fs/btrfs/super.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/time.h>
  26 #include <linux/init.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mount.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/parser.h>
  37 #include <linux/ctype.h>
  38 #include <linux/namei.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/magic.h>
  41 #include <linux/slab.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/ratelimit.h>
  44 #include "compat.h"
  45 #include "delayed-inode.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "ioctl.h"
  51 #include "print-tree.h"
  52 #include "xattr.h"
  53 #include "volumes.h"
  54 #include "version.h"
  55 #include "export.h"
  56 #include "compression.h"
  57 #include "rcu-string.h"
  58
  59 #define CREATE_TRACE_POINTS
  60 #include <trace/events/btrfs.h>
  61
  62 static const struct super_operations btrfs_super_ops;
  63 static struct file_system_type btrfs_fs_type;
  64
  65 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
  66                                       char nbuf[16])
  67 {
  68         char *errstr = NULL;
  69
  70         switch (errno) {
  71         case -EIO:
  72                 errstr = "IO failure";
  73                 break;
  74         case -ENOMEM:
  75                 errstr = "Out of memory";
  76                 break;
  77         case -EROFS:
  78                 errstr = "Readonly filesystem";
  79                 break;
  80         case -EEXIST:
  81                 errstr = "Object already exists";
  82                 break;
  83         default:
  84                 if (nbuf) {
  85                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  86                                 errstr = nbuf;
  87                 }
  88                 break;
  89         }
  90
  91         return errstr;
  92 }
  93
  94 static void __save_error_info(struct btrfs_fs_info *fs_info)
  95 {
  96         /*
  97          * today we only save the error info into ram.  Long term we'll
  98          * also send it down to the disk
  99          */
 100         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
 101 }
 102
 103 static void save_error_info(struct btrfs_fs_info *fs_info)
 104 {
 105         __save_error_info(fs_info);
 106 }
 107
 108 /* btrfs handle error by forcing the filesystem readonly */
 109 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 110 {
 111         struct super_block *sb = fs_info->sb;
 112
 113         if (sb->s_flags & MS_RDONLY)
 114                 return;
 115
 116         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 117                 sb->s_flags |= MS_RDONLY;
 118                 printk(KERN_INFO "btrfs is forced readonly\n");
 119                 /*
 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         va_list args;
 145         va_start(args, fmt);
 146
 147         /*
 148          * Special case: if the error is EROFS, and we're already
 149          * under MS_RDONLY, then it is safe here.
 150          */
 151         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 152                 return;
 153
 154         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 155         if (fmt) {
 156                 struct va_format vaf = {
 157                         .fmt = fmt,
 158                         .va = &args,
 159                 };
 160
 161                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s (%pV)\n",
 162                         sb->s_id, function, line, errstr, &vaf);
 163         } else {
 164                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 165                         sb->s_id, function, line, errstr);
 166         }
 167
 168         /* Don't go through full error handling during mount */
 169         if (sb->s_flags & MS_BORN) {
 170                 save_error_info(fs_info);
 171                 btrfs_handle_error(fs_info);
 172         }
 173         va_end(args);
 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(root->fs_info, errno, nbuf);
 264                 btrfs_printk(root->fs_info,
 265                              "%s:%d: Aborting unused transaction(%s).\n",
 266                              function, line, errstr);
 267                 return;
 268         }
 269         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(fs_info, errno, nbuf);
 292         if (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                 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                                 info->alloc_start = memparse(num, NULL);
 522                                 kfree(num);
 523                                 printk(KERN_INFO
 524                                         "btrfs: allocations start at %llu\n",
 525                                         (unsigned long long)info->alloc_start);
 526                         }
 527                         break;
 528                 case Opt_noacl:
 529                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 530                         break;
 531                 case Opt_notreelog:
 532                         printk(KERN_INFO "btrfs: disabling tree log\n");
 533                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 534                         break;
 535                 case Opt_flushoncommit:
 536                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 537                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 538                         break;
 539                 case Opt_ratio:
 540                         intarg = 0;
 541                         match_int(&args[0], &intarg);
 542                         if (intarg) {
 543                                 info->metadata_ratio = intarg;
 544                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 545                                        info->metadata_ratio);
 546                         }
 547                         break;
 548                 case Opt_discard:
 549                         btrfs_set_opt(info->mount_opt, DISCARD);
 550                         break;
 551                 case Opt_space_cache:
 552                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 553                         break;
 554                 case Opt_no_space_cache:
 555                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 556                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 557                         break;
 558                 case Opt_inode_cache:
 559                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 560                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 561                         break;
 562                 case Opt_clear_cache:
 563                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 564                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 565                         break;
 566                 case Opt_user_subvol_rm_allowed:
 567                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 568                         break;
 569                 case Opt_enospc_debug:
 570                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 571                         break;
 572                 case Opt_defrag:
 573                         printk(KERN_INFO "btrfs: enabling auto defrag\n");
 574                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 575                         break;
