fs/btrfs/super.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/time.h>
  26 #include <linux/init.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mount.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/parser.h>
  37 #include <linux/ctype.h>
  38 #include <linux/namei.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/magic.h>
  41 #include <linux/slab.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/ratelimit.h>
  44 #include <linux/btrfs.h>
  45 #include "compat.h"
  46 #include "delayed-inode.h"
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "xattr.h"
  53 #include "volumes.h"
  54 #include "export.h"
  55 #include "compression.h"
  56 #include "rcu-string.h"
  57 #include "dev-replace.h"
  58 #include "free-space-cache.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/btrfs.h>
  62
  63 static const struct super_operations btrfs_super_ops;
  64 static struct file_system_type btrfs_fs_type;
  65
  66 static const char *btrfs_decode_error(int errno)
  67 {
  68         char *errstr = "unknown";
  69
  70         switch (errno) {
  71         case -EIO:
  72                 errstr = "IO failure";
  73                 break;
  74         case -ENOMEM:
  75                 errstr = "Out of memory";
  76                 break;
  77         case -EROFS:
  78                 errstr = "Readonly filesystem";
  79                 break;
  80         case -EEXIST:
  81                 errstr = "Object already exists";
  82                 break;
  83         case -ENOSPC:
  84                 errstr = "No space left";
  85                 break;
  86         case -ENOENT:
  87                 errstr = "No such entry";
  88                 break;
  89         }
  90
  91         return errstr;
  92 }
  93
  94 static void save_error_info(struct btrfs_fs_info *fs_info)
  95 {
  96         /*
  97          * today we only save the error info into ram.  Long term we'll
  98          * also send it down to the disk
  99          */
 100         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 101 }
 102
 103 /* btrfs handle error by forcing the filesystem readonly */
 104 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 105 {
 106         struct super_block *sb = fs_info->sb;
 107
 108         if (sb->s_flags & MS_RDONLY)
 109                 return;
 110
 111         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 112                 sb->s_flags |= MS_RDONLY;
 113                 btrfs_info(fs_info, "forced readonly");
 114                 /*
 115                  * Note that a running device replace operation is not
 116                  * canceled here although there is no way to update
 117                  * the progress. It would add the risk of a deadlock,
 118                  * therefore the canceling is ommited. The only penalty
 119                  * is that some I/O remains active until the procedure
 120                  * completes. The next time when the filesystem is
 121                  * mounted writeable again, the device replace
 122                  * operation continues.
 123                  */
 124         }
 125 }
 126
 127 #ifdef CONFIG_PRINTK
 128 /*
 129  * __btrfs_std_error decodes expected errors from the caller and
 130  * invokes the approciate error response.
 131  */
 132 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 133                        unsigned int line, int errno, const char *fmt, ...)
 134 {
 135         struct super_block *sb = fs_info->sb;
 136         const char *errstr;
 137
 138         /*
 139          * Special case: if the error is EROFS, and we're already
 140          * under MS_RDONLY, then it is safe here.
 141          */
 142         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 143                 return;
 144
 145         errstr = btrfs_decode_error(errno);
 146         if (fmt) {
 147                 struct va_format vaf;
 148                 va_list args;
 149
 150                 va_start(args, fmt);
 151                 vaf.fmt = fmt;
 152                 vaf.va = &args;
 153
 154                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 155                         sb->s_id, function, line, errno, errstr, &vaf);
 156                 va_end(args);
 157         } else {
 158                 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: errno=%d %s\n",
 159                         sb->s_id, function, line, errno, errstr);
 160         }
 161
 162         /* Don't go through full error handling during mount */
 163         save_error_info(fs_info);
 164         if (sb->s_flags & MS_BORN)
 165                 btrfs_handle_error(fs_info);
 166 }
 167
 168 static const char * const logtypes[] = {
 169         "emergency",
 170         "alert",
 171         "critical",
 172         "error",
 173         "warning",
 174         "notice",
 175         "info",
 176         "debug",
 177 };
 178
 179 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 180 {
 181         struct super_block *sb = fs_info->sb;
 182         char lvl[4];
 183         struct va_format vaf;
 184         va_list args;
 185         const char *type = logtypes[4];
 186         int kern_level;
 187
 188         va_start(args, fmt);
 189
 190         kern_level = printk_get_level(fmt);
 191         if (kern_level) {
 192                 size_t size = printk_skip_level(fmt) - fmt;
 193                 memcpy(lvl, fmt,  size);
 194                 lvl[size] = '\0';
 195                 fmt += size;
 196                 type = logtypes[kern_level - '0'];
 197         } else
 198                 *lvl = '\0';
 199
 200         vaf.fmt = fmt;
 201         vaf.va = &args;
 202
 203         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 204
 205         va_end(args);
 206 }
 207
 208 #else
 209
 210 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 211                        unsigned int line, int errno, const char *fmt, ...)
 212 {
 213         struct super_block *sb = fs_info->sb;
 214
 215         /*
 216          * Special case: if the error is EROFS, and we're already
 217          * under MS_RDONLY, then it is safe here.
 218          */
 219         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 220                 return;
 221
 222         /* Don't go through full error handling during mount */
 223         if (sb->s_flags & MS_BORN) {
 224                 save_error_info(fs_info);
 225                 btrfs_handle_error(fs_info);
 226         }
 227 }
 228 #endif
 229
 230 /*
 231  * We only mark the transaction aborted and then set the file system read-only.
 232  * This will prevent new transactions from starting or trying to join this
 233  * one.
 234  *
 235  * This means that error recovery at the call site is limited to freeing
 236  * any local memory allocations and passing the error code up without
 237  * further cleanup. The transaction should complete as it normally would
 238  * in the call path but will return -EIO.
 239  *
 240  * We'll complete the cleanup in btrfs_end_transaction and
 241  * btrfs_commit_transaction.
 242  */
 243 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 244                                struct btrfs_root *root, const char *function,
 245                                unsigned int line, int errno)
 246 {
 247         /*
 248          * Report first abort since mount
 249          */
 250         if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,
 251                                 &root->fs_info->fs_state)) {
 252                 WARN(1, KERN_DEBUG "btrfs: Transaction aborted (error %d)\n",
 253                                 errno);
 254         }
 255         trans->aborted = errno;
 256         /* Nothing used. The other threads that have joined this
 257          * transaction may be able to continue. */
 258         if (!trans->blocks_used) {
 259                 const char *errstr;
 260
 261                 errstr = btrfs_decode_error(errno);
 262                 btrfs_warn(root->fs_info,
 263                            "%s:%d: Aborting unused transaction(%s).",
 264                            function, line, errstr);
 265                 return;
 266         }
 267         ACCESS_ONCE(trans->transaction->aborted) = errno;
 268         __btrfs_std_error(root->fs_info, function, line, errno, NULL);
 269 }
 270 /*
 271  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 272  * issues an alert, and either panics or BUGs, depending on mount options.
 273  */
 274 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 275                    unsigned int line, int errno, const char *fmt, ...)
