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 "tests/btrfs-tests.h"
  62
  63 #include "qgroup.h"
  64 #define CREATE_TRACE_POINTS
  65 #include <trace/events/btrfs.h>
  66
  67 static const struct super_operations btrfs_super_ops;
  68 static struct file_system_type btrfs_fs_type;
  69
  70 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  71
  72 const char *btrfs_decode_error(int errno)
  73 {
  74         char *errstr = "unknown";
  75
  76         switch (errno) {
  77         case -EIO:
  78                 errstr = "IO failure";
  79                 break;
  80         case -ENOMEM:
  81                 errstr = "Out of memory";
  82                 break;
  83         case -EROFS:
  84                 errstr = "Readonly filesystem";
  85                 break;
  86         case -EEXIST:
  87                 errstr = "Object already exists";
  88                 break;
  89         case -ENOSPC:
  90                 errstr = "No space left";
  91                 break;
  92         case -ENOENT:
  93                 errstr = "No such entry";
  94                 break;
  95         }
  96
  97         return errstr;
  98 }
  99
 100 static void save_error_info(struct btrfs_fs_info *fs_info)
 101 {
 102         /*
 103          * today we only save the error info into ram.  Long term we'll
 104          * also send it down to the disk
 105          */
 106         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 107 }
 108
 109 /* btrfs handle error by forcing the filesystem readonly */
 110 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 111 {
 112         struct super_block *sb = fs_info->sb;
 113
 114         if (sb->s_flags & MS_RDONLY)
 115                 return;
 116
 117         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 118                 sb->s_flags |= MS_RDONLY;
 119                 btrfs_info(fs_info, "forced readonly");
 120                 /*
 121                  * Note that a running device replace operation is not
 122                  * canceled here although there is no way to update
 123                  * the progress. It would add the risk of a deadlock,
 124                  * therefore the canceling is ommited. The only penalty
 125                  * is that some I/O remains active until the procedure
 126                  * completes. The next time when the filesystem is
 127                  * mounted writeable again, the device replace
 128                  * operation continues.
 129                  */
 130         }
 131 }
 132
 133 #ifdef CONFIG_PRINTK
 134 /*
 135  * __btrfs_std_error decodes expected errors from the caller and
 136  * invokes the approciate error response.
 137  */
 138 __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         const char *errstr;
 144
 145         /*
 146          * Special case: if the error is EROFS, and we're already
 147          * under MS_RDONLY, then it is safe here.
 148          */
 149         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 150                 return;
 151
 152         errstr = btrfs_decode_error(errno);
 153         if (fmt) {
 154                 struct va_format vaf;
 155                 va_list args;
 156
 157                 va_start(args, fmt);
 158                 vaf.fmt = fmt;
 159                 vaf.va = &args;
 160
 161                 printk(KERN_CRIT
 162                         "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 163                         sb->s_id, function, line, errno, errstr, &vaf);
 164                 va_end(args);
 165         } else {
 166                 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 167                         sb->s_id, function, line, errno, errstr);
 168         }
 169
 170         /* Don't go through full error handling during mount */
 171         save_error_info(fs_info);
 172         if (sb->s_flags & MS_BORN)
 173                 btrfs_handle_error(fs_info);
 174 }
 175
 176 static const char * const logtypes[] = {
 177         "emergency",
 178         "alert",
 179         "critical",
 180         "error",
 181         "warning",
 182         "notice",
 183         "info",
 184         "debug",
 185 };
 186
 187 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 188 {
 189         struct super_block *sb = fs_info->sb;
 190         char lvl[4];
 191         struct va_format vaf;
 192         va_list args;
 193         const char *type = logtypes[4];
 194         int kern_level;
 195
 196         va_start(args, fmt);
 197
 198         kern_level = printk_get_level(fmt);
 199         if (kern_level) {
 200                 size_t size = printk_skip_level(fmt) - fmt;
 201                 memcpy(lvl, fmt,  size);
 202                 lvl[size] = '\0';
 203                 fmt += size;
 204                 type = logtypes[kern_level - '0'];
 205         } else
 206                 *lvl = '\0';
 207
 208         vaf.fmt = fmt;
 209         vaf.va = &args;
 210
 211         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 212
 213         va_end(args);
 214 }
 215
 216 #else
 217
 218 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 219                        unsigned int line, int errno, const char *fmt, ...)
 220 {
 221         struct super_block *sb = fs_info->sb;
 222
 223         /*
 224          * Special case: if the error is EROFS, and we're already
 225          * under MS_RDONLY, then it is safe here.
 226          */
 227         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 228                 return;
 229
 230         /* Don't go through full error handling during mount */
 231         if (sb->s_flags & MS_BORN) {
 232                 save_error_info(fs_info);
 233                 btrfs_handle_error(fs_info);
 234         }
 235 }
 236 #endif
 237
 238 /*
 239  * We only mark the transaction aborted and then set the file system read-only.
 240  * This will prevent new transactions from starting or trying to join this
 241  * one.
 242  *
 243  * This means that error recovery at the call site is limited to freeing
 244  * any local memory allocations and passing the error code up without
 245  * further cleanup. The transaction should complete as it normally would
 246  * in the call path but will return -EIO.
 247  *
 248  * We'll complete the cleanup in btrfs_end_transaction and
 249  * btrfs_commit_transaction.
 250  */
 251 __cold
 252 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 253                                struct btrfs_root *root, const char *function,
 254                                unsigned int line, int errno)
 255 {
 256         trans->aborted = errno;
 257         /* Nothing used. The other threads that have joined this
 258          * transaction may be able to continue. */
 259         if (!trans->blocks_used && list_empty(&trans->new_bgs)) {
 260                 const char *errstr;
 261
 262                 errstr = btrfs_decode_error(errno);
 263                 btrfs_warn(root->fs_info,
 264                            "%s:%d: Aborting unused transaction(%s).",
 265                            function, line, errstr);
 266                 return;
 267         }
 268         ACCESS_ONCE(trans->transaction->aborted) = errno;
 269         /* Wake up anybody who may be waiting on this transaction */
 270         wake_up(&root->fs_info->transaction_wait);
 271         wake_up(&root->fs_info->transaction_blocked_wait);
 272         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 273 }
 274 /*
 275  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 276  * issues an alert, and either panics or BUGs, depending on mount options.
 277  */
 278 __cold
 279 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 280                    unsigned int line, int errno, const char *fmt, ...)
 281 {
 282         char *s_id = "<unknown>";
 283         const char *errstr;
 284         struct va_format vaf = { .fmt = fmt };
 285         va_list args;
 286
 287         if (fs_info)
 288                 s_id = fs_info->sb->s_id;
 289
 290         va_start(args, fmt);
 291         vaf.va = &args;
 292
 293         errstr = btrfs_decode_error(errno);
 294         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 295                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 296                         s_id, function, line, &vaf, errno, errstr);
 297
 298         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 299                    function, line, &vaf, errno, errstr);
 300         va_end(args);
 301         /* Caller calls BUG() */
 302 }
 303
 304 static void btrfs_put_super(struct super_block *sb)
 305 {
 306         close_ctree(btrfs_sb(sb)->tree_root);
 307 }
 308
 309 enum {
 310         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 311         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 312         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 313         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 314         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 315         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 316         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 317         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 318         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 319         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 320         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
 321         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
 322         Opt_datasum, Opt_treelog, Opt_noinode_cache,
 323         Opt_err,
 324 };
 325
 326 static match_table_t tokens = {
 327         {Opt_degraded, "degraded"},
 328         {Opt_subvol, "subvol=%s"},
 329         {Opt_subvolid, "subvolid=%s"},
 330         {Opt_device, "device=%s"},
 331         {Opt_nodatasum, "nodatasum"},
 332         {Opt_datasum, "datasum"},
 333         {Opt_nodatacow, "nodatacow"},
 334         {Opt_datacow, "datacow"},
 335         {Opt_nobarrier, "nobarrier"},
 336         {Opt_barrier, "barrier"},
 337         {Opt_max_inline, "max_inline=%s"},
 338         {Opt_alloc_start, "alloc_start=%s"},
 339         {Opt_thread_pool, "thread_pool=%d"},
 340         {Opt_compress, "compress"},
 341         {Opt_compress_type, "compress=%s"},
 342         {Opt_compress_force, "compress-force"},
 343         {Opt_compress_force_type, "compress-force=%s"},
 344         {Opt_ssd, "ssd"},
 345         {Opt_ssd_spread, "ssd_spread"},
 346         {Opt_nossd, "nossd"},
 347         {Opt_acl, "acl"},
 348         {Opt_noacl, "noacl"},
 349         {Opt_notreelog, "notreelog"},
 350         {Opt_treelog, "treelog"},
 351         {Opt_flushoncommit, "flushoncommit"},
 352         {Opt_noflushoncommit, "noflushoncommit"},
 353         {Opt_ratio, "metadata_ratio=%d"},
 354         {Opt_discard, "discard"},
 355         {Opt_nodiscard, "nodiscard"},
 356         {Opt_space_cache, "space_cache"},
 357         {Opt_clear_cache, "clear_cache"},
 358         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 359         {Opt_enospc_debug, "enospc_debug"},
 360         {Opt_noenospc_debug, "noenospc_debug"},
 361         {Opt_subvolrootid, "subvolrootid=%d"},
 362         {Opt_defrag, "autodefrag"},
 363         {Opt_nodefrag, "noautodefrag"},
 364         {Opt_inode_cache, "inode_cache"},
 365         {Opt_noinode_cache, "noinode_cache"},
 366         {Opt_no_space_cache, "nospace_cache"},
 367         {Opt_recovery, "recovery"},
 368         {Opt_skip_balance, "skip_balance"},
 369         {Opt_check_integrity, "check_int"},
 370         {Opt_check_integrity_including_extent_data, "check_int_data"},
 371         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 372         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 373         {Opt_fatal_errors, "fatal_errors=%s"},
 374         {Opt_commit_interval, "commit=%d"},
 375         {Opt_err, NULL},
 376 };
 377
 378 /*
 379  * Regular mount options parser.  Everything that is needed only when
 380  * reading in a new superblock is parsed here.
