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 struct kmem_cache *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, GFP_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, struct kmem_cache * 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 const 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 (bdev_read_only(sb->s_bdev)) {
1268                 printk(KERN_ERR "EXT3-fs: write access "
1269                         "unavailable, skipping orphan cleanup.\n");
1270                 return;
1271         }
1272
1273         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1274                 if (es->s_last_orphan)
1275                         jbd_debug(1, "Errors on filesystem, "
1276                                   "clearing orphan list.\n");
1277                 es->s_last_orphan = 0;
1278                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1279                 return;
1280         }
1281
1282         if (s_flags & MS_RDONLY) {
1283                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1284                        sb->s_id);
1285                 sb->s_flags &= ~MS_RDONLY;
1286         }
1287 #ifdef CONFIG_QUOTA
1288         /* Needed for iput() to work correctly and not trash data */
1289         sb->s_flags |= MS_ACTIVE;
1290         /* Turn on quotas so that they are updated correctly */
1291         for (i = 0; i < MAXQUOTAS; i++) {
1292                 if (EXT3_SB(sb)->s_qf_names[i]) {
1293                         int ret = ext3_quota_on_mount(sb, i);
1294                         if (ret < 0)
1295                                 printk(KERN_ERR
1296                                         "EXT3-fs: Cannot turn on journalled "
1297                                         "quota: error %d\n", ret);
1298                 }
1299         }
1300 #endif
1301
1302         while (es->s_last_orphan) {
1303                 struct inode *inode;
1304
1305                 if (!(inode =
1306                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1307                         es->s_last_orphan = 0;
1308                         break;
1309                 }
1310
1311                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1312                 DQUOT_INIT(inode);
1313                 if (inode->i_nlink) {
1314                         printk(KERN_DEBUG
1315                                 "%s: truncating inode %lu to %Ld bytes\n",
1316                                 __FUNCTION__, inode->i_ino, inode->i_size);
1317                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1318                                   inode->i_ino, inode->i_size);
1319                         ext3_truncate(inode);
1320                         nr_truncates++;
1321                 } else {
1322                         printk(KERN_DEBUG
1323                                 "%s: deleting unreferenced inode %lu\n",
1324                                 __FUNCTION__, inode->i_ino);
1325                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1326                                   inode->i_ino);
1327                         nr_orphans++;
1328                 }
1329                 iput(inode);  /* The delete magic happens here! */
1330         }
1331
1332 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1333
1334         if (nr_orphans)
1335                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1336                        sb->s_id, PLURAL(nr_orphans));
1337         if (nr_truncates)
1338                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1339                        sb->s_id, PLURAL(nr_truncates));
1340 #ifdef CONFIG_QUOTA
1341         /* Turn quotas off */
1342         for (i = 0; i < MAXQUOTAS; i++) {
1343                 if (sb_dqopt(sb)->files[i])
1344                         vfs_quota_off(sb, i);
1345         }
1346 #endif
1347         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1348 }
1349
1350 /*
1351  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1352  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1353  * We need to be 1 filesystem block less than the 2^32 sector limit.