 576                 case Opt_recovery:
 577                         printk(KERN_INFO "btrfs: enabling auto recovery\n");
 578                         btrfs_set_opt(info->mount_opt, RECOVERY);
 579                         break;
 580                 case Opt_skip_balance:
 581                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 582                         break;
 583 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 584                 case Opt_check_integrity_including_extent_data:
 585                         printk(KERN_INFO "btrfs: enabling check integrity"
 586                                " including extent data\n");
 587                         btrfs_set_opt(info->mount_opt,
 588                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 589                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 590                         break;
 591                 case Opt_check_integrity:
 592                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 593                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 594                         break;
 595                 case Opt_check_integrity_print_mask:
 596                         intarg = 0;
 597                         match_int(&args[0], &intarg);
 598                         if (intarg) {
 599                                 info->check_integrity_print_mask = intarg;
 600                                 printk(KERN_INFO "btrfs:"
 601                                        " check_integrity_print_mask 0x%x\n",
 602                                        info->check_integrity_print_mask);
 603                         }
 604                         break;
 605 #else
 606                 case Opt_check_integrity_including_extent_data:
 607                 case Opt_check_integrity:
 608                 case Opt_check_integrity_print_mask:
 609                         printk(KERN_ERR "btrfs: support for check_integrity*"
 610                                " not compiled in!\n");
 611                         ret = -EINVAL;
 612                         goto out;
 613 #endif
 614                 case Opt_fatal_errors:
 615                         if (strcmp(args[0].from, "panic") == 0)
 616                                 btrfs_set_opt(info->mount_opt,
 617                                               PANIC_ON_FATAL_ERROR);
 618                         else if (strcmp(args[0].from, "bug") == 0)
 619                                 btrfs_clear_opt(info->mount_opt,
 620                                               PANIC_ON_FATAL_ERROR);
 621                         else {
 622                                 ret = -EINVAL;
 623                                 goto out;
 624                         }
 625                         break;
 626                 case Opt_err:
 627                         printk(KERN_INFO "btrfs: unrecognized mount option "
 628                                "'%s'\n", p);
 629                         ret = -EINVAL;
 630                         goto out;
 631                 default:
 632                         break;
 633                 }
 634         }
 635 out:
 636         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 637                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 638         kfree(orig);
 639         return ret;
 640 }
 641
 642 /*
 643  * Parse mount options that are required early in the mount process.
 644  *
 645  * All other options will be parsed on much later in the mount process and
 646  * only when we need to allocate a new super block.
 647  */
 648 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 649                 void *holder, char **subvol_name, u64 *subvol_objectid,
 650                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
 651 {
 652         substring_t args[MAX_OPT_ARGS];
 653         char *device_name, *opts, *orig, *p;
 654         int error = 0;
 655         int intarg;
 656
 657         if (!options)
 658                 return 0;
 659
 660         /*
 661          * strsep changes the string, duplicate it because parse_options
 662          * gets called twice
 663          */
 664         opts = kstrdup(options, GFP_KERNEL);
 665         if (!opts)
 666                 return -ENOMEM;
 667         orig = opts;
 668
 669         while ((p = strsep(&opts, ",")) != NULL) {
 670                 int token;
 671                 if (!*p)
 672                         continue;
 673
 674                 token = match_token(p, tokens, args);
 675                 switch (token) {
 676                 case Opt_subvol:
 677                         kfree(*subvol_name);
 678                         *subvol_name = match_strdup(&args[0]);
 679                         break;
 680                 case Opt_subvolid:
 681                         intarg = 0;
 682                         error = match_int(&args[0], &intarg);
 683                         if (!error) {
 684                                 /* we want the original fs_tree */
 685                                 if (!intarg)
 686                                         *subvol_objectid =
 687                                                 BTRFS_FS_TREE_OBJECTID;
 688                                 else
 689                                         *subvol_objectid = intarg;
 690                         }
 691                         break;
 692                 case Opt_subvolrootid:
 693                         intarg = 0;
 694                         error = match_int(&args[0], &intarg);
 695                         if (!error) {
 696                                 /* we want the original fs_tree */
 697                                 if (!intarg)
 698                                         *subvol_rootid =
 699                                                 BTRFS_FS_TREE_OBJECTID;
 700                                 else
 701                                         *subvol_rootid = intarg;
 702                         }
 703                         break;
 704                 case Opt_device:
 705                         device_name = match_strdup(&args[0]);
 706                         if (!device_name) {
 707                                 error = -ENOMEM;
 708                                 goto out;
 709                         }
 710                         error = btrfs_scan_one_device(device_name,
 711                                         flags, holder, fs_devices);
 712                         kfree(device_name);
 713                         if (error)
 714                                 goto out;
 715                         break;
 716                 default:
 717                         break;
 718                 }
 719         }
 720
 721 out:
 722         kfree(orig);
 723         return error;
 724 }
 725
 726 static struct dentry *get_default_root(struct super_block *sb,
 727                                        u64 subvol_objectid)
 728 {
 729         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 730         struct btrfs_root *root = fs_info->tree_root;
 731         struct btrfs_root *new_root;
 732         struct btrfs_dir_item *di;
 733         struct btrfs_path *path;
 734         struct btrfs_key location;
 735         struct inode *inode;
 736         u64 dir_id;
 737         int new = 0;
 738
 739         /*
 740          * We have a specific subvol we want to mount, just setup location and
 741          * go look up the root.