 381  * XXX JDM: This needs to be cleaned up for remount.
 382  */
 383 int btrfs_parse_options(struct btrfs_root *root, char *options)
 384 {
 385         struct btrfs_fs_info *info = root->fs_info;
 386         substring_t args[MAX_OPT_ARGS];
 387         char *p, *num, *orig = NULL;
 388         u64 cache_gen;
 389         int intarg;
 390         int ret = 0;
 391         char *compress_type;
 392         bool compress_force = false;
 393
 394         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 395         if (cache_gen)
 396                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 397
 398         if (!options)
 399                 goto out;
 400
 401         /*
 402          * strsep changes the string, duplicate it because parse_options
 403          * gets called twice
 404          */
 405         options = kstrdup(options, GFP_NOFS);
 406         if (!options)
 407                 return -ENOMEM;
 408
 409         orig = options;
 410
 411         while ((p = strsep(&options, ",")) != NULL) {
 412                 int token;
 413                 if (!*p)
 414                         continue;
 415
 416                 token = match_token(p, tokens, args);
 417                 switch (token) {
 418                 case Opt_degraded:
 419                         btrfs_info(root->fs_info, "allowing degraded mounts");
 420                         btrfs_set_opt(info->mount_opt, DEGRADED);
 421                         break;
 422                 case Opt_subvol:
 423                 case Opt_subvolid:
 424                 case Opt_subvolrootid:
 425                 case Opt_device:
 426                         /*
 427                          * These are parsed by btrfs_parse_early_options
 428                          * and can be happily ignored here.
 429                          */
 430                         break;
 431                 case Opt_nodatasum:
 432                         btrfs_set_and_info(root, NODATASUM,
 433                                            "setting nodatasum");
 434                         break;
 435                 case Opt_datasum:
 436                         if (btrfs_test_opt(root, NODATASUM)) {
 437                                 if (btrfs_test_opt(root, NODATACOW))
 438                                         btrfs_info(root->fs_info, "setting datasum, datacow enabled");
 439                                 else
 440                                         btrfs_info(root->fs_info, "setting datasum");
 441                         }
 442                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 443                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 444                         break;
 445                 case Opt_nodatacow:
 446                         if (!btrfs_test_opt(root, NODATACOW)) {
 447                                 if (!btrfs_test_opt(root, COMPRESS) ||
 448                                     !btrfs_test_opt(root, FORCE_COMPRESS)) {
 449                                         btrfs_info(root->fs_info,
 450                                                    "setting nodatacow, compression disabled");
 451                                 } else {
 452                                         btrfs_info(root->fs_info, "setting nodatacow");
 453                                 }
 454                         }
 455                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 456                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 457                         btrfs_set_opt(info->mount_opt, NODATACOW);
 458                         btrfs_set_opt(info->mount_opt, NODATASUM);
 459                         break;
 460                 case Opt_datacow:
 461                         btrfs_clear_and_info(root, NODATACOW,
 462                                              "setting datacow");
 463                         break;
 464                 case Opt_compress_force:
 465                 case Opt_compress_force_type:
 466                         compress_force = true;
 467                         /* Fallthrough */
 468                 case Opt_compress:
 469                 case Opt_compress_type:
 470                         if (token == Opt_compress ||
 471                             token == Opt_compress_force ||
 472                             strcmp(args[0].from, "zlib") == 0) {
 473                                 compress_type = "zlib";
 474                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 475                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 476                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 477                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 478                         } else if (strcmp(args[0].from, "lzo") == 0) {
 479                                 compress_type = "lzo";
 480                                 info->compress_type = BTRFS_COMPRESS_LZO;
 481                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 482                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 483                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 484                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 485                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 486                                 compress_type = "no";
 487                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 488                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 489                                 compress_force = false;
 490                         } else {
 491                                 ret = -EINVAL;
 492                                 goto out;
 493                         }
 494
 495                         if (compress_force) {
 496                                 btrfs_set_and_info(root, FORCE_COMPRESS,
 497                                                    "force %s compression",
 498                                                    compress_type);
 499                         } else {
 500                                 if (!btrfs_test_opt(root, COMPRESS))
 501                                         btrfs_info(root->fs_info,
 502                                                    "btrfs: use %s compression",
 503                                                    compress_type);
 504                                 /*
 505                                  * If we remount from compress-force=xxx to
 506                                  * compress=xxx, we need clear FORCE_COMPRESS
 507                                  * flag, otherwise, there is no way for users
 508                                  * to disable forcible compression separately.
 509                                  */
 510                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 511                         }
 512                         break;
 513                 case Opt_ssd:
 514                         btrfs_set_and_info(root, SSD,
 515                                            "use ssd allocation scheme");
 516                         break;
 517                 case Opt_ssd_spread:
 518                         btrfs_set_and_info(root, SSD_SPREAD,
 519                                            "use spread ssd allocation scheme");
 520                         btrfs_set_opt(info->mount_opt, SSD);
 521                         break;
 522                 case Opt_nossd:
 523                         btrfs_set_and_info(root, NOSSD,
 524                                              "not using ssd allocation scheme");
 525                         btrfs_clear_opt(info->mount_opt, SSD);
 526                         break;
 527                 case Opt_barrier:
 528                         btrfs_clear_and_info(root, NOBARRIER,
 529                                              "turning on barriers");
 530                         break;
 531                 case Opt_nobarrier:
 532                         btrfs_set_and_info(root, NOBARRIER,
 533                                            "turning off barriers");
 534                         break;
 535                 case Opt_thread_pool:
 536                         ret = match_int(&args[0], &intarg);
 537                         if (ret) {
 538                                 goto out;
 539                         } else if (intarg > 0) {
 540                                 info->thread_pool_size = intarg;
 541                         } else {
 542                                 ret = -EINVAL;
 543                                 goto out;
 544                         }
 545                         break;
 546                 case Opt_max_inline:
 547                         num = match_strdup(&args[0]);
 548                         if (num) {
 549                                 info->max_inline = memparse(num, NULL);
 550                                 kfree(num);
 551
 552                                 if (info->max_inline) {
 553                                         info->max_inline = min_t(u64,
 554                                                 info->max_inline,
 555                                                 root->sectorsize);
 556                                 }
 557                                 btrfs_info(root->fs_info, "max_inline at %llu",
 558                                         info->max_inline);
 559                         } else {
 560                                 ret = -ENOMEM;
 561                                 goto out;
 562                         }
 563                         break;
 564                 case Opt_alloc_start:
 565                         num = match_strdup(&args[0]);
 566                         if (num) {
 567                                 mutex_lock(&info->chunk_mutex);
 568                                 info->alloc_start = memparse(num, NULL);
 569                                 mutex_unlock(&info->chunk_mutex);
 570                                 kfree(num);
 571                                 btrfs_info(root->fs_info, "allocations start at %llu",
 572                                         info->alloc_start);
 573                         } else {
 574                                 ret = -ENOMEM;
 575                                 goto out;
 576                         }
 577                         break;
 578                 case Opt_acl:
 579 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 580                         root->fs_info->sb->s_flags |= MS_POSIXACL;
 581                         break;
 582 #else
 583                         btrfs_err(root->fs_info,
 584                                 "support for ACL not compiled in!");
 585                         ret = -EINVAL;
 586                         goto out;
 587 #endif
 588                 case Opt_noacl:
 589                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 590                         break;
 591                 case Opt_notreelog:
 592                         btrfs_set_and_info(root, NOTREELOG,
 593                                            "disabling tree log");
 594                         break;
 595                 case Opt_treelog:
 596                         btrfs_clear_and_info(root, NOTREELOG,
 597                                              "enabling tree log");
 598                         break;
 599                 case Opt_flushoncommit:
 600                         btrfs_set_and_info(root, FLUSHONCOMMIT,
 601                                            "turning on flush-on-commit");
 602                         break;
 603                 case Opt_noflushoncommit:
 604                         btrfs_clear_and_info(root, FLUSHONCOMMIT,
 605                                              "turning off flush-on-commit");
 606                         break;
 607                 case Opt_ratio:
 608                         ret = match_int(&args[0], &intarg);
 609                         if (ret) {
 610                                 goto out;
 611                         } else if (intarg >= 0) {
 612                                 info->metadata_ratio = intarg;
 613                                 btrfs_info(root->fs_info, "metadata ratio %d",
 614                                        info->metadata_ratio);
 615                         } else {
 616                                 ret = -EINVAL;
 617                                 goto out;
 618                         }
 619                         break;
 620                 case Opt_discard:
 621                         btrfs_set_and_info(root, DISCARD,
 622                                            "turning on discard");
 623                         break;
 624                 case Opt_nodiscard:
 625                         btrfs_clear_and_info(root, DISCARD,
 626                                              "turning off discard");
 627                         break;
 628                 case Opt_space_cache:
 629                         btrfs_set_and_info(root, SPACE_CACHE,
 630                                            "enabling disk space caching");
 631                         break;
 632                 case Opt_rescan_uuid_tree:
