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 "delayed-inode.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "hash.h"
  52 #include "props.h"
  53 #include "xattr.h"
  54 #include "volumes.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 #include "backref.h"
  61 #include "sysfs.h"
  62 #include "tests/btrfs-tests.h"
  63
  64 #include "qgroup.h"
  65 #define CREATE_TRACE_POINTS
  66 #include <trace/events/btrfs.h>
  67
  68 static const struct super_operations btrfs_super_ops;
  69 static struct file_system_type btrfs_fs_type;
  70
  71 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  72
  73 const char *btrfs_decode_error(int errno)
  74 {
  75         char *errstr = "unknown";
  76
  77         switch (errno) {
  78         case -EIO:
  79                 errstr = "IO failure";
  80                 break;
  81         case -ENOMEM:
  82                 errstr = "Out of memory";
  83                 break;
  84         case -EROFS:
  85                 errstr = "Readonly filesystem";
  86                 break;
  87         case -EEXIST:
  88                 errstr = "Object already exists";
  89                 break;
  90         case -ENOSPC:
  91                 errstr = "No space left";
  92                 break;
  93         case -ENOENT:
  94                 errstr = "No such entry";
  95                 break;
  96         }
  97
  98         return errstr;
  99 }
 100
 101 static void save_error_info(struct btrfs_fs_info *fs_info)
 102 {
 103         /*
 104          * today we only save the error info into ram.  Long term we'll
 105          * also send it down to the disk
 106          */
 107         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 108 }
 109
 110 /* btrfs handle error by forcing the filesystem readonly */
 111 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 112 {
 113         struct super_block *sb = fs_info->sb;
 114
 115         if (sb->s_flags & MS_RDONLY)
 116                 return;
 117
 118         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 119                 sb->s_flags |= MS_RDONLY;
 120                 btrfs_info(fs_info, "forced readonly");
 121                 /*
 122                  * Note that a running device replace operation is not
 123                  * canceled here although there is no way to update
 124                  * the progress. It would add the risk of a deadlock,
 125                  * therefore the canceling is ommited. The only penalty
 126                  * is that some I/O remains active until the procedure
 127                  * completes. The next time when the filesystem is
 128                  * mounted writeable again, the device replace
 129                  * operation continues.
 130                  */
 131         }
 132 }
 133
 134 /*
 135  * __btrfs_std_error decodes expected errors from the caller and
 136  * invokes the approciate error response.
 137  */
 138 __cold
 139 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 140                        unsigned int line, int errno, const char *fmt, ...)
 141 {
 142         struct super_block *sb = fs_info->sb;
 143 #ifdef CONFIG_PRINTK
 144         const char *errstr;
 145 #endif
 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 #ifdef CONFIG_PRINTK
 155         errstr = btrfs_decode_error(errno);
 156         if (fmt) {
 157                 struct va_format vaf;
 158                 va_list args;
 159
 160                 va_start(args, fmt);
 161                 vaf.fmt = fmt;
 162                 vaf.va = &args;
 163
 164                 printk(KERN_CRIT
 165                         "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 166                         sb->s_id, function, line, errno, errstr, &vaf);
 167                 va_end(args);
 168         } else {
 169                 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 170                         sb->s_id, function, line, errno, errstr);
 171         }
 172 #endif
 173
 174         /* Don't go through full error handling during mount */
 175         save_error_info(fs_info);
 176         if (sb->s_flags & MS_BORN)
 177                 btrfs_handle_error(fs_info);
 178 }
 179
 180 #ifdef CONFIG_PRINTK
 181 static const char * const logtypes[] = {
 182         "emergency",
 183         "alert",
 184         "critical",
 185         "error",
 186         "warning",
 187         "notice",
 188         "info",
 189         "debug",
 190 };
 191
 192 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 193 {
 194         struct super_block *sb = fs_info->sb;
 195         char lvl[4];
 196         struct va_format vaf;
 197         va_list args;
 198         const char *type = logtypes[4];
 199         int kern_level;
 200
 201         va_start(args, fmt);
 202
 203         kern_level = printk_get_level(fmt);
 204         if (kern_level) {
 205                 size_t size = printk_skip_level(fmt) - fmt;
 206                 memcpy(lvl, fmt,  size);
 207                 lvl[size] = '\0';
 208                 fmt += size;
 209                 type = logtypes[kern_level - '0'];
 210         } else
 211                 *lvl = '\0';
 212
 213         vaf.fmt = fmt;
 214         vaf.va = &args;
 215
 216         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 217
 218         va_end(args);
 219 }
 220 #endif
 221
 222 /*
 223  * We only mark the transaction aborted and then set the file system read-only.
 224  * This will prevent new transactions from starting or trying to join this
 225  * one.
 226  *
 227  * This means that error recovery at the call site is limited to freeing
 228  * any local memory allocations and passing the error code up without
 229  * further cleanup. The transaction should complete as it normally would
 230  * in the call path but will return -EIO.
 231  *
 232  * We'll complete the cleanup in btrfs_end_transaction and
 233  * btrfs_commit_transaction.
 234  */
 235 __cold
 236 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 237                                struct btrfs_root *root, const char *function,
 238                                unsigned int line, int errno)
 239 {
 240         trans->aborted = errno;
 241         /* Nothing used. The other threads that have joined this
 242          * transaction may be able to continue. */
 243         if (!trans->blocks_used && list_empty(&trans->new_bgs)) {
 244                 const char *errstr;
 245
 246                 errstr = btrfs_decode_error(errno);
 247                 btrfs_warn(root->fs_info,
 248                            "%s:%d: Aborting unused transaction(%s).",
 249                            function, line, errstr);
 250                 return;
 251         }
 252         ACCESS_ONCE(trans->transaction->aborted) = errno;
 253         /* Wake up anybody who may be waiting on this transaction */
 254         wake_up(&root->fs_info->transaction_wait);
 255         wake_up(&root->fs_info->transaction_blocked_wait);
 256         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 257 }
 258 /*
 259  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 260  * issues an alert, and either panics or BUGs, depending on mount options.
 261  */
 262 __cold
 263 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 264                    unsigned int line, int errno, const char *fmt, ...)
 265 {
 266         char *s_id = "<unknown>";
 267         const char *errstr;
 268         struct va_format vaf = { .fmt = fmt };
 269         va_list args;
 270
 271         if (fs_info)
 272                 s_id = fs_info->sb->s_id;
 273
 274         va_start(args, fmt);
 275         vaf.va = &args;
 276
 277         errstr = btrfs_decode_error(errno);
 278         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 279                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 280                         s_id, function, line, &vaf, errno, errstr);
 281
 282         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 283                    function, line, &vaf, errno, errstr);
 284         va_end(args);
 285         /* Caller calls BUG() */
 286 }
 287
 288 static void btrfs_put_super(struct super_block *sb)
 289 {
 290         close_ctree(btrfs_sb(sb)->tree_root);
 291 }
 292
 293 enum {
 294         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 295         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 296         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 297         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 298         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 299         Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
 300         Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
 301         Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
 302         Opt_skip_balance, Opt_check_integrity,
 303         Opt_check_integrity_including_extent_data,
 304         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 305         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
 306         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
 307         Opt_datasum, Opt_treelog, Opt_noinode_cache,
 308 #ifdef CONFIG_BTRFS_DEBUG
 309         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 310 #endif
 311         Opt_err,
 312 };
 313
 314 static const match_table_t tokens = {
 315         {Opt_degraded, "degraded"},
 316         {Opt_subvol, "subvol=%s"},
 317         {Opt_subvolid, "subvolid=%s"},
 318         {Opt_device, "device=%s"},
 319         {Opt_nodatasum, "nodatasum"},
 320         {Opt_datasum, "datasum"},
 321         {Opt_nodatacow, "nodatacow"},
 322         {Opt_datacow, "datacow"},
 323         {Opt_nobarrier, "nobarrier"},
 324         {Opt_barrier, "barrier"},
 325         {Opt_max_inline, "max_inline=%s"},
 326         {Opt_alloc_start, "alloc_start=%s"},
 327         {Opt_thread_pool, "thread_pool=%d"},
 328         {Opt_compress, "compress"},
 329         {Opt_compress_type, "compress=%s"},
 330         {Opt_compress_force, "compress-force"},
 331         {Opt_compress_force_type, "compress-force=%s"},
 332         {Opt_ssd, "ssd"},
 333         {Opt_ssd_spread, "ssd_spread"},
 334         {Opt_nossd, "nossd"},
 335         {Opt_acl, "acl"},
 336         {Opt_noacl, "noacl"},
 337         {Opt_notreelog, "notreelog"},
 338         {Opt_treelog, "treelog"},
 339         {Opt_flushoncommit, "flushoncommit"},
 340         {Opt_noflushoncommit, "noflushoncommit"},
 341         {Opt_ratio, "metadata_ratio=%d"},
 342         {Opt_discard, "discard"},
 343         {Opt_nodiscard, "nodiscard"},
 344         {Opt_space_cache, "space_cache"},
 345         {Opt_space_cache_version, "space_cache=%s"},
 346         {Opt_clear_cache, "clear_cache"},
 347         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 348         {Opt_enospc_debug, "enospc_debug"},
 349         {Opt_noenospc_debug, "noenospc_debug"},
 350         {Opt_subvolrootid, "subvolrootid=%d"},
 351         {Opt_defrag, "autodefrag"},
 352         {Opt_nodefrag, "noautodefrag"},
 353         {Opt_inode_cache, "inode_cache"},
 354         {Opt_noinode_cache, "noinode_cache"},
 355         {Opt_no_space_cache, "nospace_cache"},
 356         {Opt_recovery, "recovery"},
 357         {Opt_skip_balance, "skip_balance"},
 358         {Opt_check_integrity, "check_int"},
 359         {Opt_check_integrity_including_extent_data, "check_int_data"},
 360         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 361         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 362         {Opt_fatal_errors, "fatal_errors=%s"},
 363         {Opt_commit_interval, "commit=%d"},
 364 #ifdef CONFIG_BTRFS_DEBUG
 365         {Opt_fragment_data, "fragment=data"},
 366         {Opt_fragment_metadata, "fragment=metadata"},
 367         {Opt_fragment_all, "fragment=all"},
 368 #endif
 369         {Opt_err, NULL},
 370 };
 371
 372 /*
 373  * Regular mount options parser.  Everything that is needed only when
 374  * reading in a new superblock is parsed here.