 276 {
 277         char *s_id = "<unknown>";
 278         const char *errstr;
 279         struct va_format vaf = { .fmt = fmt };
 280         va_list args;
 281
 282         if (fs_info)
 283                 s_id = fs_info->sb->s_id;
 284
 285         va_start(args, fmt);
 286         vaf.va = &args;
 287
 288         errstr = btrfs_decode_error(errno);
 289         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 290                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 291                         s_id, function, line, &vaf, errno, errstr);
 292
 293         printk(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 294                s_id, function, line, &vaf, errno, errstr);
 295         va_end(args);
 296         /* Caller calls BUG() */
 297 }
 298
 299 static void btrfs_put_super(struct super_block *sb)
 300 {
 301         (void)close_ctree(btrfs_sb(sb)->tree_root);
 302         /* FIXME: need to fix VFS to return error? */
 303         /* AV: return it _where_?  ->put_super() can be triggered by any number
 304          * of async events, up to and including delivery of SIGKILL to the
 305          * last process that kept it busy.  Or segfault in the aforementioned
 306          * process...  Whom would you report that to?
 307          */
 308 }
 309
 310 enum {
 311         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 312         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 313         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 314         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 315         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 316         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 317         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 318         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 319         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 320         Opt_check_integrity_print_mask, Opt_fatal_errors,
 321         Opt_err,
 322 };
 323
 324 static match_table_t tokens = {
 325         {Opt_degraded, "degraded"},
 326         {Opt_subvol, "subvol=%s"},
 327         {Opt_subvolid, "subvolid=%d"},
 328         {Opt_device, "device=%s"},
 329         {Opt_nodatasum, "nodatasum"},
 330         {Opt_nodatacow, "nodatacow"},
 331         {Opt_nobarrier, "nobarrier"},
 332         {Opt_max_inline, "max_inline=%s"},
 333         {Opt_alloc_start, "alloc_start=%s"},
 334         {Opt_thread_pool, "thread_pool=%d"},
 335         {Opt_compress, "compress"},
 336         {Opt_compress_type, "compress=%s"},
 337         {Opt_compress_force, "compress-force"},
 338         {Opt_compress_force_type, "compress-force=%s"},
 339         {Opt_ssd, "ssd"},
 340         {Opt_ssd_spread, "ssd_spread"},
 341         {Opt_nossd, "nossd"},
 342         {Opt_noacl, "noacl"},
 343         {Opt_notreelog, "notreelog"},
 344         {Opt_flushoncommit, "flushoncommit"},
 345         {Opt_ratio, "metadata_ratio=%d"},
 346         {Opt_discard, "discard"},
 347         {Opt_space_cache, "space_cache"},
 348         {Opt_clear_cache, "clear_cache"},
 349         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 350         {Opt_enospc_debug, "enospc_debug"},
 351         {Opt_subvolrootid, "subvolrootid=%d"},
 352         {Opt_defrag, "autodefrag"},
 353         {Opt_inode_cache, "inode_cache"},
 354         {Opt_no_space_cache, "nospace_cache"},
 355         {Opt_recovery, "recovery"},
 356         {Opt_skip_balance, "skip_balance"},
 357         {Opt_check_integrity, "check_int"},
 358         {Opt_check_integrity_including_extent_data, "check_int_data"},
 359         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 360         {Opt_fatal_errors, "fatal_errors=%s"},
 361         {Opt_err, NULL},
 362 };
 363
 364 /*
 365  * Regular mount options parser.  Everything that is needed only when
 366  * reading in a new superblock is parsed here.
 367  * XXX JDM: This needs to be cleaned up for remount.
 368  */
 369 int btrfs_parse_options(struct btrfs_root *root, char *options)
 370 {
 371         struct btrfs_fs_info *info = root->fs_info;
 372         substring_t args[MAX_OPT_ARGS];
 373         char *p, *num, *orig = NULL;
 374         u64 cache_gen;
 375         int intarg;
 376         int ret = 0;
 377         char *compress_type;
 378         bool compress_force = false;
 379
 380         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 381         if (cache_gen)
 382                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 383
 384         if (!options)
 385                 goto out;
 386
 387         /*
 388          * strsep changes the string, duplicate it because parse_options
 389          * gets called twice
 390          */
 391         options = kstrdup(options, GFP_NOFS);
 392         if (!options)
 393                 return -ENOMEM;
 394
 395         orig = options;
 396
 397         while ((p = strsep(&options, ",")) != NULL) {
 398                 int token;
 399                 if (!*p)
 400                         continue;
 401
 402                 token = match_token(p, tokens, args);
 403                 switch (token) {
 404                 case Opt_degraded:
 405                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 406                         btrfs_set_opt(info->mount_opt, DEGRADED);
 407                         break;
 408                 case Opt_subvol:
 409                 case Opt_subvolid:
 410                 case Opt_subvolrootid:
 411                 case Opt_device:
 412                         /*
 413                          * These are parsed by btrfs_parse_early_options
 414                          * and can be happily ignored here.
 415                          */
 416                         break;
 417                 case Opt_nodatasum:
 418                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 419                         btrfs_set_opt(info->mount_opt, NODATASUM);
 420                         break;
 421                 case Opt_nodatacow:
 422                         if (!btrfs_test_opt(root, COMPRESS) ||
 423                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
 424                                         printk(KERN_INFO "btrfs: setting nodatacow, compression disabled\n");
 425                         } else {
 426                                 printk(KERN_INFO "btrfs: setting nodatacow\n");
 427                         }
 428                         info->compress_type = BTRFS_COMPRESS_NONE;
 429                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 430                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 431                         btrfs_set_opt(info->mount_opt, NODATACOW);
 432                         btrfs_set_opt(info->mount_opt, NODATASUM);
 433                         break;
 434                 case Opt_compress_force:
 435                 case Opt_compress_force_type:
 436                         compress_force = true;
 437                         /* Fallthrough */
 438                 case Opt_compress:
 439                 case Opt_compress_type:
 440                         if (token == Opt_compress ||
 441                             token == Opt_compress_force ||
 442                             strcmp(args[0].from, "zlib") == 0) {
 443                                 compress_type = "zlib";
 444                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 445                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 446                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 447                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 448                         } else if (strcmp(args[0].from, "lzo") == 0) {
 449                                 compress_type = "lzo";
 450                                 info->compress_type = BTRFS_COMPRESS_LZO;
 451                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 452                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 453                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 454                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 455                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 456                                 compress_type = "no";
 457                                 info->compress_type = BTRFS_COMPRESS_NONE;
 458                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 459                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 460                                 compress_force = false;
 461                         } else {
 462                                 ret = -EINVAL;
 463                                 goto out;
 464                         }
 465
 466                         if (compress_force) {
 467                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 468                                 pr_info("btrfs: force %s compression\n",
 469                                         compress_type);
 470                         } else
 471                                 pr_info("btrfs: use %s compression\n",
 472                                         compress_type);
 473                         break;
 474                 case Opt_ssd:
 475                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 476                         btrfs_set_opt(info->mount_opt, SSD);
 477                         break;
 478                 case Opt_ssd_spread:
 479                         printk(KERN_INFO "btrfs: use spread ssd "
 480                                "allocation scheme\n");
 481                         btrfs_set_opt(info->mount_opt, SSD);
 482                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 483                         break;
 484                 case Opt_nossd:
 485                         printk(KERN_INFO "btrfs: not using ssd allocation "
 486                                "scheme\n");
 487                         btrfs_set_opt(info->mount_opt, NOSSD);
 488                         btrfs_clear_opt(info->mount_opt, SSD);
 489                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 490                         break;
 491                 case Opt_nobarrier:
 492                         printk(KERN_INFO "btrfs: turning off barriers\n");
 493                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 494                         break;
 495                 case Opt_thread_pool:
 496                         intarg = 0;
 497                         match_int(&args[0], &intarg);
 498                         if (intarg)
 499                                 info->thread_pool_size = intarg;
 500                         break;
 501                 case Opt_max_inline:
 502                         num = match_strdup(&args[0]);
 503                         if (num) {
 504                                 info->max_inline = memparse(num, NULL);