1354  */
1355 static loff_t ext3_max_size(int bits)
1356 {
1357         loff_t res = EXT3_NDIR_BLOCKS;
1358         /* This constant is calculated to be the largest file size for a
1359          * dense, 4k-blocksize file such that the total number of
1360          * sectors in the file, including data and all indirect blocks,
1361          * does not exceed 2^32. */
1362         const loff_t upper_limit = 0x1ff7fffd000LL;
1363
1364         res += 1LL << (bits-2);
1365         res += 1LL << (2*(bits-2));
1366         res += 1LL << (3*(bits-2));
1367         res <<= bits;
1368         if (res > upper_limit)
1369                 res = upper_limit;
1370         return res;
1371 }
1372
1373 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1374                                     ext3_fsblk_t logic_sb_block,
1375                                     int nr)
1376 {
1377         struct ext3_sb_info *sbi = EXT3_SB(sb);
1378         unsigned long bg, first_meta_bg;
1379         int has_super = 0;
1380
1381         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1382
1383         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1384             nr < first_meta_bg)
1385                 return (logic_sb_block + nr + 1);
1386         bg = sbi->s_desc_per_block * nr;
1387         if (ext3_bg_has_super(sb, bg))
1388                 has_super = 1;
1389         return (has_super + ext3_group_first_block_no(sb, bg));
1390 }
1391
1392
1393 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1394 {
1395         struct buffer_head * bh;
1396         struct ext3_super_block *es = NULL;
1397         struct ext3_sb_info *sbi;
1398         ext3_fsblk_t block;
1399         ext3_fsblk_t sb_block = get_sb_block(&data);
1400         ext3_fsblk_t logic_sb_block;
1401         unsigned long offset = 0;
1402         unsigned int journal_inum = 0;
1403         unsigned long journal_devnum = 0;
1404         unsigned long def_mount_opts;
1405         struct inode *root;
1406         int blocksize;
1407         int hblock;
1408         int db_count;
1409         int i;
1410         int needs_recovery;
1411         __le32 features;
1412
1413         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1414         if (!sbi)
1415                 return -ENOMEM;
1416         sb->s_fs_info = sbi;
1417         sbi->s_mount_opt = 0;
1418         sbi->s_resuid = EXT3_DEF_RESUID;
1419         sbi->s_resgid = EXT3_DEF_RESGID;
1420
1421         unlock_kernel();
1422
1423         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1424         if (!blocksize) {
1425                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1426                 goto out_fail;
1427         }
1428
1429         /*
1430          * The ext3 superblock will not be buffer aligned for other than 1kB
1431          * block sizes.  We need to calculate the offset from buffer start.
1432          */
1433         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1434                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1435                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1436         } else {
1437                 logic_sb_block = sb_block;
1438         }
1439
1440         if (!(bh = sb_bread(sb, logic_sb_block))) {
1441                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1442                 goto out_fail;
1443         }
1444         /*
1445          * Note: s_es must be initialized as soon as possible because
1446          *       some ext3 macro-instructions depend on its value
1447          */
1448         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1449         sbi->s_es = es;
1450         sb->s_magic = le16_to_cpu(es->s_magic);
1451         if (sb->s_magic != EXT3_SUPER_MAGIC)
1452                 goto cantfind_ext3;
1453
1454         /* Set defaults before we parse the mount options */
1455         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1456         if (def_mount_opts & EXT3_DEFM_DEBUG)
1457                 set_opt(sbi->s_mount_opt, DEBUG);
1458         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1459                 set_opt(sbi->s_mount_opt, GRPID);
1460         if (def_mount_opts & EXT3_DEFM_UID16)
1461                 set_opt(sbi->s_mount_opt, NO_UID32);
1462 #ifdef CONFIG_EXT3_FS_XATTR
1463         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1464                 set_opt(sbi->s_mount_opt, XATTR_USER);
1465 #endif
1466 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1467         if (def_mount_opts & EXT3_DEFM_ACL)
1468                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1469 #endif
1470         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1471                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1472         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1473                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1474         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1475                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1476
1477         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1478                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1479         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1480                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1481         else
1482                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1483
1484         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1485         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1486
1487         set_opt(sbi->s_mount_opt, RESERVATION);
1488
1489         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1490                             NULL, 0))
1491                 goto failed_mount;
1492
1493         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1494                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1495
1496         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1497             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1498              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1499              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1500                 printk(KERN_WARNING
1501                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1502                        "running e2fsck is recommended\n");
1503         /*
1504          * Check feature flags regardless of the revision level, since we
1505          * previously didn't change the revision level when setting the flags,
1506          * so there is a chance incompat flags are set on a rev 0 filesystem.
1507          */
1508         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1509         if (features) {
1510                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1511                        "unsupported optional features (%x).\n",
1512                        sb->s_id, le32_to_cpu(features));
1513                 goto failed_mount;
1514         }
1515         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1516         if (!(sb->s_flags & MS_RDONLY) && features) {
1517                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1518                        "unsupported optional features (%x).\n",
1519                        sb->s_id, le32_to_cpu(features));
1520                 goto failed_mount;
1521         }
1522         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1523
1524         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1525             blocksize > EXT3_MAX_BLOCK_SIZE) {
1526                 printk(KERN_ERR
1527                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1528                        blocksize, sb->s_id);
1529                 goto failed_mount;
1530         }
1531
1532         hblock = bdev_hardsect_size(sb->s_bdev);
1533         if (sb->s_blocksize != blocksize) {
1534                 /*
1535                  * Make sure the blocksize for the filesystem is larger
1536                  * than the hardware sectorsize for the machine.