 375  * XXX JDM: This needs to be cleaned up for remount.
 376  */
 377 int btrfs_parse_options(struct btrfs_root *root, char *options)
 378 {
 379         struct btrfs_fs_info *info = root->fs_info;
 380         substring_t args[MAX_OPT_ARGS];
 381         char *p, *num, *orig = NULL;
 382         u64 cache_gen;
 383         int intarg;
 384         int ret = 0;
 385         char *compress_type;
 386         bool compress_force = false;
 387         enum btrfs_compression_type saved_compress_type;
 388         bool saved_compress_force;
 389         int no_compress = 0;
 390
 391         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 392         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE))
 393                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 394         else if (cache_gen)
 395                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 396
 397         if (!options)
 398                 goto out;
 399
 400         /*
 401          * strsep changes the string, duplicate it because parse_options
 402          * gets called twice
 403          */
 404         options = kstrdup(options, GFP_NOFS);
 405         if (!options)
 406                 return -ENOMEM;
 407
 408         orig = options;
 409
 410         while ((p = strsep(&options, ",")) != NULL) {
 411                 int token;
 412                 if (!*p)
 413                         continue;
 414
 415                 token = match_token(p, tokens, args);
 416                 switch (token) {
 417                 case Opt_degraded:
 418                         btrfs_info(root->fs_info, "allowing degraded mounts");
 419                         btrfs_set_opt(info->mount_opt, DEGRADED);
 420                         break;
 421                 case Opt_subvol:
 422                 case Opt_subvolid:
 423                 case Opt_subvolrootid:
 424                 case Opt_device:
 425                         /*
 426                          * These are parsed by btrfs_parse_early_options
 427                          * and can be happily ignored here.
 428                          */
 429                         break;
 430                 case Opt_nodatasum:
 431                         btrfs_set_and_info(root, NODATASUM,
 432                                            "setting nodatasum");
 433                         break;
 434                 case Opt_datasum:
 435                         if (btrfs_test_opt(root, NODATASUM)) {
 436                                 if (btrfs_test_opt(root, NODATACOW))
 437                                         btrfs_info(root->fs_info, "setting datasum, datacow enabled");
 438                                 else
 439                                         btrfs_info(root->fs_info, "setting datasum");
 440                         }
 441                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 442                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 443                         break;
 444                 case Opt_nodatacow:
 445                         if (!btrfs_test_opt(root, NODATACOW)) {
 446                                 if (!btrfs_test_opt(root, COMPRESS) ||
 447                                     !btrfs_test_opt(root, FORCE_COMPRESS)) {
 448                                         btrfs_info(root->fs_info,
 449                                                    "setting nodatacow, compression disabled");
 450                                 } else {
 451                                         btrfs_info(root->fs_info, "setting nodatacow");
 452                                 }
 453                         }
 454                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 455                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 456                         btrfs_set_opt(info->mount_opt, NODATACOW);
 457                         btrfs_set_opt(info->mount_opt, NODATASUM);
 458                         break;
 459                 case Opt_datacow:
 460                         btrfs_clear_and_info(root, NODATACOW,
 461                                              "setting datacow");
 462                         break;
 463                 case Opt_compress_force:
 464                 case Opt_compress_force_type:
 465                         compress_force = true;
 466                         /* Fallthrough */
 467                 case Opt_compress:
 468                 case Opt_compress_type:
 469                         saved_compress_type = btrfs_test_opt(root, COMPRESS) ?
 470                                 info->compress_type : BTRFS_COMPRESS_NONE;
 471                         saved_compress_force =
 472                                 btrfs_test_opt(root, FORCE_COMPRESS);
 473                         if (token == Opt_compress ||
 474                             token == Opt_compress_force ||
 475                             strcmp(args[0].from, "zlib") == 0) {
 476                                 compress_type = "zlib";
 477                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 478                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 479                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 480                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 481                                 no_compress = 0;
 482                         } else if (strcmp(args[0].from, "lzo") == 0) {
 483                                 compress_type = "lzo";
 484                                 info->compress_type = BTRFS_COMPRESS_LZO;
 485                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 486                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 487                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 488                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 489                                 btrfs_sysfs_feature_update(root->fs_info,
 490                                         BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO,
 491                                         FEAT_INCOMPAT);
 492                                 no_compress = 0;
 493                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 494                                 compress_type = "no";
 495                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 496                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 497                                 compress_force = false;
 498                                 no_compress++;
 499                         } else {
 500                                 ret = -EINVAL;
 501                                 goto out;
 502                         }
 503
 504                         if (compress_force) {
 505                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 506                         } else {
 507                                 /*
 508                                  * If we remount from compress-force=xxx to
 509                                  * compress=xxx, we need clear FORCE_COMPRESS
 510                                  * flag, otherwise, there is no way for users
 511                                  * to disable forcible compression separately.
 512                                  */
 513                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 514                         }
 515                         if ((btrfs_test_opt(root, COMPRESS) &&
 516                              (info->compress_type != saved_compress_type ||
 517                               compress_force != saved_compress_force)) ||
 518                             (!btrfs_test_opt(root, COMPRESS) &&
 519                              no_compress == 1)) {
 520                                 btrfs_info(root->fs_info,
 521                                            "%s %s compression",
 522                                            (compress_force) ? "force" : "use",
 523                                            compress_type);
 524                         }
 525                         compress_force = false;
 526                         break;
 527                 case Opt_ssd:
 528                         btrfs_set_and_info(root, SSD,
 529                                            "use ssd allocation scheme");
 530                         break;
 531                 case Opt_ssd_spread:
 532                         btrfs_set_and_info(root, SSD_SPREAD,
 533                                            "use spread ssd allocation scheme");
 534                         btrfs_set_opt(info->mount_opt, SSD);
 535                         break;
 536                 case Opt_nossd:
 537                         btrfs_set_and_info(root, NOSSD,
 538                                              "not using ssd allocation scheme");
 539                         btrfs_clear_opt(info->mount_opt, SSD);
 540                         break;
 541                 case Opt_barrier:
 542                         btrfs_clear_and_info(root, NOBARRIER,
 543                                              "turning on barriers");
 544                         break;
 545                 case Opt_nobarrier:
 546                         btrfs_set_and_info(root, NOBARRIER,
 547                                            "turning off barriers");
 548                         break;
 549                 case Opt_thread_pool:
 550                         ret = match_int(&args[0], &intarg);
 551                         if (ret) {
 552                                 goto out;
 553                         } else if (intarg > 0) {
 554                                 info->thread_pool_size = intarg;
 555                         } else {
 556                                 ret = -EINVAL;
 557                                 goto out;
 558                         }
 559                         break;
 560                 case Opt_max_inline:
 561                         num = match_strdup(&args[0]);
 562                         if (num) {
 563                                 info->max_inline = memparse(num, NULL);
 564                                 kfree(num);
 565
 566                                 if (info->max_inline) {
 567                                         info->max_inline = min_t(u64,
 568                                                 info->max_inline,
 569                                                 root->sectorsize);
 570                                 }
 571                                 btrfs_info(root->fs_info, "max_inline at %llu",
 572                                         info->max_inline);
 573                         } else {
 574                                 ret = -ENOMEM;
 575                                 goto out;
 576                         }
 577                         break;
 578                 case Opt_alloc_start:
 579                         num = match_strdup(&args[0]);
 580                         if (num) {
 581                                 mutex_lock(&info->chunk_mutex);
 582                                 info->alloc_start = memparse(num, NULL);
 583                                 mutex_unlock(&info->chunk_mutex);
 584                                 kfree(num);
 585                                 btrfs_info(root->fs_info, "allocations start at %llu",
 586                                         info->alloc_start);
 587                         } else {
 588                                 ret = -ENOMEM;
 589                                 goto out;
 590                         }
 591                         break;
 592                 case Opt_acl:
 593 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 594                         root->fs_info->sb->s_flags |= MS_POSIXACL;
 595                         break;
 596 #else
 597                         btrfs_err(root->fs_info,
 598                                 "support for ACL not compiled in!");
 599                         ret = -EINVAL;
 600                         goto out;
 601 #endif
 602                 case Opt_noacl:
 603                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 604                         break;
 605                 case Opt_notreelog:
 606                         btrfs_set_and_info(root, NOTREELOG,
 607                                            "disabling tree log");
 608                         break;
 609                 case Opt_treelog:
 610                         btrfs_clear_and_info(root, NOTREELOG,
 611                                              "enabling tree log");
 612                         break;
 613                 case Opt_flushoncommit:
 614                         btrfs_set_and_info(root, FLUSHONCOMMIT,
 615                                            "turning on flush-on-commit");
 616                         break;
 617                 case Opt_noflushoncommit:
 618                         btrfs_clear_and_info(root, FLUSHONCOMMIT,
 619                                              "turning off flush-on-commit");
 620                         break;
 621                 case Opt_ratio:
 622                         ret = match_int(&args[0], &intarg);
 623                         if (ret) {
 624                                 goto out;
 625                         } else if (intarg >= 0) {
 626                                 info->metadata_ratio = intarg;
 627                                 btrfs_info(root->fs_info, "metadata ratio %d",
 628                                        info->metadata_ratio);
 629                         } else {
 630                                 ret = -EINVAL;
 631                                 goto out;
 632                         }
 633                         break;
 634                 case Opt_discard:
 635                         btrfs_set_and_info(root, DISCARD,
 636                                            "turning on discard");
 637                         break;
 638                 case Opt_nodiscard:
