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