 633                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 634                         break;
 635                 case Opt_no_space_cache:
 636                         btrfs_clear_and_info(root, SPACE_CACHE,
 637                                              "disabling disk space caching");
 638                         break;
 639                 case Opt_inode_cache:
 640                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 641                                            "enabling inode map caching");
 642                         break;
 643                 case Opt_noinode_cache:
 644                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 645                                              "disabling inode map caching");
 646                         break;
 647                 case Opt_clear_cache:
 648                         btrfs_set_and_info(root, CLEAR_CACHE,
 649                                            "force clearing of disk cache");
 650                         break;
 651                 case Opt_user_subvol_rm_allowed:
 652                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 653                         break;
 654                 case Opt_enospc_debug:
 655                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 656                         break;
 657                 case Opt_noenospc_debug:
 658                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 659                         break;
 660                 case Opt_defrag:
 661                         btrfs_set_and_info(root, AUTO_DEFRAG,
 662                                            "enabling auto defrag");
 663                         break;
 664                 case Opt_nodefrag:
 665                         btrfs_clear_and_info(root, AUTO_DEFRAG,
 666                                              "disabling auto defrag");
 667                         break;
 668                 case Opt_recovery:
 669                         btrfs_info(root->fs_info, "enabling auto recovery");
 670                         btrfs_set_opt(info->mount_opt, RECOVERY);
 671                         break;
 672                 case Opt_skip_balance:
 673                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 674                         break;
 675 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 676                 case Opt_check_integrity_including_extent_data:
 677                         btrfs_info(root->fs_info,
 678                                    "enabling check integrity including extent data");
 679                         btrfs_set_opt(info->mount_opt,
 680                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 681                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 682                         break;
 683                 case Opt_check_integrity:
 684                         btrfs_info(root->fs_info, "enabling check integrity");
 685                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 686                         break;
 687                 case Opt_check_integrity_print_mask:
 688                         ret = match_int(&args[0], &intarg);
 689                         if (ret) {
 690                                 goto out;
 691                         } else if (intarg >= 0) {
 692                                 info->check_integrity_print_mask = intarg;
 693                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
 694                                        info->check_integrity_print_mask);
 695                         } else {
 696                                 ret = -EINVAL;
 697                                 goto out;
 698                         }
 699                         break;
 700 #else
 701                 case Opt_check_integrity_including_extent_data:
 702                 case Opt_check_integrity:
 703                 case Opt_check_integrity_print_mask:
 704                         btrfs_err(root->fs_info,
 705                                 "support for check_integrity* not compiled in!");
 706                         ret = -EINVAL;
 707                         goto out;
 708 #endif
 709                 case Opt_fatal_errors:
 710                         if (strcmp(args[0].from, "panic") == 0)
 711                                 btrfs_set_opt(info->mount_opt,
 712                                               PANIC_ON_FATAL_ERROR);
 713                         else if (strcmp(args[0].from, "bug") == 0)
 714                                 btrfs_clear_opt(info->mount_opt,
 715                                               PANIC_ON_FATAL_ERROR);
 716                         else {
 717                                 ret = -EINVAL;
 718                                 goto out;
 719                         }
 720                         break;
 721                 case Opt_commit_interval:
 722                         intarg = 0;
 723                         ret = match_int(&args[0], &intarg);
 724                         if (ret < 0) {
 725                                 btrfs_err(root->fs_info, "invalid commit interval");
 726                                 ret = -EINVAL;
 727                                 goto out;
 728                         }
 729                         if (intarg > 0) {
 730                                 if (intarg > 300) {
 731                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
 732                                                         intarg);
 733                                 }
 734                                 info->commit_interval = intarg;
 735                         } else {
 736                                 btrfs_info(root->fs_info, "using default commit interval %ds",
 737                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
 738                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 739                         }
 740                         break;
 741                 case Opt_err:
 742                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
 743                         ret = -EINVAL;
 744                         goto out;
 745                 default:
 746                         break;
 747                 }
 748         }
 749 out:
 750         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 751                 btrfs_info(root->fs_info, "disk space caching is enabled");
 752         kfree(orig);
 753         return ret;
 754 }
 755
 756 /*
 757  * Parse mount options that are required early in the mount process.
 758  *
 759  * All other options will be parsed on much later in the mount process and
 760  * only when we need to allocate a new super block.
 761  */
 762 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 763                 void *holder, char **subvol_name, u64 *subvol_objectid,
 764                 struct btrfs_fs_devices **fs_devices)
 765 {
 766         substring_t args[MAX_OPT_ARGS];
 767         char *device_name, *opts, *orig, *p;
 768         char *num = NULL;
 769         int error = 0;
 770
 771         if (!options)
 772                 return 0;
 773
 774         /*
 775          * strsep changes the string, duplicate it because parse_options
 776          * gets called twice
 777          */
 778         opts = kstrdup(options, GFP_KERNEL);
 779         if (!opts)
 780                 return -ENOMEM;
 781         orig = opts;
 782
 783         while ((p = strsep(&opts, ",")) != NULL) {
 784                 int token;
 785                 if (!*p)
 786                         continue;
 787
 788                 token = match_token(p, tokens, args);
 789                 switch (token) {
 790                 case Opt_subvol:
 791                         kfree(*subvol_name);
 792                         *subvol_name = match_strdup(&args[0]);
 793                         if (!*subvol_name) {
 794                                 error = -ENOMEM;
 795                                 goto out;
 796                         }
 797                         break;
 798                 case Opt_subvolid:
 799                         num = match_strdup(&args[0]);
 800                         if (num) {
 801                                 *subvol_objectid = memparse(num, NULL);
 802                                 kfree(num);
 803                                 /* we want the original fs_tree */
 804                                 if (!*subvol_objectid)
 805                                         *subvol_objectid =
 806                                                 BTRFS_FS_TREE_OBJECTID;
 807                         } else {
 808                                 error = -EINVAL;
 809                                 goto out;
 810                         }
 811                         break;
 812                 case Opt_subvolrootid:
 813                         printk(KERN_WARNING
 814                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
 815                                 "no effect\n");
 816                         break;
 817                 case Opt_device:
 818                         device_name = match_strdup(&args[0]);
 819                         if (!device_name) {
 820                                 error = -ENOMEM;
 821                                 goto out;
 822                         }
 823                         error = btrfs_scan_one_device(device_name,
 824                                         flags, holder, fs_devices);
 825                         kfree(device_name);
 826                         if (error)
 827                                 goto out;
 828                         break;
 829                 default:
 830                         break;
 831                 }
 832         }
 833
 834 out:
 835         kfree(orig);
 836         return error;
 837 }
 838
 839 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 840                                            u64 subvol_objectid)
 841 {
 842         struct btrfs_root *root = fs_info->tree_root;
 843         struct btrfs_root *fs_root;
 844         struct btrfs_root_ref *root_ref;
 845         struct btrfs_inode_ref *inode_ref;
 846         struct btrfs_key key;
 847         struct btrfs_path *path = NULL;
 848         char *name = NULL, *ptr;
 849         u64 dirid;
 850         int len;
 851         int ret;
 852
 853         path = btrfs_alloc_path();
 854         if (!path) {
 855                 ret = -ENOMEM;
 856                 goto err;
 857         }
 858         path->leave_spinning = 1;
 859
 860         name = kmalloc(PATH_MAX, GFP_NOFS);
 861         if (!name) {
 862                 ret = -ENOMEM;
 863                 goto err;
 864         }
 865         ptr = name + PATH_MAX - 1;
 866         ptr[0] = '\0';
 867
 868         /*
 869          * Walk up the subvolume trees in the tree of tree roots by root
 870          * backrefs until we hit the top-level subvolume.
 871          */
 872         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 873                 key.objectid = subvol_objectid;
 874                 key.type = BTRFS_ROOT_BACKREF_KEY;
 875                 key.offset = (u64)-1;
 876
 877                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 878                 if (ret < 0) {
 879                         goto err;
 880                 } else if (ret > 0) {
 881                         ret = btrfs_previous_item(root, path, subvol_objectid,
 882                                                   BTRFS_ROOT_BACKREF_KEY);
 883                         if (ret < 0) {
 884                                 goto err;
 885                         } else if (ret > 0) {
 886                                 ret = -ENOENT;
 887                                 goto err;
 888                         }
 889                 }
 890
 891                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 892                 subvol_objectid = key.offset;
 893
 894                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 895                                           struct btrfs_root_ref);
 896                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 897                 ptr -= len + 1;
 898                 if (ptr < name) {
 899                         ret = -ENAMETOOLONG;
 900                         goto err;
 901                 }
 902                 read_extent_buffer(path->nodes[0], ptr + 1,
 903                                    (unsigned long)(root_ref + 1), len);
 904                 ptr[0] = '/';
 905                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
 906                 btrfs_release_path(path);
 907
 908                 key.objectid = subvol_objectid;
 909                 key.type = BTRFS_ROOT_ITEM_KEY;
 910                 key.offset = (u64)-1;
 911                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
 912                 if (IS_ERR(fs_root)) {
 913                         ret = PTR_ERR(fs_root);
 914                         goto err;
 915                 }
 916
 917                 /*
 918                  * Walk up the filesystem tree by inode refs until we hit the
 919                  * root directory.
