fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd.h>
  24 #include <linux/ext3_fs.h>
  25 #include <linux/ext3_jbd.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38
  39 #include <asm/uaccess.h>
  40
  41 #include "xattr.h"
  42 #include "acl.h"
  43 #include "namei.h"
  44
  45 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  46                              unsigned long journal_devnum);
  47 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  48                                unsigned int);
  49 static void ext3_commit_super (struct super_block * sb,
  50                                struct ext3_super_block * es,
  51                                int sync);
  52 static void ext3_mark_recovery_complete(struct super_block * sb,
  53                                         struct ext3_super_block * es);
  54 static void ext3_clear_journal_err(struct super_block * sb,
  55                                    struct ext3_super_block * es);
  56 static int ext3_sync_fs(struct super_block *sb, int wait);
  57 static const char *ext3_decode_error(struct super_block * sb, int errno,
  58                                      char nbuf[16]);
  59 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  60 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  61 static void ext3_unlockfs(struct super_block *sb);
  62 static void ext3_write_super (struct super_block * sb);
  63 static void ext3_write_super_lockfs(struct super_block *sb);
  64
  65 /*
  66  * Wrappers for journal_start/end.
  67  *
  68  * The only special thing we need to do here is to make sure that all
  69  * journal_end calls result in the superblock being marked dirty, so
  70  * that sync() will call the filesystem's write_super callback if
  71  * appropriate.
  72  */
  73 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  74 {
  75         journal_t *journal;
  76
  77         if (sb->s_flags & MS_RDONLY)
  78                 return ERR_PTR(-EROFS);
  79
  80         /* Special case here: if the journal has aborted behind our
  81          * backs (eg. EIO in the commit thread), then we still need to
  82          * take the FS itself readonly cleanly. */
  83         journal = EXT3_SB(sb)->s_journal;
  84         if (is_journal_aborted(journal)) {
  85                 ext3_abort(sb, __FUNCTION__,
  86                            "Detected aborted journal");
  87                 return ERR_PTR(-EROFS);
  88         }
  89
  90         return journal_start(journal, nblocks);
  91 }
  92
  93 /*
  94  * The only special thing we need to do here is to make sure that all
  95  * journal_stop calls result in the superblock being marked dirty, so
  96  * that sync() will call the filesystem's write_super callback if
  97  * appropriate.
  98  */
  99 int __ext3_journal_stop(const char *where, handle_t *handle)
 100 {
 101         struct super_block *sb;
 102         int err;
 103         int rc;
 104
 105         sb = handle->h_transaction->t_journal->j_private;
 106         err = handle->h_err;
 107         rc = journal_stop(handle);
 108
 109         if (!err)
 110                 err = rc;
 111         if (err)
 112                 __ext3_std_error(sb, where, err);
 113         return err;
 114 }
 115
 116 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 117                 struct buffer_head *bh, handle_t *handle, int err)
 118 {
 119         char nbuf[16];
 120         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 121
 122         if (bh)
 123                 BUFFER_TRACE(bh, "abort");
 124
 125         if (!handle->h_err)
 126                 handle->h_err = err;
 127
 128         if (is_handle_aborted(handle))
 129                 return;
 130
 131         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 132                caller, errstr, err_fn);
 133
 134         journal_abort_handle(handle);
 135 }
 136
 137 /* Deal with the reporting of failure conditions on a filesystem such as
 138  * inconsistencies detected or read IO failures.
 139  *
 140  * On ext2, we can store the error state of the filesystem in the
 141  * superblock.  That is not possible on ext3, because we may have other
 142  * write ordering constraints on the superblock which prevent us from
 143  * writing it out straight away; and given that the journal is about to
 144  * be aborted, we can't rely on the current, or future, transactions to
 145  * write out the superblock safely.
 146  *
 147  * We'll just use the journal_abort() error code to record an error in
 148  * the journal instead.  On recovery, the journal will compain about
 149  * that error until we've noted it down and cleared it.
 150  */
 151
 152 static void ext3_handle_error(struct super_block *sb)
 153 {
 154         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 155
 156         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 157         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 158
 159         if (sb->s_flags & MS_RDONLY)
 160                 return;
 161
 162         if (!test_opt (sb, ERRORS_CONT)) {
 163                 journal_t *journal = EXT3_SB(sb)->s_journal;
 164
 165                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 166                 if (journal)
 167                         journal_abort(journal, -EIO);
 168         }
 169         if (test_opt (sb, ERRORS_RO)) {
 170                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 171                 sb->s_flags |= MS_RDONLY;
 172         }
 173         ext3_commit_super(sb, es, 1);
 174         if (test_opt(sb, ERRORS_PANIC))
 175                 panic("EXT3-fs (device %s): panic forced after error\n",
 176                         sb->s_id);
 177 }
 178
 179 void ext3_error (struct super_block * sb, const char * function,
 180                  const char * fmt, ...)
 181 {
 182         va_list args;
 183
 184         va_start(args, fmt);
 185         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 186         vprintk(fmt, args);
 187         printk("\n");
 188         va_end(args);
 189
 190         ext3_handle_error(sb);
 191 }
 192
 193 static const char *ext3_decode_error(struct super_block * sb, int errno,
 194                                      char nbuf[16])
 195 {
 196         char *errstr = NULL;
 197
 198         switch (errno) {
 199         case -EIO:
 200                 errstr = "IO failure";
 201                 break;
 202         case -ENOMEM:
 203                 errstr = "Out of memory";
 204                 break;
 205         case -EROFS:
 206                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 207                         errstr = "Journal has aborted";
 208                 else
 209                         errstr = "Readonly filesystem";
 210                 break;
 211         default:
 212                 /* If the caller passed in an extra buffer for unknown
 213                  * errors, textualise them now.  Else we just return
 214                  * NULL. */
 215                 if (nbuf) {
 216                         /* Check for truncated error codes... */
 217                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 218                                 errstr = nbuf;
 219                 }
 220                 break;
 221         }
 222
 223         return errstr;
 224 }
 225
 226 /* __ext3_std_error decodes expected errors from journaling functions
 227  * automatically and invokes the appropriate error response.  */
 228
 229 void __ext3_std_error (struct super_block * sb, const char * function,
 230                        int errno)
 231 {
 232         char nbuf[16];
 233         const char *errstr;
 234
 235         /* Special case: if the error is EROFS, and we're not already
 236          * inside a transaction, then there's really no point in logging
 237          * an error. */
 238         if (errno == -EROFS && journal_current_handle() == NULL &&
 239             (sb->s_flags & MS_RDONLY))
 240                 return;
 241
 242         errstr = ext3_decode_error(sb, errno, nbuf);
 243         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 244                 sb->s_id, function, errstr);
 245
 246         ext3_handle_error(sb);
 247 }
 248
 249 /*
 250  * ext3_abort is a much stronger failure handler than ext3_error.  The
 251  * abort function may be used to deal with unrecoverable failures such
 252  * as journal IO errors or ENOMEM at a critical moment in log management.
 253  *
 254  * We unconditionally force the filesystem into an ABORT|READONLY state,
 255  * unless the error response on the fs has been set to panic in which
 256  * case we take the easy way out and panic immediately.
 257  */
 258
 259 void ext3_abort (struct super_block * sb, const char * function,
 260                  const char * fmt, ...)
 261 {
 262         va_list args;
 263
 264         printk (KERN_CRIT "ext3_abort called.\n");
 265
 266         va_start(args, fmt);
 267         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 268         vprintk(fmt, args);
 269         printk("\n");
 270         va_end(args);
 271
 272         if (test_opt(sb, ERRORS_PANIC))
 273                 panic("EXT3-fs panic from previous error\n");
 274
 275         if (sb->s_flags & MS_RDONLY)
 276                 return;
 277
 278         printk(KERN_CRIT "Remounting filesystem read-only\n");
 279         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 280         sb->s_flags |= MS_RDONLY;
 281         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 282         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 283 }
 284
 285 void ext3_warning (struct super_block * sb, const char * function,
 286                    const char * fmt, ...)
 287 {
 288         va_list args;
 289
 290         va_start(args, fmt);
 291         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 292                sb->s_id, function);
 293         vprintk(fmt, args);
 294         printk("\n");
 295         va_end(args);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = NULL;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        NEXT_ORPHAN(inode));
 384         }
 385 }
 386
 387 static void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 407         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 408         percpu_counter_destroy(&sbi->s_dirs_counter);
 409         brelse(sbi->s_sbh);
 410 #ifdef CONFIG_QUOTA
 411         for (i = 0; i < MAXQUOTAS; i++)
 412                 kfree(sbi->s_qf_names[i]);
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static kmem_cache_t *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->i_block_alloc_info = NULL;
 456         ei->vfs_inode.i_version = 1;
 457         return &ei->vfs_inode;
 458 }
 459
 460 static void ext3_destroy_inode(struct inode *inode)
 461 {
 462         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 463 }
 464
 465 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 466 {
 467         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 468
 469         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 470             SLAB_CTOR_CONSTRUCTOR) {
 471                 INIT_LIST_HEAD(&ei->i_orphan);
 472 #ifdef CONFIG_EXT3_FS_XATTR
 473                 init_rwsem(&ei->xattr_sem);
 474 #endif
 475                 mutex_init(&ei->truncate_mutex);
 476                 inode_init_once(&ei->vfs_inode);
 477         }
 478 }
 479
 480 static int init_inodecache(void)
 481 {
 482         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 483                                              sizeof(struct ext3_inode_info),
 484                                              0, (SLAB_RECLAIM_ACCOUNT|
 485                                                 SLAB_MEM_SPREAD),
 486                                              init_once, NULL);
 487         if (ext3_inode_cachep == NULL)
 488                 return -ENOMEM;
 489         return 0;
 490 }
 491
 492 static void destroy_inodecache(void)
 493 {
 494         kmem_cache_destroy(ext3_inode_cachep);
 495 }
 496
 497 static void ext3_clear_inode(struct inode *inode)
 498 {
 499         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 500 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 501         if (EXT3_I(inode)->i_acl &&
 502                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 503                 posix_acl_release(EXT3_I(inode)->i_acl);
 504                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 505         }
 506         if (EXT3_I(inode)->i_default_acl &&
 507                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 508                 posix_acl_release(EXT3_I(inode)->i_default_acl);
 509                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 510         }
 511 #endif
 512         ext3_discard_reservation(inode);
 513         EXT3_I(inode)->i_block_alloc_info = NULL;
 514         if (unlikely(rsv))
 515                 kfree(rsv);
 516 }
 517
 518 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 519 {
 520 #if defined(CONFIG_QUOTA)
 521         struct ext3_sb_info *sbi = EXT3_SB(sb);
 522
 523         if (sbi->s_jquota_fmt)
 524                 seq_printf(seq, ",jqfmt=%s",
 525                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 526
 527         if (sbi->s_qf_names[USRQUOTA])
 528                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 529
 530         if (sbi->s_qf_names[GRPQUOTA])
 531                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 532
 533         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 534                 seq_puts(seq, ",usrquota");
 535
 536         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 537                 seq_puts(seq, ",grpquota");
 538 #endif
 539 }
 540
 541 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 542 {
 543         struct super_block *sb = vfs->mnt_sb;
 544
 545         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 546                 seq_puts(seq, ",data=journal");
 547         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 548                 seq_puts(seq, ",data=ordered");
 549         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 550                 seq_puts(seq, ",data=writeback");
 551
 552         ext3_show_quota_options(seq, sb);
 553
 554         return 0;
 555 }
 556
 557
 558 static struct dentry *ext3_get_dentry(struct super_block *sb, void *vobjp)
 559 {
 560         __u32 *objp = vobjp;
 561         unsigned long ino = objp[0];
 562         __u32 generation = objp[1];
 563         struct inode *inode;
 564         struct dentry *result;
 565
 566         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 567                 return ERR_PTR(-ESTALE);
 568         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 569                 return ERR_PTR(-ESTALE);
 570
 571         /* iget isn't really right if the inode is currently unallocated!!