 742          */
 743         if (subvol_objectid) {
 744                 location.objectid = subvol_objectid;
 745                 location.type = BTRFS_ROOT_ITEM_KEY;
 746                 location.offset = (u64)-1;
 747                 goto find_root;
 748         }
 749
 750         path = btrfs_alloc_path();
 751         if (!path)
 752                 return ERR_PTR(-ENOMEM);
 753         path->leave_spinning = 1;
 754
 755         /*
 756          * Find the "default" dir item which points to the root item that we
 757          * will mount by default if we haven't been given a specific subvolume
 758          * to mount.
 759          */
 760         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 761         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 762         if (IS_ERR(di)) {
 763                 btrfs_free_path(path);
 764                 return ERR_CAST(di);
 765         }
 766         if (!di) {
 767                 /*
 768                  * Ok the default dir item isn't there.  This is weird since
 769                  * it's always been there, but don't freak out, just try and
 770                  * mount to root most subvolume.
 771                  */
 772                 btrfs_free_path(path);
 773                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 774                 new_root = fs_info->fs_root;
 775                 goto setup_root;
 776         }
 777
 778         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 779         btrfs_free_path(path);
 780
 781 find_root:
 782         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 783         if (IS_ERR(new_root))
 784                 return ERR_CAST(new_root);
 785
 786         if (btrfs_root_refs(&new_root->root_item) == 0)
 787                 return ERR_PTR(-ENOENT);
 788
 789         dir_id = btrfs_root_dirid(&new_root->root_item);
 790 setup_root:
 791         location.objectid = dir_id;
 792         location.type = BTRFS_INODE_ITEM_KEY;
 793         location.offset = 0;
 794
 795         inode = btrfs_iget(sb, &location, new_root, &new);
 796         if (IS_ERR(inode))
 797                 return ERR_CAST(inode);
 798
 799         /*
 800          * If we're just mounting the root most subvol put the inode and return
 801          * a reference to the dentry.  We will have already gotten a reference
 802          * to the inode in btrfs_fill_super so we're good to go.
 803          */
 804         if (!new && sb->s_root->d_inode == inode) {
 805                 iput(inode);
 806                 return dget(sb->s_root);
 807         }
 808
 809         return d_obtain_alias(inode);
 810 }
 811
 812 static int btrfs_fill_super(struct super_block *sb,
 813                             struct btrfs_fs_devices *fs_devices,
 814                             void *data, int silent)
 815 {
 816         struct inode *inode;
 817         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 818         struct btrfs_key key;
 819         int err;
 820
 821         sb->s_maxbytes = MAX_LFS_FILESIZE;
 822         sb->s_magic = BTRFS_SUPER_MAGIC;
 823         sb->s_op = &btrfs_super_ops;
 824         sb->s_d_op = &btrfs_dentry_operations;
 825         sb->s_export_op = &btrfs_export_ops;
 826         sb->s_xattr = btrfs_xattr_handlers;
 827         sb->s_time_gran = 1;
 828 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 829         sb->s_flags |= MS_POSIXACL;
 830 #endif
 831         sb->s_flags |= MS_I_VERSION;
 832         err = open_ctree(sb, fs_devices, (char *)data);
 833         if (err) {
 834                 printk("btrfs: open_ctree failed\n");
 835                 return err;
 836         }
 837
 838         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 839         key.type = BTRFS_INODE_ITEM_KEY;
 840         key.offset = 0;
 841         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 842         if (IS_ERR(inode)) {
 843                 err = PTR_ERR(inode);
 844                 goto fail_close;
 845         }
 846
 847         sb->s_root = d_make_root(inode);
 848         if (!sb->s_root) {
 849                 err = -ENOMEM;
 850                 goto fail_close;
 851         }
 852
 853         save_mount_options(sb, data);
 854         cleancache_init_fs(sb);
 855         sb->s_flags |= MS_ACTIVE;
 856         return 0;
 857
 858 fail_close:
 859         close_ctree(fs_info->tree_root);
 860         return err;
 861 }
 862
 863 int btrfs_sync_fs(struct super_block *sb, int wait)
 864 {
 865         struct btrfs_trans_handle *trans;
 866         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 867         struct btrfs_root *root = fs_info->tree_root;
 868
 869         trace_btrfs_sync_fs(wait);
 870
 871         if (!wait) {
 872                 filemap_flush(fs_info->btree_inode->i_mapping);
 873                 return 0;
 874         }
 875
 876         btrfs_wait_ordered_extents(root, 0);
 877
 878         trans = btrfs_attach_transaction(root);
 879         if (IS_ERR(trans)) {
 880                 /* no transaction, don't bother */
 881                 if (PTR_ERR(trans) == -ENOENT)
 882                         return 0;
 883                 return PTR_ERR(trans);
 884         }
 885         return btrfs_commit_transaction(trans, root);
 886 }
 887
 888 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 889 {
 890         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 891         struct btrfs_root *root = info->tree_root;
 892         char *compress_type;
 893
 894         if (btrfs_test_opt(root, DEGRADED))
 895                 seq_puts(seq, ",degraded");
 896         if (btrfs_test_opt(root, NODATASUM))
 897                 seq_puts(seq, ",nodatasum");
 898         if (btrfs_test_opt(root, NODATACOW))
 899                 seq_puts(seq, ",nodatacow");
 900         if (btrfs_test_opt(root, NOBARRIER))
 901                 seq_puts(seq, ",nobarrier");
 902         if (info->max_inline != 8192 * 1024)
 903                 seq_printf(seq, ",max_inline=%llu",
 904                            (unsigned long long)info->max_inline);
 905         if (info->alloc_start != 0)