 639                         btrfs_clear_and_info(root, DISCARD,
 640                                              "turning off discard");
 641                         break;
 642                 case Opt_space_cache:
 643                 case Opt_space_cache_version:
 644                         if (token == Opt_space_cache ||
 645                             strcmp(args[0].from, "v1") == 0) {
 646                                 btrfs_clear_opt(root->fs_info->mount_opt,
 647                                                 FREE_SPACE_TREE);
 648                                 btrfs_set_and_info(root, SPACE_CACHE,
 649                                                    "enabling disk space caching");
 650                         } else if (strcmp(args[0].from, "v2") == 0) {
 651                                 btrfs_clear_opt(root->fs_info->mount_opt,
 652                                                 SPACE_CACHE);
 653                                 btrfs_set_and_info(root, FREE_SPACE_TREE,
 654                                                    "enabling free space tree");
 655                         } else {
 656                                 ret = -EINVAL;
 657                                 goto out;
 658                         }
 659                         break;
 660                 case Opt_rescan_uuid_tree:
 661                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 662                         break;
 663                 case Opt_no_space_cache:
 664                         if (btrfs_test_opt(root, SPACE_CACHE)) {
 665                                 btrfs_clear_and_info(root, SPACE_CACHE,
 666                                                      "disabling disk space caching");
 667                         }
 668                         if (btrfs_test_opt(root, FREE_SPACE_TREE)) {
 669                                 btrfs_clear_and_info(root, FREE_SPACE_TREE,
 670                                                      "disabling free space tree");
 671                         }
 672                         break;
 673                 case Opt_inode_cache:
 674                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 675                                            "enabling inode map caching");
 676                         break;
 677                 case Opt_noinode_cache:
 678                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 679                                              "disabling inode map caching");
 680                         break;
 681                 case Opt_clear_cache:
 682                         btrfs_set_and_info(root, CLEAR_CACHE,
 683                                            "force clearing of disk cache");
 684                         break;
 685                 case Opt_user_subvol_rm_allowed:
 686                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 687                         break;
 688                 case Opt_enospc_debug:
 689                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 690                         break;
 691                 case Opt_noenospc_debug:
 692                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 693                         break;
 694                 case Opt_defrag:
 695                         btrfs_set_and_info(root, AUTO_DEFRAG,
 696                                            "enabling auto defrag");
 697                         break;
 698                 case Opt_nodefrag:
 699                         btrfs_clear_and_info(root, AUTO_DEFRAG,
 700                                              "disabling auto defrag");
 701                         break;
 702                 case Opt_recovery:
 703                         btrfs_info(root->fs_info, "enabling auto recovery");
 704                         btrfs_set_opt(info->mount_opt, RECOVERY);
 705                         break;
 706                 case Opt_skip_balance:
 707                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 708                         break;
 709 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 710                 case Opt_check_integrity_including_extent_data:
 711                         btrfs_info(root->fs_info,
 712                                    "enabling check integrity including extent data");
 713                         btrfs_set_opt(info->mount_opt,
 714                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 715                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 716                         break;
 717                 case Opt_check_integrity:
 718                         btrfs_info(root->fs_info, "enabling check integrity");
 719                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 720                         break;
 721                 case Opt_check_integrity_print_mask:
 722                         ret = match_int(&args[0], &intarg);
 723                         if (ret) {
 724                                 goto out;
 725                         } else if (intarg >= 0) {
 726                                 info->check_integrity_print_mask = intarg;
 727                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
 728                                        info->check_integrity_print_mask);
 729                         } else {
 730                                 ret = -EINVAL;
 731                                 goto out;
 732                         }
 733                         break;
 734 #else
 735                 case Opt_check_integrity_including_extent_data:
 736                 case Opt_check_integrity:
 737                 case Opt_check_integrity_print_mask:
 738                         btrfs_err(root->fs_info,
 739                                 "support for check_integrity* not compiled in!");
 740                         ret = -EINVAL;
 741                         goto out;
 742 #endif
 743                 case Opt_fatal_errors:
 744                         if (strcmp(args[0].from, "panic") == 0)
 745                                 btrfs_set_opt(info->mount_opt,
 746                                               PANIC_ON_FATAL_ERROR);
 747                         else if (strcmp(args[0].from, "bug") == 0)
 748                                 btrfs_clear_opt(info->mount_opt,
 749                                               PANIC_ON_FATAL_ERROR);
 750                         else {
 751                                 ret = -EINVAL;
 752                                 goto out;
 753                         }
 754                         break;
 755                 case Opt_commit_interval:
 756                         intarg = 0;
 757                         ret = match_int(&args[0], &intarg);
 758                         if (ret < 0) {
 759                                 btrfs_err(root->fs_info, "invalid commit interval");
 760                                 ret = -EINVAL;
 761                                 goto out;
 762                         }
 763                         if (intarg > 0) {
 764                                 if (intarg > 300) {
 765                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
 766                                                         intarg);
 767                                 }
 768                                 info->commit_interval = intarg;
 769                         } else {
 770                                 btrfs_info(root->fs_info, "using default commit interval %ds",
 771                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
 772                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 773                         }
 774                         break;
 775 #ifdef CONFIG_BTRFS_DEBUG
 776                 case Opt_fragment_all:
 777                         btrfs_info(root->fs_info, "fragmenting all space");
 778                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 779                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 780                         break;
 781                 case Opt_fragment_metadata:
 782                         btrfs_info(root->fs_info, "fragmenting metadata");
 783                         btrfs_set_opt(info->mount_opt,
 784                                       FRAGMENT_METADATA);
 785                         break;
 786                 case Opt_fragment_data:
 787                         btrfs_info(root->fs_info, "fragmenting data");
 788                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 789                         break;
 790 #endif
 791                 case Opt_err:
 792                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
 793                         ret = -EINVAL;
 794                         goto out;
 795                 default:
 796                         break;
 797                 }
 798         }
 799 out:
 800         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE) &&
 801             !btrfs_test_opt(root, FREE_SPACE_TREE) &&
 802             !btrfs_test_opt(root, CLEAR_CACHE)) {
 803                 btrfs_err(root->fs_info, "cannot disable free space tree");
 804                 ret = -EINVAL;
 805
 806         }
 807         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 808                 btrfs_info(root->fs_info, "disk space caching is enabled");
 809         if (!ret && btrfs_test_opt(root, FREE_SPACE_TREE))
 810                 btrfs_info(root->fs_info, "using free space tree");
 811         kfree(orig);
 812         return ret;
 813 }
 814
 815 /*
 816  * Parse mount options that are required early in the mount process.
 817  *
 818  * All other options will be parsed on much later in the mount process and
 819  * only when we need to allocate a new super block.
 820  */
 821 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 822                 void *holder, char **subvol_name, u64 *subvol_objectid,
 823                 struct btrfs_fs_devices **fs_devices)
 824 {
 825         substring_t args[MAX_OPT_ARGS];
 826         char *device_name, *opts, *orig, *p;
 827         char *num = NULL;
 828         int error = 0;
 829
 830         if (!options)
 831                 return 0;
 832
 833         /*
 834          * strsep changes the string, duplicate it because parse_options
 835          * gets called twice
 836          */
 837         opts = kstrdup(options, GFP_KERNEL);
 838         if (!opts)
 839                 return -ENOMEM;
 840         orig = opts;
 841
 842         while ((p = strsep(&opts, ",")) != NULL) {
 843                 int token;
 844                 if (!*p)
 845                         continue;
 846
 847                 token = match_token(p, tokens, args);
 848                 switch (token) {
 849                 case Opt_subvol:
 850                         kfree(*subvol_name);
 851                         *subvol_name = match_strdup(&args[0]);
 852                         if (!*subvol_name) {
 853                                 error = -ENOMEM;
 854                                 goto out;
 855                         }
 856                         break;
 857                 case Opt_subvolid:
 858                         num = match_strdup(&args[0]);
 859                         if (num) {
 860                                 *subvol_objectid = memparse(num, NULL);
 861                                 kfree(num);
 862                                 /* we want the original fs_tree */
 863                                 if (!*subvol_objectid)
 864                                         *subvol_objectid =
 865                                                 BTRFS_FS_TREE_OBJECTID;
 866                         } else {
 867                                 error = -EINVAL;
 868                                 goto out;
 869                         }
 870                         break;
 871                 case Opt_subvolrootid:
 872                         printk(KERN_WARNING
 873                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
 874                                 "no effect\n");
 875                         break;
 876                 case Opt_device:
 877                         device_name = match_strdup(&args[0]);
 878                         if (!device_name) {
 879                                 error = -ENOMEM;
 880                                 goto out;
 881                         }
 882                         error = btrfs_scan_one_device(device_name,
 883                                         flags, holder, fs_devices);
 884                         kfree(device_name);
 885                         if (error)
 886                                 goto out;
 887                         break;
 888                 default:
 889                         break;
 890                 }
 891         }
 892
 893 out:
 894         kfree(orig);
 895         return error;
 896 }
 897
 898 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 899                                            u64 subvol_objectid)
 900 {
 901         struct btrfs_root *root = fs_info->tree_root;
 902         struct btrfs_root *fs_root;
 903         struct btrfs_root_ref *root_ref;
 904         struct btrfs_inode_ref *inode_ref;
 905         struct btrfs_key key;
 906         struct btrfs_path *path = NULL;
 907         char *name = NULL, *ptr;
 908         u64 dirid;
 909         int len;
 910         int ret;
 911
 912         path = btrfs_alloc_path();
 913         if (!path) {
 914                 ret = -ENOMEM;
 915                 goto err;
 916         }
 917         path->leave_spinning = 1;
 918
 919         name = kmalloc(PATH_MAX, GFP_NOFS);
 920         if (!name) {
 921                 ret = -ENOMEM;
 922                 goto err;
 923         }
 924         ptr = name + PATH_MAX - 1;
 925         ptr[0] = '\0';
 926
 927         /*
 928          * Walk up the subvolume trees in the tree of tree roots by root
 929          * backrefs until we hit the top-level subvolume.