 505                                 kfree(num);
 506
 507                                 if (info->max_inline) {
 508                                         info->max_inline = max_t(u64,
 509                                                 info->max_inline,
 510                                                 root->sectorsize);
 511                                 }
 512                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 513                                         (unsigned long long)info->max_inline);
 514                         }
 515                         break;
 516                 case Opt_alloc_start:
 517                         num = match_strdup(&args[0]);
 518                         if (num) {
 519                                 mutex_lock(&info->chunk_mutex);
 520                                 info->alloc_start = memparse(num, NULL);
 521                                 mutex_unlock(&info->chunk_mutex);
 522                                 kfree(num);
 523                                 printk(KERN_INFO
 524                                         "btrfs: allocations start at %llu\n",
 525                                         (unsigned long long)info->alloc_start);
 526                         }
 527                         break;
 528                 case Opt_noacl:
 529                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 530                         break;
 531                 case Opt_notreelog:
 532                         printk(KERN_INFO "btrfs: disabling tree log\n");
 533                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 534                         break;
 535                 case Opt_flushoncommit:
 536                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 537                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 538                         break;
 539                 case Opt_ratio:
 540                         intarg = 0;
 541                         match_int(&args[0], &intarg);
 542                         if (intarg) {
 543                                 info->metadata_ratio = intarg;
 544                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 545                                        info->metadata_ratio);
 546                         }
 547                         break;
 548                 case Opt_discard:
 549                         btrfs_set_opt(info->mount_opt, DISCARD);
 550                         break;
 551                 case Opt_space_cache:
 552                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 553                         break;
 554                 case Opt_no_space_cache:
 555                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 556                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 557                         break;
 558                 case Opt_inode_cache:
 559                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 560                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 561                         break;
 562                 case Opt_clear_cache:
 563                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 564                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 565                         break;
 566                 case Opt_user_subvol_rm_allowed:
 567                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 568                         break;
 569                 case Opt_enospc_debug:
 570                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 571                         break;
 572                 case Opt_defrag:
 573                         printk(KERN_INFO "btrfs: enabling auto defrag\n");
 574                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 575                         break;
 576                 case Opt_recovery:
 577                         printk(KERN_INFO "btrfs: enabling auto recovery\n");
 578                         btrfs_set_opt(info->mount_opt, RECOVERY);
 579                         break;
 580                 case Opt_skip_balance:
 581                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 582                         break;
 583 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 584                 case Opt_check_integrity_including_extent_data:
 585                         printk(KERN_INFO "btrfs: enabling check integrity"
 586                                " including extent data\n");
 587                         btrfs_set_opt(info->mount_opt,
 588                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 589                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 590                         break;
 591                 case Opt_check_integrity:
 592                         printk(KERN_INFO "btrfs: enabling check integrity\n");
 593                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 594                         break;
 595                 case Opt_check_integrity_print_mask:
 596                         intarg = 0;
 597                         match_int(&args[0], &intarg);
 598                         if (intarg) {
 599                                 info->check_integrity_print_mask = intarg;
 600                                 printk(KERN_INFO "btrfs:"
 601                                        " check_integrity_print_mask 0x%x\n",
 602                                        info->check_integrity_print_mask);
 603                         }
 604                         break;
 605 #else
 606                 case Opt_check_integrity_including_extent_data:
 607                 case Opt_check_integrity:
 608                 case Opt_check_integrity_print_mask:
 609                         printk(KERN_ERR "btrfs: support for check_integrity*"
 610                                " not compiled in!\n");
 611                         ret = -EINVAL;
 612                         goto out;
 613 #endif
 614                 case Opt_fatal_errors:
 615                         if (strcmp(args[0].from, "panic") == 0)
 616                                 btrfs_set_opt(info->mount_opt,
 617                                               PANIC_ON_FATAL_ERROR);
 618                         else if (strcmp(args[0].from, "bug") == 0)
 619                                 btrfs_clear_opt(info->mount_opt,
 620                                               PANIC_ON_FATAL_ERROR);
 621                         else {
 622                                 ret = -EINVAL;
 623                                 goto out;
 624                         }
 625                         break;
 626                 case Opt_err:
 627                         printk(KERN_INFO "btrfs: unrecognized mount option "
 628                                "'%s'\n", p);
 629                         ret = -EINVAL;
 630                         goto out;
 631                 default:
 632                         break;
 633                 }
 634         }
 635 out:
 636         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 637                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 638         kfree(orig);
 639         return ret;
 640 }
 641
 642 /*
 643  * Parse mount options that are required early in the mount process.
 644  *
 645  * All other options will be parsed on much later in the mount process and
 646  * only when we need to allocate a new super block.
 647  */
 648 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 649                 void *holder, char **subvol_name, u64 *subvol_objectid,
 650                 struct btrfs_fs_devices **fs_devices)
 651 {
 652         substring_t args[MAX_OPT_ARGS];
 653         char *device_name, *opts, *orig, *p;
 654         int error = 0;
 655         int intarg;
 656
 657         if (!options)
 658                 return 0;
 659
 660         /*
 661          * strsep changes the string, duplicate it because parse_options
 662          * gets called twice
 663          */
 664         opts = kstrdup(options, GFP_KERNEL);
 665         if (!opts)
 666                 return -ENOMEM;
 667         orig = opts;
 668
 669         while ((p = strsep(&opts, ",")) != NULL) {
 670                 int token;
 671                 if (!*p)
 672                         continue;
 673
 674                 token = match_token(p, tokens, args);
 675                 switch (token) {
 676                 case Opt_subvol:
 677                         kfree(*subvol_name);
 678                         *subvol_name = match_strdup(&args[0]);
 679                         break;
 680                 case Opt_subvolid:
 681                         intarg = 0;
 682                         error = match_int(&args[0], &intarg);
 683                         if (!error) {
 684                                 /* we want the original fs_tree */
 685                                 if (!intarg)
 686                                         *subvol_objectid =
 687                                                 BTRFS_FS_TREE_OBJECTID;
 688                                 else
 689                                         *subvol_objectid = intarg;
 690                         }
 691                         break;
 692                 case Opt_subvolrootid:
 693                         printk(KERN_WARNING
 694                                 "btrfs: 'subvolrootid' mount option is deprecated and has no effect\n");
 695                         break;
 696                 case Opt_device:
 697                         device_name = match_strdup(&args[0]);
 698                         if (!device_name) {
 699                                 error = -ENOMEM;
 700                                 goto out;
 701                         }
 702                         error = btrfs_scan_one_device(device_name,
 703                                         flags, holder, fs_devices);
 704                         kfree(device_name);
 705                         if (error)
 706                                 goto out;
 707                         break;
 708                 default:
 709                         break;
 710                 }
 711         }
 712
 713 out:
 714         kfree(orig);
 715         return error;
 716 }
 717
 718 static struct dentry *get_default_root(struct super_block *sb,
 719                                        u64 subvol_objectid)
 720 {
 721         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 722         struct btrfs_root *root = fs_info->tree_root;
 723         struct btrfs_root *new_root;
 724         struct btrfs_dir_item *di;
 725         struct btrfs_path *path;
 726         struct btrfs_key location;
 727         struct inode *inode;
 728         u64 dir_id;
 729         int new = 0;
 730
 731         /*
 732          * We have a specific subvol we want to mount, just setup location and
 733          * go look up the root.
