fs/btrfs/super.c

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