 920                  */
 921                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
 922                         key.objectid = dirid;
 923                         key.type = BTRFS_INODE_REF_KEY;
 924                         key.offset = (u64)-1;
 925
 926                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
 927                         if (ret < 0) {
 928                                 goto err;
 929                         } else if (ret > 0) {
 930                                 ret = btrfs_previous_item(fs_root, path, dirid,
 931                                                           BTRFS_INODE_REF_KEY);
 932                                 if (ret < 0) {
 933                                         goto err;
 934                                 } else if (ret > 0) {
 935                                         ret = -ENOENT;
 936                                         goto err;
 937                                 }
 938                         }
 939
 940                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 941                         dirid = key.offset;
 942
 943                         inode_ref = btrfs_item_ptr(path->nodes[0],
 944                                                    path->slots[0],
 945                                                    struct btrfs_inode_ref);
 946                         len = btrfs_inode_ref_name_len(path->nodes[0],
 947                                                        inode_ref);
 948                         ptr -= len + 1;
 949                         if (ptr < name) {
 950                                 ret = -ENAMETOOLONG;
 951                                 goto err;
 952                         }
 953                         read_extent_buffer(path->nodes[0], ptr + 1,
 954                                            (unsigned long)(inode_ref + 1), len);
 955                         ptr[0] = '/';
 956                         btrfs_release_path(path);
 957                 }
 958         }
 959
 960         btrfs_free_path(path);
 961         if (ptr == name + PATH_MAX - 1) {
 962                 name[0] = '/';
 963                 name[1] = '\0';
 964         } else {
 965                 memmove(name, ptr, name + PATH_MAX - ptr);
 966         }
 967         return name;
 968
 969 err:
 970         btrfs_free_path(path);
 971         kfree(name);
 972         return ERR_PTR(ret);
 973 }
 974
 975 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
 976 {
 977         struct btrfs_root *root = fs_info->tree_root;
 978         struct btrfs_dir_item *di;
 979         struct btrfs_path *path;
 980         struct btrfs_key location;
 981         u64 dir_id;
 982
 983         path = btrfs_alloc_path();
 984         if (!path)
 985                 return -ENOMEM;
 986         path->leave_spinning = 1;
 987
 988         /*
 989          * Find the "default" dir item which points to the root item that we
 990          * will mount by default if we haven't been given a specific subvolume
 991          * to mount.
 992          */
 993         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 994         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 995         if (IS_ERR(di)) {
 996                 btrfs_free_path(path);
 997                 return PTR_ERR(di);
 998         }
 999         if (!di) {
1000                 /*
1001                  * Ok the default dir item isn't there.  This is weird since
1002                  * it's always been there, but don't freak out, just try and
1003                  * mount the top-level subvolume.
1004                  */
1005                 btrfs_free_path(path);
1006                 *objectid = BTRFS_FS_TREE_OBJECTID;
1007                 return 0;
1008         }
1009
1010         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1011         btrfs_free_path(path);
1012         *objectid = location.objectid;
1013         return 0;
1014 }
1015
1016 static int btrfs_fill_super(struct super_block *sb,
1017                             struct btrfs_fs_devices *fs_devices,
1018                             void *data, int silent)
1019 {
1020         struct inode *inode;
1021         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1022         struct btrfs_key key;
1023         int err;
1024
1025         sb->s_maxbytes = MAX_LFS_FILESIZE;
1026         sb->s_magic = BTRFS_SUPER_MAGIC;
1027         sb->s_op = &btrfs_super_ops;
1028         sb->s_d_op = &btrfs_dentry_operations;
1029         sb->s_export_op = &btrfs_export_ops;
1030         sb->s_xattr = btrfs_xattr_handlers;
1031         sb->s_time_gran = 1;
1032 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1033         sb->s_flags |= MS_POSIXACL;
1034 #endif
1035         sb->s_flags |= MS_I_VERSION;
1036         sb->s_iflags |= SB_I_CGROUPWB;
1037         err = open_ctree(sb, fs_devices, (char *)data);
1038         if (err) {
1039                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
1040                 return err;
1041         }
1042
1043         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1044         key.type = BTRFS_INODE_ITEM_KEY;
1045         key.offset = 0;
1046         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1047         if (IS_ERR(inode)) {
1048                 err = PTR_ERR(inode);
1049                 goto fail_close;
1050         }
1051
1052         sb->s_root = d_make_root(inode);
1053         if (!sb->s_root) {
1054                 err = -ENOMEM;
1055                 goto fail_close;
1056         }
1057
1058         save_mount_options(sb, data);
1059         cleancache_init_fs(sb);
1060         sb->s_flags |= MS_ACTIVE;
1061         return 0;
1062
1063 fail_close:
1064         close_ctree(fs_info->tree_root);
1065         return err;
1066 }
1067
1068 int btrfs_sync_fs(struct super_block *sb, int wait)
1069 {
1070         struct btrfs_trans_handle *trans;
1071         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1072         struct btrfs_root *root = fs_info->tree_root;
1073
1074         trace_btrfs_sync_fs(wait);
1075
1076         if (!wait) {
1077                 filemap_flush(fs_info->btree_inode->i_mapping);
1078                 return 0;
1079         }
1080
1081         btrfs_wait_ordered_roots(fs_info, -1);
1082
1083         trans = btrfs_attach_transaction_barrier(root);
1084         if (IS_ERR(trans)) {
1085                 /* no transaction, don't bother */
1086                 if (PTR_ERR(trans) == -ENOENT) {
1087                         /*
1088                          * Exit unless we have some pending changes
1089                          * that need to go through commit
1090                          */
1091                         if (fs_info->pending_changes == 0)
1092                                 return 0;
1093                         /*
1094                          * A non-blocking test if the fs is frozen. We must not
1095                          * start a new transaction here otherwise a deadlock
1096                          * happens. The pending operations are delayed to the
1097                          * next commit after thawing.
1098                          */
1099                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1100                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1101                         else
1102                                 return 0;
1103                         trans = btrfs_start_transaction(root, 0);
1104                 }
1105                 if (IS_ERR(trans))
1106                         return PTR_ERR(trans);
1107         }
1108         return btrfs_commit_transaction(trans, root);
1109 }
1110
1111 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1112 {
1113         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1114         struct btrfs_root *root = info->tree_root;
1115         char *compress_type;
1116
1117         if (btrfs_test_opt(root, DEGRADED))
1118                 seq_puts(seq, ",degraded");
1119         if (btrfs_test_opt(root, NODATASUM))
1120                 seq_puts(seq, ",nodatasum");
1121         if (btrfs_test_opt(root, NODATACOW))
1122                 seq_puts(seq, ",nodatacow");
1123         if (btrfs_test_opt(root, NOBARRIER))
1124                 seq_puts(seq, ",nobarrier");
1125         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1126                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1127         if (info->alloc_start != 0)
1128                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1129         if (info->thread_pool_size !=  min_t(unsigned long,
1130                                              num_online_cpus() + 2, 8))
1131                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1132         if (btrfs_test_opt(root, COMPRESS)) {
1133                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1134                         compress_type = "zlib";
1135                 else
1136                         compress_type = "lzo";
1137                 if (btrfs_test_opt(root, FORCE_COMPRESS))
1138                         seq_printf(seq, ",compress-force=%s", compress_type);
1139                 else
1140                         seq_printf(seq, ",compress=%s", compress_type);
1141         }
1142         if (btrfs_test_opt(root, NOSSD))
1143                 seq_puts(seq, ",nossd");
1144         if (btrfs_test_opt(root, SSD_SPREAD))
1145                 seq_puts(seq, ",ssd_spread");
1146         else if (btrfs_test_opt(root, SSD))
1147                 seq_puts(seq, ",ssd");
1148         if (btrfs_test_opt(root, NOTREELOG))
1149                 seq_puts(seq, ",notreelog");
1150         if (btrfs_test_opt(root, FLUSHONCOMMIT))
1151                 seq_puts(seq, ",flushoncommit");
1152         if (btrfs_test_opt(root, DISCARD))
1153                 seq_puts(seq, ",discard");
1154         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1155                 seq_puts(seq, ",noacl");
1156         if (btrfs_test_opt(root, SPACE_CACHE))
1157                 seq_puts(seq, ",space_cache");
1158         else
1159                 seq_puts(seq, ",nospace_cache");
1160         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1161                 seq_puts(seq, ",rescan_uuid_tree");
1162         if (btrfs_test_opt(root, CLEAR_CACHE))
1163                 seq_puts(seq, ",clear_cache");
1164         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1165                 seq_puts(seq, ",user_subvol_rm_allowed");
1166         if (btrfs_test_opt(root, ENOSPC_DEBUG))
1167                 seq_puts(seq, ",enospc_debug");
1168         if (btrfs_test_opt(root, AUTO_DEFRAG))
1169                 seq_puts(seq, ",autodefrag");
1170         if (btrfs_test_opt(root, INODE_MAP_CACHE))
1171                 seq_puts(seq, ",inode_cache");
1172         if (btrfs_test_opt(root, SKIP_BALANCE))
1173                 seq_puts(seq, ",skip_balance");
1174         if (btrfs_test_opt(root, RECOVERY))
1175                 seq_puts(seq, ",recovery");
1176 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1177         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1178                 seq_puts(seq, ",check_int_data");
1179         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1180                 seq_puts(seq, ",check_int");
1181         if (info->check_integrity_print_mask)
1182                 seq_printf(seq, ",check_int_print_mask=%d",
1183                                 info->check_integrity_print_mask);
1184 #endif
1185         if (info->metadata_ratio)
1186                 seq_printf(seq, ",metadata_ratio=%d",
1187                                 info->metadata_ratio);
1188         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1189                 seq_puts(seq, ",fatal_errors=panic");
1190         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1191                 seq_printf(seq, ",commit=%d", info->commit_interval);
1192         seq_printf(seq, ",subvolid=%llu",
1193                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1194         seq_puts(seq, ",subvol=");
1195         seq_dentry(seq, dentry, " \t\n\\");
1196         return 0;
1197 }
1198
1199 static int btrfs_test_super(struct super_block *s, void *data)
1200 {
1201         struct btrfs_fs_info *p = data;
1202         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1203
1204         return fs_info->fs_devices == p->fs_devices;
1205 }
1206
1207 static int btrfs_set_super(struct super_block *s, void *data)
1208 {
1209         int err = set_anon_super(s, data);
1210         if (!err)
1211                 s->s_fs_info = data;
1212         return err;
1213 }
1214
1215 /*
1216  * subvolumes are identified by ino 256
1217  */
1218 static inline int is_subvolume_inode(struct inode *inode)
1219 {
1220         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1221                 return 1;
1222         return 0;
1223 }
1224
1225 /*
1226  * This will add subvolid=0 to the argument string while removing any subvol=
1227  * and subvolid= arguments to make sure we get the top-level root for path
1228  * walking to the subvol we want.