 930          */
 931         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 932                 key.objectid = subvol_objectid;
 933                 key.type = BTRFS_ROOT_BACKREF_KEY;
 934                 key.offset = (u64)-1;
 935
 936                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 937                 if (ret < 0) {
 938                         goto err;
 939                 } else if (ret > 0) {
 940                         ret = btrfs_previous_item(root, path, subvol_objectid,
 941                                                   BTRFS_ROOT_BACKREF_KEY);
 942                         if (ret < 0) {
 943                                 goto err;
 944                         } else if (ret > 0) {
 945                                 ret = -ENOENT;
 946                                 goto err;
 947                         }
 948                 }
 949
 950                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 951                 subvol_objectid = key.offset;
 952
 953                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 954                                           struct btrfs_root_ref);
 955                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 956                 ptr -= len + 1;
 957                 if (ptr < name) {
 958                         ret = -ENAMETOOLONG;
 959                         goto err;
 960                 }
 961                 read_extent_buffer(path->nodes[0], ptr + 1,
 962                                    (unsigned long)(root_ref + 1), len);
 963                 ptr[0] = '/';
 964                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
 965                 btrfs_release_path(path);
 966
 967                 key.objectid = subvol_objectid;
 968                 key.type = BTRFS_ROOT_ITEM_KEY;
 969                 key.offset = (u64)-1;
 970                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
 971                 if (IS_ERR(fs_root)) {
 972                         ret = PTR_ERR(fs_root);
 973                         goto err;
 974                 }
 975
 976                 /*
 977                  * Walk up the filesystem tree by inode refs until we hit the
 978                  * root directory.
 979                  */
 980                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
 981                         key.objectid = dirid;
 982                         key.type = BTRFS_INODE_REF_KEY;
 983                         key.offset = (u64)-1;
 984
 985                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
 986                         if (ret < 0) {
 987                                 goto err;
 988                         } else if (ret > 0) {
 989                                 ret = btrfs_previous_item(fs_root, path, dirid,
 990                                                           BTRFS_INODE_REF_KEY);
 991                                 if (ret < 0) {
 992                                         goto err;
 993                                 } else if (ret > 0) {
 994                                         ret = -ENOENT;
 995                                         goto err;
 996                                 }
 997                         }
 998
 999                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1000                         dirid = key.offset;
1001
1002                         inode_ref = btrfs_item_ptr(path->nodes[0],
1003                                                    path->slots[0],
1004                                                    struct btrfs_inode_ref);
1005                         len = btrfs_inode_ref_name_len(path->nodes[0],
1006                                                        inode_ref);
1007                         ptr -= len + 1;
1008                         if (ptr < name) {
1009                                 ret = -ENAMETOOLONG;
1010                                 goto err;
1011                         }
1012                         read_extent_buffer(path->nodes[0], ptr + 1,
1013                                            (unsigned long)(inode_ref + 1), len);
1014                         ptr[0] = '/';
1015                         btrfs_release_path(path);
1016                 }
1017         }
1018
1019         btrfs_free_path(path);
1020         if (ptr == name + PATH_MAX - 1) {
1021                 name[0] = '/';
1022                 name[1] = '\0';
1023         } else {
1024                 memmove(name, ptr, name + PATH_MAX - ptr);
1025         }
1026         return name;
1027
1028 err:
1029         btrfs_free_path(path);
1030         kfree(name);
1031         return ERR_PTR(ret);
1032 }
1033
1034 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1035 {
1036         struct btrfs_root *root = fs_info->tree_root;
1037         struct btrfs_dir_item *di;
1038         struct btrfs_path *path;
1039         struct btrfs_key location;
1040         u64 dir_id;
1041
1042         path = btrfs_alloc_path();
1043         if (!path)
1044                 return -ENOMEM;
1045         path->leave_spinning = 1;
1046
1047         /*
1048          * Find the "default" dir item which points to the root item that we
1049          * will mount by default if we haven't been given a specific subvolume
1050          * to mount.
1051          */
1052         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1053         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1054         if (IS_ERR(di)) {
1055                 btrfs_free_path(path);
1056                 return PTR_ERR(di);
1057         }
1058         if (!di) {
1059                 /*
1060                  * Ok the default dir item isn't there.  This is weird since
1061                  * it's always been there, but don't freak out, just try and
1062                  * mount the top-level subvolume.
1063                  */
1064                 btrfs_free_path(path);
1065                 *objectid = BTRFS_FS_TREE_OBJECTID;
1066                 return 0;
1067         }
1068
1069         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1070         btrfs_free_path(path);
1071         *objectid = location.objectid;
1072         return 0;
1073 }
1074
1075 static int btrfs_fill_super(struct super_block *sb,
1076                             struct btrfs_fs_devices *fs_devices,
1077                             void *data, int silent)
1078 {
1079         struct inode *inode;
1080         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1081         struct btrfs_key key;
1082         int err;
1083
1084         sb->s_maxbytes = MAX_LFS_FILESIZE;
1085         sb->s_magic = BTRFS_SUPER_MAGIC;
1086         sb->s_op = &btrfs_super_ops;
1087         sb->s_d_op = &btrfs_dentry_operations;
1088         sb->s_export_op = &btrfs_export_ops;
1089         sb->s_xattr = btrfs_xattr_handlers;
1090         sb->s_time_gran = 1;
1091 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1092         sb->s_flags |= MS_POSIXACL;
1093 #endif
1094         sb->s_flags |= MS_I_VERSION;
1095         sb->s_iflags |= SB_I_CGROUPWB;
1096         err = open_ctree(sb, fs_devices, (char *)data);
1097         if (err) {
1098                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
1099                 return err;
1100         }
1101
1102         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1103         key.type = BTRFS_INODE_ITEM_KEY;
1104         key.offset = 0;
1105         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1106         if (IS_ERR(inode)) {
1107                 err = PTR_ERR(inode);
1108                 goto fail_close;
1109         }
1110
1111         sb->s_root = d_make_root(inode);
1112         if (!sb->s_root) {
1113                 err = -ENOMEM;
1114                 goto fail_close;
1115         }
1116
1117         save_mount_options(sb, data);
1118         cleancache_init_fs(sb);
1119         sb->s_flags |= MS_ACTIVE;
1120         return 0;
1121
1122 fail_close:
1123         close_ctree(fs_info->tree_root);
1124         return err;
1125 }
1126
1127 int btrfs_sync_fs(struct super_block *sb, int wait)
1128 {
1129         struct btrfs_trans_handle *trans;
1130         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1131         struct btrfs_root *root = fs_info->tree_root;
1132
1133         trace_btrfs_sync_fs(wait);
1134
1135         if (!wait) {
1136                 filemap_flush(fs_info->btree_inode->i_mapping);
1137                 return 0;
1138         }
1139
1140         btrfs_wait_ordered_roots(fs_info, -1);
1141
1142         trans = btrfs_attach_transaction_barrier(root);
1143         if (IS_ERR(trans)) {
1144                 /* no transaction, don't bother */
1145                 if (PTR_ERR(trans) == -ENOENT) {
1146                         /*
1147                          * Exit unless we have some pending changes
1148                          * that need to go through commit
1149                          */
1150                         if (fs_info->pending_changes == 0)
1151                                 return 0;
1152                         /*
1153                          * A non-blocking test if the fs is frozen. We must not
1154                          * start a new transaction here otherwise a deadlock
1155                          * happens. The pending operations are delayed to the
1156                          * next commit after thawing.