1537                  */
1538                 if (blocksize < hblock) {
1539                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1540                                "device blocksize %d.\n", blocksize, hblock);
1541                         goto failed_mount;
1542                 }
1543
1544                 brelse (bh);
1545                 sb_set_blocksize(sb, blocksize);
1546                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1547                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1548                 bh = sb_bread(sb, logic_sb_block);
1549                 if (!bh) {
1550                         printk(KERN_ERR
1551                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1552                         goto failed_mount;
1553                 }
1554                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1555                 sbi->s_es = es;
1556                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1557                         printk (KERN_ERR
1558                                 "EXT3-fs: Magic mismatch, very weird !\n");
1559                         goto failed_mount;
1560                 }
1561         }
1562
1563         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1564
1565         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1566                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1567                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1568         } else {
1569                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1570                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1571                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1572                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1573                     (sbi->s_inode_size > blocksize)) {
1574                         printk (KERN_ERR
1575                                 "EXT3-fs: unsupported inode size: %d\n",
1576                                 sbi->s_inode_size);
1577                         goto failed_mount;
1578                 }
1579         }
1580         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1581                                    le32_to_cpu(es->s_log_frag_size);
1582         if (blocksize != sbi->s_frag_size) {
1583                 printk(KERN_ERR
1584                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1585                        sbi->s_frag_size, blocksize);
1586                 goto failed_mount;
1587         }
1588         sbi->s_frags_per_block = 1;
1589         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1590         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1591         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1592         if (EXT3_INODE_SIZE(sb) == 0)
1593                 goto cantfind_ext3;
1594         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1595         if (sbi->s_inodes_per_block == 0)
1596                 goto cantfind_ext3;
1597         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1598                                         sbi->s_inodes_per_block;
1599         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1600         sbi->s_sbh = bh;
1601         sbi->s_mount_state = le16_to_cpu(es->s_state);
1602         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1603         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1604         for (i=0; i < 4; i++)
1605                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1606         sbi->s_def_hash_version = es->s_def_hash_version;
1607
1608         if (sbi->s_blocks_per_group > blocksize * 8) {
1609                 printk (KERN_ERR
1610                         "EXT3-fs: #blocks per group too big: %lu\n",
1611                         sbi->s_blocks_per_group);
1612                 goto failed_mount;
1613         }
1614         if (sbi->s_frags_per_group > blocksize * 8) {
1615                 printk (KERN_ERR
1616                         "EXT3-fs: #fragments per group too big: %lu\n",
1617                         sbi->s_frags_per_group);
1618                 goto failed_mount;
1619         }
1620         if (sbi->s_inodes_per_group > blocksize * 8) {
1621                 printk (KERN_ERR
1622                         "EXT3-fs: #inodes per group too big: %lu\n",
1623                         sbi->s_inodes_per_group);
1624                 goto failed_mount;
1625         }
1626
1627         if (le32_to_cpu(es->s_blocks_count) >
1628                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1629                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1630                         " too large to mount safely\n", sb->s_id);
1631                 if (sizeof(sector_t) < 8)
1632                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1633                                         "enabled\n");
1634                 goto failed_mount;
1635         }
1636
1637         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1638                 goto cantfind_ext3;
1639         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1640                                le32_to_cpu(es->s_first_data_block) - 1)
1641                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1642         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1643                    EXT3_DESC_PER_BLOCK(sb);
1644         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1645                                     GFP_KERNEL);
1646         if (sbi->s_group_desc == NULL) {
1647                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1648                 goto failed_mount;
1649         }
1650
1651         bgl_lock_init(&sbi->s_blockgroup_lock);
1652
1653         for (i = 0; i < db_count; i++) {
1654                 block = descriptor_loc(sb, logic_sb_block, i);
1655                 sbi->s_group_desc[i] = sb_bread(sb, block);
1656                 if (!sbi->s_group_desc[i]) {
1657                         printk (KERN_ERR "EXT3-fs: "
1658                                 "can't read group descriptor %d\n", i);
1659                         db_count = i;