1229  */
1230 static char *setup_root_args(char *args)
1231 {
1232         char *buf, *dst, *sep;
1233
1234         if (!args)
1235                 return kstrdup("subvolid=0", GFP_NOFS);
1236
1237         /* The worst case is that we add ",subvolid=0" to the end. */
1238         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1239         if (!buf)
1240                 return NULL;
1241
1242         while (1) {
1243                 sep = strchrnul(args, ',');
1244                 if (!strstarts(args, "subvol=") &&
1245                     !strstarts(args, "subvolid=")) {
1246                         memcpy(dst, args, sep - args);
1247                         dst += sep - args;
1248                         *dst++ = ',';
1249                 }
1250                 if (*sep)
1251                         args = sep + 1;
1252                 else
1253                         break;
1254         }
1255         strcpy(dst, "subvolid=0");
1256
1257         return buf;
1258 }
1259
1260 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1261                                    int flags, const char *device_name,
1262                                    char *data)
1263 {
1264         struct dentry *root;
1265         struct vfsmount *mnt = NULL;
1266         char *newargs;
1267         int ret;
1268
1269         newargs = setup_root_args(data);
1270         if (!newargs) {
1271                 root = ERR_PTR(-ENOMEM);
1272                 goto out;
1273         }
1274
1275         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1276         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1277                 if (flags & MS_RDONLY) {
1278                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1279                                              device_name, newargs);
1280                 } else {
1281                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1282                                              device_name, newargs);
1283                         if (IS_ERR(mnt)) {
1284                                 root = ERR_CAST(mnt);
1285                                 mnt = NULL;
1286                                 goto out;
1287                         }
1288
1289                         down_write(&mnt->mnt_sb->s_umount);
1290                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1291                         up_write(&mnt->mnt_sb->s_umount);
1292                         if (ret < 0) {
1293                                 root = ERR_PTR(ret);
1294                                 goto out;
1295                         }
1296                 }
1297         }
1298         if (IS_ERR(mnt)) {
1299                 root = ERR_CAST(mnt);
1300                 mnt = NULL;
1301                 goto out;
1302         }
1303
1304         if (!subvol_name) {
1305                 if (!subvol_objectid) {
1306                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1307                                                           &subvol_objectid);
1308                         if (ret) {
1309                                 root = ERR_PTR(ret);
1310                                 goto out;
1311                         }
1312                 }
1313                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1314                                                             subvol_objectid);
1315                 if (IS_ERR(subvol_name)) {
1316                         root = ERR_CAST(subvol_name);
1317                         subvol_name = NULL;
1318                         goto out;
1319                 }
1320
1321         }
1322
1323         root = mount_subtree(mnt, subvol_name);
1324         /* mount_subtree() drops our reference on the vfsmount. */
1325         mnt = NULL;
1326
1327         if (!IS_ERR(root)) {
1328                 struct super_block *s = root->d_sb;
1329                 struct inode *root_inode = d_inode(root);
1330                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1331
1332                 ret = 0;
1333                 if (!is_subvolume_inode(root_inode)) {
1334                         pr_err("BTRFS: '%s' is not a valid subvolume\n",
1335                                subvol_name);
1336                         ret = -EINVAL;
1337                 }
1338                 if (subvol_objectid && root_objectid != subvol_objectid) {
1339                         /*
1340                          * This will also catch a race condition where a
1341                          * subvolume which was passed by ID is renamed and
1342                          * another subvolume is renamed over the old location.
1343                          */
1344                         pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1345                                subvol_name, subvol_objectid);
1346                         ret = -EINVAL;
1347                 }
1348                 if (ret) {
1349                         dput(root);
1350                         root = ERR_PTR(ret);
1351                         deactivate_locked_super(s);
1352                 }
1353         }
1354
1355 out:
1356         mntput(mnt);
1357         kfree(newargs);
1358         kfree(subvol_name);
1359         return root;
1360 }
1361
1362 static int parse_security_options(char *orig_opts,
1363                                   struct security_mnt_opts *sec_opts)
1364 {
1365         char *secdata = NULL;
1366         int ret = 0;
1367
1368         secdata = alloc_secdata();
1369         if (!secdata)
1370                 return -ENOMEM;
1371         ret = security_sb_copy_data(orig_opts, secdata);
1372         if (ret) {
1373                 free_secdata(secdata);
1374                 return ret;
1375         }
1376         ret = security_sb_parse_opts_str(secdata, sec_opts);
1377         free_secdata(secdata);
1378         return ret;
1379 }
1380
1381 static int setup_security_options(struct btrfs_fs_info *fs_info,
1382                                   struct super_block *sb,
1383                                   struct security_mnt_opts *sec_opts)
1384 {
1385         int ret = 0;
1386
1387         /*
1388          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1389          * is valid.
1390          */
1391         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1392         if (ret)
1393                 return ret;
1394
1395 #ifdef CONFIG_SECURITY
1396         if (!fs_info->security_opts.num_mnt_opts) {
1397                 /* first time security setup, copy sec_opts to fs_info */
1398                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1399         } else {
1400                 /*
1401                  * Since SELinux(the only one supports security_mnt_opts) does
1402                  * NOT support changing context during remount/mount same sb,
1403                  * This must be the same or part of the same security options,
1404                  * just free it.
1405                  */
1406                 security_free_mnt_opts(sec_opts);
1407         }
1408 #endif
1409         return ret;
1410 }
1411
1412 /*
1413  * Find a superblock for the given device / mount point.
1414  *
1415  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1416  *        for multiple device setup.  Make sure to keep it in sync.
1417  */
1418 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1419                 const char *device_name, void *data)
1420 {
1421         struct block_device *bdev = NULL;
1422         struct super_block *s;
1423         struct btrfs_fs_devices *fs_devices = NULL;
1424         struct btrfs_fs_info *fs_info = NULL;
1425         struct security_mnt_opts new_sec_opts;
1426         fmode_t mode = FMODE_READ;
1427         char *subvol_name = NULL;
1428         u64 subvol_objectid = 0;
1429         int error = 0;
1430
1431         if (!(flags & MS_RDONLY))
1432                 mode |= FMODE_WRITE;
1433
1434         error = btrfs_parse_early_options(data, mode, fs_type,
1435                                           &subvol_name, &subvol_objectid,
1436                                           &fs_devices);
1437         if (error) {
1438                 kfree(subvol_name);
1439                 return ERR_PTR(error);
1440         }
1441
1442         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1443                 /* mount_subvol() will free subvol_name. */
1444                 return mount_subvol(subvol_name, subvol_objectid, flags,
1445                                     device_name, data);
1446         }
1447
1448         security_init_mnt_opts(&new_sec_opts);
1449         if (data) {
1450                 error = parse_security_options(data, &new_sec_opts);
1451                 if (error)
1452                         return ERR_PTR(error);
1453         }
1454
1455         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1456         if (error)
1457                 goto error_sec_opts;
1458
1459         /*
1460          * Setup a dummy root and fs_info for test/set super.  This is because
1461          * we don't actually fill this stuff out until open_ctree, but we need
1462          * it for searching for existing supers, so this lets us do that and
1463          * then open_ctree will properly initialize everything later.