 734          */
 735         if (subvol_objectid) {
 736                 location.objectid = subvol_objectid;
 737                 location.type = BTRFS_ROOT_ITEM_KEY;
 738                 location.offset = (u64)-1;
 739                 goto find_root;
 740         }
 741
 742         path = btrfs_alloc_path();
 743         if (!path)
 744                 return ERR_PTR(-ENOMEM);
 745         path->leave_spinning = 1;
 746
 747         /*
 748          * Find the "default" dir item which points to the root item that we
 749          * will mount by default if we haven't been given a specific subvolume
 750          * to mount.
 751          */
 752         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 753         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 754         if (IS_ERR(di)) {
 755                 btrfs_free_path(path);
 756                 return ERR_CAST(di);
 757         }
 758         if (!di) {
 759                 /*
 760                  * Ok the default dir item isn't there.  This is weird since
 761                  * it's always been there, but don't freak out, just try and
 762                  * mount to root most subvolume.
 763                  */
 764                 btrfs_free_path(path);
 765                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 766                 new_root = fs_info->fs_root;
 767                 goto setup_root;
 768         }
 769
 770         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 771         btrfs_free_path(path);
 772
 773 find_root:
 774         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 775         if (IS_ERR(new_root))
 776                 return ERR_CAST(new_root);
 777
 778         dir_id = btrfs_root_dirid(&new_root->root_item);
 779 setup_root:
 780         location.objectid = dir_id;
 781         location.type = BTRFS_INODE_ITEM_KEY;
 782         location.offset = 0;
 783
 784         inode = btrfs_iget(sb, &location, new_root, &new);
 785         if (IS_ERR(inode))
 786                 return ERR_CAST(inode);
 787
 788         /*
 789          * If we're just mounting the root most subvol put the inode and return
 790          * a reference to the dentry.  We will have already gotten a reference
 791          * to the inode in btrfs_fill_super so we're good to go.
 792          */
 793         if (!new && sb->s_root->d_inode == inode) {
 794                 iput(inode);
 795                 return dget(sb->s_root);
 796         }
 797
 798         return d_obtain_alias(inode);
 799 }
 800
 801 static int btrfs_fill_super(struct super_block *sb,
 802                             struct btrfs_fs_devices *fs_devices,
 803                             void *data, int silent)
 804 {
 805         struct inode *inode;
 806         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 807         struct btrfs_key key;
 808         int err;
 809
 810         sb->s_maxbytes = MAX_LFS_FILESIZE;
 811         sb->s_magic = BTRFS_SUPER_MAGIC;
 812         sb->s_op = &btrfs_super_ops;
 813         sb->s_d_op = &btrfs_dentry_operations;
 814         sb->s_export_op = &btrfs_export_ops;
 815         sb->s_xattr = btrfs_xattr_handlers;
 816         sb->s_time_gran = 1;
 817 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 818         sb->s_flags |= MS_POSIXACL;
 819 #endif
 820         sb->s_flags |= MS_I_VERSION;
 821         err = open_ctree(sb, fs_devices, (char *)data);
 822         if (err) {
 823                 printk("btrfs: open_ctree failed\n");
 824                 return err;
 825         }
 826
 827         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 828         key.type = BTRFS_INODE_ITEM_KEY;
 829         key.offset = 0;
 830         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 831         if (IS_ERR(inode)) {
 832                 err = PTR_ERR(inode);
 833                 goto fail_close;
 834         }
 835
 836         sb->s_root = d_make_root(inode);
 837         if (!sb->s_root) {
 838                 err = -ENOMEM;
 839                 goto fail_close;
 840         }
 841
 842         save_mount_options(sb, data);
 843         cleancache_init_fs(sb);
 844         sb->s_flags |= MS_ACTIVE;
 845         return 0;
 846
 847 fail_close:
 848         close_ctree(fs_info->tree_root);
 849         return err;
 850 }
 851
 852 int btrfs_sync_fs(struct super_block *sb, int wait)
 853 {
 854         struct btrfs_trans_handle *trans;
 855         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 856         struct btrfs_root *root = fs_info->tree_root;
 857
 858         trace_btrfs_sync_fs(wait);
 859
 860         if (!wait) {
 861                 filemap_flush(fs_info->btree_inode->i_mapping);
 862                 return 0;
 863         }
 864
 865         btrfs_wait_all_ordered_extents(fs_info, 0);
 866
 867         trans = btrfs_attach_transaction_barrier(root);
 868         if (IS_ERR(trans)) {
 869                 /* no transaction, don't bother */
 870                 if (PTR_ERR(trans) == -ENOENT)
 871                         return 0;
 872                 return PTR_ERR(trans);
 873         }
 874         return btrfs_commit_transaction(trans, root);
 875 }
 876
 877 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 878 {
 879         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 880         struct btrfs_root *root = info->tree_root;
 881         char *compress_type;
 882
 883         if (btrfs_test_opt(root, DEGRADED))
 884                 seq_puts(seq, ",degraded");
 885         if (btrfs_test_opt(root, NODATASUM))
 886                 seq_puts(seq, ",nodatasum");
 887         if (btrfs_test_opt(root, NODATACOW))
 888                 seq_puts(seq, ",nodatacow");
 889         if (btrfs_test_opt(root, NOBARRIER))
 890                 seq_puts(seq, ",nobarrier");
 891         if (info->max_inline != 8192 * 1024)
 892                 seq_printf(seq, ",max_inline=%llu",
 893                            (unsigned long long)info->max_inline);
 894         if (info->alloc_start != 0)
 895                 seq_printf(seq, ",alloc_start=%llu",
 896                            (unsigned long long)info->alloc_start);
 897         if (info->thread_pool_size !=  min_t(unsigned long,
 898                                              num_online_cpus() + 2, 8))
 899                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 900         if (btrfs_test_opt(root, COMPRESS)) {
 901                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 902                         compress_type = "zlib";
 903                 else
 904                         compress_type = "lzo";
 905                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 906                         seq_printf(seq, ",compress-force=%s", compress_type);
 907                 else
 908                         seq_printf(seq, ",compress=%s", compress_type);
 909         }
 910         if (btrfs_test_opt(root, NOSSD))