 572          *
 573          * ext3_read_inode will return a bad_inode if the inode had been
 574          * deleted, so we should be safe.
 575          *
 576          * Currently we don't know the generation for parent directory, so
 577          * a generation of 0 means "accept any"
 578          */
 579         inode = iget(sb, ino);
 580         if (inode == NULL)
 581                 return ERR_PTR(-ENOMEM);
 582         if (is_bad_inode(inode) ||
 583             (generation && inode->i_generation != generation)) {
 584                 iput(inode);
 585                 return ERR_PTR(-ESTALE);
 586         }
 587         /* now to find a dentry.
 588          * If possible, get a well-connected one
 589          */
 590         result = d_alloc_anon(inode);
 591         if (!result) {
 592                 iput(inode);
 593                 return ERR_PTR(-ENOMEM);
 594         }
 595         return result;
 596 }
 597
 598 #ifdef CONFIG_QUOTA
 599 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 600 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 601
 602 static int ext3_dquot_initialize(struct inode *inode, int type);
 603 static int ext3_dquot_drop(struct inode *inode);
 604 static int ext3_write_dquot(struct dquot *dquot);
 605 static int ext3_acquire_dquot(struct dquot *dquot);
 606 static int ext3_release_dquot(struct dquot *dquot);
 607 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 608 static int ext3_write_info(struct super_block *sb, int type);
 609 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 610 static int ext3_quota_on_mount(struct super_block *sb, int type);
 611 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 612                                size_t len, loff_t off);
 613 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 614                                 const char *data, size_t len, loff_t off);
 615
 616 static struct dquot_operations ext3_quota_operations = {
 617         .initialize     = ext3_dquot_initialize,
 618         .drop           = ext3_dquot_drop,
 619         .alloc_space    = dquot_alloc_space,
 620         .alloc_inode    = dquot_alloc_inode,
 621         .free_space     = dquot_free_space,
 622         .free_inode     = dquot_free_inode,
 623         .transfer       = dquot_transfer,
 624         .write_dquot    = ext3_write_dquot,
 625         .acquire_dquot  = ext3_acquire_dquot,
 626         .release_dquot  = ext3_release_dquot,
 627         .mark_dirty     = ext3_mark_dquot_dirty,
 628         .write_info     = ext3_write_info
 629 };
 630
 631 static struct quotactl_ops ext3_qctl_operations = {
 632         .quota_on       = ext3_quota_on,
 633         .quota_off      = vfs_quota_off,
 634         .quota_sync     = vfs_quota_sync,
 635         .get_info       = vfs_get_dqinfo,
 636         .set_info       = vfs_set_dqinfo,
 637         .get_dqblk      = vfs_get_dqblk,
 638         .set_dqblk      = vfs_set_dqblk
 639 };
 640 #endif
 641
 642 static struct super_operations ext3_sops = {
 643         .alloc_inode    = ext3_alloc_inode,
 644         .destroy_inode  = ext3_destroy_inode,
 645         .read_inode     = ext3_read_inode,
 646         .write_inode    = ext3_write_inode,
 647         .dirty_inode    = ext3_dirty_inode,
 648         .delete_inode   = ext3_delete_inode,
 649         .put_super      = ext3_put_super,
 650         .write_super    = ext3_write_super,
 651         .sync_fs        = ext3_sync_fs,
 652         .write_super_lockfs = ext3_write_super_lockfs,
 653         .unlockfs       = ext3_unlockfs,
 654         .statfs         = ext3_statfs,
 655         .remount_fs     = ext3_remount,
 656         .clear_inode    = ext3_clear_inode,
 657         .show_options   = ext3_show_options,
 658 #ifdef CONFIG_QUOTA
 659         .quota_read     = ext3_quota_read,
 660         .quota_write    = ext3_quota_write,
 661 #endif
 662 };
 663
 664 static struct export_operations ext3_export_ops = {
 665         .get_parent = ext3_get_parent,
 666         .get_dentry = ext3_get_dentry,
 667 };
 668
 669 enum {
 670         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 671         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 672         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 673         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 674         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 675         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 676         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 677         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 678         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 679         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 680         Opt_grpquota
 681 };
 682
 683 static match_table_t tokens = {
 684         {Opt_bsd_df, "bsddf"},
 685         {Opt_minix_df, "minixdf"},
 686         {Opt_grpid, "grpid"},
 687         {Opt_grpid, "bsdgroups"},
 688         {Opt_nogrpid, "nogrpid"},
 689         {Opt_nogrpid, "sysvgroups"},
 690         {Opt_resgid, "resgid=%u"},
 691         {Opt_resuid, "resuid=%u"},
 692         {Opt_sb, "sb=%u"},
 693         {Opt_err_cont, "errors=continue"},
 694         {Opt_err_panic, "errors=panic"},
 695         {Opt_err_ro, "errors=remount-ro"},
 696         {Opt_nouid32, "nouid32"},
 697         {Opt_nocheck, "nocheck"},
 698         {Opt_nocheck, "check=none"},
 699         {Opt_debug, "debug"},
 700         {Opt_oldalloc, "oldalloc"},
 701         {Opt_orlov, "orlov"},
 702         {Opt_user_xattr, "user_xattr"},
 703         {Opt_nouser_xattr, "nouser_xattr"},
 704         {Opt_acl, "acl"},
 705         {Opt_noacl, "noacl"},
 706         {Opt_reservation, "reservation"},
 707         {Opt_noreservation, "noreservation"},
 708         {Opt_noload, "noload"},
 709         {Opt_nobh, "nobh"},
 710         {Opt_bh, "bh"},
 711         {Opt_commit, "commit=%u"},
 712         {Opt_journal_update, "journal=update"},
 713         {Opt_journal_inum, "journal=%u"},
 714         {Opt_journal_dev, "journal_dev=%u"},
 715         {Opt_abort, "abort"},
 716         {Opt_data_journal, "data=journal"},
 717         {Opt_data_ordered, "data=ordered"},
 718         {Opt_data_writeback, "data=writeback"},
 719         {Opt_offusrjquota, "usrjquota="},
 720         {Opt_usrjquota, "usrjquota=%s"},
 721         {Opt_offgrpjquota, "grpjquota="},
 722         {Opt_grpjquota, "grpjquota=%s"},
 723         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 724         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 725         {Opt_grpquota, "grpquota"},
 726         {Opt_noquota, "noquota"},
 727         {Opt_quota, "quota"},
 728         {Opt_usrquota, "usrquota"},
 729         {Opt_barrier, "barrier=%u"},
 730         {Opt_err, NULL},
 731         {Opt_resize, "resize"},
 732 };
 733
 734 static ext3_fsblk_t get_sb_block(void **data)
 735 {
 736         ext3_fsblk_t    sb_block;
 737         char            *options = (char *) *data;
 738
 739         if (!options || strncmp(options, "sb=", 3) != 0)
 740                 return 1;       /* Default location */
 741         options += 3;
 742         /*todo: use simple_strtoll with >32bit ext3 */
 743         sb_block = simple_strtoul(options, &options, 0);
 744         if (*options && *options != ',') {
 745                 printk("EXT3-fs: Invalid sb specification: %s\n",
 746                        (char *) *data);
 747                 return 1;
 748         }
 749         if (*options == ',')
 750                 options++;
 751         *data = (void *) options;
 752         return sb_block;
 753 }
 754
 755 static int parse_options (char *options, struct super_block *sb,
 756                           unsigned int *inum, unsigned long *journal_devnum,
 757                           ext3_fsblk_t *n_blocks_count, int is_remount)
 758 {
 759         struct ext3_sb_info *sbi = EXT3_SB(sb);
 760         char * p;
 761         substring_t args[MAX_OPT_ARGS];
 762         int data_opt = 0;
 763         int option;
 764 #ifdef CONFIG_QUOTA
 765         int qtype;
 766         char *qname;
 767 #endif
 768
 769         if (!options)
 770                 return 1;
 771
 772         while ((p = strsep (&options, ",")) != NULL) {
 773                 int token;
 774                 if (!*p)
 775                         continue;
 776
 777                 token = match_token(p, tokens, args);
 778                 switch (token) {
 779                 case Opt_bsd_df:
 780                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 781                         break;
 782                 case Opt_minix_df:
 783                         set_opt (sbi->s_mount_opt, MINIX_DF);
 784                         break;
 785                 case Opt_grpid:
 786                         set_opt (sbi->s_mount_opt, GRPID);
 787                         break;
 788                 case Opt_nogrpid:
 789                         clear_opt (sbi->s_mount_opt, GRPID);
 790                         break;
 791                 case Opt_resuid:
 792                         if (match_int(&args[0], &option))
 793                                 return 0;
 794                         sbi->s_resuid = option;
 795                         break;
 796                 case Opt_resgid:
 797                         if (match_int(&args[0], &option))
 798                                 return 0;
 799                         sbi->s_resgid = option;
 800                         break;
 801                 case Opt_sb:
 802                         /* handled by get_sb_block() instead of here */
 803                         /* *sb_block = match_int(&args[0]); */
 804                         break;
 805                 case Opt_err_panic:
 806                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 807                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 808                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 809                         break;
 810                 case Opt_err_ro:
 811                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 812                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 813                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 814                         break;
 815                 case Opt_err_cont:
 816                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 817                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 818                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 819                         break;
 820                 case Opt_nouid32:
 821                         set_opt (sbi->s_mount_opt, NO_UID32);
 822                         break;
 823                 case Opt_nocheck:
 824                         clear_opt (sbi->s_mount_opt, CHECK);
 825                         break;
 826                 case Opt_debug:
 827                         set_opt (sbi->s_mount_opt, DEBUG);
 828                         break;
 829                 case Opt_oldalloc:
 830                         set_opt (sbi->s_mount_opt, OLDALLOC);
 831                         break;
 832                 case Opt_orlov:
 833                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 834                         break;
 835 #ifdef CONFIG_EXT3_FS_XATTR
 836                 case Opt_user_xattr:
 837                         set_opt (sbi->s_mount_opt, XATTR_USER);
 838                         break;
 839                 case Opt_nouser_xattr:
 840                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 841                         break;
 842 #else
 843                 case Opt_user_xattr:
 844                 case Opt_nouser_xattr:
 845                         printk("EXT3 (no)user_xattr options not supported\n");
 846                         break;
 847 #endif
 848 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 849                 case Opt_acl:
 850                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 851                         break;
 852                 case Opt_noacl:
 853                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 854                         break;
 855 #else
 856                 case Opt_acl:
 857                 case Opt_noacl:
 858                         printk("EXT3 (no)acl options not supported\n");
 859                         break;
 860 #endif
 861                 case Opt_reservation:
 862                         set_opt(sbi->s_mount_opt, RESERVATION);
 863                         break;
 864                 case Opt_noreservation:
 865                         clear_opt(sbi->s_mount_opt, RESERVATION);
 866                         break;
 867                 case Opt_journal_update:
 868                         /* @@@ FIXME */
 869                         /* Eventually we will want to be able to create
 870                            a journal file here.  For now, only allow the
 871                            user to specify an existing inode to be the