1464          */
1465         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1466         if (!fs_info) {
1467                 error = -ENOMEM;
1468                 goto error_sec_opts;
1469         }
1470
1471         fs_info->fs_devices = fs_devices;
1472
1473         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1474         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1475         security_init_mnt_opts(&fs_info->security_opts);
1476         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1477                 error = -ENOMEM;
1478                 goto error_fs_info;
1479         }
1480
1481         error = btrfs_open_devices(fs_devices, mode, fs_type);
1482         if (error)
1483                 goto error_fs_info;
1484
1485         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1486                 error = -EACCES;
1487                 goto error_close_devices;
1488         }
1489
1490         bdev = fs_devices->latest_bdev;
1491         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1492                  fs_info);
1493         if (IS_ERR(s)) {
1494                 error = PTR_ERR(s);
1495                 goto error_close_devices;
1496         }
1497
1498         if (s->s_root) {
1499                 btrfs_close_devices(fs_devices);
1500                 free_fs_info(fs_info);
1501                 if ((flags ^ s->s_flags) & MS_RDONLY)
1502                         error = -EBUSY;
1503         } else {
1504                 char b[BDEVNAME_SIZE];
1505
1506                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1507                 btrfs_sb(s)->bdev_holder = fs_type;
1508                 error = btrfs_fill_super(s, fs_devices, data,
1509                                          flags & MS_SILENT ? 1 : 0);
1510         }
1511         if (error) {
1512                 deactivate_locked_super(s);
1513                 goto error_sec_opts;
1514         }
1515
1516         fs_info = btrfs_sb(s);
1517         error = setup_security_options(fs_info, s, &new_sec_opts);
1518         if (error) {
1519                 deactivate_locked_super(s);
1520                 goto error_sec_opts;
1521         }
1522
1523         return dget(s->s_root);
1524
1525 error_close_devices:
1526         btrfs_close_devices(fs_devices);
1527 error_fs_info:
1528         free_fs_info(fs_info);
1529 error_sec_opts:
1530         security_free_mnt_opts(&new_sec_opts);
1531         return ERR_PTR(error);
1532 }
1533
1534 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1535                                      int new_pool_size, int old_pool_size)
1536 {
1537         if (new_pool_size == old_pool_size)
1538                 return;
1539
1540         fs_info->thread_pool_size = new_pool_size;
1541
1542         btrfs_info(fs_info, "resize thread pool %d -> %d",
1543                old_pool_size, new_pool_size);
1544
1545         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1546         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1547         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1548         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1549         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1550         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1551         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1552                                 new_pool_size);
1553         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1554         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1555         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1556         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1557         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1558                                 new_pool_size);
1559 }
1560
1561 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1562 {
1563         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1564 }
1565
1566 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1567                                        unsigned long old_opts, int flags)
1568 {
1569         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1570             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1571              (flags & MS_RDONLY))) {
1572                 /* wait for any defraggers to finish */
1573                 wait_event(fs_info->transaction_wait,
1574                            (atomic_read(&fs_info->defrag_running) == 0));
1575                 if (flags & MS_RDONLY)
1576                         sync_filesystem(fs_info->sb);
1577         }
1578 }
1579
1580 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1581                                          unsigned long old_opts)
1582 {
1583         /*
1584          * We need cleanup all defragable inodes if the autodefragment is
1585          * close or the fs is R/O.
1586          */
1587         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1588             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1589              (fs_info->sb->s_flags & MS_RDONLY))) {
1590                 btrfs_cleanup_defrag_inodes(fs_info);
1591         }
1592
1593         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1594 }
1595
1596 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1597 {
1598         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1599         struct btrfs_root *root = fs_info->tree_root;
1600         unsigned old_flags = sb->s_flags;
1601         unsigned long old_opts = fs_info->mount_opt;
1602         unsigned long old_compress_type = fs_info->compress_type;
1603         u64 old_max_inline = fs_info->max_inline;
1604         u64 old_alloc_start = fs_info->alloc_start;
1605         int old_thread_pool_size = fs_info->thread_pool_size;
1606         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1607         int ret;
1608
1609         sync_filesystem(sb);
1610         btrfs_remount_prepare(fs_info);
1611
1612         if (data) {
1613                 struct security_mnt_opts new_sec_opts;
1614
1615                 security_init_mnt_opts(&new_sec_opts);
1616                 ret = parse_security_options(data, &new_sec_opts);
1617                 if (ret)
1618                         goto restore;
1619                 ret = setup_security_options(fs_info, sb,
1620                                              &new_sec_opts);
1621                 if (ret) {
1622                         security_free_mnt_opts(&new_sec_opts);
1623                         goto restore;
1624                 }
1625         }
1626
1627         ret = btrfs_parse_options(root, data);
1628         if (ret) {
1629                 ret = -EINVAL;
1630                 goto restore;
1631         }
1632
1633         btrfs_remount_begin(fs_info, old_opts, *flags);
1634         btrfs_resize_thread_pool(fs_info,
1635                 fs_info->thread_pool_size, old_thread_pool_size);
1636
1637         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1638                 goto out;
1639
1640         if (*flags & MS_RDONLY) {
1641                 /*
1642                  * this also happens on 'umount -rf' or on shutdown, when
1643                  * the filesystem is busy.
1644                  */
1645                 cancel_work_sync(&fs_info->async_reclaim_work);
1646
1647                 /* wait for the uuid_scan task to finish */
1648                 down(&fs_info->uuid_tree_rescan_sem);
1649                 /* avoid complains from lockdep et al. */
1650                 up(&fs_info->uuid_tree_rescan_sem);
1651
1652                 sb->s_flags |= MS_RDONLY;
1653
1654                 /*
1655                  * Setting MS_RDONLY will put the cleaner thread to
1656                  * sleep at the next loop if it's already active.
1657                  * If it's already asleep, we'll leave unused block
1658                  * groups on disk until we're mounted read-write again
1659                  * unless we clean them up here.
1660                  */
1661                 btrfs_delete_unused_bgs(fs_info);
1662
1663                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1664                 btrfs_scrub_cancel(fs_info);
1665                 btrfs_pause_balance(fs_info);
1666
1667                 ret = btrfs_commit_super(root);
1668                 if (ret)
1669                         goto restore;
1670         } else {
1671                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1672                         btrfs_err(fs_info,
1673                                 "Remounting read-write after error is not allowed");
1674                         ret = -EINVAL;
1675                         goto restore;
1676                 }
1677                 if (fs_info->fs_devices->rw_devices == 0) {
1678                         ret = -EACCES;
1679                         goto restore;
1680                 }
1681
1682                 if (fs_info->fs_devices->missing_devices >
1683                      fs_info->num_tolerated_disk_barrier_failures &&
1684                     !(*flags & MS_RDONLY)) {
1685                         btrfs_warn(fs_info,
1686                                 "too many missing devices, writeable remount is not allowed");
1687                         ret = -EACCES;
1688                         goto restore;
1689                 }
1690
1691                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1692                         ret = -EINVAL;
1693                         goto restore;
1694                 }
1695
1696                 ret = btrfs_cleanup_fs_roots(fs_info);
1697                 if (ret)
1698                         goto restore;
1699
1700                 /* recover relocation */
1701                 mutex_lock(&fs_info->cleaner_mutex);
1702                 ret = btrfs_recover_relocation(root);
1703                 mutex_unlock(&fs_info->cleaner_mutex);
1704                 if (ret)
1705                         goto restore;
1706
1707                 ret = btrfs_resume_balance_async(fs_info);
1708                 if (ret)
1709                         goto restore;
1710
1711                 ret = btrfs_resume_dev_replace_async(fs_info);
1712                 if (ret) {
1713                         btrfs_warn(fs_info, "failed to resume dev_replace");
1714                         goto restore;
1715                 }
1716
1717                 if (!fs_info->uuid_root) {
1718                         btrfs_info(fs_info, "creating UUID tree");
1719                         ret = btrfs_create_uuid_tree(fs_info);
1720                         if (ret) {
1721                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1722                                 goto restore;
1723                         }
1724                 }
1725                 sb->s_flags &= ~MS_RDONLY;
1726         }
1727 out:
1728         wake_up_process(fs_info->transaction_kthread);
1729         btrfs_remount_cleanup(fs_info, old_opts);
1730         return 0;
1731
1732 restore:
1733         /* We've hit an error - don't reset MS_RDONLY */
1734         if (sb->s_flags & MS_RDONLY)
1735                 old_flags |= MS_RDONLY;
1736         sb->s_flags = old_flags;
1737         fs_info->mount_opt = old_opts;
1738         fs_info->compress_type = old_compress_type;
1739         fs_info->max_inline = old_max_inline;
1740         mutex_lock(&fs_info->chunk_mutex);
1741         fs_info->alloc_start = old_alloc_start;
1742         mutex_unlock(&fs_info->chunk_mutex);
1743         btrfs_resize_thread_pool(fs_info,
1744                 old_thread_pool_size, fs_info->thread_pool_size);
1745         fs_info->metadata_ratio = old_metadata_ratio;
1746         btrfs_remount_cleanup(fs_info, old_opts);
1747         return ret;
1748 }
1749
1750 /* Used to sort the devices by max_avail(descending sort) */
1751 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1752                                        const void *dev_info2)
1753 {
1754         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1755             ((struct btrfs_device_info *)dev_info2)->max_avail)
1756                 return -1;
1757         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1758                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1759                 return 1;
1760         else
1761         return 0;
1762 }
1763
1764 /*
1765  * sort the devices by max_avail, in which max free extent size of each device
1766  * is stored.(Descending Sort)
1767  */
1768 static inline void btrfs_descending_sort_devices(
1769                                         struct btrfs_device_info *devices,
1770                                         size_t nr_devices)
1771 {
1772         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1773              btrfs_cmp_device_free_bytes, NULL);
1774 }
1775
1776 /*
1777  * The helper to calc the free space on the devices that can be used to store
1778  * file data.