 906                 seq_printf(seq, ",alloc_start=%llu",
 907                            (unsigned long long)info->alloc_start);
 908         if (info->thread_pool_size !=  min_t(unsigned long,
 909                                              num_online_cpus() + 2, 8))
 910                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 911         if (btrfs_test_opt(root, COMPRESS)) {
 912                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 913                         compress_type = "zlib";
 914                 else
 915                         compress_type = "lzo";
 916                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 917                         seq_printf(seq, ",compress-force=%s", compress_type);
 918                 else
 919                         seq_printf(seq, ",compress=%s", compress_type);
 920         }
 921         if (btrfs_test_opt(root, NOSSD))
 922                 seq_puts(seq, ",nossd");
 923         if (btrfs_test_opt(root, SSD_SPREAD))
 924                 seq_puts(seq, ",ssd_spread");
 925         else if (btrfs_test_opt(root, SSD))
 926                 seq_puts(seq, ",ssd");
 927         if (btrfs_test_opt(root, NOTREELOG))
 928                 seq_puts(seq, ",notreelog");
 929         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 930                 seq_puts(seq, ",flushoncommit");
 931         if (btrfs_test_opt(root, DISCARD))
 932                 seq_puts(seq, ",discard");
 933         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 934                 seq_puts(seq, ",noacl");
 935         if (btrfs_test_opt(root, SPACE_CACHE))
 936                 seq_puts(seq, ",space_cache");
 937         else
 938                 seq_puts(seq, ",nospace_cache");
 939         if (btrfs_test_opt(root, CLEAR_CACHE))
 940                 seq_puts(seq, ",clear_cache");
 941         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 942                 seq_puts(seq, ",user_subvol_rm_allowed");
 943         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 944                 seq_puts(seq, ",enospc_debug");
 945         if (btrfs_test_opt(root, AUTO_DEFRAG))
 946                 seq_puts(seq, ",autodefrag");
 947         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 948                 seq_puts(seq, ",inode_cache");
 949         if (btrfs_test_opt(root, SKIP_BALANCE))
 950                 seq_puts(seq, ",skip_balance");
 951         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 952                 seq_puts(seq, ",fatal_errors=panic");
 953         return 0;
 954 }
 955
 956 static int btrfs_test_super(struct super_block *s, void *data)
 957 {
 958         struct btrfs_fs_info *p = data;
 959         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 960
 961         return fs_info->fs_devices == p->fs_devices;
 962 }
 963
 964 static int btrfs_set_super(struct super_block *s, void *data)
 965 {
 966         int err = set_anon_super(s, data);
 967         if (!err)
 968                 s->s_fs_info = data;
 969         return err;
 970 }
 971
 972 /*
 973  * subvolumes are identified by ino 256
 974  */
 975 static inline int is_subvolume_inode(struct inode *inode)
 976 {
 977         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 978                 return 1;
 979         return 0;
 980 }
 981
 982 /*
 983  * This will strip out the subvol=%s argument for an argument string and add
 984  * subvolid=0 to make sure we get the actual tree root for path walking to the
 985  * subvol we want.
 986  */
 987 static char *setup_root_args(char *args)
 988 {
 989         unsigned len = strlen(args) + 2 + 1;
 990         char *src, *dst, *buf;
 991
 992         /*
 993          * We need the same args as before, but with this substitution:
 994          * s!subvol=[^,]+!subvolid=0!
 995          *
 996          * Since the replacement string is up to 2 bytes longer than the
 997          * original, allocate strlen(args) + 2 + 1 bytes.
 998          */
 999
1000         src = strstr(args, "subvol=");
1001         /* This shouldn't happen, but just in case.. */
1002         if (!src)
1003                 return NULL;
1004
1005         buf = dst = kmalloc(len, GFP_NOFS);
1006         if (!buf)
1007                 return NULL;
1008
1009         /*
1010          * If the subvol= arg is not at the start of the string,
1011          * copy whatever precedes it into buf.
1012          */
1013         if (src != args) {
1014                 *src++ = '\0';
1015                 strcpy(buf, args);
1016                 dst += strlen(args);
1017         }
1018
1019         strcpy(dst, "subvolid=0");
1020         dst += strlen("subvolid=0");
1021
1022         /*
1023          * If there is a "," after the original subvol=... string,
1024          * copy that suffix into our buffer.  Otherwise, we're done.
1025          */
1026         src = strchr(src, ',');
1027         if (src)
1028                 strcpy(dst, src);
1029
1030         return buf;
1031 }
1032
1033 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1034                                    const char *device_name, char *data)
1035 {
1036         struct dentry *root;
1037         struct vfsmount *mnt;
1038         char *newargs;
1039
1040         newargs = setup_root_args(data);
1041         if (!newargs)
1042                 return ERR_PTR(-ENOMEM);
1043         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1044                              newargs);
1045         kfree(newargs);
1046         if (IS_ERR(mnt))
1047                 return ERR_CAST(mnt);
1048
1049         root = mount_subtree(mnt, subvol_name);
1050
1051         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1052                 struct super_block *s = root->d_sb;
1053                 dput(root);
1054                 root = ERR_PTR(-EINVAL);
1055                 deactivate_locked_super(s);
1056                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1057                                 subvol_name);
1058         }
1059
1060         return root;
1061 }
1062
1063 /*
1064  * Find a superblock for the given device / mount point.