 872                            journal file. */
 873                         if (is_remount) {
 874                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 875                                        "journal on remount\n");
 876                                 return 0;
 877                         }
 878                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 879                         break;
 880                 case Opt_journal_inum:
 881                         if (is_remount) {
 882                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 883                                        "journal on remount\n");
 884                                 return 0;
 885                         }
 886                         if (match_int(&args[0], &option))
 887                                 return 0;
 888                         *inum = option;
 889                         break;
 890                 case Opt_journal_dev:
 891                         if (is_remount) {
 892                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 893                                        "journal on remount\n");
 894                                 return 0;
 895                         }
 896                         if (match_int(&args[0], &option))
 897                                 return 0;
 898                         *journal_devnum = option;
 899                         break;
 900                 case Opt_noload:
 901                         set_opt (sbi->s_mount_opt, NOLOAD);
 902                         break;
 903                 case Opt_commit:
 904                         if (match_int(&args[0], &option))
 905                                 return 0;
 906                         if (option < 0)
 907                                 return 0;
 908                         if (option == 0)
 909                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 910                         sbi->s_commit_interval = HZ * option;
 911                         break;
 912                 case Opt_data_journal:
 913                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 914                         goto datacheck;
 915                 case Opt_data_ordered:
 916                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 917                         goto datacheck;
 918                 case Opt_data_writeback:
 919                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 920                 datacheck:
 921                         if (is_remount) {
 922                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 923                                                 != data_opt) {
 924                                         printk(KERN_ERR
 925                                                 "EXT3-fs: cannot change data "
 926                                                 "mode on remount\n");
 927                                         return 0;
 928                                 }
 929                         } else {
 930                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 931                                 sbi->s_mount_opt |= data_opt;
 932                         }
 933                         break;
 934 #ifdef CONFIG_QUOTA
 935                 case Opt_usrjquota:
 936                         qtype = USRQUOTA;
 937                         goto set_qf_name;
 938                 case Opt_grpjquota:
 939                         qtype = GRPQUOTA;
 940 set_qf_name:
 941                         if (sb_any_quota_enabled(sb)) {
 942                                 printk(KERN_ERR
 943                                         "EXT3-fs: Cannot change journalled "
 944                                         "quota options when quota turned on.\n");
 945                                 return 0;
 946                         }
 947                         qname = match_strdup(&args[0]);
 948                         if (!qname) {
 949                                 printk(KERN_ERR
 950                                         "EXT3-fs: not enough memory for "
 951                                         "storing quotafile name.\n");
 952                                 return 0;
 953                         }
 954                         if (sbi->s_qf_names[qtype] &&
 955                             strcmp(sbi->s_qf_names[qtype], qname)) {
 956                                 printk(KERN_ERR
 957                                         "EXT3-fs: %s quota file already "
 958                                         "specified.\n", QTYPE2NAME(qtype));
 959                                 kfree(qname);
 960                                 return 0;
 961                         }
 962                         sbi->s_qf_names[qtype] = qname;
 963                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 964                                 printk(KERN_ERR
 965                                         "EXT3-fs: quotafile must be on "
 966                                         "filesystem root.\n");
 967                                 kfree(sbi->s_qf_names[qtype]);
 968                                 sbi->s_qf_names[qtype] = NULL;
 969                                 return 0;
 970                         }
 971                         set_opt(sbi->s_mount_opt, QUOTA);
 972                         break;
 973                 case Opt_offusrjquota:
 974                         qtype = USRQUOTA;
 975                         goto clear_qf_name;
 976                 case Opt_offgrpjquota:
 977                         qtype = GRPQUOTA;
 978 clear_qf_name:
 979                         if (sb_any_quota_enabled(sb)) {
 980                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 981                                         "journalled quota options when "
 982                                         "quota turned on.\n");
 983                                 return 0;
 984                         }
 985                         /*
 986                          * The space will be released later when all options
 987                          * are confirmed to be correct
 988                          */
 989                         sbi->s_qf_names[qtype] = NULL;
 990                         break;
 991                 case Opt_jqfmt_vfsold:
 992                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 993                         break;
 994                 case Opt_jqfmt_vfsv0:
 995                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 996                         break;
 997                 case Opt_quota:
 998                 case Opt_usrquota:
 999                         set_opt(sbi->s_mount_opt, QUOTA);
1000                         set_opt(sbi->s_mount_opt, USRQUOTA);
1001                         break;
1002                 case Opt_grpquota:
1003                         set_opt(sbi->s_mount_opt, QUOTA);
1004                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1005                         break;
1006                 case Opt_noquota:
1007                         if (sb_any_quota_enabled(sb)) {
1008                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1009                                         "options when quota turned on.\n");
1010                                 return 0;
1011                         }
1012                         clear_opt(sbi->s_mount_opt, QUOTA);
1013                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1014                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1015                         break;
1016 #else
1017                 case Opt_quota:
1018                 case Opt_usrquota:
1019                 case Opt_grpquota:
1020                 case Opt_usrjquota:
1021                 case Opt_grpjquota:
1022                 case Opt_offusrjquota:
1023                 case Opt_offgrpjquota:
1024                 case Opt_jqfmt_vfsold:
1025                 case Opt_jqfmt_vfsv0:
1026                         printk(KERN_ERR
1027                                 "EXT3-fs: journalled quota options not "
1028                                 "supported.\n");
1029                         break;
1030                 case Opt_noquota:
1031                         break;
1032 #endif
1033                 case Opt_abort:
1034                         set_opt(sbi->s_mount_opt, ABORT);
1035                         break;
1036                 case Opt_barrier:
1037                         if (match_int(&args[0], &option))
1038                                 return 0;
1039                         if (option)
1040                                 set_opt(sbi->s_mount_opt, BARRIER);
1041                         else
1042                                 clear_opt(sbi->s_mount_opt, BARRIER);
1043                         break;
1044                 case Opt_ignore:
1045                         break;
1046                 case Opt_resize:
1047                         if (!is_remount) {
1048                                 printk("EXT3-fs: resize option only available "
1049                                         "for remount\n");
1050                                 return 0;
1051                         }
1052                         if (match_int(&args[0], &option) != 0)
1053                                 return 0;
1054                         *n_blocks_count = option;
1055                         break;
1056                 case Opt_nobh:
1057                         set_opt(sbi->s_mount_opt, NOBH);
1058                         break;
1059                 case Opt_bh:
1060                         clear_opt(sbi->s_mount_opt, NOBH);
1061                         break;
1062                 default:
1063                         printk (KERN_ERR
1064                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1065                                 "or missing value\n", p);
1066                         return 0;
1067                 }
1068         }
1069 #ifdef CONFIG_QUOTA
1070         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1071                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1072                      sbi->s_qf_names[USRQUOTA])
1073                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1074
1075                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1076                      sbi->s_qf_names[GRPQUOTA])
1077                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1078
1079                 if ((sbi->s_qf_names[USRQUOTA] &&
1080                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1081                     (sbi->s_qf_names[GRPQUOTA] &&
1082                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1083                         printk(KERN_ERR "EXT3-fs: old and new quota "
1084                                         "format mixing.\n");
1085                         return 0;
1086                 }
1087
1088                 if (!sbi->s_jquota_fmt) {
1089                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1090                                         "not specified.\n");
1091                         return 0;
1092                 }
1093         } else {
1094                 if (sbi->s_jquota_fmt) {
1095                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1096                                         "specified with no journalling "
1097                                         "enabled.\n");
1098                         return 0;
1099                 }
1100         }
1101 #endif
1102         return 1;
1103 }
1104
1105 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1106                             int read_only)
1107 {
1108         struct ext3_sb_info *sbi = EXT3_SB(sb);
1109         int res = 0;
1110
1111         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1112                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1113                         "forcing read-only mode\n");
1114                 res = MS_RDONLY;
1115         }
1116         if (read_only)
1117                 return res;
1118         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1119                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1120                         "running e2fsck is recommended\n");
1121         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1122                 printk (KERN_WARNING
1123                         "EXT3-fs warning: mounting fs with errors, "
1124                         "running e2fsck is recommended\n");
1125         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1126                  le16_to_cpu(es->s_mnt_count) >=
1127                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1128                 printk (KERN_WARNING
1129                         "EXT3-fs warning: maximal mount count reached, "
1130                         "running e2fsck is recommended\n");
1131         else if (le32_to_cpu(es->s_checkinterval) &&
1132                 (le32_to_cpu(es->s_lastcheck) +