 911                 seq_puts(seq, ",nossd");
 912         if (btrfs_test_opt(root, SSD_SPREAD))
 913                 seq_puts(seq, ",ssd_spread");
 914         else if (btrfs_test_opt(root, SSD))
 915                 seq_puts(seq, ",ssd");
 916         if (btrfs_test_opt(root, NOTREELOG))
 917                 seq_puts(seq, ",notreelog");
 918         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 919                 seq_puts(seq, ",flushoncommit");
 920         if (btrfs_test_opt(root, DISCARD))
 921                 seq_puts(seq, ",discard");
 922         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 923                 seq_puts(seq, ",noacl");
 924         if (btrfs_test_opt(root, SPACE_CACHE))
 925                 seq_puts(seq, ",space_cache");
 926         else
 927                 seq_puts(seq, ",nospace_cache");
 928         if (btrfs_test_opt(root, CLEAR_CACHE))
 929                 seq_puts(seq, ",clear_cache");
 930         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 931                 seq_puts(seq, ",user_subvol_rm_allowed");
 932         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 933                 seq_puts(seq, ",enospc_debug");
 934         if (btrfs_test_opt(root, AUTO_DEFRAG))
 935                 seq_puts(seq, ",autodefrag");
 936         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 937                 seq_puts(seq, ",inode_cache");
 938         if (btrfs_test_opt(root, SKIP_BALANCE))
 939                 seq_puts(seq, ",skip_balance");
 940         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
 941                 seq_puts(seq, ",fatal_errors=panic");
 942         return 0;
 943 }
 944
 945 static int btrfs_test_super(struct super_block *s, void *data)
 946 {
 947         struct btrfs_fs_info *p = data;
 948         struct btrfs_fs_info *fs_info = btrfs_sb(s);
 949
 950         return fs_info->fs_devices == p->fs_devices;
 951 }
 952
 953 static int btrfs_set_super(struct super_block *s, void *data)
 954 {
 955         int err = set_anon_super(s, data);
 956         if (!err)
 957                 s->s_fs_info = data;
 958         return err;
 959 }
 960
 961 /*
 962  * subvolumes are identified by ino 256
 963  */
 964 static inline int is_subvolume_inode(struct inode *inode)
 965 {
 966         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 967                 return 1;
 968         return 0;
 969 }
 970
 971 /*
 972  * This will strip out the subvol=%s argument for an argument string and add
 973  * subvolid=0 to make sure we get the actual tree root for path walking to the
 974  * subvol we want.
 975  */
 976 static char *setup_root_args(char *args)
 977 {
 978         unsigned len = strlen(args) + 2 + 1;
 979         char *src, *dst, *buf;
 980
 981         /*
 982          * We need the same args as before, but with this substitution:
 983          * s!subvol=[^,]+!subvolid=0!
 984          *
 985          * Since the replacement string is up to 2 bytes longer than the
 986          * original, allocate strlen(args) + 2 + 1 bytes.
 987          */
 988
 989         src = strstr(args, "subvol=");
 990         /* This shouldn't happen, but just in case.. */
 991         if (!src)
 992                 return NULL;
 993
 994         buf = dst = kmalloc(len, GFP_NOFS);
 995         if (!buf)
 996                 return NULL;
 997
 998         /*
 999          * If the subvol= arg is not at the start of the string,
1000          * copy whatever precedes it into buf.
1001          */
1002         if (src != args) {
1003                 *src++ = '\0';
1004                 strcpy(buf, args);
1005                 dst += strlen(args);
1006         }
1007
1008         strcpy(dst, "subvolid=0");
1009         dst += strlen("subvolid=0");
1010
1011         /*
1012          * If there is a "," after the original subvol=... string,
1013          * copy that suffix into our buffer.  Otherwise, we're done.
1014          */
1015         src = strchr(src, ',');
1016         if (src)
1017                 strcpy(dst, src);
1018
1019         return buf;
1020 }
1021
1022 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1023                                    const char *device_name, char *data)
1024 {
1025         struct dentry *root;
1026         struct vfsmount *mnt;
1027         char *newargs;
1028
1029         newargs = setup_root_args(data);
1030         if (!newargs)
1031                 return ERR_PTR(-ENOMEM);
1032         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1033                              newargs);
1034         kfree(newargs);
1035         if (IS_ERR(mnt))
1036                 return ERR_CAST(mnt);
1037
1038         root = mount_subtree(mnt, subvol_name);
1039
1040         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1041                 struct super_block *s = root->d_sb;
1042                 dput(root);
1043                 root = ERR_PTR(-EINVAL);
1044                 deactivate_locked_super(s);
1045                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
1046                                 subvol_name);
1047         }
1048
1049         return root;
1050 }
1051
1052 /*
1053  * Find a superblock for the given device / mount point.
1054  *
1055  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1056  *        for multiple device setup.  Make sure to keep it in sync.
1057  */
1058 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1059                 const char *device_name, void *data)
1060 {
1061         struct block_device *bdev = NULL;
1062         struct super_block *s;
1063         struct dentry *root;
1064         struct btrfs_fs_devices *fs_devices = NULL;
1065         struct btrfs_fs_info *fs_info = NULL;
1066         fmode_t mode = FMODE_READ;
1067         char *subvol_name = NULL;
1068         u64 subvol_objectid = 0;
1069         int error = 0;
1070
1071         if (!(flags & MS_RDONLY))
1072                 mode |= FMODE_WRITE;
1073
1074         error = btrfs_parse_early_options(data, mode, fs_type,
1075                                           &subvol_name, &subvol_objectid,
1076                                           &fs_devices);
1077         if (error) {
1078                 kfree(subvol_name);
1079                 return ERR_PTR(error);
1080         }
1081
1082         if (subvol_name) {
1083                 root = mount_subvol(subvol_name, flags, device_name, data);
1084                 kfree(subvol_name);
1085                 return root;
1086         }
1087
1088         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1089         if (error)
1090                 return ERR_PTR(error);
1091
1092         /*
1093          * Setup a dummy root and fs_info for test/set super.  This is because
1094          * we don't actually fill this stuff out until open_ctree, but we need
1095          * it for searching for existing supers, so this lets us do that and
1096          * then open_ctree will properly initialize everything later.