1065  *
1066  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1067  *        for multiple device setup.  Make sure to keep it in sync.
1068  */
1069 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1070                 const char *device_name, void *data)
1071 {
1072         struct block_device *bdev = NULL;
1073         struct super_block *s;
1074         struct dentry *root;
1075         struct btrfs_fs_devices *fs_devices = NULL;
1076         struct btrfs_fs_info *fs_info = NULL;
1077         fmode_t mode = FMODE_READ;
1078         char *subvol_name = NULL;
1079         u64 subvol_objectid = 0;
1080         u64 subvol_rootid = 0;
1081         int error = 0;
1082
1083         if (!(flags & MS_RDONLY))
1084                 mode |= FMODE_WRITE;
1085
1086         error = btrfs_parse_early_options(data, mode, fs_type,
1087                                           &subvol_name, &subvol_objectid,
1088                                           &subvol_rootid, &fs_devices);
1089         if (error) {
1090                 kfree(subvol_name);
1091                 return ERR_PTR(error);
1092         }
1093
1094         if (subvol_name) {
1095                 root = mount_subvol(subvol_name, flags, device_name, data);
1096                 kfree(subvol_name);
1097                 return root;
1098         }
1099
1100         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1101         if (error)
1102                 return ERR_PTR(error);
1103
1104         /*
1105          * Setup a dummy root and fs_info for test/set super.  This is because
1106          * we don't actually fill this stuff out until open_ctree, but we need
1107          * it for searching for existing supers, so this lets us do that and
1108          * then open_ctree will properly initialize everything later.
1109          */
1110         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1111         if (!fs_info)
1112                 return ERR_PTR(-ENOMEM);
1113
1114         fs_info->fs_devices = fs_devices;
1115
1116         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1117         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1118         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1119                 error = -ENOMEM;
1120                 goto error_fs_info;
1121         }
1122
1123         error = btrfs_open_devices(fs_devices, mode, fs_type);
1124         if (error)
1125                 goto error_fs_info;
1126
1127         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1128                 error = -EACCES;
1129                 goto error_close_devices;
1130         }
1131
1132         bdev = fs_devices->latest_bdev;
1133         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1134                  fs_info);
1135         if (IS_ERR(s)) {
1136                 error = PTR_ERR(s);
1137                 goto error_close_devices;
1138         }
1139
1140         if (s->s_root) {
1141                 btrfs_close_devices(fs_devices);
1142                 free_fs_info(fs_info);
1143                 if ((flags ^ s->s_flags) & MS_RDONLY)
1144                         error = -EBUSY;
1145         } else {
1146                 char b[BDEVNAME_SIZE];
1147
1148                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1149                 btrfs_sb(s)->bdev_holder = fs_type;
1150                 error = btrfs_fill_super(s, fs_devices, data,
1151                                          flags & MS_SILENT ? 1 : 0);
1152         }
1153
1154         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1155         if (IS_ERR(root))
1156                 deactivate_locked_super(s);
1157
1158         return root;
1159
1160 error_close_devices:
1161         btrfs_close_devices(fs_devices);
1162 error_fs_info:
1163         free_fs_info(fs_info);
1164         return ERR_PTR(error);
1165 }
1166
1167 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1168 {
1169         spin_lock_irq(&workers->lock);
1170         workers->max_workers = new_limit;
1171         spin_unlock_irq(&workers->lock);
1172 }
1173
1174 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1175                                      int new_pool_size, int old_pool_size)
1176 {
1177         if (new_pool_size == old_pool_size)
1178                 return;
1179
1180         fs_info->thread_pool_size = new_pool_size;
1181
1182         printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1183                old_pool_size, new_pool_size);
1184
1185         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1186         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1187         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1188         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1189         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1190         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1191         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1192         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1193         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1194         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1195         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1196         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1197         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1198         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1199                               new_pool_size);
1200 }
1201
1202 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1203 {
1204         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1205         struct btrfs_root *root = fs_info->tree_root;
1206         unsigned old_flags = sb->s_flags;
1207         unsigned long old_opts = fs_info->mount_opt;
1208         unsigned long old_compress_type = fs_info->compress_type;
1209         u64 old_max_inline = fs_info->max_inline;
1210         u64 old_alloc_start = fs_info->alloc_start;
1211         int old_thread_pool_size = fs_info->thread_pool_size;
1212         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1213         int ret;
1214
1215         ret = btrfs_parse_options(root, data);
1216         if (ret) {
1217                 ret = -EINVAL;
1218                 goto restore;
1219         }
1220
1221         btrfs_resize_thread_pool(fs_info,
1222                 fs_info->thread_pool_size, old_thread_pool_size);
1223
1224         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1225                 return 0;
1226
1227         if (*flags & MS_RDONLY) {
1228                 sb->s_flags |= MS_RDONLY;
1229
1230                 ret = btrfs_commit_super(root);
1231                 if (ret)
1232                         goto restore;