1779  */
1780 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1781 {
1782         struct btrfs_fs_info *fs_info = root->fs_info;
1783         struct btrfs_device_info *devices_info;
1784         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1785         struct btrfs_device *device;
1786         u64 skip_space;
1787         u64 type;
1788         u64 avail_space;
1789         u64 used_space;
1790         u64 min_stripe_size;
1791         int min_stripes = 1, num_stripes = 1;
1792         int i = 0, nr_devices;
1793         int ret;
1794
1795         /*
1796          * We aren't under the device list lock, so this is racey-ish, but good
1797          * enough for our purposes.
1798          */
1799         nr_devices = fs_info->fs_devices->open_devices;
1800         if (!nr_devices) {
1801                 smp_mb();
1802                 nr_devices = fs_info->fs_devices->open_devices;
1803                 ASSERT(nr_devices);
1804                 if (!nr_devices) {
1805                         *free_bytes = 0;
1806                         return 0;
1807                 }
1808         }
1809
1810         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1811                                GFP_NOFS);
1812         if (!devices_info)
1813                 return -ENOMEM;
1814
1815         /* calc min stripe number for data space alloction */
1816         type = btrfs_get_alloc_profile(root, 1);
1817         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1818                 min_stripes = 2;
1819                 num_stripes = nr_devices;
1820         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1821                 min_stripes = 2;
1822                 num_stripes = 2;
1823         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1824                 min_stripes = 4;
1825                 num_stripes = 4;
1826         }
1827
1828         if (type & BTRFS_BLOCK_GROUP_DUP)
1829                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1830         else
1831                 min_stripe_size = BTRFS_STRIPE_LEN;
1832
1833         if (fs_info->alloc_start)
1834                 mutex_lock(&fs_devices->device_list_mutex);
1835         rcu_read_lock();
1836         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1837                 if (!device->in_fs_metadata || !device->bdev ||
1838                     device->is_tgtdev_for_dev_replace)
1839                         continue;
1840
1841                 if (i >= nr_devices)
1842                         break;
1843
1844                 avail_space = device->total_bytes - device->bytes_used;
1845
1846                 /* align with stripe_len */
1847                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1848                 avail_space *= BTRFS_STRIPE_LEN;
1849
1850                 /*
1851                  * In order to avoid overwritting the superblock on the drive,
1852                  * btrfs starts at an offset of at least 1MB when doing chunk
1853                  * allocation.
1854                  */
1855                 skip_space = 1024 * 1024;
1856
1857                 /* user can set the offset in fs_info->alloc_start. */
1858                 if (fs_info->alloc_start &&
1859                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1860                     device->total_bytes) {
1861                         rcu_read_unlock();
1862                         skip_space = max(fs_info->alloc_start, skip_space);
1863
1864                         /*
1865                          * btrfs can not use the free space in
1866                          * [0, skip_space - 1], we must subtract it from the
1867                          * total. In order to implement it, we account the used
1868                          * space in this range first.
1869                          */
1870                         ret = btrfs_account_dev_extents_size(device, 0,
1871                                                              skip_space - 1,
1872                                                              &used_space);
1873                         if (ret) {
1874                                 kfree(devices_info);
1875                                 mutex_unlock(&fs_devices->device_list_mutex);
1876                                 return ret;
1877                         }
1878
1879                         rcu_read_lock();
1880
1881                         /* calc the free space in [0, skip_space - 1] */
1882                         skip_space -= used_space;
1883                 }
1884
1885                 /*
1886                  * we can use the free space in [0, skip_space - 1], subtract
1887                  * it from the total.
1888                  */
1889                 if (avail_space && avail_space >= skip_space)
1890                         avail_space -= skip_space;
1891                 else
1892                         avail_space = 0;
1893
1894                 if (avail_space < min_stripe_size)
1895                         continue;
1896
1897                 devices_info[i].dev = device;
1898                 devices_info[i].max_avail = avail_space;
1899
1900                 i++;
1901         }
1902         rcu_read_unlock();
1903         if (fs_info->alloc_start)
1904                 mutex_unlock(&fs_devices->device_list_mutex);
1905
1906         nr_devices = i;
1907
1908         btrfs_descending_sort_devices(devices_info, nr_devices);
1909
1910         i = nr_devices - 1;
1911         avail_space = 0;
1912         while (nr_devices >= min_stripes) {
1913                 if (num_stripes > nr_devices)
1914                         num_stripes = nr_devices;
1915
1916                 if (devices_info[i].max_avail >= min_stripe_size) {
1917                         int j;
1918                         u64 alloc_size;
1919
1920                         avail_space += devices_info[i].max_avail * num_stripes;
1921                         alloc_size = devices_info[i].max_avail;
1922                         for (j = i + 1 - num_stripes; j <= i; j++)
1923                                 devices_info[j].max_avail -= alloc_size;
1924                 }
1925                 i--;
1926                 nr_devices--;
1927         }
1928
1929         kfree(devices_info);
1930         *free_bytes = avail_space;
1931         return 0;
1932 }
1933
1934 /*
1935  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
1936  *
1937  * If there's a redundant raid level at DATA block groups, use the respective
1938  * multiplier to scale the sizes.
1939  *
1940  * Unused device space usage is based on simulating the chunk allocator
1941  * algorithm that respects the device sizes, order of allocations and the
1942  * 'alloc_start' value, this is a close approximation of the actual use but
1943  * there are other factors that may change the result (like a new metadata
1944  * chunk).
1945  *
1946  * FIXME: not accurate for mixed block groups, total and free/used are ok,
1947  * available appears slightly larger.
1948  */
1949 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1950 {
1951         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1952         struct btrfs_super_block *disk_super = fs_info->super_copy;
1953         struct list_head *head = &fs_info->space_info;
1954         struct btrfs_space_info *found;
1955         u64 total_used = 0;
1956         u64 total_free_data = 0;
1957         int bits = dentry->d_sb->s_blocksize_bits;
1958         __be32 *fsid = (__be32 *)fs_info->fsid;
1959         unsigned factor = 1;
1960         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
1961         int ret;
1962
1963         /*
1964          * holding chunk_muext to avoid allocating new chunks, holding
1965          * device_list_mutex to avoid the device being removed
1966          */
1967         rcu_read_lock();
1968         list_for_each_entry_rcu(found, head, list) {
1969                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1970                         int i;
1971
1972                         total_free_data += found->disk_total - found->disk_used;
1973                         total_free_data -=
1974                                 btrfs_account_ro_block_groups_free_space(found);
1975
1976                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1977                                 if (!list_empty(&found->block_groups[i])) {
1978                                         switch (i) {
1979                                         case BTRFS_RAID_DUP:
1980                                         case BTRFS_RAID_RAID1:
1981                                         case BTRFS_RAID_RAID10:
1982                                                 factor = 2;
1983                                         }
1984                                 }
1985                         }
1986                 }
1987
1988                 total_used += found->disk_used;
1989         }
1990
1991         rcu_read_unlock();
1992
1993         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
1994         buf->f_blocks >>= bits;
1995         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
1996
1997         /* Account global block reserve as used, it's in logical size already */
1998         spin_lock(&block_rsv->lock);
1999         buf->f_bfree -= block_rsv->size >> bits;
2000         spin_unlock(&block_rsv->lock);
2001
2002         buf->f_bavail = div_u64(total_free_data, factor);
2003         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2004         if (ret)
2005                 return ret;
2006         buf->f_bavail += div_u64(total_free_data, factor);
2007         buf->f_bavail = buf->f_bavail >> bits;
2008
2009         buf->f_type = BTRFS_SUPER_MAGIC;
2010         buf->f_bsize = dentry->d_sb->s_blocksize;
2011         buf->f_namelen = BTRFS_NAME_LEN;
2012
2013         /* We treat it as constant endianness (it doesn't matter _which_)
2014            because we want the fsid to come out the same whether mounted
2015            on a big-endian or little-endian host */
2016         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2017         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2018         /* Mask in the root object ID too, to disambiguate subvols */
2019         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2020         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2021
2022         return 0;
2023 }
2024
2025 static void btrfs_kill_super(struct super_block *sb)
2026 {
2027         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2028         kill_anon_super(sb);
2029         free_fs_info(fs_info);
2030 }
2031
2032 static struct file_system_type btrfs_fs_type = {
2033         .owner          = THIS_MODULE,
2034         .name           = "btrfs",
2035         .mount          = btrfs_mount,
2036         .kill_sb        = btrfs_kill_super,
2037         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2038 };
2039 MODULE_ALIAS_FS("btrfs");
2040
2041 static int btrfs_control_open(struct inode *inode, struct file *file)
2042 {
2043         /*
2044          * The control file's private_data is used to hold the
2045          * transaction when it is started and is used to keep
2046          * track of whether a transaction is already in progress.