1133                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1134                 printk (KERN_WARNING
1135                         "EXT3-fs warning: checktime reached, "
1136                         "running e2fsck is recommended\n");
1137 #if 0
1138                 /* @@@ We _will_ want to clear the valid bit if we find
1139                    inconsistencies, to force a fsck at reboot.  But for
1140                    a plain journaled filesystem we can keep it set as
1141                    valid forever! :) */
1142         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1143 #endif
1144         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1145                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1146         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1147         es->s_mtime = cpu_to_le32(get_seconds());
1148         ext3_update_dynamic_rev(sb);
1149         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1150
1151         ext3_commit_super(sb, es, 1);
1152         if (test_opt(sb, DEBUG))
1153                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1154                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1155                         sb->s_blocksize,
1156                         sbi->s_groups_count,
1157                         EXT3_BLOCKS_PER_GROUP(sb),
1158                         EXT3_INODES_PER_GROUP(sb),
1159                         sbi->s_mount_opt);
1160
1161         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1162         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1163                 char b[BDEVNAME_SIZE];
1164
1165                 printk("external journal on %s\n",
1166                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1167         } else {
1168                 printk("internal journal\n");
1169         }
1170         return res;
1171 }
1172
1173 /* Called at mount-time, super-block is locked */
1174 static int ext3_check_descriptors (struct super_block * sb)
1175 {
1176         struct ext3_sb_info *sbi = EXT3_SB(sb);
1177         ext3_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1178         ext3_fsblk_t last_block;
1179         struct ext3_group_desc * gdp = NULL;
1180         int desc_block = 0;
1181         int i;
1182
1183         ext3_debug ("Checking group descriptors");
1184
1185         for (i = 0; i < sbi->s_groups_count; i++)
1186         {
1187                 if (i == sbi->s_groups_count - 1)
1188                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1189                 else
1190                         last_block = first_block +
1191                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1192
1193                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1194                         gdp = (struct ext3_group_desc *)
1195                                         sbi->s_group_desc[desc_block++]->b_data;
1196                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1197                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1198                 {
1199                         ext3_error (sb, "ext3_check_descriptors",
1200                                     "Block bitmap for group %d"
1201                                     " not in group (block %lu)!",
1202                                     i, (unsigned long)
1203                                         le32_to_cpu(gdp->bg_block_bitmap));
1204                         return 0;
1205                 }
1206                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1207                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1208                 {
1209                         ext3_error (sb, "ext3_check_descriptors",
1210                                     "Inode bitmap for group %d"
1211                                     " not in group (block %lu)!",
1212                                     i, (unsigned long)
1213                                         le32_to_cpu(gdp->bg_inode_bitmap));
1214                         return 0;
1215                 }
1216                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1217                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >
1218                     last_block)
1219                 {
1220                         ext3_error (sb, "ext3_check_descriptors",
1221                                     "Inode table for group %d"
1222                                     " not in group (block %lu)!",
1223                                     i, (unsigned long)
1224                                         le32_to_cpu(gdp->bg_inode_table));
1225                         return 0;
1226                 }
1227                 first_block += EXT3_BLOCKS_PER_GROUP(sb);
1228                 gdp++;
1229         }
1230
1231         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1232         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1233         return 1;
1234 }
1235
1236
1237 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1238  * the superblock) which were deleted from all directories, but held open by
1239  * a process at the time of a crash.  We walk the list and try to delete these
1240  * inodes at recovery time (only with a read-write filesystem).
1241  *
1242  * In order to keep the orphan inode chain consistent during traversal (in
1243  * case of crash during recovery), we link each inode into the superblock
1244  * orphan list_head and handle it the same way as an inode deletion during
1245  * normal operation (which journals the operations for us).
1246  *
1247  * We only do an iget() and an iput() on each inode, which is very safe if we
1248  * accidentally point at an in-use or already deleted inode.  The worst that
1249  * can happen in this case is that we get a "bit already cleared" message from
1250  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1251  * e2fsck was run on this filesystem, and it must have already done the orphan
1252  * inode cleanup for us, so we can safely abort without any further action.
1253  */
1254 static void ext3_orphan_cleanup (struct super_block * sb,
1255                                  struct ext3_super_block * es)
1256 {
1257         unsigned int s_flags = sb->s_flags;
1258         int nr_orphans = 0, nr_truncates = 0;
1259 #ifdef CONFIG_QUOTA
1260         int i;
1261 #endif
1262         if (!es->s_last_orphan) {
1263                 jbd_debug(4, "no orphan inodes to clean up\n");
1264                 return;
1265         }
1266
1267         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1268                 if (es->s_last_orphan)
1269                         jbd_debug(1, "Errors on filesystem, "
1270                                   "clearing orphan list.\n");
1271                 es->s_last_orphan = 0;
1272                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1273                 return;
1274         }
1275
1276         if (s_flags & MS_RDONLY) {
1277                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1278                        sb->s_id);
1279                 sb->s_flags &= ~MS_RDONLY;
1280         }
1281 #ifdef CONFIG_QUOTA
1282         /* Needed for iput() to work correctly and not trash data */
1283         sb->s_flags |= MS_ACTIVE;
1284         /* Turn on quotas so that they are updated correctly */
1285         for (i = 0; i < MAXQUOTAS; i++) {
1286                 if (EXT3_SB(sb)->s_qf_names[i]) {
1287                         int ret = ext3_quota_on_mount(sb, i);
1288                         if (ret < 0)
1289                                 printk(KERN_ERR
1290                                         "EXT3-fs: Cannot turn on journalled "
1291                                         "quota: error %d\n", ret);
1292                 }
1293         }
1294 #endif
1295
1296         while (es->s_last_orphan) {
1297                 struct inode *inode;
1298
1299                 if (!(inode =
1300                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1301                         es->s_last_orphan = 0;
1302                         break;
1303                 }
1304
1305                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1306                 DQUOT_INIT(inode);
1307                 if (inode->i_nlink) {
1308                         printk(KERN_DEBUG
1309                                 "%s: truncating inode %lu to %Ld bytes\n",
1310                                 __FUNCTION__, inode->i_ino, inode->i_size);
1311                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1312                                   inode->i_ino, inode->i_size);
1313                         ext3_truncate(inode);
1314                         nr_truncates++;
1315                 } else {
1316                         printk(KERN_DEBUG
1317                                 "%s: deleting unreferenced inode %lu\n",
1318                                 __FUNCTION__, inode->i_ino);
1319                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1320                                   inode->i_ino);
1321                         nr_orphans++;
1322                 }
1323                 iput(inode);  /* The delete magic happens here! */
1324         }
1325
1326 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1327
1328         if (nr_orphans)
1329                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1330                        sb->s_id, PLURAL(nr_orphans));
1331         if (nr_truncates)
1332                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1333                        sb->s_id, PLURAL(nr_truncates));
1334 #ifdef CONFIG_QUOTA
1335         /* Turn quotas off */
1336         for (i = 0; i < MAXQUOTAS; i++) {
1337                 if (sb_dqopt(sb)->files[i])
1338                         vfs_quota_off(sb, i);
1339         }
1340 #endif
1341         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1342 }
1343
1344 #define log2(n) ffz(~(n))
1345
1346 /*
1347  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1348  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1349  * We need to be 1 filesystem block less than the 2^32 sector limit.
1350  */
1351 static loff_t ext3_max_size(int bits)
1352 {
1353         loff_t res = EXT3_NDIR_BLOCKS;
1354         /* This constant is calculated to be the largest file size for a
1355          * dense, 4k-blocksize file such that the total number of
1356          * sectors in the file, including data and all indirect blocks,
1357          * does not exceed 2^32. */
1358         const loff_t upper_limit = 0x1ff7fffd000LL;
1359
1360         res += 1LL << (bits-2);
1361         res += 1LL << (2*(bits-2));
1362         res += 1LL << (3*(bits-2));
1363         res <<= bits;
1364         if (res > upper_limit)
1365                 res = upper_limit;
1366         return res;
1367 }
1368
1369 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1370                                     ext3_fsblk_t logic_sb_block,
1371                                     int nr)
1372 {
1373         struct ext3_sb_info *sbi = EXT3_SB(sb);
1374         unsigned long bg, first_meta_bg;
1375         int has_super = 0;
1376
1377         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1378
1379         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1380             nr < first_meta_bg)
1381                 return (logic_sb_block + nr + 1);
1382         bg = sbi->s_desc_per_block * nr;
1383         if (ext3_bg_has_super(sb, bg))
1384                 has_super = 1;
1385         return (has_super + ext3_group_first_block_no(sb, bg));
1386 }
1387
1388
1389 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1390 {
1391         struct buffer_head * bh;
1392         struct ext3_super_block *es = NULL;
1393         struct ext3_sb_info *sbi;
1394         ext3_fsblk_t block;
1395         ext3_fsblk_t sb_block = get_sb_block(&data);
1396         ext3_fsblk_t logic_sb_block;
1397         unsigned long offset = 0;
1398         unsigned int journal_inum = 0;
1399         unsigned long journal_devnum = 0;
1400         unsigned long def_mount_opts;
1401         struct inode *root;
1402         int blocksize;
1403         int hblock;
1404         int db_count;
1405         int i;
1406         int needs_recovery;
1407         __le32 features;
1408
1409         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1410         if (!sbi)
1411                 return -ENOMEM;
1412         sb->s_fs_info = sbi;
1413         sbi->s_mount_opt = 0;
1414         sbi->s_resuid = EXT3_DEF_RESUID;
1415         sbi->s_resgid = EXT3_DEF_RESGID;
1416
1417         unlock_kernel();
1418
1419         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1420         if (!blocksize) {
1421                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1422                 goto out_fail;
1423         }
1424
1425         /*
1426          * The ext3 superblock will not be buffer aligned for other than 1kB
1427          * block sizes.  We need to calculate the offset from buffer start.