1233         } else {
1234                 if (fs_info->fs_devices->rw_devices == 0) {
1235                         ret = -EACCES;
1236                         goto restore;
1237                 }
1238
1239                 if (fs_info->fs_devices->missing_devices >
1240                      fs_info->num_tolerated_disk_barrier_failures &&
1241                     !(*flags & MS_RDONLY)) {
1242                         printk(KERN_WARNING
1243                                "Btrfs: too many missing devices, writeable remount is not allowed\n");
1244                         ret = -EACCES;
1245                         goto restore;
1246                 }
1247
1248                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1249                         ret = -EINVAL;
1250                         goto restore;
1251                 }
1252
1253                 ret = btrfs_cleanup_fs_roots(fs_info);
1254                 if (ret)
1255                         goto restore;
1256
1257                 /* recover relocation */
1258                 ret = btrfs_recover_relocation(root);
1259                 if (ret)
1260                         goto restore;
1261
1262                 ret = btrfs_resume_balance_async(fs_info);
1263                 if (ret)
1264                         goto restore;
1265
1266                 sb->s_flags &= ~MS_RDONLY;
1267         }
1268
1269         return 0;
1270
1271 restore:
1272         /* We've hit an error - don't reset MS_RDONLY */
1273         if (sb->s_flags & MS_RDONLY)
1274                 old_flags |= MS_RDONLY;
1275         sb->s_flags = old_flags;
1276         fs_info->mount_opt = old_opts;
1277         fs_info->compress_type = old_compress_type;
1278         fs_info->max_inline = old_max_inline;
1279         fs_info->alloc_start = old_alloc_start;
1280         btrfs_resize_thread_pool(fs_info,
1281                 old_thread_pool_size, fs_info->thread_pool_size);
1282         fs_info->metadata_ratio = old_metadata_ratio;
1283         return ret;
1284 }
1285
1286 /* Used to sort the devices by max_avail(descending sort) */
1287 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1288                                        const void *dev_info2)
1289 {
1290         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1291             ((struct btrfs_device_info *)dev_info2)->max_avail)
1292                 return -1;
1293         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1294                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1295                 return 1;
1296         else
1297         return 0;
1298 }
1299
1300 /*
1301  * sort the devices by max_avail, in which max free extent size of each device
1302  * is stored.(Descending Sort)
1303  */
1304 static inline void btrfs_descending_sort_devices(
1305                                         struct btrfs_device_info *devices,
1306                                         size_t nr_devices)
1307 {
1308         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1309              btrfs_cmp_device_free_bytes, NULL);
1310 }
1311
1312 /*
1313  * The helper to calc the free space on the devices that can be used to store
1314  * file data.
1315  */
1316 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1317 {
1318         struct btrfs_fs_info *fs_info = root->fs_info;
1319         struct btrfs_device_info *devices_info;
1320         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1321         struct btrfs_device *device;
1322         u64 skip_space;
1323         u64 type;
1324         u64 avail_space;
1325         u64 used_space;
1326         u64 min_stripe_size;
1327         int min_stripes = 1, num_stripes = 1;
1328         int i = 0, nr_devices;
1329         int ret;
1330
1331         nr_devices = fs_info->fs_devices->open_devices;
1332         BUG_ON(!nr_devices);
1333
1334         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1335                                GFP_NOFS);
1336         if (!devices_info)
1337                 return -ENOMEM;
1338
1339         /* calc min stripe number for data space alloction */
1340         type = btrfs_get_alloc_profile(root, 1);
1341         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1342                 min_stripes = 2;
1343                 num_stripes = nr_devices;
1344         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1345                 min_stripes = 2;
1346                 num_stripes = 2;
1347         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1348                 min_stripes = 4;
1349                 num_stripes = 4;
1350         }
1351
1352         if (type & BTRFS_BLOCK_GROUP_DUP)
1353                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1354         else
1355                 min_stripe_size = BTRFS_STRIPE_LEN;
1356
1357         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1358                 if (!device->in_fs_metadata || !device->bdev ||
1359                     device->is_tgtdev_for_dev_replace)
1360                         continue;
1361
1362                 avail_space = device->total_bytes - device->bytes_used;
1363
1364                 /* align with stripe_len */
1365                 do_div(avail_space, BTRFS_STRIPE_LEN);
1366                 avail_space *= BTRFS_STRIPE_LEN;
1367
1368                 /*
1369                  * In order to avoid overwritting the superblock on the drive,
1370                  * btrfs starts at an offset of at least 1MB when doing chunk
1371                  * allocation.
1372                  */
1373                 skip_space = 1024 * 1024;
1374
1375                 /* user can set the offset in fs_info->alloc_start. */
1376                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1377                     device->total_bytes)
1378                         skip_space = max(fs_info->alloc_start, skip_space);
1379
1380                 /*
1381                  * btrfs can not use the free space in [0, skip_space - 1],
1382                  * we must subtract it from the total. In order to implement
1383                  * it, we account the used space in this range first.
1384                  */
1385                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1386                                                      &used_space);
1387                 if (ret) {
1388                         kfree(devices_info);
1389                         return ret;
1390                 }
1391
1392                 /* calc the free space in [0, skip_space - 1] */
1393                 skip_space -= used_space;
1394
1395                 /*
1396                  * we can use the free space in [0, skip_space - 1], subtract
1397                  * it from the total.