1097          */
1098         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1099         if (!fs_info)
1100                 return ERR_PTR(-ENOMEM);
1101
1102         fs_info->fs_devices = fs_devices;
1103
1104         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1105         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1106         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1107                 error = -ENOMEM;
1108                 goto error_fs_info;
1109         }
1110
1111         error = btrfs_open_devices(fs_devices, mode, fs_type);
1112         if (error)
1113                 goto error_fs_info;
1114
1115         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1116                 error = -EACCES;
1117                 goto error_close_devices;
1118         }
1119
1120         bdev = fs_devices->latest_bdev;
1121         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1122                  fs_info);
1123         if (IS_ERR(s)) {
1124                 error = PTR_ERR(s);
1125                 goto error_close_devices;
1126         }
1127
1128         if (s->s_root) {
1129                 btrfs_close_devices(fs_devices);
1130                 free_fs_info(fs_info);
1131                 if ((flags ^ s->s_flags) & MS_RDONLY)
1132                         error = -EBUSY;
1133         } else {
1134                 char b[BDEVNAME_SIZE];
1135
1136                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1137                 btrfs_sb(s)->bdev_holder = fs_type;
1138                 error = btrfs_fill_super(s, fs_devices, data,
1139                                          flags & MS_SILENT ? 1 : 0);
1140         }
1141
1142         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1143         if (IS_ERR(root))
1144                 deactivate_locked_super(s);
1145
1146         return root;
1147
1148 error_close_devices:
1149         btrfs_close_devices(fs_devices);
1150 error_fs_info:
1151         free_fs_info(fs_info);
1152         return ERR_PTR(error);
1153 }
1154
1155 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1156 {
1157         spin_lock_irq(&workers->lock);
1158         workers->max_workers = new_limit;
1159         spin_unlock_irq(&workers->lock);
1160 }
1161
1162 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1163                                      int new_pool_size, int old_pool_size)
1164 {
1165         if (new_pool_size == old_pool_size)
1166                 return;
1167
1168         fs_info->thread_pool_size = new_pool_size;
1169
1170         printk(KERN_INFO "btrfs: resize thread pool %d -> %d\n",
1171                old_pool_size, new_pool_size);
1172
1173         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1174         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1175         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1176         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1177         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1178         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1179         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1180         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1181         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1182         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1183         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1184         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1185         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1186         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1187                               new_pool_size);
1188 }
1189
1190 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1191 {
1192         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1193 }
1194
1195 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1196                                        unsigned long old_opts, int flags)
1197 {
1198         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1199             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1200              (flags & MS_RDONLY))) {
1201                 /* wait for any defraggers to finish */
1202                 wait_event(fs_info->transaction_wait,
1203                            (atomic_read(&fs_info->defrag_running) == 0));
1204                 if (flags & MS_RDONLY)
1205                         sync_filesystem(fs_info->sb);
1206         }
1207 }
1208
1209 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1210                                          unsigned long old_opts)
1211 {
1212         /*
1213          * We need cleanup all defragable inodes if the autodefragment is
1214          * close or the fs is R/O.
1215          */
1216         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1217             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1218              (fs_info->sb->s_flags & MS_RDONLY))) {
1219                 btrfs_cleanup_defrag_inodes(fs_info);
1220         }
1221
1222         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1223 }
1224
1225 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1226 {
1227         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1228         struct btrfs_root *root = fs_info->tree_root;
1229         unsigned old_flags = sb->s_flags;
1230         unsigned long old_opts = fs_info->mount_opt;
1231         unsigned long old_compress_type = fs_info->compress_type;
1232         u64 old_max_inline = fs_info->max_inline;
1233         u64 old_alloc_start = fs_info->alloc_start;
1234         int old_thread_pool_size = fs_info->thread_pool_size;
1235         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1236         int ret;
1237
1238         btrfs_remount_prepare(fs_info);
1239
1240         ret = btrfs_parse_options(root, data);
1241         if (ret) {
1242                 ret = -EINVAL;
1243                 goto restore;
1244         }
1245
1246         btrfs_remount_begin(fs_info, old_opts, *flags);
1247         btrfs_resize_thread_pool(fs_info,
1248                 fs_info->thread_pool_size, old_thread_pool_size);
1249
1250         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1251                 goto out;
1252
1253         if (*flags & MS_RDONLY) {
1254                 /*
1255                  * this also happens on 'umount -rf' or on shutdown, when
1256                  * the filesystem is busy.
1257                  */
1258                 sb->s_flags |= MS_RDONLY;
1259
1260                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1261                 btrfs_scrub_cancel(fs_info);
1262                 btrfs_pause_balance(fs_info);
1263
1264                 ret = btrfs_commit_super(root);
1265                 if (ret)
1266                         goto restore;
1267         } else {
1268                 if (fs_info->fs_devices->rw_devices == 0) {
1269                         ret = -EACCES;
1270                         goto restore;
1271                 }
1272
1273                 if (fs_info->fs_devices->missing_devices >
1274                      fs_info->num_tolerated_disk_barrier_failures &&
1275                     !(*flags & MS_RDONLY)) {
1276                         printk(KERN_WARNING
1277                                "Btrfs: too many missing devices, writeable remount is not allowed\n");
1278                         ret = -EACCES;
1279                         goto restore;
1280                 }
1281
1282                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1283                         ret = -EINVAL;
1284                         goto restore;
1285                 }
1286
1287                 ret = btrfs_cleanup_fs_roots(fs_info);
1288                 if (ret)
1289                         goto restore;
1290
1291                 /* recover relocation */
1292                 ret = btrfs_recover_relocation(root);
1293                 if (ret)
1294                         goto restore;
1295
1296                 ret = btrfs_resume_balance_async(fs_info);
1297                 if (ret)
1298                         goto restore;
1299
1300                 ret = btrfs_resume_dev_replace_async(fs_info);
1301                 if (ret) {
1302                         pr_warn("btrfs: failed to resume dev_replace\n");
1303                         goto restore;
1304                 }
1305                 sb->s_flags &= ~MS_RDONLY;
1306         }
1307 out:
1308         btrfs_remount_cleanup(fs_info, old_opts);
1309         return 0;
1310
1311 restore:
1312         /* We've hit an error - don't reset MS_RDONLY */
1313         if (sb->s_flags & MS_RDONLY)
1314                 old_flags |= MS_RDONLY;
1315         sb->s_flags = old_flags;
1316         fs_info->mount_opt = old_opts;
1317         fs_info->compress_type = old_compress_type;
1318         fs_info->max_inline = old_max_inline;
1319         mutex_lock(&fs_info->chunk_mutex);
1320         fs_info->alloc_start = old_alloc_start;
1321         mutex_unlock(&fs_info->chunk_mutex);
1322         btrfs_resize_thread_pool(fs_info,
1323                 old_thread_pool_size, fs_info->thread_pool_size);
1324         fs_info->metadata_ratio = old_metadata_ratio;
1325         btrfs_remount_cleanup(fs_info, old_opts);
1326         return ret;
1327 }
1328
1329 /* Used to sort the devices by max_avail(descending sort) */
1330 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1331                                        const void *dev_info2)
1332 {
1333         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1334             ((struct btrfs_device_info *)dev_info2)->max_avail)
1335                 return -1;
1336         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1337                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1338                 return 1;
1339         else
1340         return 0;
1341 }
1342
1343 /*
1344  * sort the devices by max_avail, in which max free extent size of each device
1345  * is stored.(Descending Sort)
1346  */
1347 static inline void btrfs_descending_sort_devices(
1348                                         struct btrfs_device_info *devices,
1349                                         size_t nr_devices)
1350 {
1351         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1352              btrfs_cmp_device_free_bytes, NULL);
1353 }
1354
1355 /*
1356  * The helper to calc the free space on the devices that can be used to store
1357  * file data.