1428          */
1429         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1430                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1431                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1432         } else {
1433                 logic_sb_block = sb_block;
1434         }
1435
1436         if (!(bh = sb_bread(sb, logic_sb_block))) {
1437                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1438                 goto out_fail;
1439         }
1440         /*
1441          * Note: s_es must be initialized as soon as possible because
1442          *       some ext3 macro-instructions depend on its value
1443          */
1444         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1445         sbi->s_es = es;
1446         sb->s_magic = le16_to_cpu(es->s_magic);
1447         if (sb->s_magic != EXT3_SUPER_MAGIC)
1448                 goto cantfind_ext3;
1449
1450         /* Set defaults before we parse the mount options */
1451         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1452         if (def_mount_opts & EXT3_DEFM_DEBUG)
1453                 set_opt(sbi->s_mount_opt, DEBUG);
1454         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1455                 set_opt(sbi->s_mount_opt, GRPID);
1456         if (def_mount_opts & EXT3_DEFM_UID16)
1457                 set_opt(sbi->s_mount_opt, NO_UID32);
1458         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1459                 set_opt(sbi->s_mount_opt, XATTR_USER);
1460         if (def_mount_opts & EXT3_DEFM_ACL)
1461                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1462         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1463                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1464         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1465                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1466         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1467                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1468
1469         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1470                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1471         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1472                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1473         else
1474                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1475
1476         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1477         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1478
1479         set_opt(sbi->s_mount_opt, RESERVATION);
1480
1481         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1482                             NULL, 0))
1483                 goto failed_mount;
1484
1485         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1486                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1487
1488         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1489             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1490              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1491              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1492                 printk(KERN_WARNING
1493                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1494                        "running e2fsck is recommended\n");
1495         /*
1496          * Check feature flags regardless of the revision level, since we
1497          * previously didn't change the revision level when setting the flags,
1498          * so there is a chance incompat flags are set on a rev 0 filesystem.
1499          */
1500         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1501         if (features) {
1502                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1503                        "unsupported optional features (%x).\n",
1504                        sb->s_id, le32_to_cpu(features));
1505                 goto failed_mount;
1506         }
1507         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1508         if (!(sb->s_flags & MS_RDONLY) && features) {
1509                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1510                        "unsupported optional features (%x).\n",
1511                        sb->s_id, le32_to_cpu(features));
1512                 goto failed_mount;
1513         }
1514         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1515
1516         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1517             blocksize > EXT3_MAX_BLOCK_SIZE) {
1518                 printk(KERN_ERR
1519                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1520                        blocksize, sb->s_id);
1521                 goto failed_mount;
1522         }
1523
1524         hblock = bdev_hardsect_size(sb->s_bdev);
1525         if (sb->s_blocksize != blocksize) {
1526                 /*
1527                  * Make sure the blocksize for the filesystem is larger
1528                  * than the hardware sectorsize for the machine.
1529                  */
1530                 if (blocksize < hblock) {
1531                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1532                                "device blocksize %d.\n", blocksize, hblock);
1533                         goto failed_mount;
1534                 }
1535
1536                 brelse (bh);
1537                 sb_set_blocksize(sb, blocksize);
1538                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1539                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1540                 bh = sb_bread(sb, logic_sb_block);
1541                 if (!bh) {
1542                         printk(KERN_ERR
1543                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1544                         goto failed_mount;
1545                 }
1546                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1547                 sbi->s_es = es;
1548                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1549                         printk (KERN_ERR
1550                                 "EXT3-fs: Magic mismatch, very weird !\n");
1551                         goto failed_mount;
1552                 }
1553         }
1554
1555         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1556
1557         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1558                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1559                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1560         } else {
1561                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1562                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1563                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1564                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1565                     (sbi->s_inode_size > blocksize)) {
1566                         printk (KERN_ERR
1567                                 "EXT3-fs: unsupported inode size: %d\n",
1568                                 sbi->s_inode_size);
1569                         goto failed_mount;
1570                 }
1571         }
1572         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1573                                    le32_to_cpu(es->s_log_frag_size);
1574         if (blocksize != sbi->s_frag_size) {
1575                 printk(KERN_ERR
1576                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1577                        sbi->s_frag_size, blocksize);
1578                 goto failed_mount;
1579         }
1580         sbi->s_frags_per_block = 1;
1581         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1582         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1583         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1584         if (EXT3_INODE_SIZE(sb) == 0)
1585                 goto cantfind_ext3;
1586         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1587         if (sbi->s_inodes_per_block == 0)
1588                 goto cantfind_ext3;
1589         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1590                                         sbi->s_inodes_per_block;
1591         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1592         sbi->s_sbh = bh;
1593         sbi->s_mount_state = le16_to_cpu(es->s_state);
1594         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1595         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1596         for (i=0; i < 4; i++)
1597                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1598         sbi->s_def_hash_version = es->s_def_hash_version;
1599
1600         if (sbi->s_blocks_per_group > blocksize * 8) {
1601                 printk (KERN_ERR
1602                         "EXT3-fs: #blocks per group too big: %lu\n",
1603                         sbi->s_blocks_per_group);
1604                 goto failed_mount;
1605         }
1606         if (sbi->s_frags_per_group > blocksize * 8) {
1607                 printk (KERN_ERR
1608                         "EXT3-fs: #fragments per group too big: %lu\n",
1609                         sbi->s_frags_per_group);
1610                 goto failed_mount;
1611         }
1612         if (sbi->s_inodes_per_group > blocksize * 8) {
1613                 printk (KERN_ERR
1614                         "EXT3-fs: #inodes per group too big: %lu\n",
1615                         sbi->s_inodes_per_group);
1616                 goto failed_mount;
1617         }
1618
1619         if (le32_to_cpu(es->s_blocks_count) >
1620                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1621                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1622                         " too large to mount safely\n", sb->s_id);
1623                 if (sizeof(sector_t) < 8)
1624                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1625                                         "enabled\n");
1626                 goto failed_mount;
1627         }
1628
1629         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1630                 goto cantfind_ext3;
1631         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1632                                le32_to_cpu(es->s_first_data_block) - 1)
1633                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1634         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1635                    EXT3_DESC_PER_BLOCK(sb);
1636         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1637                                     GFP_KERNEL);
1638         if (sbi->s_group_desc == NULL) {
1639                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1640                 goto failed_mount;
1641         }
1642
1643         bgl_lock_init(&sbi->s_blockgroup_lock);
1644
1645         for (i = 0; i < db_count; i++) {
1646                 block = descriptor_loc(sb, logic_sb_block, i);
1647                 sbi->s_group_desc[i] = sb_bread(sb, block);
1648                 if (!sbi->s_group_desc[i]) {
1649                         printk (KERN_ERR "EXT3-fs: "
1650                                 "can't read group descriptor %d\n", i);
1651                         db_count = i;
1652                         goto failed_mount2;
1653                 }
1654         }
1655         if (!ext3_check_descriptors (sb)) {
1656                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1657                 goto failed_mount2;
1658         }
1659         sbi->s_gdb_count = db_count;
1660         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1661         spin_lock_init(&sbi->s_next_gen_lock);
1662
1663         percpu_counter_init(&sbi->s_freeblocks_counter,
1664                 ext3_count_free_blocks(sb));
1665         percpu_counter_init(&sbi->s_freeinodes_counter,
1666                 ext3_count_free_inodes(sb));
1667         percpu_counter_init(&sbi->s_dirs_counter,
1668                 ext3_count_dirs(sb));
1669
1670         /* per fileystem reservation list head & lock */
1671         spin_lock_init(&sbi->s_rsv_window_lock);
1672         sbi->s_rsv_window_root = RB_ROOT;
1673         /* Add a single, static dummy reservation to the start of the
1674          * reservation window list --- it gives us a placeholder for
1675          * append-at-start-of-list which makes the allocation logic
1676          * _much_ simpler. */
1677         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1678         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1679         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1680         sbi->s_rsv_window_head.rsv_goal_size = 0;
1681         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1682
1683         /*
1684          * set up enough so that it can read an inode
1685          */
1686         sb->s_op = &ext3_sops;
1687         sb->s_export_op = &ext3_export_ops;
1688         sb->s_xattr = ext3_xattr_handlers;
1689 #ifdef CONFIG_QUOTA
1690         sb->s_qcop = &ext3_qctl_operations;
1691         sb->dq_op = &ext3_quota_operations;
1692 #endif
1693         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1694
1695         sb->s_root = NULL;
1696
1697         needs_recovery = (es->s_last_orphan != 0 ||
1698                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1699                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1700
1701         /*
1702          * The first inode we look at is the journal inode.  Don't try
1703          * root first: it may be modified in the journal!
1704          */
1705         if (!test_opt(sb, NOLOAD) &&
1706             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1707                 if (ext3_load_journal(sb, es, journal_devnum))
1708                         goto failed_mount3;
1709         } else if (journal_inum) {
1710                 if (ext3_create_journal(sb, es, journal_inum))
1711                         goto failed_mount3;
1712         } else {
1713                 if (!silent)
1714                         printk (KERN_ERR
1715                                 "ext3: No journal on filesystem on %s\n",
1716                                 sb->s_id);
1717                 goto failed_mount3;
1718         }
1719
1720         /* We have now updated the journal if required, so we can
1721          * validate the data journaling mode. */
1722         switch (test_opt(sb, DATA_FLAGS)) {
1723         case 0:
1724                 /* No mode set, assume a default based on the journal
1725                    capabilities: ORDERED_DATA if the journal can
1726                    cope, else JOURNAL_DATA */
1727                 if (journal_check_available_features
1728                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1729                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1730                 else
1731                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1732                 break;
1733
1734         case EXT3_MOUNT_ORDERED_DATA:
1735         case EXT3_MOUNT_WRITEBACK_DATA:
1736                 if (!journal_check_available_features
1737                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1738                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1739                                "requested data journaling mode\n");
1740                         goto failed_mount4;
1741                 }
1742         default:
1743                 break;
1744         }
1745
1746         if (test_opt(sb, NOBH)) {
1747                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1748                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1749                                 "its supported only with writeback mode\n");
1750                         clear_opt(sbi->s_mount_opt, NOBH);
1751                 }
1752         }
1753         /*
1754          * The journal_load will have done any necessary log recovery,
1755          * so we can safely mount the rest of the filesystem now.
