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