1157                          */
1158                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1159                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1160                         else
1161                                 return 0;
1162                         trans = btrfs_start_transaction(root, 0);
1163                 }
1164                 if (IS_ERR(trans))
1165                         return PTR_ERR(trans);
1166         }
1167         return btrfs_commit_transaction(trans, root);
1168 }
1169
1170 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1171 {
1172         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1173         struct btrfs_root *root = info->tree_root;
1174         char *compress_type;
1175
1176         if (btrfs_test_opt(root, DEGRADED))
1177                 seq_puts(seq, ",degraded");
1178         if (btrfs_test_opt(root, NODATASUM))
1179                 seq_puts(seq, ",nodatasum");
1180         if (btrfs_test_opt(root, NODATACOW))
1181                 seq_puts(seq, ",nodatacow");
1182         if (btrfs_test_opt(root, NOBARRIER))
1183                 seq_puts(seq, ",nobarrier");
1184         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1185                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1186         if (info->alloc_start != 0)
1187                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1188         if (info->thread_pool_size !=  min_t(unsigned long,
1189                                              num_online_cpus() + 2, 8))
1190                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1191         if (btrfs_test_opt(root, COMPRESS)) {
1192                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1193                         compress_type = "zlib";
1194                 else
1195                         compress_type = "lzo";
1196                 if (btrfs_test_opt(root, FORCE_COMPRESS))
1197                         seq_printf(seq, ",compress-force=%s", compress_type);
1198                 else
1199                         seq_printf(seq, ",compress=%s", compress_type);
1200         }
1201         if (btrfs_test_opt(root, NOSSD))
1202                 seq_puts(seq, ",nossd");
1203         if (btrfs_test_opt(root, SSD_SPREAD))
1204                 seq_puts(seq, ",ssd_spread");
1205         else if (btrfs_test_opt(root, SSD))
1206                 seq_puts(seq, ",ssd");
1207         if (btrfs_test_opt(root, NOTREELOG))
1208                 seq_puts(seq, ",notreelog");
1209         if (btrfs_test_opt(root, FLUSHONCOMMIT))
1210                 seq_puts(seq, ",flushoncommit");
1211         if (btrfs_test_opt(root, DISCARD))
1212                 seq_puts(seq, ",discard");
1213         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1214                 seq_puts(seq, ",noacl");
1215         if (btrfs_test_opt(root, SPACE_CACHE))
1216                 seq_puts(seq, ",space_cache");
1217         else if (btrfs_test_opt(root, FREE_SPACE_TREE))
1218                 seq_puts(seq, ",space_cache=v2");
1219         else
1220                 seq_puts(seq, ",nospace_cache");
1221         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1222                 seq_puts(seq, ",rescan_uuid_tree");
1223         if (btrfs_test_opt(root, CLEAR_CACHE))
1224                 seq_puts(seq, ",clear_cache");
1225         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1226                 seq_puts(seq, ",user_subvol_rm_allowed");
1227         if (btrfs_test_opt(root, ENOSPC_DEBUG))
1228                 seq_puts(seq, ",enospc_debug");
1229         if (btrfs_test_opt(root, AUTO_DEFRAG))
1230                 seq_puts(seq, ",autodefrag");
1231         if (btrfs_test_opt(root, INODE_MAP_CACHE))
1232                 seq_puts(seq, ",inode_cache");
1233         if (btrfs_test_opt(root, SKIP_BALANCE))
1234                 seq_puts(seq, ",skip_balance");
1235         if (btrfs_test_opt(root, RECOVERY))
1236                 seq_puts(seq, ",recovery");
1237 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1238         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1239                 seq_puts(seq, ",check_int_data");
1240         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1241                 seq_puts(seq, ",check_int");
1242         if (info->check_integrity_print_mask)
1243                 seq_printf(seq, ",check_int_print_mask=%d",
1244                                 info->check_integrity_print_mask);
1245 #endif
1246         if (info->metadata_ratio)
1247                 seq_printf(seq, ",metadata_ratio=%d",
1248                                 info->metadata_ratio);
1249         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1250                 seq_puts(seq, ",fatal_errors=panic");
1251         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1252                 seq_printf(seq, ",commit=%d", info->commit_interval);
1253 #ifdef CONFIG_BTRFS_DEBUG
1254         if (btrfs_test_opt(root, FRAGMENT_DATA))
1255                 seq_puts(seq, ",fragment=data");
1256         if (btrfs_test_opt(root, FRAGMENT_METADATA))
1257                 seq_puts(seq, ",fragment=metadata");
1258 #endif
1259         seq_printf(seq, ",subvolid=%llu",
1260                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1261         seq_puts(seq, ",subvol=");
1262         seq_dentry(seq, dentry, " \t\n\\");
1263         return 0;
1264 }
1265
1266 static int btrfs_test_super(struct super_block *s, void *data)
1267 {
1268         struct btrfs_fs_info *p = data;
1269         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1270
1271         return fs_info->fs_devices == p->fs_devices;
1272 }
1273
1274 static int btrfs_set_super(struct super_block *s, void *data)
1275 {
1276         int err = set_anon_super(s, data);
1277         if (!err)
1278                 s->s_fs_info = data;
1279         return err;
1280 }
1281
1282 /*
1283  * subvolumes are identified by ino 256
1284  */
1285 static inline int is_subvolume_inode(struct inode *inode)
1286 {
1287         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1288                 return 1;
1289         return 0;
1290 }
1291
1292 /*
1293  * This will add subvolid=0 to the argument string while removing any subvol=
1294  * and subvolid= arguments to make sure we get the top-level root for path
1295  * walking to the subvol we want.
1296  */
1297 static char *setup_root_args(char *args)
1298 {
1299         char *buf, *dst, *sep;
1300
1301         if (!args)
1302                 return kstrdup("subvolid=0", GFP_NOFS);
1303
1304         /* The worst case is that we add ",subvolid=0" to the end. */
1305         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1306         if (!buf)
1307                 return NULL;
1308
1309         while (1) {
1310                 sep = strchrnul(args, ',');
1311                 if (!strstarts(args, "subvol=") &&
1312                     !strstarts(args, "subvolid=")) {
1313                         memcpy(dst, args, sep - args);
1314                         dst += sep - args;
1315                         *dst++ = ',';
1316                 }
1317                 if (*sep)
1318                         args = sep + 1;
1319                 else
1320                         break;
1321         }
1322         strcpy(dst, "subvolid=0");
1323
1324         return buf;
1325 }
1326
1327 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1328                                    int flags, const char *device_name,
1329                                    char *data)
1330 {
1331         struct dentry *root;
1332         struct vfsmount *mnt = NULL;
1333         char *newargs;
1334         int ret;
1335
1336         newargs = setup_root_args(data);
1337         if (!newargs) {
1338                 root = ERR_PTR(-ENOMEM);
1339                 goto out;
1340         }
1341
1342         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1343         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1344                 if (flags & MS_RDONLY) {
1345                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1346                                              device_name, newargs);
1347                 } else {
1348                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1349                                              device_name, newargs);
1350                         if (IS_ERR(mnt)) {
1351                                 root = ERR_CAST(mnt);
1352                                 mnt = NULL;
1353                                 goto out;
1354                         }
1355
1356                         down_write(&mnt->mnt_sb->s_umount);
1357                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1358                         up_write(&mnt->mnt_sb->s_umount);
1359                         if (ret < 0) {
1360                                 root = ERR_PTR(ret);
1361                                 goto out;
1362                         }
1363                 }
1364         }
1365         if (IS_ERR(mnt)) {
1366                 root = ERR_CAST(mnt);
1367                 mnt = NULL;
1368                 goto out;
1369         }
1370
1371         if (!subvol_name) {
1372                 if (!subvol_objectid) {
1373                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1374                                                           &subvol_objectid);
1375                         if (ret) {
1376                                 root = ERR_PTR(ret);
1377                                 goto out;
1378                         }
1379                 }
1380                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1381                                                             subvol_objectid);
1382                 if (IS_ERR(subvol_name)) {
1383                         root = ERR_CAST(subvol_name);
1384                         subvol_name = NULL;
1385                         goto out;
1386                 }
1387
1388         }
1389
1390         root = mount_subtree(mnt, subvol_name);
1391         /* mount_subtree() drops our reference on the vfsmount. */
1392         mnt = NULL;
1393
1394         if (!IS_ERR(root)) {
1395                 struct super_block *s = root->d_sb;
1396                 struct inode *root_inode = d_inode(root);
1397                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1398
1399                 ret = 0;
1400                 if (!is_subvolume_inode(root_inode)) {
1401                         pr_err("BTRFS: '%s' is not a valid subvolume\n",
1402                                subvol_name);
1403                         ret = -EINVAL;
1404                 }
1405                 if (subvol_objectid && root_objectid != subvol_objectid) {
1406                         /*
1407                          * This will also catch a race condition where a
1408                          * subvolume which was passed by ID is renamed and
1409                          * another subvolume is renamed over the old location.
1410                          */
1411                         pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1412                                subvol_name, subvol_objectid);
1413                         ret = -EINVAL;
1414                 }
1415                 if (ret) {
1416                         dput(root);
1417                         root = ERR_PTR(ret);
1418                         deactivate_locked_super(s);
1419                 }
1420         }
1421
1422 out:
1423         mntput(mnt);
1424         kfree(newargs);
1425         kfree(subvol_name);
1426         return root;
1427 }
1428
1429 static int parse_security_options(char *orig_opts,
1430                                   struct security_mnt_opts *sec_opts)
1431 {
1432         char *secdata = NULL;
1433         int ret = 0;
1434
1435         secdata = alloc_secdata();
1436         if (!secdata)
1437                 return -ENOMEM;
1438         ret = security_sb_copy_data(orig_opts, secdata);
1439         if (ret) {
1440                 free_secdata(secdata);
1441                 return ret;
1442         }
1443         ret = security_sb_parse_opts_str(secdata, sec_opts);
1444         free_secdata(secdata);
1445         return ret;
1446 }
1447
1448 static int setup_security_options(struct btrfs_fs_info *fs_info,
1449                                   struct super_block *sb,
1450                                   struct security_mnt_opts *sec_opts)
1451 {
1452         int ret = 0;
1453
1454         /*
1455          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1456          * is valid.
1457          */
1458         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1459         if (ret)
1460                 return ret;
1461
1462 #ifdef CONFIG_SECURITY
1463         if (!fs_info->security_opts.num_mnt_opts) {
1464                 /* first time security setup, copy sec_opts to fs_info */
1465                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1466         } else {
1467                 /*
1468                  * Since SELinux(the only one supports security_mnt_opts) does
1469                  * NOT support changing context during remount/mount same sb,
1470                  * This must be the same or part of the same security options,
1471                  * just free it.
1472                  */
1473                 security_free_mnt_opts(sec_opts);
1474         }
1475 #endif
1476         return ret;
1477 }
1478
1479 /*
1480  * Find a superblock for the given device / mount point.
1481  *
1482  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1483  *        for multiple device setup.  Make sure to keep it in sync.