1756          */
1757
1758         root = iget(sb, EXT3_ROOT_INO);
1759         sb->s_root = d_alloc_root(root);
1760         if (!sb->s_root) {
1761                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1762                 iput(root);
1763                 goto failed_mount4;
1764         }
1765         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1766                 dput(sb->s_root);
1767                 sb->s_root = NULL;
1768                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1769                 goto failed_mount4;
1770         }
1771
1772         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1773         /*
1774          * akpm: core read_super() calls in here with the superblock locked.
1775          * That deadlocks, because orphan cleanup needs to lock the superblock
1776          * in numerous places.  Here we just pop the lock - it's relatively
1777          * harmless, because we are now ready to accept write_super() requests,
1778          * and aviro says that's the only reason for hanging onto the
1779          * superblock lock.
1780          */
1781         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1782         ext3_orphan_cleanup(sb, es);
1783         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1784         if (needs_recovery)
1785                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1786         ext3_mark_recovery_complete(sb, es);
1787         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1788                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1789                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1790                 "writeback");
1791
1792         lock_kernel();
1793         return 0;
1794
1795 cantfind_ext3:
1796         if (!silent)
1797                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1798                        sb->s_id);
1799         goto failed_mount;
1800
1801 failed_mount4:
1802         journal_destroy(sbi->s_journal);
1803 failed_mount3:
1804         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1805         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1806         percpu_counter_destroy(&sbi->s_dirs_counter);
1807 failed_mount2:
1808         for (i = 0; i < db_count; i++)
1809                 brelse(sbi->s_group_desc[i]);
1810         kfree(sbi->s_group_desc);
1811 failed_mount:
1812 #ifdef CONFIG_QUOTA
1813         for (i = 0; i < MAXQUOTAS; i++)
1814                 kfree(sbi->s_qf_names[i]);
1815 #endif
1816         ext3_blkdev_remove(sbi);
1817         brelse(bh);
1818 out_fail:
1819         sb->s_fs_info = NULL;
1820         kfree(sbi);
1821         lock_kernel();
1822         return -EINVAL;
1823 }
1824
1825 /*
1826  * Setup any per-fs journal parameters now.  We'll do this both on
1827  * initial mount, once the journal has been initialised but before we've
1828  * done any recovery; and again on any subsequent remount.
1829  */
1830 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1831 {
1832         struct ext3_sb_info *sbi = EXT3_SB(sb);
1833
1834         if (sbi->s_commit_interval)
1835                 journal->j_commit_interval = sbi->s_commit_interval;
1836         /* We could also set up an ext3-specific default for the commit
1837          * interval here, but for now we'll just fall back to the jbd
1838          * default. */
1839
1840         spin_lock(&journal->j_state_lock);
1841         if (test_opt(sb, BARRIER))
1842                 journal->j_flags |= JFS_BARRIER;
1843         else
1844                 journal->j_flags &= ~JFS_BARRIER;
1845         spin_unlock(&journal->j_state_lock);
1846 }
1847
1848 static journal_t *ext3_get_journal(struct super_block *sb,
1849                                    unsigned int journal_inum)
1850 {
1851         struct inode *journal_inode;
1852         journal_t *journal;
1853
1854         /* First, test for the existence of a valid inode on disk.  Bad
1855          * things happen if we iget() an unused inode, as the subsequent
1856          * iput() will try to delete it. */
1857
1858         journal_inode = iget(sb, journal_inum);
1859         if (!journal_inode) {
1860                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1861                 return NULL;
1862         }
1863         if (!journal_inode->i_nlink) {
1864                 make_bad_inode(journal_inode);
1865                 iput(journal_inode);
1866                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1867                 return NULL;
1868         }
1869
1870         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1871                   journal_inode, journal_inode->i_size);
1872         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1873                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1874                 iput(journal_inode);
1875                 return NULL;
1876         }
1877
1878         journal = journal_init_inode(journal_inode);
1879         if (!journal) {
1880                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1881                 iput(journal_inode);
1882                 return NULL;
1883         }
1884         journal->j_private = sb;
1885         ext3_init_journal_params(sb, journal);
1886         return journal;
1887 }
1888
1889 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1890                                        dev_t j_dev)
1891 {
1892         struct buffer_head * bh;
1893         journal_t *journal;
1894         ext3_fsblk_t start;
1895         ext3_fsblk_t len;
1896         int hblock, blocksize;
1897         ext3_fsblk_t sb_block;
1898         unsigned long offset;
1899         struct ext3_super_block * es;
1900         struct block_device *bdev;
1901
1902         bdev = ext3_blkdev_get(j_dev);
1903         if (bdev == NULL)
1904                 return NULL;
1905
1906         if (bd_claim(bdev, sb)) {
1907                 printk(KERN_ERR
1908                         "EXT3: failed to claim external journal device.\n");
1909                 blkdev_put(bdev);
1910                 return NULL;
1911         }
1912
1913         blocksize = sb->s_blocksize;
1914         hblock = bdev_hardsect_size(bdev);
1915         if (blocksize < hblock) {
1916                 printk(KERN_ERR
1917                         "EXT3-fs: blocksize too small for journal device.\n");
1918                 goto out_bdev;
1919         }
1920
1921         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1922         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1923         set_blocksize(bdev, blocksize);
1924         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1925                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1926                        "external journal\n");
1927                 goto out_bdev;
1928         }
1929
1930         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1931         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1932             !(le32_to_cpu(es->s_feature_incompat) &
1933               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1934                 printk(KERN_ERR "EXT3-fs: external journal has "
1935                                         "bad superblock\n");
1936                 brelse(bh);
1937                 goto out_bdev;
1938         }
1939
1940         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1941                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1942                 brelse(bh);
1943                 goto out_bdev;
1944         }
1945
1946         len = le32_to_cpu(es->s_blocks_count);
1947         start = sb_block + 1;
1948         brelse(bh);     /* we're done with the superblock */
1949
1950         journal = journal_init_dev(bdev, sb->s_bdev,
1951                                         start, len, blocksize);
1952         if (!journal) {
1953                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1954                 goto out_bdev;
1955         }
1956         journal->j_private = sb;
1957         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1958         wait_on_buffer(journal->j_sb_buffer);
1959         if (!buffer_uptodate(journal->j_sb_buffer)) {
1960                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1961                 goto out_journal;
1962         }
1963         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1964                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1965                                         "user (unsupported) - %d\n",
1966                         be32_to_cpu(journal->j_superblock->s_nr_users));
1967                 goto out_journal;
1968         }
1969         EXT3_SB(sb)->journal_bdev = bdev;
1970         ext3_init_journal_params(sb, journal);
1971         return journal;
1972 out_journal:
1973         journal_destroy(journal);
1974 out_bdev:
1975         ext3_blkdev_put(bdev);
1976         return NULL;
1977 }
1978
1979 static int ext3_load_journal(struct super_block *sb,
1980                              struct ext3_super_block *es,
1981                              unsigned long journal_devnum)
1982 {
1983         journal_t *journal;
1984         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
1985         dev_t journal_dev;
1986         int err = 0;
1987         int really_read_only;
1988
1989         if (journal_devnum &&
1990             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1991                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
1992                         "numbers have changed\n");
1993                 journal_dev = new_decode_dev(journal_devnum);
1994         } else
1995                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1996
1997         really_read_only = bdev_read_only(sb->s_bdev);
1998
1999         /*
2000          * Are we loading a blank journal or performing recovery after a
2001          * crash?  For recovery, we need to check in advance whether we
2002          * can get read-write access to the device.
2003          */
2004
2005         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2006                 if (sb->s_flags & MS_RDONLY) {
2007                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2008                                         "required on readonly filesystem.\n");
2009                         if (really_read_only) {
2010                                 printk(KERN_ERR "EXT3-fs: write access "
2011                                         "unavailable, cannot proceed.\n");
2012                                 return -EROFS;
2013                         }
2014                         printk (KERN_INFO "EXT3-fs: write access will "
2015                                         "be enabled during recovery.\n");
2016                 }
2017         }
2018
2019         if (journal_inum && journal_dev) {
2020                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2021                        "and inode journals!\n");
2022                 return -EINVAL;
2023         }
2024
2025         if (journal_inum) {
2026                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2027                         return -EINVAL;
2028         } else {
2029                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2030                         return -EINVAL;
2031         }
2032
2033         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2034                 err = journal_update_format(journal);
2035                 if (err)  {
2036                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2037                         journal_destroy(journal);
2038                         return err;
2039                 }
2040         }
2041
2042         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2043                 err = journal_wipe(journal, !really_read_only);
2044         if (!err)
2045                 err = journal_load(journal);
2046
2047         if (err) {
2048                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2049                 journal_destroy(journal);
2050                 return err;
2051         }
2052
2053         EXT3_SB(sb)->s_journal = journal;
2054         ext3_clear_journal_err(sb, es);
2055
2056         if (journal_devnum &&
2057             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2058                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2059                 sb->s_dirt = 1;
2060
2061                 /* Make sure we flush the recovery flag to disk. */
2062                 ext3_commit_super(sb, es, 1);
2063         }
2064
2065         return 0;
2066 }
2067
2068 static int ext3_create_journal(struct super_block * sb,
2069                                struct ext3_super_block * es,
2070                                unsigned int journal_inum)
2071 {
2072         journal_t *journal;
2073
2074         if (sb->s_flags & MS_RDONLY) {
2075                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2076                                 "create journal.\n");
2077                 return -EROFS;
2078         }
2079
2080         if (!(journal = ext3_get_journal(sb, journal_inum)))
2081                 return -EINVAL;
2082
2083         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2084                journal_inum);
2085
2086         if (journal_create(journal)) {
2087                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2088                 journal_destroy(journal);
2089                 return -EIO;
2090         }
2091
2092         EXT3_SB(sb)->s_journal = journal;
2093
2094         ext3_update_dynamic_rev(sb);
2095         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2096         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2097
2098         es->s_journal_inum = cpu_to_le32(journal_inum);
2099         sb->s_dirt = 1;
2100
2101         /* Make sure we flush the recovery flag to disk. */
2102         ext3_commit_super(sb, es, 1);
2103
2104         return 0;
2105 }
2106
2107 static void ext3_commit_super (struct super_block * sb,
2108                                struct ext3_super_block * es,
2109                                int sync)
2110 {
2111         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2112
2113         if (!sbh)
2114                 return;
2115         es->s_wtime = cpu_to_le32(get_seconds());
2116         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2117         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2118         BUFFER_TRACE(sbh, "marking dirty");
2119         mark_buffer_dirty(sbh);
2120         if (sync)
2121                 sync_dirty_buffer(sbh);
2122 }
2123
2124
2125 /*
2126  * Have we just finished recovery?  If so, and if we are mounting (or
2127  * remounting) the filesystem readonly, then we will end up with a
2128  * consistent fs on disk.  Record that fact.
