->disk_total - found->disk_used;
1512                         total_free_data -=
1513                                 btrfs_account_ro_block_groups_free_space(found);
1514                 }
1515
1516                 total_used += found->disk_used;
1517         }
1518         rcu_read_unlock();
1519
1520         buf->f_namelen = BTRFS_NAME_LEN;
1521         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1522         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1523         buf->f_bsize = dentry->d_sb->s_blocksize;
1524         buf->f_type = BTRFS_SUPER_MAGIC;
1525         buf->f_bavail = total_free_data;
1526         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1527         if (ret) {
1528                 mutex_unlock(&fs_info->chunk_mutex);
1529                 return ret;
1530         }
1531         buf->f_bavail += total_free_data;
1532         buf->f_bavail = buf->f_bavail >> bits;
1533         mutex_unlock(&fs_info->chunk_mutex);
1534
1535         /* We treat it as constant endianness (it doesn't matter _which_)
1536            because we want the fsid to come out the same whether mounted
1537            on a big-endian or little-endian host */
1538         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1539         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1540         /* Mask in the root object ID too, to disambiguate subvols */
1541         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1542         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1543
1544         return 0;
1545 }
1546
1547 static void btrfs_kill_super(struct super_block *sb)
1548 {
1549         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1550         kill_anon_super(sb);
1551         free_fs_info(fs_info);
1552 }
1553
1554 static struct file_system_type btrfs_fs_type = {
1555         .owner          = THIS_MODULE,
1556         .name           = "btrfs",
1557         .mount          = btrfs_mount,
1558         .kill_sb        = btrfs_kill_super,
1559         .fs_flags       = FS_REQUIRES_DEV,
1560 };
1561 MODULE_ALIAS_FS("btrfs");
1562
1563 /*
1564  * used by btrfsctl to scan devices when no FS is mounted
1565  */
1566 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1567                                 unsigned long arg)
1568 {
1569         struct btrfs_ioctl_vol_args *vol;
1570         struct btrfs_fs_devices *fs_devices;
1571         int ret = -ENOTTY;
1572
1573         if (!capable(CAP_SYS_ADMIN))
1574                 return -EPERM;
1575
1576         vol = memdup_user((void __user *)arg, sizeof(*vol));
1577         if (IS_ERR(vol))
1578                 return PTR_ERR(vol);
1579
1580         switch (cmd) {
1581         case BTRFS_IOC_SCAN_DEV:
1582                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1583                                             &btrfs_fs_type, &fs_devices);
1584                 break;
1585         case BTRFS_IOC_DEVICES_READY:
1586                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1587                                             &btrfs_fs_type, &fs_devices);
1588                 if (ret)
1589                         break;
1590                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1591                 break;
1592         }
1593
1594         kfree(vol);
1595         return ret;
1596 }
1597
1598 static int btrfs_freeze(struct super_block *sb)
1599 {
1600         struct btrfs_trans_handle *trans;
1601         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1602
1603         trans = btrfs_attach_transaction_barrier(root);
1604         if (IS_ERR(trans)) {
1605                 /* no transaction, don't bother */
1606                 if (PTR_ERR(trans) == -ENOENT)
1607                         return 0;
1608                 return PTR_ERR(trans);
1609         }
1610         return btrfs_commit_transaction(trans, root);
1611 }
1612
1613 static int btrfs_unfreeze(struct super_block *sb)
1614 {
1615         return 0;
1616 }
1617
1618 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1619 {
1620         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1621         struct btrfs_fs_devices *cur_devices;
1622         struct btrfs_device *dev, *first_dev = NULL;
1623         struct list_head *head;
1624         struct rcu_string *name;
1625
1626         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1627         cur_devices = fs_info->fs_devices;
1628         while (cur_devices) {
1629                 head = &cur_devices->devices;
1630                 list_for_each_entry(dev, head, dev_list) {
1631                         if (dev->missing)
1632                                 continue;
1633                         if (!first_dev || dev->devid < first_dev->devid)