1358  */
1359 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1360 {
1361         struct btrfs_fs_info *fs_info = root->fs_info;
1362         struct btrfs_device_info *devices_info;
1363         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1364         struct btrfs_device *device;
1365         u64 skip_space;
1366         u64 type;
1367         u64 avail_space;
1368         u64 used_space;
1369         u64 min_stripe_size;
1370         int min_stripes = 1, num_stripes = 1;
1371         int i = 0, nr_devices;
1372         int ret;
1373
1374         nr_devices = fs_info->fs_devices->open_devices;
1375         BUG_ON(!nr_devices);
1376
1377         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1378                                GFP_NOFS);
1379         if (!devices_info)
1380                 return -ENOMEM;
1381
1382         /* calc min stripe number for data space alloction */
1383         type = btrfs_get_alloc_profile(root, 1);
1384         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1385                 min_stripes = 2;
1386                 num_stripes = nr_devices;
1387         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1388                 min_stripes = 2;
1389                 num_stripes = 2;
1390         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1391                 min_stripes = 4;
1392                 num_stripes = 4;
1393         }
1394
1395         if (type & BTRFS_BLOCK_GROUP_DUP)
1396                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1397         else
1398                 min_stripe_size = BTRFS_STRIPE_LEN;
1399
1400         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1401                 if (!device->in_fs_metadata || !device->bdev ||
1402                     device->is_tgtdev_for_dev_replace)
1403                         continue;
1404
1405                 avail_space = device->total_bytes - device->bytes_used;
1406
1407                 /* align with stripe_len */
1408                 do_div(avail_space, BTRFS_STRIPE_LEN);
1409                 avail_space *= BTRFS_STRIPE_LEN;
1410
1411                 /*
1412                  * In order to avoid overwritting the superblock on the drive,
1413                  * btrfs starts at an offset of at least 1MB when doing chunk
1414                  * allocation.
1415                  */
1416                 skip_space = 1024 * 1024;
1417
1418                 /* user can set the offset in fs_info->alloc_start. */
1419                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1420                     device->total_bytes)
1421                         skip_space = max(fs_info->alloc_start, skip_space);
1422
1423                 /*
1424                  * btrfs can not use the free space in [0, skip_space - 1],
1425                  * we must subtract it from the total. In order to implement
1426                  * it, we account the used space in this range first.
1427                  */
1428                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1429                                                      &used_space);
1430                 if (ret) {
1431                         kfree(devices_info);
1432                         return ret;
1433                 }
1434
1435                 /* calc the free space in [0, skip_space - 1] */
1436                 skip_space -= used_space;
1437
1438                 /*
1439                  * we can use the free space in [0, skip_space - 1], subtract
1440                  * it from the total.
1441                  */
1442                 if (avail_space && avail_space >= skip_space)
1443                         avail_space -= skip_space;
1444                 else
1445                         avail_space = 0;
1446
1447                 if (avail_space < min_stripe_size)
1448                         continue;
1449
1450                 devices_info[i].dev = device;
1451                 devices_info[i].max_avail = avail_space;
1452
1453                 i++;
1454         }
1455
1456         nr_devices = i;
1457
1458         btrfs_descending_sort_devices(devices_info, nr_devices);
1459
1460         i = nr_devices - 1;
1461         avail_space = 0;
1462         while (nr_devices >= min_stripes) {
1463                 if (num_stripes > nr_devices)
1464                         num_stripes = nr_devices;
1465
1466                 if (devices_info[i].max_avail >= min_stripe_size) {
1467                         int j;
1468                         u64 alloc_size;
1469
1470                         avail_space += devices_info[i].max_avail * num_stripes;
1471                         alloc_size = devices_info[i].max_avail;
1472                         for (j = i + 1 - num_stripes; j <= i; j++)
1473                                 devices_info[j].max_avail -= alloc_size;
1474                 }
1475                 i--;
1476                 nr_devices--;
1477         }
1478
1479         kfree(devices_info);
1480         *free_bytes = avail_space;
1481         return 0;
1482 }
1483
1484 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1485 {
1486         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1487         struct btrfs_super_block *disk_super = fs_info->super_copy;
1488         struct list_head *head = &fs_info->space_info;
1489         struct btrfs_space_info *found;
1490         u64 total_used = 0;
1491         u64 total_free_data = 0;
1492         int bits = dentry->d_sb->s_blocksize_bits;
1493         __be32 *fsid = (__be32 *)fs_info->fsid;
1494         int ret;
1495
1496         /* holding chunk_muext to avoid allocating new chunks */
1497         mutex_lock(&fs_info->chunk_mutex);
1498         rcu_read_lock();
1499         list_for_each_entry_rcu(found, head, list) {
1500                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1501                         total_free_data += found->disk_total - found->disk_used;
1502                         total_free_data -=
1503                                 btrfs_account_ro_block_groups_free_space(found);
1504                 }
1505
1506                 total_used += found->disk_used;
1507         }
1508         rcu_read_unlock();
1509
1510         buf->f_namelen = BTRFS_NAME_LEN;
1511         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1512         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1513         buf->f_bsize = dentry->d_sb->s_blocksize;
1514         buf->f_type = BTRFS_SUPER_MAGIC;
1515         buf->f_bavail = total_free_data;
1516         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1517         if (ret) {
1518                 mutex_unlock(&fs_info->chunk_mutex);
1519                 return ret;
1520         }
1521         buf->f_bavail += total_free_data;
1522         buf->f_bavail = buf->f_bavail >> bits;
1523         mutex_unlock(&fs_info->chunk_mutex);
1524
1525         /* We treat it as constant endianness (it doesn't matter _which_)
1526            because we want the fsid to come out the same whether mounted
1527            on a big-endian or little-endian host */
1528         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1529         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1530         /* Mask in the root object ID too, to disambiguate subvols */
1531         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1532         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1533
1534         return 0;
1535 }
1536
1537 static void btrfs_kill_super(struct super_block *sb)
1538 {
1539         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1540         kill_anon_super(sb);
1541         free_fs_info(fs_info);
1542 }
1543
1544 static struct file_system_type btrfs_fs_type = {
1545         .owner          = THIS_MODULE,
1546         .name           = "btrfs",
1547         .mount          = btrfs_mount,