2129  */
2130 static void ext3_mark_recovery_complete(struct super_block * sb,
2131                                         struct ext3_super_block * es)
2132 {
2133         journal_t *journal = EXT3_SB(sb)->s_journal;
2134
2135         journal_lock_updates(journal);
2136         journal_flush(journal);
2137         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2138             sb->s_flags & MS_RDONLY) {
2139                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2140                 sb->s_dirt = 0;
2141                 ext3_commit_super(sb, es, 1);
2142         }
2143         journal_unlock_updates(journal);
2144 }
2145
2146 /*
2147  * If we are mounting (or read-write remounting) a filesystem whose journal
2148  * has recorded an error from a previous lifetime, move that error to the
2149  * main filesystem now.
2150  */
2151 static void ext3_clear_journal_err(struct super_block * sb,
2152                                    struct ext3_super_block * es)
2153 {
2154         journal_t *journal;
2155         int j_errno;
2156         const char *errstr;
2157
2158         journal = EXT3_SB(sb)->s_journal;
2159
2160         /*
2161          * Now check for any error status which may have been recorded in the
2162          * journal by a prior ext3_error() or ext3_abort()
2163          */
2164
2165         j_errno = journal_errno(journal);
2166         if (j_errno) {
2167                 char nbuf[16];
2168
2169                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2170                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2171                              "from previous mount: %s", errstr);
2172                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2173                              "filesystem check.");
2174
2175                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2176                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2177                 ext3_commit_super (sb, es, 1);
2178
2179                 journal_clear_err(journal);
2180         }
2181 }
2182
2183 /*
2184  * Force the running and committing transactions to commit,
2185  * and wait on the commit.
2186  */
2187 int ext3_force_commit(struct super_block *sb)
2188 {
2189         journal_t *journal;
2190         int ret;
2191
2192         if (sb->s_flags & MS_RDONLY)
2193                 return 0;
2194
2195         journal = EXT3_SB(sb)->s_journal;
2196         sb->s_dirt = 0;
2197         ret = ext3_journal_force_commit(journal);
2198         return ret;
2199 }
2200
2201 /*
2202  * Ext3 always journals updates to the superblock itself, so we don't
2203  * have to propagate any other updates to the superblock on disk at this
2204  * point.  Just start an async writeback to get the buffers on their way
2205  * to the disk.
2206  *
2207  * This implicitly triggers the writebehind on sync().
2208  */
2209
2210 static void ext3_write_super (struct super_block * sb)
2211 {
2212         if (mutex_trylock(&sb->s_lock) != 0)
2213                 BUG();
2214         sb->s_dirt = 0;
2215 }
2216
2217 static int ext3_sync_fs(struct super_block *sb, int wait)
2218 {
2219         tid_t target;
2220
2221         sb->s_dirt = 0;
2222         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2223                 if (wait)
2224                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2225         }
2226         return 0;
2227 }
2228
2229 /*
2230  * LVM calls this function before a (read-only) snapshot is created.  This
2231  * gives us a chance to flush the journal completely and mark the fs clean.
2232  */
2233 static void ext3_write_super_lockfs(struct super_block *sb)
2234 {
2235         sb->s_dirt = 0;
2236
2237         if (!(sb->s_flags & MS_RDONLY)) {
2238                 journal_t *journal = EXT3_SB(sb)->s_journal;
2239
2240                 /* Now we set up the journal barrier. */
2241                 journal_lock_updates(journal);
2242                 journal_flush(journal);
2243
2244                 /* Journal blocked and flushed, clear needs_recovery flag. */
2245                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2246                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2247         }
2248 }
2249
2250 /*
2251  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2252  * flag here, even though the filesystem is not technically dirty yet.
2253  */
2254 static void ext3_unlockfs(struct super_block *sb)
2255 {
2256         if (!(sb->s_flags & MS_RDONLY)) {
2257                 lock_super(sb);
2258                 /* Reser the needs_recovery flag before the fs is unlocked. */
2259                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2260                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2261                 unlock_super(sb);
2262                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2263         }
2264 }
2265
2266 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2267 {
2268         struct ext3_super_block * es;
2269         struct ext3_sb_info *sbi = EXT3_SB(sb);
2270         ext3_fsblk_t n_blocks_count = 0;
2271         unsigned long old_sb_flags;
2272         struct ext3_mount_options old_opts;
2273         int err;
2274 #ifdef CONFIG_QUOTA
2275         int i;
2276 #endif
2277
2278         /* Store the original options */
2279         old_sb_flags = sb->s_flags;
2280         old_opts.s_mount_opt = sbi->s_mount_opt;
2281         old_opts.s_resuid = sbi->s_resuid;
2282         old_opts.s_resgid = sbi->s_resgid;
2283         old_opts.s_commit_interval = sbi->s_commit_interval;
2284 #ifdef CONFIG_QUOTA
2285         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2286         for (i = 0; i < MAXQUOTAS; i++)
2287                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2288 #endif
2289
2290         /*
2291          * Allow the "check" option to be passed as a remount option.
2292          */
2293         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2294                 err = -EINVAL;
2295                 goto restore_opts;
2296         }
2297
2298         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2299                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2300
2301         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2302                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2303
2304         es = sbi->s_es;
2305
2306         ext3_init_journal_params(sb, sbi->s_journal);
2307
2308         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2309                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2310                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2311                         err = -EROFS;
2312                         goto restore_opts;
2313                 }
2314
2315                 if (*flags & MS_RDONLY) {
2316                         /*
2317                          * First of all, the unconditional stuff we have to do
2318                          * to disable replay of the journal when we next remount
2319                          */
2320                         sb->s_flags |= MS_RDONLY;
2321
2322                         /*
2323                          * OK, test if we are remounting a valid rw partition
2324                          * readonly, and if so set the rdonly flag and then
2325                          * mark the partition as valid again.
2326                          */
2327                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2328                             (sbi->s_mount_state & EXT3_VALID_FS))
2329                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2330
2331                         ext3_mark_recovery_complete(sb, es);
2332                 } else {
2333                         __le32 ret;
2334                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2335                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2336                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2337                                        "remount RDWR because of unsupported "
2338                                        "optional features (%x).\n",
2339                                        sb->s_id, le32_to_cpu(ret));
2340                                 err = -EROFS;
2341                                 goto restore_opts;
2342                         }
2343                         /*
2344                          * Mounting a RDONLY partition read-write, so reread
2345                          * and store the current valid flag.  (It may have
2346                          * been changed by e2fsck since we originally mounted
2347                          * the partition.)
2348                          */
2349                         ext3_clear_journal_err(sb, es);
2350                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2351                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2352                                 goto restore_opts;
2353                         if (!ext3_setup_super (sb, es, 0))
2354                                 sb->s_flags &= ~MS_RDONLY;
2355                 }
2356         }
2357 #ifdef CONFIG_QUOTA
2358         /* Release old quota file names */
2359         for (i = 0; i < MAXQUOTAS; i++)
2360                 if (old_opts.s_qf_names[i] &&
2361                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2362                         kfree(old_opts.s_qf_names[i]);
2363 #endif
2364         return 0;
2365 restore_opts:
2366         sb->s_flags = old_sb_flags;
2367         sbi->s_mount_opt = old_opts.s_mount_opt;
2368         sbi->s_resuid = old_opts.s_resuid;
2369         sbi->s_resgid = old_opts.s_resgid;
2370         sbi->s_commit_interval = old_opts.s_commit_interval;
2371 #ifdef CONFIG_QUOTA
2372         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2373         for (i = 0; i < MAXQUOTAS; i++) {
2374                 if (sbi->s_qf_names[i] &&
2375                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2376                         kfree(sbi->s_qf_names[i]);
2377                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2378         }
2379 #endif
2380         return err;
2381 }
2382
2383 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2384 {
2385         struct super_block *sb = dentry->d_sb;
2386         struct ext3_sb_info *sbi = EXT3_SB(sb);
2387         struct ext3_super_block *es = sbi->s_es;
2388         ext3_fsblk_t overhead;
2389         int i;
2390
2391         if (test_opt (sb, MINIX_DF))
2392                 overhead = 0;
2393         else {
2394                 unsigned long ngroups;
2395                 ngroups = EXT3_SB(sb)->s_groups_count;
2396                 smp_rmb();
2397
2398                 /*
2399                  * Compute the overhead (FS structures)
2400                  */
2401
2402                 /*
2403                  * All of the blocks before first_data_block are
2404                  * overhead
2405                  */
2406                 overhead = le32_to_cpu(es->s_first_data_block);
2407
2408                 /*
2409                  * Add the overhead attributed to the superblock and
2410                  * block group descriptors.  If the sparse superblocks
2411                  * feature is turned on, then not all groups have this.
2412                  */
2413                 for (i = 0; i < ngroups; i++) {
2414                         overhead += ext3_bg_has_super(sb, i) +
2415                                 ext3_bg_num_gdb(sb, i);
2416                         cond_resched();
2417                 }
2418
2419                 /*
2420                  * Every block group has an inode bitmap, a block
2421                  * bitmap, and an inode table.