2047          */
2048         file->private_data = NULL;
2049         return 0;
2050 }
2051
2052 /*
2053  * used by btrfsctl to scan devices when no FS is mounted
2054  */
2055 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2056                                 unsigned long arg)
2057 {
2058         struct btrfs_ioctl_vol_args *vol;
2059         struct btrfs_fs_devices *fs_devices;
2060         int ret = -ENOTTY;
2061
2062         if (!capable(CAP_SYS_ADMIN))
2063                 return -EPERM;
2064
2065         vol = memdup_user((void __user *)arg, sizeof(*vol));
2066         if (IS_ERR(vol))
2067                 return PTR_ERR(vol);
2068
2069         switch (cmd) {
2070         case BTRFS_IOC_SCAN_DEV:
2071                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2072                                             &btrfs_fs_type, &fs_devices);
2073                 break;
2074         case BTRFS_IOC_DEVICES_READY:
2075                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2076                                             &btrfs_fs_type, &fs_devices);
2077                 if (ret)
2078                         break;
2079                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2080                 break;
2081         }
2082
2083         kfree(vol);
2084         return ret;
2085 }
2086
2087 static int btrfs_freeze(struct super_block *sb)
2088 {
2089         struct btrfs_trans_handle *trans;
2090         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2091
2092         trans = btrfs_attach_transaction_barrier(root);
2093         if (IS_ERR(trans)) {
2094                 /* no transaction, don't bother */
2095                 if (PTR_ERR(trans) == -ENOENT)
2096                         return 0;
2097                 return PTR_ERR(trans);
2098         }
2099         return btrfs_commit_transaction(trans, root);
2100 }
2101
2102 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2103 {
2104         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2105         struct btrfs_fs_devices *cur_devices;
2106         struct btrfs_device *dev, *first_dev = NULL;
2107         struct list_head *head;
2108         struct rcu_string *name;
2109
2110         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2111         cur_devices = fs_info->fs_devices;
2112         while (cur_devices) {
2113                 head = &cur_devices->devices;
2114                 list_for_each_entry(dev, head, dev_list) {
2115                         if (dev->missing)
2116                                 continue;
2117                         if (!dev->name)
2118                                 continue;
2119                         if (!first_dev || dev->devid < first_dev->devid)
2120                                 first_dev = dev;
2121                 }
2122                 cur_devices = cur_devices->seed;
2123         }
2124
2125         if (first_dev) {
2126                 rcu_read_lock();
2127                 name = rcu_dereference(first_dev->name);
2128                 seq_escape(m, name->str, " \t\n\\");
2129                 rcu_read_unlock();
2130         } else {
2131                 WARN_ON(1);
2132         }
2133         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2134         return 0;
2135 }
2136
2137 static const struct super_operations btrfs_super_ops = {
2138         .drop_inode     = btrfs_drop_inode,
2139         .evict_inode    = btrfs_evict_inode,
2140         .put_super      = btrfs_put_super,
2141         .sync_fs        = btrfs_sync_fs,
2142         .show_options   = btrfs_show_options,
2143         .show_devname   = btrfs_show_devname,
2144         .write_inode    = btrfs_write_inode,
2145         .alloc_inode    = btrfs_alloc_inode,
2146         .destroy_inode  = btrfs_destroy_inode,
2147         .statfs         = btrfs_statfs,
2148         .remount_fs     = btrfs_remount,
2149         .freeze_fs      = btrfs_freeze,
2150 };
2151
2152 static const struct file_operations btrfs_ctl_fops = {
2153         .open = btrfs_control_open,
2154         .unlocked_ioctl  = btrfs_control_ioctl,
2155         .compat_ioctl = btrfs_control_ioctl,
2156         .owner   = THIS_MODULE,
2157         .llseek = noop_llseek,
2158 };
2159
2160 static struct miscdevice btrfs_misc = {
2161         .minor          = BTRFS_MINOR,
2162         .name           = "btrfs-control",
2163         .fops           = &btrfs_ctl_fops
2164 };
2165
2166 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2167 MODULE_ALIAS("devname:btrfs-control");
2168
2169 static int btrfs_interface_init(void)
2170 {
2171         return misc_register(&btrfs_misc);
2172 }
2173
2174 static void btrfs_interface_exit(void)
2175 {
2176         misc_deregister(&btrfs_misc);
2177 }
2178
2179 static void btrfs_print_info(void)
2180 {
2181         printk(KERN_INFO "Btrfs loaded"
2182 #ifdef CONFIG_BTRFS_DEBUG
2183                         ", debug=on"
2184 #endif
2185 #ifdef CONFIG_BTRFS_ASSERT
2186                         ", assert=on"
2187 #endif
2188 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2189                         ", integrity-checker=on"
2190 #endif
2191                         "\n");
2192 }
2193
2194 static int btrfs_run_sanity_tests(void)
2195 {
2196         int ret;
2197
2198         ret = btrfs_init_test_fs();
2199         if (ret)
2200                 return ret;
2201
2202         ret = btrfs_test_free_space_cache();
2203         if (ret)
2204                 goto out;
2205         ret = btrfs_test_extent_buffer_operations();
2206         if (ret)
2207                 goto out;
2208         ret = btrfs_test_extent_io();
2209         if (ret)
2210                 goto out;
2211         ret = btrfs_test_inodes();
2212         if (ret)
2213                 goto out;
2214         ret = btrfs_test_qgroups();
2215 out:
2216         btrfs_destroy_test_fs();
2217         return ret;
2218 }
2219
2220 static int __init init_btrfs_fs(void)
2221 {
2222         int err;
2223
2224         err = btrfs_hash_init();
2225         if (err)
2226                 return err;
2227
2228         btrfs_props_init();
2229
2230         err = btrfs_init_sysfs();
2231         if (err)
2232                 goto free_hash;
2233
2234         btrfs_init_compress();
2235
2236         err = btrfs_init_cachep();
2237         if (err)
2238                 goto free_compress;
2239
2240         err = extent_io_init();
2241         if (err)
2242                 goto free_cachep;
2243
2244         err = extent_map_init();
2245         if (err)
2246                 goto free_extent_io;
2247
2248         err = ordered_data_init();
2249         if (err)
2250                 goto free_extent_map;
2251
2252         err = btrfs_delayed_inode_init();
2253         if (err)
2254                 goto free_ordered_data;
2255
2256         err = btrfs_auto_defrag_init();
2257         if (err)
2258                 goto free_delayed_inode;
2259
2260         err = btrfs_delayed_ref_init();
2261         if (err)
2262                 goto free_auto_defrag;
2263
2264         err = btrfs_prelim_ref_init();
2265         if (err)
2266                 goto free_delayed_ref;
2267
2268         err = btrfs_end_io_wq_init();
2269         if (err)
2270                 goto free_prelim_ref;
2271
2272         err = btrfs_interface_init();
2273         if (err)
2274                 goto free_end_io_wq;
2275
2276         btrfs_init_lockdep();
2277
2278         btrfs_print_info();
2279
2280         err = btrfs_run_sanity_tests();
2281         if (err)
2282                 goto unregister_ioctl;
2283
2284         err = register_filesystem(&btrfs_fs_type);
2285         if (err)
2286                 goto unregister_ioctl;
2287
2288         return 0;
2289
2290 unregister_ioctl:
2291         btrfs_interface_exit();
2292 free_end_io_wq:
2293         btrfs_end_io_wq_exit();
2294 free_prelim_ref:
2295         btrfs_prelim_ref_exit();
2296 free_delayed_ref:
2297         btrfs_delayed_ref_exit();
2298 free_auto_defrag:
2299         btrfs_auto_defrag_exit();
2300 free_delayed_inode:
2301         btrfs_delayed_inode_exit();
2302 free_ordered_data:
2303         ordered_data_exit();
2304 free_extent_map:
2305         extent_map_exit();
2306 free_extent_io:
2307         extent_io_exit();
2308 free_cachep:
2309         btrfs_destroy_cachep();
2310 free_compress:
2311         btrfs_exit_compress();
2312         btrfs_exit_sysfs();
2313 free_hash:
2314         btrfs_hash_exit();
2315         return err;
2316 }
2317
2318 static void __exit exit_btrfs_fs(void)
2319 {
2320         btrfs_destroy_cachep();
2321         btrfs_delayed_ref_exit();
2322         btrfs_auto_defrag_exit();
2323         btrfs_delayed_inode_exit();
2324         btrfs_prelim_ref_exit();
2325         ordered_data_exit();
2326         extent_map_exit();
2327         extent_io_exit();
2328         btrfs_interface_exit();
2329         btrfs_end_io_wq_exit();
2330         unregister_filesystem(&btrfs_fs_type);
2331         btrfs_exit_sysfs();
2332         btrfs_cleanup_fs_uuids();
2333         btrfs_exit_compress();
2334         btrfs_hash_exit();
2335 }
2336
2337 late_initcall(init_btrfs_fs);
2338 module_exit(exit_btrfs_fs)
2339
2340 MODULE_LICENSE("GPL");