1548         .kill_sb        = btrfs_kill_super,
1549         .fs_flags       = FS_REQUIRES_DEV,
1550 };
1551 MODULE_ALIAS_FS("btrfs");
1552
1553 /*
1554  * used by btrfsctl to scan devices when no FS is mounted
1555  */
1556 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1557                                 unsigned long arg)
1558 {
1559         struct btrfs_ioctl_vol_args *vol;
1560         struct btrfs_fs_devices *fs_devices;
1561         int ret = -ENOTTY;
1562
1563         if (!capable(CAP_SYS_ADMIN))
1564                 return -EPERM;
1565
1566         vol = memdup_user((void __user *)arg, sizeof(*vol));
1567         if (IS_ERR(vol))
1568                 return PTR_ERR(vol);
1569
1570         switch (cmd) {
1571         case BTRFS_IOC_SCAN_DEV:
1572                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1573                                             &btrfs_fs_type, &fs_devices);
1574                 break;
1575         case BTRFS_IOC_DEVICES_READY:
1576                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1577                                             &btrfs_fs_type, &fs_devices);
1578                 if (ret)
1579                         break;
1580                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1581                 break;
1582         }
1583
1584         kfree(vol);
1585         return ret;
1586 }
1587
1588 static int btrfs_freeze(struct super_block *sb)
1589 {
1590         struct btrfs_trans_handle *trans;
1591         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1592
1593         trans = btrfs_attach_transaction_barrier(root);
1594         if (IS_ERR(trans)) {
1595                 /* no transaction, don't bother */
1596                 if (PTR_ERR(trans) == -ENOENT)
1597                         return 0;
1598                 return PTR_ERR(trans);
1599         }
1600         return btrfs_commit_transaction(trans, root);
1601 }
1602
1603 static int btrfs_unfreeze(struct super_block *sb)
1604 {
1605         return 0;
1606 }
1607
1608 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1609 {
1610         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1611         struct btrfs_fs_devices *cur_devices;
1612         struct btrfs_device *dev, *first_dev = NULL;
1613         struct list_head *head;
1614         struct rcu_string *name;
1615
1616         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1617         cur_devices = fs_info->fs_devices;
1618         while (cur_devices) {
1619                 head = &cur_devices->devices;
1620                 list_for_each_entry(dev, head, dev_list) {
1621                         if (dev->missing)
1622                                 continue;
1623                         if (!first_dev || dev->devid < first_dev->devid)
1624                                 first_dev = dev;
1625                 }
1626                 cur_devices = cur_devices->seed;
1627         }
1628
1629         if (first_dev) {
1630                 rcu_read_lock();
1631                 name = rcu_dereference(first_dev->name);
1632                 seq_escape(m, name->str, " \t\n\\");
1633                 rcu_read_unlock();
1634         } else {
1635                 WARN_ON(1);
1636         }
1637         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1638         return 0;
1639 }
1640
1641 static const struct super_operations btrfs_super_ops = {
1642         .drop_inode     = btrfs_drop_inode,
1643         .evict_inode    = btrfs_evict_inode,
1644         .put_super      = btrfs_put_super,
1645         .sync_fs        = btrfs_sync_fs,
1646         .show_options   = btrfs_show_options,
1647         .show_devname   = btrfs_show_devname,
1648         .write_inode    = btrfs_write_inode,
1649         .alloc_inode    = btrfs_alloc_inode,
1650         .destroy_inode  = btrfs_destroy_inode,
1651         .statfs         = btrfs_statfs,
1652         .remount_fs     = btrfs_remount,
1653         .freeze_fs      = btrfs_freeze,
1654         .unfreeze_fs    = btrfs_unfreeze,
1655 };
1656
1657 static const struct file_operations btrfs_ctl_fops = {
1658         .unlocked_ioctl  = btrfs_control_ioctl,
1659         .compat_ioctl = btrfs_control_ioctl,
1660         .owner   = THIS_MODULE,
1661         .llseek = noop_llseek,
1662 };
1663
1664 static struct miscdevice btrfs_misc = {
1665         .minor          = BTRFS_MINOR,
1666         .name           = "btrfs-control",
1667         .fops           = &btrfs_ctl_fops
1668 };
1669
1670 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1671 MODULE_ALIAS("devname:btrfs-control");
1672
1673 static int btrfs_interface_init(void)
1674 {
1675         return misc_register(&btrfs_misc);
1676 }
1677
1678 static void btrfs_interface_exit(void)
1679 {
1680         if (misc_deregister(&btrfs_misc) < 0)
1681                 printk(KERN_INFO "btrfs: misc_deregister failed for control device\n");
1682 }
1683
1684 static void btrfs_print_info(void)
1685 {
1686         printk(KERN_INFO "Btrfs loaded"
1687 #ifdef CONFIG_BTRFS_DEBUG
1688                         ", debug=on"
1689 #endif
1690 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1691                         ", integrity-checker=on"
1692 #endif
1693                         "\n");
1694 }
1695
1696 static int __init init_btrfs_fs(void)
1697 {
1698         int err;
1699
1700         err = btrfs_init_sysfs();
1701         if (err)
1702                 return err;
1703
1704         btrfs_init_compress();
1705
1706         err = btrfs_init_cachep();
1707         if (err)
1708                 goto free_compress;
1709
1710         err = extent_io_init();
1711         if (err)
1712                 goto free_cachep;
1713
1714         err = extent_map_init();
1715         if (err)
1716                 goto free_extent_io;
1717
1718         err = ordered_data_init();
1719         if (err)
1720                 goto free_extent_map;
1721
1722         err = btrfs_delayed_inode_init();
1723         if (err)
1724                 goto free_ordered_data;
1725
1726         err = btrfs_auto_defrag_init();
1727         if (err)
1728                 goto free_delayed_inode;
1729
1730         err = btrfs_delayed_ref_init();
1731         if (err)
1732                 goto free_auto_defrag;
1733
1734         err = btrfs_interface_init();
1735         if (err)
1736                 goto free_delayed_ref;
1737
1738         err = register_filesystem(&btrfs_fs_type);
1739         if (err)
1740                 goto unregister_ioctl;
1741
1742         btrfs_init_lockdep();
1743
1744         btrfs_print_info();
1745         btrfs_test_free_space_cache();
1746
1747         return 0;
1748
1749 unregister_ioctl:
1750         btrfs_interface_exit();
1751 free_delayed_ref:
1752         btrfs_delayed_ref_exit();
1753 free_auto_defrag:
1754         btrfs_auto_defrag_exit();
1755 free_delayed_inode:
1756         btrfs_delayed_inode_exit();
1757 free_ordered_data:
1758         ordered_data_exit();
1759 free_extent_map:
1760         extent_map_exit();
1761 free_extent_io:
1762         extent_io_exit();
1763 free_cachep:
1764         btrfs_destroy_cachep();
1765 free_compress:
1766         btrfs_exit_compress();
1767         btrfs_exit_sysfs();
1768         return err;
1769 }
1770
1771 static void __exit exit_btrfs_fs(void)
1772 {
1773         btrfs_destroy_cachep();
1774         btrfs_delayed_ref_exit();
1775         btrfs_auto_defrag_exit();
1776         btrfs_delayed_inode_exit();
1777         ordered_data_exit();
1778         extent_map_exit();
1779         extent_io_exit();
1780         btrfs_interface_exit();
1781         unregister_filesystem(&btrfs_fs_type);
1782         btrfs_exit_sysfs();
1783         btrfs_cleanup_fs_uuids();
1784         btrfs_exit_compress();
1785 }
1786
1787 module_init(init_btrfs_fs)
1788 module_exit(exit_btrfs_fs)
1789
1790 MODULE_LICENSE("GPL");