1484  */
1485 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1486                 const char *device_name, void *data)
1487 {
1488         struct block_device *bdev = NULL;
1489         struct super_block *s;
1490         struct btrfs_fs_devices *fs_devices = NULL;
1491         struct btrfs_fs_info *fs_info = NULL;
1492         struct security_mnt_opts new_sec_opts;
1493         fmode_t mode = FMODE_READ;
1494         char *subvol_name = NULL;
1495         u64 subvol_objectid = 0;
1496         int error = 0;
1497
1498         if (!(flags & MS_RDONLY))
1499                 mode |= FMODE_WRITE;
1500
1501         error = btrfs_parse_early_options(data, mode, fs_type,
1502                                           &subvol_name, &subvol_objectid,
1503                                           &fs_devices);
1504         if (error) {
1505                 kfree(subvol_name);
1506                 return ERR_PTR(error);
1507         }
1508
1509         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1510                 /* mount_subvol() will free subvol_name. */
1511                 return mount_subvol(subvol_name, subvol_objectid, flags,
1512                                     device_name, data);
1513         }
1514
1515         security_init_mnt_opts(&new_sec_opts);
1516         if (data) {
1517                 error = parse_security_options(data, &new_sec_opts);
1518                 if (error)
1519                         return ERR_PTR(error);
1520         }
1521
1522         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1523         if (error)
1524                 goto error_sec_opts;
1525
1526         /*
1527          * Setup a dummy root and fs_info for test/set super.  This is because
1528          * we don't actually fill this stuff out until open_ctree, but we need
1529          * it for searching for existing supers, so this lets us do that and
1530          * then open_ctree will properly initialize everything later.
1531          */
1532         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1533         if (!fs_info) {
1534                 error = -ENOMEM;
1535                 goto error_sec_opts;
1536         }
1537
1538         fs_info->fs_devices = fs_devices;
1539
1540         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1541         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1542         security_init_mnt_opts(&fs_info->security_opts);
1543         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1544                 error = -ENOMEM;
1545                 goto error_fs_info;
1546         }
1547
1548         error = btrfs_open_devices(fs_devices, mode, fs_type);
1549         if (error)
1550                 goto error_fs_info;
1551
1552         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1553                 error = -EACCES;
1554                 goto error_close_devices;
1555         }
1556
1557         bdev = fs_devices->latest_bdev;
1558         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1559                  fs_info);
1560         if (IS_ERR(s)) {
1561                 error = PTR_ERR(s);
1562                 goto error_close_devices;
1563         }
1564
1565         if (s->s_root) {
1566                 btrfs_close_devices(fs_devices);
1567                 free_fs_info(fs_info);
1568                 if ((flags ^ s->s_flags) & MS_RDONLY)
1569                         error = -EBUSY;
1570         } else {
1571                 char b[BDEVNAME_SIZE];
1572
1573                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1574                 btrfs_sb(s)->bdev_holder = fs_type;
1575                 error = btrfs_fill_super(s, fs_devices, data,
1576                                          flags & MS_SILENT ? 1 : 0);
1577         }
1578         if (error) {
1579                 deactivate_locked_super(s);
1580                 goto error_sec_opts;
1581         }
1582
1583         fs_info = btrfs_sb(s);
1584         error = setup_security_options(fs_info, s, &new_sec_opts);
1585         if (error) {
1586                 deactivate_locked_super(s);
1587                 goto error_sec_opts;
1588         }
1589
1590         return dget(s->s_root);
1591
1592 error_close_devices:
1593         btrfs_close_devices(fs_devices);
1594 error_fs_info:
1595         free_fs_info(fs_info);
1596 error_sec_opts:
1597         security_free_mnt_opts(&new_sec_opts);
1598         return ERR_PTR(error);
1599 }
1600
1601 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1602                                      int new_pool_size, int old_pool_size)
1603 {
1604         if (new_pool_size == old_pool_size)
1605                 return;
1606
1607         fs_info->thread_pool_size = new_pool_size;
1608
1609         btrfs_info(fs_info, "resize thread pool %d -> %d",
1610                old_pool_size, new_pool_size);
1611
1612         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1613         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1614         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1615         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1616         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1617         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1618         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1619                                 new_pool_size);
1620         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1621         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1622         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1623         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1624         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1625                                 new_pool_size);
1626 }
1627
1628 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1629 {
1630         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1631 }
1632
1633 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1634                                        unsigned long old_opts, int flags)
1635 {
1636         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1637             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1638              (flags & MS_RDONLY))) {
1639                 /* wait for any defraggers to finish */
1640                 wait_event(fs_info->transaction_wait,
1641                            (atomic_read(&fs_info->defrag_running) == 0));
1642                 if (flags & MS_RDONLY)
1643                         sync_filesystem(fs_info->sb);
1644         }
1645 }
1646
1647 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1648                                          unsigned long old_opts)
1649 {
1650         /*
1651          * We need cleanup all defragable inodes if the autodefragment is
1652          * close or the fs is R/O.
1653          */
1654         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1655             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1656              (fs_info->sb->s_flags & MS_RDONLY))) {
1657                 btrfs_cleanup_defrag_inodes(fs_info);
1658         }
1659
1660         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1661 }
1662
1663 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1664 {
1665         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1666         struct btrfs_root *root = fs_info->tree_root;
1667         unsigned old_flags = sb->s_flags;
1668         unsigned long old_opts = fs_info->mount_opt;
1669         unsigned long old_compress_type = fs_info->compress_type;
1670         u64 old_max_inline = fs_info->max_inline;
1671         u64 old_alloc_start = fs_info->alloc_start;
1672         int old_thread_pool_size = fs_info->thread_pool_size;
1673         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1674         int ret;
1675
1676         sync_filesystem(sb);
1677         btrfs_remount_prepare(fs_info);
1678
1679         if (data) {
1680                 struct security_mnt_opts new_sec_opts;
1681
1682                 security_init_mnt_opts(&new_sec_opts);
1683                 ret = parse_security_options(data, &new_sec_opts);
1684                 if (ret)
1685                         goto restore;
1686                 ret = setup_security_options(fs_info, sb,
1687                                              &new_sec_opts);
1688                 if (ret) {
1689                         security_free_mnt_opts(&new_sec_opts);
1690                         goto restore;
1691                 }
1692         }
1693
1694         ret = btrfs_parse_options(root, data);
1695         if (ret) {
1696                 ret = -EINVAL;
1697                 goto restore;
1698         }
1699
1700         btrfs_remount_begin(fs_info, old_opts, *flags);
1701         btrfs_resize_thread_pool(fs_info,
1702                 fs_info->thread_pool_size, old_thread_pool_size);
1703
1704         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1705                 goto out;
1706
1707         if (*flags & MS_RDONLY) {
1708                 /*
1709                  * this also happens on 'umount -rf' or on shutdown, when
1710                  * the filesystem is busy.
1711                  */
1712                 cancel_work_sync(&fs_info->async_reclaim_work);
1713
1714                 /* wait for the uuid_scan task to finish */
1715                 down(&fs_info->uuid_tree_rescan_sem);
1716                 /* avoid complains from lockdep et al. */
1717                 up(&fs_info->uuid_tree_rescan_sem);
1718
1719                 sb->s_flags |= MS_RDONLY;
1720
1721                 /*
1722                  * Setting MS_RDONLY will put the cleaner thread to
1723                  * sleep at the next loop if it's already active.
1724                  * If it's already asleep, we'll leave unused block
1725                  * groups on disk until we're mounted read-write again
1726                  * unless we clean them up here.
1727                  */
1728                 btrfs_delete_unused_bgs(fs_info);
1729
1730                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1731                 btrfs_scrub_cancel(fs_info);
1732                 btrfs_pause_balance(fs_info);
1733
1734                 ret = btrfs_commit_super(root);
1735                 if (ret)
1736                         goto restore;
1737         } else {
1738                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1739                         btrfs_err(fs_info,
1740                                 "Remounting read-write after error is not allowed");
1741                         ret = -EINVAL;
1742                         goto restore;
1743                 }
1744                 if (fs_info->fs_devices->rw_devices == 0) {
1745                         ret = -EACCES;
1746                         goto restore;
1747                 }
1748
1749                 if (fs_info->fs_devices->missing_devices >
1750                      fs_info->num_tolerated_disk_barrier_failures &&
1751                     !(*flags & MS_RDONLY)) {
1752                         btrfs_warn(fs_info,
1753                                 "too many missing devices, writeable remount is not allowed");
1754                         ret = -EACCES;
1755                         goto restore;
1756                 }
1757
1758                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1759                         ret = -EINVAL;
1760                         goto restore;
1761                 }
1762
1763                 ret = btrfs_cleanup_fs_roots(fs_info);
1764                 if (ret)
1765                         goto restore;
1766
1767                 /* recover relocation */
1768                 mutex_lock(&fs_info->cleaner_mutex);
1769                 ret = btrfs_recover_relocation(root);
1770                 mutex_unlock(&fs_info->cleaner_mutex);
1771                 if (ret)
1772                         goto restore;
1773
1774                 ret = btrfs_resume_balance_async(fs_info);
1775                 if (ret)
1776                         goto restore;
1777
1778                 ret = btrfs_resume_dev_replace_async(fs_info);
1779                 if (ret) {
1780                         btrfs_warn(fs_info, "failed to resume dev_replace");
1781                         goto restore;
1782                 }
1783
1784                 if (!fs_info->uuid_root) {
1785                         btrfs_info(fs_info, "creating UUID tree");
1786                         ret = btrfs_create_uuid_tree(fs_info);
1787                         if (ret) {
1788                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1789                                 goto restore;
1790                         }
1791                 }
1792                 sb->s_flags &= ~MS_RDONLY;
1793         }
1794 out:
1795         wake_up_process(fs_info->transaction_kthread);
1796         btrfs_remount_cleanup(fs_info, old_opts);
1797         return 0;
1798
1799 restore:
1800         /* We've hit an error - don't reset MS_RDONLY */
1801         if (sb->s_flags & MS_RDONLY)
1802                 old_flags |= MS_RDONLY;
1803         sb->s_flags = old_flags;
1804         fs_info->mount_opt = old_opts;
1805         fs_info->compress_type = old_compress_type;
1806         fs_info->max_inline = old_max_inline;
1807         mutex_lock(&fs_info->chunk_mutex);
1808         fs_info->alloc_start = old_alloc_start;
1809         mutex_unlock(&fs_info->chunk_mutex);
1810         btrfs_resize_thread_pool(fs_info,
1811                 old_thread_pool_size, fs_info->thread_pool_size);
1812         fs_info->metadata_ratio = old_metadata_ratio;
1813         btrfs_remount_cleanup(fs_info, old_opts);
1814         return ret;
1815 }
1816
1817 /* Used to sort the devices by max_avail(descending sort) */
1818 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1819                                        const void *dev_info2)
1820 {
1821         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1822             ((struct btrfs_device_info *)dev_info2)->max_avail)
1823                 return -1;
1824         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1825                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1826                 return 1;
1827         else
1828         return 0;
1829 }
1830
1831 /*
1832  * sort the devices by max_avail, in which max free extent size of each device
1833  * is stored.(Descending Sort)
1834  */
1835 static inline void btrfs_descending_sort_devices(
1836                                         struct btrfs_device_info *devices,
1837                                         size_t nr_devices)
1838 {
1839         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1840              btrfs_cmp_device_free_bytes, NULL);
1841 }
1842
1843 /*
1844  * The helper to calc the free space on the devices that can be used to store
1845  * file data.