1398                  */
1399                 if (avail_space && avail_space >= skip_space)
1400                         avail_space -= skip_space;
1401                 else
1402                         avail_space = 0;
1403
1404                 if (avail_space < min_stripe_size)
1405                         continue;
1406
1407                 devices_info[i].dev = device;
1408                 devices_info[i].max_avail = avail_space;
1409
1410                 i++;
1411         }
1412
1413         nr_devices = i;
1414
1415         btrfs_descending_sort_devices(devices_info, nr_devices);
1416
1417         i = nr_devices - 1;
1418         avail_space = 0;
1419         while (nr_devices >= min_stripes) {
1420                 if (num_stripes > nr_devices)
1421                         num_stripes = nr_devices;
1422
1423                 if (devices_info[i].max_avail >= min_stripe_size) {
1424                         int j;
1425                         u64 alloc_size;
1426
1427                         avail_space += devices_info[i].max_avail * num_stripes;
1428                         alloc_size = devices_info[i].max_avail;
1429                         for (j = i + 1 - num_stripes; j <= i; j++)
1430                                 devices_info[j].max_avail -= alloc_size;
1431                 }
1432                 i--;
1433                 nr_devices--;
1434         }
1435
1436         kfree(devices_info);
1437         *free_bytes = avail_space;
1438         return 0;
1439 }
1440
1441 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1442 {
1443         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1444         struct btrfs_super_block *disk_super = fs_info->super_copy;
1445         struct list_head *head = &fs_info->space_info;
1446         struct btrfs_space_info *found;
1447         u64 total_used = 0;
1448         u64 total_free_data = 0;
1449         int bits = dentry->d_sb->s_blocksize_bits;
1450         __be32 *fsid = (__be32 *)fs_info->fsid;
1451         int ret;
1452
1453         /* holding chunk_muext to avoid allocating new chunks */
1454         mutex_lock(&fs_info->chunk_mutex);
1455         rcu_read_lock();
1456         list_for_each_entry_rcu(found, head, list) {
1457                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1458                         total_free_data += found->disk_total - found->disk_used;
1459                         total_free_data -=
1460                                 btrfs_account_ro_block_groups_free_space(found);
1461                 }
1462
1463                 total_used += found->disk_used;
1464         }
1465         rcu_read_unlock();
1466
1467         buf->f_namelen = BTRFS_NAME_LEN;
1468         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1469         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1470         buf->f_bsize = dentry->d_sb->s_blocksize;
1471         buf->f_type = BTRFS_SUPER_MAGIC;
1472         buf->f_bavail = total_free_data;
1473         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1474         if (ret) {
1475                 mutex_unlock(&fs_info->chunk_mutex);
1476                 return ret;
1477         }
1478         buf->f_bavail += total_free_data;
1479         buf->f_bavail = buf->f_bavail >> bits;
1480         mutex_unlock(&fs_info->chunk_mutex);
1481
1482         /* We treat it as constant endianness (it doesn't matter _which_)
1483            because we want the fsid to come out the same whether mounted
1484            on a big-endian or little-endian host */
1485         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1486         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1487         /* Mask in the root object ID too, to disambiguate subvols */
1488         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1489         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1490
1491         return 0;
1492 }
1493
1494 static void btrfs_kill_super(struct super_block *sb)
1495 {
1496         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1497         kill_anon_super(sb);
1498         free_fs_info(fs_info);
1499 }
1500
1501 static struct file_system_type btrfs_fs_type = {
1502         .owner          = THIS_MODULE,
1503         .name           = "btrfs",
1504         .mount          = btrfs_mount,
1505         .kill_sb        = btrfs_kill_super,
1506         .fs_flags       = FS_REQUIRES_DEV,
1507 };
1508
1509 /*
1510  * used by btrfsctl to scan devices when no FS is mounted
1511  */
1512 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1513                                 unsigned long arg)
1514 {
1515         struct btrfs_ioctl_vol_args *vol;
1516         struct btrfs_fs_devices *fs_devices;
1517         int ret = -ENOTTY;
1518
1519         if (!capable(CAP_SYS_ADMIN))
1520                 return -EPERM;
1521
1522         vol = memdup_user((void __user *)arg, sizeof(*vol));
1523         if (IS_ERR(vol))
1524                 return PTR_ERR(vol);
1525
1526         switch (cmd) {
1527         case BTRFS_IOC_SCAN_DEV:
1528                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1529                                             &btrfs_fs_type, &fs_devices);
1530                 break;
1531         case BTRFS_IOC_DEVICES_READY:
1532                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1533                                             &btrfs_fs_type, &fs_devices);
1534                 if (ret)
1535                         break;
1536                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1537                 break;
1538         }
1539
1540         kfree(vol);
1541         return ret;
1542 }