1634                                 first_dev = dev;
1635                 }
1636                 cur_devices = cur_devices->seed;
1637         }
1638
1639         if (first_dev) {
1640                 rcu_read_lock();
1641                 name = rcu_dereference(first_dev->name);
1642                 seq_escape(m, name->str, " \t\n\\");
1643                 rcu_read_unlock();
1644         } else {
1645                 WARN_ON(1);
1646         }
1647         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1648         return 0;
1649 }
1650
1651 static const struct super_operations btrfs_super_ops = {
1652         .drop_inode     = btrfs_drop_inode,
1653         .evict_inode    = btrfs_evict_inode,
1654         .put_super      = btrfs_put_super,
1655         .sync_fs        = btrfs_sync_fs,
1656         .show_options   = btrfs_show_options,
1657         .show_devname   = btrfs_show_devname,
1658         .write_inode    = btrfs_write_inode,
1659         .alloc_inode    = btrfs_alloc_inode,
1660         .destroy_inode  = btrfs_destroy_inode,
1661         .statfs         = btrfs_statfs,
1662         .remount_fs     = btrfs_remount,
1663         .freeze_fs      = btrfs_freeze,
1664         .unfreeze_fs    = btrfs_unfreeze,
1665 };
1666
1667 static const struct file_operations btrfs_ctl_fops = {
1668         .unlocked_ioctl  = btrfs_control_ioctl,
1669         .compat_ioctl = btrfs_control_ioctl,
1670         .owner   = THIS_MODULE,
1671         .llseek = noop_llseek,
1672 };
1673
1674 static struct miscdevice btrfs_misc = {
1675         .minor          = BTRFS_MINOR,
1676         .name           = "btrfs-control",
1677         .fops           = &btrfs_ctl_fops
1678 };
1679
1680 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1681 MODULE_ALIAS("devname:btrfs-control");
1682
1683 static int btrfs_interface_init(void)
1684 {
1685         return misc_register(&btrfs_misc);
1686 }
1687
1688 static void btrfs_interface_exit(void)
1689 {
1690         if (misc_deregister(&btrfs_misc) < 0)
1691                 printk(KERN_INFO "btrfs: misc_deregister failed for control device\n");
1692 }
1693
1694 static int __init init_btrfs_fs(void)
1695 {
1696         int err;
1697
1698         err = btrfs_init_sysfs();
1699         if (err)
1700                 return err;
1701
1702         btrfs_init_compress();
1703
1704         err = btrfs_init_cachep();
1705         if (err)
1706                 goto free_compress;
1707
1708         err = extent_io_init();
1709         if (err)
1710                 goto free_cachep;
1711
1712         err = extent_map_init();
1713         if (err)
1714                 goto free_extent_io;
1715
1716         err = ordered_data_init();
1717         if (err)
1718                 goto free_extent_map;
1719
1720         err = btrfs_delayed_inode_init();
1721         if (err)
1722                 goto free_ordered_data;
1723
1724         err = btrfs_auto_defrag_init();
1725         if (err)
1726                 goto free_delayed_inode;
1727
1728         err = btrfs_delayed_ref_init();
1729         if (err)
1730                 goto free_auto_defrag;
1731
1732         err = btrfs_interface_init();
1733         if (err)
1734                 goto free_delayed_ref;
1735
1736         err = register_filesystem(&btrfs_fs_type);
1737         if (err)
1738                 goto unregister_ioctl;
1739
1740         btrfs_init_lockdep();
1741
1742         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1743         return 0;
1744
1745 unregister_ioctl:
1746         btrfs_interface_exit();
1747 free_delayed_ref:
1748         btrfs_delayed_ref_exit();
1749 free_auto_defrag:
1750         btrfs_auto_defrag_exit();
1751 free_delayed_inode:
1752         btrfs_delayed_inode_exit();
1753 free_ordered_data:
1754         ordered_data_exit();
1755 free_extent_map:
1756         extent_map_exit();
1757 free_extent_io:
1758         extent_io_exit();
1759 free_cachep:
1760         btrfs_destroy_cachep();
1761 free_compress:
1762         btrfs_exit_compress();
1763         btrfs_exit_sysfs();
1764         return err;
1765 }
1766
1767 static void __exit exit_btrfs_fs(void)
1768 {
1769         btrfs_destroy_cachep();
1770         btrfs_delayed_ref_exit();
1771         btrfs_auto_defrag_exit();
1772         btrfs_delayed_inode_exit();
1773         ordered_data_exit();
1774         extent_map_exit();
1775         extent_io_exit();
1776         btrfs_interface_exit();
1777         unregister_filesystem(&btrfs_fs_type);
1778         btrfs_exit_sysfs();
1779         btrfs_cleanup_fs_uuids();
1780         btrfs_exit_compress();
1781 }
1782
1783 module_init(init_btrfs_fs)
1784 module_exit(exit_btrfs_fs)
1785
1786 MODULE_LICENSE("GPL");