1846  */
1847 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1848 {
1849         struct btrfs_fs_info *fs_info = root->fs_info;
1850         struct btrfs_device_info *devices_info;
1851         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1852         struct btrfs_device *device;
1853         u64 skip_space;
1854         u64 type;
1855         u64 avail_space;
1856         u64 used_space;
1857         u64 min_stripe_size;
1858         int min_stripes = 1, num_stripes = 1;
1859         int i = 0, nr_devices;
1860         int ret;
1861
1862         /*
1863          * We aren't under the device list lock, so this is racey-ish, but good
1864          * enough for our purposes.
1865          */
1866         nr_devices = fs_info->fs_devices->open_devices;
1867         if (!nr_devices) {
1868                 smp_mb();
1869                 nr_devices = fs_info->fs_devices->open_devices;
1870                 ASSERT(nr_devices);
1871                 if (!nr_devices) {
1872                         *free_bytes = 0;
1873                         return 0;
1874                 }
1875         }
1876
1877         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1878                                GFP_NOFS);
1879         if (!devices_info)
1880                 return -ENOMEM;
1881
1882         /* calc min stripe number for data space alloction */
1883         type = btrfs_get_alloc_profile(root, 1);
1884         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1885                 min_stripes = 2;
1886                 num_stripes = nr_devices;
1887         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1888                 min_stripes = 2;
1889                 num_stripes = 2;
1890         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1891                 min_stripes = 4;
1892                 num_stripes = 4;
1893         }
1894
1895         if (type & BTRFS_BLOCK_GROUP_DUP)
1896                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1897         else
1898                 min_stripe_size = BTRFS_STRIPE_LEN;
1899
1900         if (fs_info->alloc_start)
1901                 mutex_lock(&fs_devices->device_list_mutex);
1902         rcu_read_lock();
1903         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1904                 if (!device->in_fs_metadata || !device->bdev ||
1905                     device->is_tgtdev_for_dev_replace)
1906                         continue;
1907
1908                 if (i >= nr_devices)
1909                         break;
1910
1911                 avail_space = device->total_bytes - device->bytes_used;
1912
1913                 /* align with stripe_len */
1914                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1915                 avail_space *= BTRFS_STRIPE_LEN;
1916
1917                 /*
1918                  * In order to avoid overwritting the superblock on the drive,
1919                  * btrfs starts at an offset of at least 1MB when doing chunk
1920                  * allocation.
1921                  */
1922                 skip_space = SZ_1M;
1923
1924                 /* user can set the offset in fs_info->alloc_start. */
1925                 if (fs_info->alloc_start &&
1926                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1927                     device->total_bytes) {
1928                         rcu_read_unlock();
1929                         skip_space = max(fs_info->alloc_start, skip_space);
1930
1931                         /*
1932                          * btrfs can not use the free space in
1933                          * [0, skip_space - 1], we must subtract it from the
1934                          * total. In order to implement it, we account the used
1935                          * space in this range first.
1936                          */
1937                         ret = btrfs_account_dev_extents_size(device, 0,
1938                                                              skip_space - 1,
1939                                                              &used_space);
1940                         if (ret) {
1941                                 kfree(devices_info);
1942                                 mutex_unlock(&fs_devices->device_list_mutex);
1943                                 return ret;
1944                         }
1945
1946                         rcu_read_lock();
1947
1948                         /* calc the free space in [0, skip_space - 1] */
1949                         skip_space -= used_space;
1950                 }
1951
1952                 /*
1953                  * we can use the free space in [0, skip_space - 1], subtract
1954                  * it from the total.
1955                  */
1956                 if (avail_space && avail_space >= skip_space)
1957                         avail_space -= skip_space;
1958                 else
1959                         avail_space = 0;
1960
1961                 if (avail_space < min_stripe_size)
1962                         continue;
1963
1964                 devices_info[i].dev = device;
1965                 devices_info[i].max_avail = avail_space;
1966
1967                 i++;
1968         }
1969         rcu_read_unlock();
1970         if (fs_info->alloc_start)
1971                 mutex_unlock(&fs_devices->device_list_mutex);
1972
1973         nr_devices = i;
1974
1975         btrfs_descending_sort_devices(devices_info, nr_devices);
1976
1977         i = nr_devices - 1;
1978         avail_space = 0;
1979         while (nr_devices >= min_stripes) {
1980                 if (num_stripes > nr_devices)
1981                         num_stripes = nr_devices;
1982
1983                 if (devices_info[i].max_avail >= min_stripe_size) {
1984                         int j;
1985                         u64 alloc_size;
1986
1987                         avail_space += devices_info[i].max_avail * num_stripes;
1988                         alloc_size = devices_info[i].max_avail;
1989                         for (j = i + 1 - num_stripes; j <= i; j++)
1990                                 devices_info[j].max_avail -= alloc_size;
1991                 }
1992                 i--;
1993                 nr_devices--;
1994         }
1995
1996         kfree(devices_info);
1997         *free_bytes = avail_space;
1998         return 0;
1999 }
2000
2001 /*
2002  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2003  *
2004  * If there's a redundant raid level at DATA block groups, use the respective
2005  * multiplier to scale the sizes.
2006  *
2007  * Unused device space usage is based on simulating the chunk allocator
2008  * algorithm that respects the device sizes, order of allocations and the
2009  * 'alloc_start' value, this is a close approximation of the actual use but
2010  * there are other factors that may change the result (like a new metadata
2011  * chunk).
2012  *
2013  * If metadata is exhausted, f_bavail will be 0.
2014  *
2015  * FIXME: not accurate for mixed block groups, total and free/used are ok,
2016  * available appears slightly larger.
2017  */
2018 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2019 {
2020         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2021         struct btrfs_super_block *disk_super = fs_info->super_copy;
2022         struct list_head *head = &fs_info->space_info;
2023         struct btrfs_space_info *found;
2024         u64 total_used = 0;
2025         u64 total_free_data = 0;
2026         u64 total_free_meta = 0;
2027         int bits = dentry->d_sb->s_blocksize_bits;
2028         __be32 *fsid = (__be32 *)fs_info->fsid;
2029         unsigned factor = 1;
2030         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2031         int ret;
2032         u64 thresh = 0;
2033
2034         /*
2035          * holding chunk_muext to avoid allocating new chunks, holding
2036          * device_list_mutex to avoid the device being removed
2037          */
2038         rcu_read_lock();
2039         list_for_each_entry_rcu(found, head, list) {
2040                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2041                         int i;
2042
2043                         total_free_data += found->disk_total - found->disk_used;
2044                         total_free_data -=
2045                                 btrfs_account_ro_block_groups_free_space(found);
2046
2047                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2048                                 if (!list_empty(&found->block_groups[i])) {
2049                                         switch (i) {
2050                                         case BTRFS_RAID_DUP:
2051                                         case BTRFS_RAID_RAID1:
2052                                         case BTRFS_RAID_RAID10:
2053                                                 factor = 2;
2054                                         }
2055                                 }
2056                         }
2057                 }
2058                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2059                         total_free_meta += found->disk_total - found->disk_used;
2060
2061                 total_used += found->disk_used;
2062         }
2063
2064         rcu_read_unlock();
2065
2066         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2067         buf->f_blocks >>= bits;
2068         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2069
2070         /* Account global block reserve as used, it's in logical size already */
2071         spin_lock(&block_rsv->lock);
2072         buf->f_bfree -= block_rsv->size >> bits;
2073         spin_unlock(&block_rsv->lock);
2074
2075         buf->f_bavail = div_u64(total_free_data, factor);
2076         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2077         if (ret)
2078                 return ret;
2079         buf->f_bavail += div_u64(total_free_data, factor);
2080         buf->f_bavail = buf->f_bavail >> bits;
2081
2082         /*
2083          * We calculate the remaining metadata space minus global reserve. If
2084          * this is (supposedly) smaller than zero, there's no space. But this
2085          * does not hold in practice, the exhausted state happens where's still
2086          * some positive delta. So we apply some guesswork and compare the
2087          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2088          *
2089          * We probably cannot calculate the exact threshold value because this
2090          * depends on the internal reservations requested by various
2091          * operations, so some operations that consume a few metadata will
2092          * succeed even if the Avail is zero. But this is better than the other
2093          * way around.