1543
1544 static int btrfs_freeze(struct super_block *sb)
1545 {
1546         struct btrfs_trans_handle *trans;
1547         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1548
1549         trans = btrfs_attach_transaction(root);
1550         if (IS_ERR(trans)) {
1551                 /* no transaction, don't bother */
1552                 if (PTR_ERR(trans) == -ENOENT)
1553                         return 0;
1554                 return PTR_ERR(trans);
1555         }
1556         return btrfs_commit_transaction(trans, root);
1557 }
1558
1559 static int btrfs_unfreeze(struct super_block *sb)
1560 {
1561         return 0;
1562 }
1563
1564 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1565 {
1566         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1567         struct btrfs_fs_devices *cur_devices;
1568         struct btrfs_device *dev, *first_dev = NULL;
1569         struct list_head *head;
1570         struct rcu_string *name;
1571
1572         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1573         cur_devices = fs_info->fs_devices;
1574         while (cur_devices) {
1575                 head = &cur_devices->devices;
1576                 list_for_each_entry(dev, head, dev_list) {
1577                         if (dev->missing)
1578                                 continue;
1579                         if (!first_dev || dev->devid < first_dev->devid)
1580                                 first_dev = dev;
1581                 }
1582                 cur_devices = cur_devices->seed;
1583         }
1584
1585         if (first_dev) {
1586                 rcu_read_lock();
1587                 name = rcu_dereference(first_dev->name);
1588                 seq_escape(m, name->str, " \t\n\\");
1589                 rcu_read_unlock();
1590         } else {
1591                 WARN_ON(1);
1592         }
1593         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1594         return 0;
1595 }
1596
1597 static const struct super_operations btrfs_super_ops = {
1598         .drop_inode     = btrfs_drop_inode,
1599         .evict_inode    = btrfs_evict_inode,
1600         .put_super      = btrfs_put_super,
1601         .sync_fs        = btrfs_sync_fs,
1602         .show_options   = btrfs_show_options,
1603         .show_devname   = btrfs_show_devname,
1604         .write_inode    = btrfs_write_inode,
1605         .alloc_inode    = btrfs_alloc_inode,
1606         .destroy_inode  = btrfs_destroy_inode,
1607         .statfs         = btrfs_statfs,
1608         .remount_fs     = btrfs_remount,
1609         .freeze_fs      = btrfs_freeze,
1610         .unfreeze_fs    = btrfs_unfreeze,
1611 };
1612
1613 static const struct file_operations btrfs_ctl_fops = {
1614         .unlocked_ioctl  = btrfs_control_ioctl,
1615         .compat_ioctl = btrfs_control_ioctl,
1616         .owner   = THIS_MODULE,
1617         .llseek = noop_llseek,
1618 };
1619
1620 static struct miscdevice btrfs_misc = {
1621         .minor          = BTRFS_MINOR,
1622         .name           = "btrfs-control",
1623         .fops           = &btrfs_ctl_fops
1624 };
1625
1626 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1627 MODULE_ALIAS("devname:btrfs-control");
1628
1629 static int btrfs_interface_init(void)
1630 {
1631         return misc_register(&btrfs_misc);
1632 }
1633
1634 static void btrfs_interface_exit(void)
1635 {
1636         if (misc_deregister(&btrfs_misc) < 0)
1637                 printk(KERN_INFO "btrfs: misc_deregister failed for control device\n");
1638 }
1639
1640 static int __init init_btrfs_fs(void)
1641 {
1642         int err;
1643
1644         err = btrfs_init_sysfs();
1645         if (err)
1646                 return err;
1647
1648         btrfs_init_compress();
1649
1650         err = btrfs_init_cachep();
1651         if (err)
1652                 goto free_compress;
1653
1654         err = extent_io_init();
1655         if (err)
1656                 goto free_cachep;
1657
1658         err = extent_map_init();
1659         if (err)
1660                 goto free_extent_io;
1661
1662         err = ordered_data_init();
1663         if (err)
1664                 goto free_extent_map;
1665
1666         err = btrfs_delayed_inode_init();
1667         if (err)
1668                 goto free_ordered_data;
1669
1670         err = btrfs_interface_init();
1671         if (err)
1672                 goto free_delayed_inode;
1673
1674         err = register_filesystem(&btrfs_fs_type);
1675         if (err)
1676                 goto unregister_ioctl;
1677
1678         btrfs_init_lockdep();
1679
1680         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1681         return 0;
1682
1683 unregister_ioctl:
1684         btrfs_interface_exit();
1685 free_delayed_inode:
1686         btrfs_delayed_inode_exit();
1687 free_ordered_data:
1688         ordered_data_exit();
1689 free_extent_map:
1690         extent_map_exit();
1691 free_extent_io:
1692         extent_io_exit();
1693 free_cachep:
1694         btrfs_destroy_cachep();
1695 free_compress:
1696         btrfs_exit_compress();
1697         btrfs_exit_sysfs();
1698         return err;
1699 }
1700
1701 static void __exit exit_btrfs_fs(void)
1702 {
1703         btrfs_destroy_cachep();
1704         btrfs_delayed_inode_exit();
1705         ordered_data_exit();
1706         extent_map_exit();
1707         extent_io_exit();
1708         btrfs_interface_exit();
1709         unregister_filesystem(&btrfs_fs_type);
1710         btrfs_exit_sysfs();
1711         btrfs_cleanup_fs_uuids();
1712         btrfs_exit_compress();
1713 }
1714
1715 module_init(init_btrfs_fs)
1716 module_exit(exit_btrfs_fs)
1717
1718 MODULE_LICENSE("GPL");