2422                  */
2423                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2424         }
2425
2426         buf->f_type = EXT3_SUPER_MAGIC;
2427         buf->f_bsize = sb->s_blocksize;
2428         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2429         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2430         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2431         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2432                 buf->f_bavail = 0;
2433         buf->f_files = le32_to_cpu(es->s_inodes_count);
2434         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2435         buf->f_namelen = EXT3_NAME_LEN;
2436         return 0;
2437 }
2438
2439 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2440  * is locked for write. Otherwise the are possible deadlocks:
2441  * Process 1                         Process 2
2442  * ext3_create()                     quota_sync()
2443  *   journal_start()                   write_dquot()
2444  *   DQUOT_INIT()                        down(dqio_mutex)
2445  *     down(dqio_mutex)                    journal_start()
2446  *
2447  */
2448
2449 #ifdef CONFIG_QUOTA
2450
2451 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2452 {
2453         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2454 }
2455
2456 static int ext3_dquot_initialize(struct inode *inode, int type)
2457 {
2458         handle_t *handle;
2459         int ret, err;
2460
2461         /* We may create quota structure so we need to reserve enough blocks */
2462         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2463         if (IS_ERR(handle))
2464                 return PTR_ERR(handle);
2465         ret = dquot_initialize(inode, type);
2466         err = ext3_journal_stop(handle);
2467         if (!ret)
2468                 ret = err;
2469         return ret;
2470 }
2471
2472 static int ext3_dquot_drop(struct inode *inode)
2473 {
2474         handle_t *handle;
2475         int ret, err;
2476
2477         /* We may delete quota structure so we need to reserve enough blocks */
2478         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2479         if (IS_ERR(handle))
2480                 return PTR_ERR(handle);
2481         ret = dquot_drop(inode);
2482         err = ext3_journal_stop(handle);
2483         if (!ret)
2484                 ret = err;
2485         return ret;
2486 }
2487
2488 static int ext3_write_dquot(struct dquot *dquot)
2489 {
2490         int ret, err;
2491         handle_t *handle;
2492         struct inode *inode;
2493
2494         inode = dquot_to_inode(dquot);
2495         handle = ext3_journal_start(inode,
2496                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2497         if (IS_ERR(handle))
2498                 return PTR_ERR(handle);
2499         ret = dquot_commit(dquot);
2500         err = ext3_journal_stop(handle);
2501         if (!ret)
2502                 ret = err;
2503         return ret;
2504 }
2505
2506 static int ext3_acquire_dquot(struct dquot *dquot)
2507 {
2508         int ret, err;
2509         handle_t *handle;
2510
2511         handle = ext3_journal_start(dquot_to_inode(dquot),
2512                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2513         if (IS_ERR(handle))
2514                 return PTR_ERR(handle);
2515         ret = dquot_acquire(dquot);
2516         err = ext3_journal_stop(handle);
2517         if (!ret)
2518                 ret = err;
2519         return ret;
2520 }
2521
2522 static int ext3_release_dquot(struct dquot *dquot)
2523 {
2524         int ret, err;
2525         handle_t *handle;
2526
2527         handle = ext3_journal_start(dquot_to_inode(dquot),
2528                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2529         if (IS_ERR(handle))
2530                 return PTR_ERR(handle);
2531         ret = dquot_release(dquot);
2532         err = ext3_journal_stop(handle);
2533         if (!ret)
2534                 ret = err;
2535         return ret;
2536 }
2537
2538 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2539 {
2540         /* Are we journalling quotas? */
2541         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2542             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2543                 dquot_mark_dquot_dirty(dquot);
2544                 return ext3_write_dquot(dquot);
2545         } else {
2546                 return dquot_mark_dquot_dirty(dquot);
2547         }
2548 }
2549
2550 static int ext3_write_info(struct super_block *sb, int type)
2551 {
2552         int ret, err;
2553         handle_t *handle;
2554
2555         /* Data block + inode block */
2556         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2557         if (IS_ERR(handle))
2558                 return PTR_ERR(handle);
2559         ret = dquot_commit_info(sb, type);
2560         err = ext3_journal_stop(handle);
2561         if (!ret)
2562                 ret = err;
2563         return ret;
2564 }
2565
2566 /*
2567  * Turn on quotas during mount time - we need to find
2568  * the quota file and such...
2569  */
2570 static int ext3_quota_on_mount(struct super_block *sb, int type)
2571 {
2572         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2573                         EXT3_SB(sb)->s_jquota_fmt, type);
2574 }
2575
2576 /*
2577  * Standard function to be called on quota_on
2578  */
2579 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2580                          char *path)
2581 {
2582         int err;
2583         struct nameidata nd;
2584
2585         if (!test_opt(sb, QUOTA))
2586                 return -EINVAL;
2587         /* Not journalling quota? */
2588         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2589             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2590                 return vfs_quota_on(sb, type, format_id, path);
2591         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2592         if (err)
2593                 return err;
2594         /* Quotafile not on the same filesystem? */
2595         if (nd.mnt->mnt_sb != sb) {
2596                 path_release(&nd);
2597                 return -EXDEV;
2598         }
2599         /* Quotafile not of fs root? */
2600         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2601                 printk(KERN_WARNING
2602                         "EXT3-fs: Quota file not on filesystem root. "
2603                         "Journalled quota will not work.\n");
2604         path_release(&nd);
2605         return vfs_quota_on(sb, type, format_id, path);
2606 }
2607
2608 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2609  * acquiring the locks... As quota files are never truncated and quota code
2610  * itself serializes the operations (and noone else should touch the files)
2611  * we don't have to be afraid of races */
2612 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2613                                size_t len, loff_t off)
2614 {
2615         struct inode *inode = sb_dqopt(sb)->files[type];
2616         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2617         int err = 0;
2618         int offset = off & (sb->s_blocksize - 1);
2619         int tocopy;
2620         size_t toread;
2621         struct buffer_head *bh;
2622         loff_t i_size = i_size_read(inode);
2623
2624         if (off > i_size)
2625                 return 0;
2626         if (off+len > i_size)
2627                 len = i_size-off;
2628         toread = len;
2629         while (toread > 0) {
2630                 tocopy = sb->s_blocksize - offset < toread ?
2631                                 sb->s_blocksize - offset : toread;
2632                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2633                 if (err)
2634                         return err;
2635                 if (!bh)        /* A hole? */
2636                         memset(data, 0, tocopy);
2637                 else
2638                         memcpy(data, bh->b_data+offset, tocopy);
2639                 brelse(bh);
2640                 offset = 0;
2641                 toread -= tocopy;
2642                 data += tocopy;
2643                 blk++;
2644         }
2645         return len;
2646 }
2647
2648 /* Write to quotafile (we know the transaction is already started and has
2649  * enough credits) */
2650 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2651                                 const char *data, size_t len, loff_t off)
2652 {
2653         struct inode *inode = sb_dqopt(sb)->files[type];
2654         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2655         int err = 0;
2656         int offset = off & (sb->s_blocksize - 1);
2657         int tocopy;
2658         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2659         size_t towrite = len;
2660         struct buffer_head *bh;
2661         handle_t *handle = journal_current_handle();
2662
2663         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2664         while (towrite > 0) {
2665                 tocopy = sb->s_blocksize - offset < towrite ?
2666                                 sb->s_blocksize - offset : towrite;
2667                 bh = ext3_bread(handle, inode, blk, 1, &err);
2668                 if (!bh)
2669                         goto out;
2670                 if (journal_quota) {
2671                         err = ext3_journal_get_write_access(handle, bh);
2672                         if (err) {
2673                                 brelse(bh);
2674                                 goto out;
2675                         }
2676                 }
2677                 lock_buffer(bh);
2678                 memcpy(bh->b_data+offset, data, tocopy);
2679                 flush_dcache_page(bh->b_page);
2680                 unlock_buffer(bh);
2681                 if (journal_quota)
2682                         err = ext3_journal_dirty_metadata(handle, bh);
2683                 else {
2684                         /* Always do at least ordered writes for quotas */
2685                         err = ext3_journal_dirty_data(handle, bh);
2686                         mark_buffer_dirty(bh);
2687                 }
2688                 brelse(bh);
2689                 if (err)
2690                         goto out;
2691                 offset = 0;
2692                 towrite -= tocopy;
2693                 data += tocopy;
2694                 blk++;
2695         }
2696 out:
2697         if (len == towrite)
2698                 return err;
2699         if (inode->i_size < off+len-towrite) {
2700                 i_size_write(inode, off+len-towrite);
2701                 EXT3_I(inode)->i_disksize = inode->i_size;
2702         }
2703         inode->i_version++;
2704         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2705         ext3_mark_inode_dirty(handle, inode);
2706         mutex_unlock(&inode->i_mutex);
2707         return len - towrite;
2708 }
2709
2710 #endif
2711
2712 static int ext3_get_sb(struct file_system_type *fs_type,
2713         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2714 {
2715         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2716 }
2717
2718 static struct file_system_type ext3_fs_type = {
2719         .owner          = THIS_MODULE,
2720         .name           = "ext3",
2721         .get_sb         = ext3_get_sb,
2722         .kill_sb        = kill_block_super,
2723         .fs_flags       = FS_REQUIRES_DEV,
2724 };
2725
2726 static int __init init_ext3_fs(void)
2727 {
2728         int err = init_ext3_xattr();
2729         if (err)
2730                 return err;
2731         err = init_inodecache();
2732         if (err)
2733                 goto out1;
2734         err = register_filesystem(&ext3_fs_type);
2735         if (err)
2736                 goto out;
2737         return 0;
2738 out:
2739         destroy_inodecache();
2740 out1:
2741         exit_ext3_xattr();
2742         return err;
2743 }
2744
2745 static void __exit exit_ext3_fs(void)
2746 {
2747         unregister_filesystem(&ext3_fs_type);
2748         destroy_inodecache();
2749         exit_ext3_xattr();
2750 }
2751
2752 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2753 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2754 MODULE_LICENSE("GPL");
2755 module_init(init_ext3_fs)
2756 module_exit(exit_ext3_fs)