2094          */
2095         thresh = 4 * 1024 * 1024;
2096
2097         if (total_free_meta - thresh < block_rsv->size)
2098                 buf->f_bavail = 0;
2099
2100         buf->f_type = BTRFS_SUPER_MAGIC;
2101         buf->f_bsize = dentry->d_sb->s_blocksize;
2102         buf->f_namelen = BTRFS_NAME_LEN;
2103
2104         /* We treat it as constant endianness (it doesn't matter _which_)
2105            because we want the fsid to come out the same whether mounted
2106            on a big-endian or little-endian host */
2107         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2108         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2109         /* Mask in the root object ID too, to disambiguate subvols */
2110         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2111         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2112
2113         return 0;
2114 }
2115
2116 static void btrfs_kill_super(struct super_block *sb)
2117 {
2118         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2119         kill_anon_super(sb);
2120         free_fs_info(fs_info);
2121 }
2122
2123 static struct file_system_type btrfs_fs_type = {
2124         .owner          = THIS_MODULE,
2125         .name           = "btrfs",
2126         .mount          = btrfs_mount,
2127         .kill_sb        = btrfs_kill_super,
2128         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2129 };
2130 MODULE_ALIAS_FS("btrfs");
2131
2132 static int btrfs_control_open(struct inode *inode, struct file *file)
2133 {
2134         /*
2135          * The control file's private_data is used to hold the
2136          * transaction when it is started and is used to keep
2137          * track of whether a transaction is already in progress.
2138          */
2139         file->private_data = NULL;
2140         return 0;
2141 }
2142
2143 /*
2144  * used by btrfsctl to scan devices when no FS is mounted
2145  */
2146 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2147                                 unsigned long arg)
2148 {
2149         struct btrfs_ioctl_vol_args *vol;
2150         struct btrfs_fs_devices *fs_devices;
2151         int ret = -ENOTTY;
2152
2153         if (!capable(CAP_SYS_ADMIN))
2154                 return -EPERM;
2155
2156         vol = memdup_user((void __user *)arg, sizeof(*vol));
2157         if (IS_ERR(vol))
2158                 return PTR_ERR(vol);
2159
2160         switch (cmd) {
2161         case BTRFS_IOC_SCAN_DEV:
2162                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2163                                             &btrfs_fs_type, &fs_devices);
2164                 break;
2165         case BTRFS_IOC_DEVICES_READY:
2166                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2167                                             &btrfs_fs_type, &fs_devices);
2168                 if (ret)
2169                         break;
2170                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2171                 break;
2172         }
2173
2174         kfree(vol);
2175         return ret;
2176 }
2177
2178 static int btrfs_freeze(struct super_block *sb)
2179 {
2180         struct btrfs_trans_handle *trans;
2181         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2182
2183         trans = btrfs_attach_transaction_barrier(root);
2184         if (IS_ERR(trans)) {
2185                 /* no transaction, don't bother */
2186                 if (PTR_ERR(trans) == -ENOENT)
2187                         return 0;
2188                 return PTR_ERR(trans);
2189         }
2190         return btrfs_commit_transaction(trans, root);
2191 }
2192
2193 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2194 {
2195         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2196         struct btrfs_fs_devices *cur_devices;
2197         struct btrfs_device *dev, *first_dev = NULL;
2198         struct list_head *head;
2199         struct rcu_string *name;
2200
2201         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2202         cur_devices = fs_info->fs_devices;
2203         while (cur_devices) {
2204                 head = &cur_devices->devices;
2205                 list_for_each_entry(dev, head, dev_list) {
2206                         if (dev->missing)
2207                                 continue;
2208                         if (!dev->name)
2209                                 continue;
2210                         if (!first_dev || dev->devid < first_dev->devid)
2211                                 first_dev = dev;
2212                 }
2213                 cur_devices = cur_devices->seed;
2214         }
2215
2216         if (first_dev) {
2217                 rcu_read_lock();
2218                 name = rcu_dereference(first_dev->name);
2219                 seq_escape(m, name->str, " \t\n\\");
2220                 rcu_read_unlock();
2221         } else {
2222                 WARN_ON(1);
2223         }
2224         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2225         return 0;
2226 }
2227
2228 static const struct super_operations btrfs_super_ops = {
2229         .drop_inode     = btrfs_drop_inode,
2230         .evict_inode    = btrfs_evict_inode,
2231         .put_super      = btrfs_put_super,
2232         .sync_fs        = btrfs_sync_fs,
2233         .show_options   = btrfs_show_options,
2234         .show_devname   = btrfs_show_devname,
2235         .write_inode    = btrfs_write_inode,
2236         .alloc_inode    = btrfs_alloc_inode,
2237         .destroy_inode  = btrfs_destroy_inode,
2238         .statfs         = btrfs_statfs,
2239         .remount_fs     = btrfs_remount,
2240         .freeze_fs      = btrfs_freeze,
2241 };
2242
2243 static const struct file_operations btrfs_ctl_fops = {
2244         .open = btrfs_control_open,
2245         .unlocked_ioctl  = btrfs_control_ioctl,
2246         .compat_ioctl = btrfs_control_ioctl,
2247         .owner   = THIS_MODULE,
2248         .llseek = noop_llseek,
2249 };
2250
2251 static struct miscdevice btrfs_misc = {
2252         .minor          = BTRFS_MINOR,
2253         .name           = "btrfs-control",
2254         .fops           = &btrfs_ctl_fops
2255 };
2256
2257 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2258 MODULE_ALIAS("devname:btrfs-control");
2259
2260 static int btrfs_interface_init(void)
2261 {
2262         return misc_register(&btrfs_misc);
2263 }
2264
2265 static void btrfs_interface_exit(void)
2266 {
2267         misc_deregister(&btrfs_misc);
2268 }
2269
2270 static void btrfs_print_info(void)
2271 {
2272         printk(KERN_INFO "Btrfs loaded"
2273 #ifdef CONFIG_BTRFS_DEBUG
2274                         ", debug=on"
2275 #endif
2276 #ifdef CONFIG_BTRFS_ASSERT
2277                         ", assert=on"
2278 #endif
2279 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2280                         ", integrity-checker=on"
2281 #endif
2282                         "\n");
2283 }
2284
2285 static int btrfs_run_sanity_tests(void)
2286 {
2287         int ret;
2288
2289         ret = btrfs_init_test_fs();
2290         if (ret)
2291                 return ret;
2292
2293         ret = btrfs_test_free_space_cache();
2294         if (ret)
2295                 goto out;
2296         ret = btrfs_test_extent_buffer_operations();
2297         if (ret)
2298                 goto out;
2299         ret = btrfs_test_extent_io();
2300         if (ret)
2301                 goto out;
2302         ret = btrfs_test_inodes();
2303         if (ret)
2304                 goto out;
2305         ret = btrfs_test_qgroups();
2306         if (ret)
2307                 goto out;
2308         ret = btrfs_test_free_space_tree();
2309 out:
2310         btrfs_destroy_test_fs();
2311         return ret;
2312 }
2313
2314 static int __init init_btrfs_fs(void)
2315 {
2316         int err;
2317
2318         err = btrfs_hash_init();
2319         if (err)
2320                 return err;
2321
2322         btrfs_props_init();
2323
2324         err = btrfs_init_sysfs();
2325         if (err)
2326                 goto free_hash;
2327
2328         btrfs_init_compress();
2329
2330         err = btrfs_init_cachep();
2331         if (err)
2332                 goto free_compress;
2333
2334         err = extent_io_init();
2335         if (err)
2336                 goto free_cachep;
2337
2338         err = extent_map_init();
2339         if (err)
2340                 goto free_extent_io;
2341
2342         err = ordered_data_init();
2343         if (err)
2344                 goto free_extent_map;
2345
2346         err = btrfs_delayed_inode_init();
2347         if (err)
2348                 goto free_ordered_data;
2349
2350         err = btrfs_auto_defrag_init();
2351         if (err)
2352                 goto free_delayed_inode;
2353
2354         err = btrfs_delayed_ref_init();
2355         if (err)
2356                 goto free_auto_defrag;
2357
2358         err = btrfs_prelim_ref_init();
2359         if (err)
2360                 goto free_delayed_ref;
2361
2362         err = btrfs_end_io_wq_init();
2363         if (err)
2364                 goto free_prelim_ref;
2365
2366         err = btrfs_interface_init();
2367         if (err)
2368                 goto free_end_io_wq;
2369
2370         btrfs_init_lockdep();
2371
2372         btrfs_print_info();
2373
2374         err = btrfs_run_sanity_tests();
2375         if (err)
2376                 goto unregister_ioctl;
2377
2378         err = register_filesystem(&btrfs_fs_type);
2379         if (err)
2380                 goto unregister_ioctl;
2381
2382         return 0;
2383
2384 unregister_ioctl:
2385         btrfs_interface_exit();
2386 free_end_io_wq:
2387         btrfs_end_io_wq_exit();
2388 free_prelim_ref:
2389         btrfs_prelim_ref_exit();
2390 free_delayed_ref:
2391         btrfs_delayed_ref_exit();
2392 free_auto_defrag:
2393         btrfs_auto_defrag_exit();
2394 free_delayed_inode:
2395         btrfs_delayed_inode_exit();
2396 free_ordered_data:
2397         ordered_data_exit();
2398 free_extent_map:
2399         extent_map_exit();
2400 free_extent_io:
2401         extent_io_exit();
2402 free_cachep:
2403         btrfs_destroy_cachep();
2404 free_compress:
2405         btrfs_exit_compress();
2406         btrfs_exit_sysfs();
2407 free_hash:
2408         btrfs_hash_exit();
2409         return err;
2410 }
2411
2412 static void __exit exit_btrfs_fs(void)
2413 {
2414         btrfs_destroy_cachep();
2415         btrfs_delayed_ref_exit();
2416         btrfs_auto_defrag_exit();
2417         btrfs_delayed_inode_exit();
2418         btrfs_prelim_ref_exit();
2419         ordered_data_exit();
2420         extent_map_exit();
2421         extent_io_exit();
2422         btrfs_interface_exit();
2423         btrfs_end_io_wq_exit();
2424         unregister_filesystem(&btrfs_fs_type);
2425         btrfs_exit_sysfs();
2426         btrfs_cleanup_fs_uuids();
2427         btrfs_exit_compress();
2428         btrfs_hash_exit();
2429 }
2430
2431 late_initcall(init_btrfs_fs);
2432 module_exit(exit_btrfs_fs)
2433
2434 MODULE_LICENSE("GPL");