1660                         goto failed_mount2;
1661                 }
1662         }
1663         if (!ext3_check_descriptors (sb)) {
1664                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1665                 goto failed_mount2;
1666         }
1667         sbi->s_gdb_count = db_count;
1668         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1669         spin_lock_init(&sbi->s_next_gen_lock);
1670
1671         percpu_counter_init(&sbi->s_freeblocks_counter,
1672                 ext3_count_free_blocks(sb));
1673         percpu_counter_init(&sbi->s_freeinodes_counter,
1674                 ext3_count_free_inodes(sb));
1675         percpu_counter_init(&sbi->s_dirs_counter,
1676                 ext3_count_dirs(sb));
1677
1678         /* per fileystem reservation list head & lock */
1679         spin_lock_init(&sbi->s_rsv_window_lock);
1680         sbi->s_rsv_window_root = RB_ROOT;
1681         /* Add a single, static dummy reservation to the start of the
1682          * reservation window list --- it gives us a placeholder for
1683          * append-at-start-of-list which makes the allocation logic
1684          * _much_ simpler. */
1685         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1686         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1687         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1688         sbi->s_rsv_window_head.rsv_goal_size = 0;
1689         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1690
1691         /*
1692          * set up enough so that it can read an inode
1693          */
1694         sb->s_op = &ext3_sops;
1695         sb->s_export_op = &ext3_export_ops;
1696         sb->s_xattr = ext3_xattr_handlers;
1697 #ifdef CONFIG_QUOTA
1698         sb->s_qcop = &ext3_qctl_operations;
1699         sb->dq_op = &ext3_quota_operations;
1700 #endif
1701         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1702
1703         sb->s_root = NULL;
1704
1705         needs_recovery = (es->s_last_orphan != 0 ||
1706                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1707                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1708
1709         /*
1710          * The first inode we look at is the journal inode.  Don't try
1711          * root first: it may be modified in the journal!
1712          */
1713         if (!test_opt(sb, NOLOAD) &&
1714             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1715                 if (ext3_load_journal(sb, es, journal_devnum))
1716                         goto failed_mount3;
1717         } else if (journal_inum) {
1718                 if (ext3_create_journal(sb, es, journal_inum))
1719                         goto failed_mount3;
1720         } else {
1721                 if (!silent)
1722                         printk (KERN_ERR
1723                                 "ext3: No journal on filesystem on %s\n",
1724                                 sb->s_id);
1725                 goto failed_mount3;
1726         }
1727
1728         /* We have now updated the journal if required, so we can
1729          * validate the data journaling mode. */
1730         switch (test_opt(sb, DATA_FLAGS)) {
1731         case 0:
1732                 /* No mode set, assume a default based on the journal
1733                    capabilities: ORDERED_DATA if the journal can
1734                    cope, else JOURNAL_DATA */
1735                 if (journal_check_available_features
1736                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1737                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1738                 else
1739                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1740                 break;
1741
1742         case EXT3_MOUNT_ORDERED_DATA:
1743         case EXT3_MOUNT_WRITEBACK_DATA:
1744                 if (!journal_check_available_features
1745                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1746                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1747                                "requested data journaling mode\n");
1748                         goto failed_mount4;
1749                 }
1750         default:
1751                 break;
1752         }
1753
1754         if (test_opt(sb, NOBH)) {
1755                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1756                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1757                                 "its supported only with writeback mode\n");
1758                         clear_opt(sbi->s_mount_opt, NOBH);
1759                 }
1760         }
1761         /*
1762          * The journal_load will have done any necessary log recovery,
1763          * so we can safely mount the rest of the filesystem now.
1764          */
1765
1766         root = iget(sb, EXT3_ROOT_INO);
1767         sb->s_root = d_alloc_root(root);
1768         if (!sb->s_root) {
1769                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1770                 iput(root);
1771                 goto failed_mount4;
1772         }
1773         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1774                 dput(sb->s_root);
1775                 sb->s_root = NULL;
1776                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1777                 goto failed_mount4;
1778         }
1779
1780         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1781         /*
1782          * akpm: core read_super() calls in here with the superblock locked.
1783          * That deadlocks, because orphan cleanup needs to lock the superblock
1784          * in numerous places.  Here we just pop the lock - it's relatively
1785          * harmless, because we are now ready to accept write_super() requests,
1786          * and aviro says that's the only reason for hanging onto the
1787          * superblock lock.
1788          */
1789         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1790         ext3_orphan_cleanup(sb, es);
1791         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1792         if (needs_recovery)
1793                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1794         ext3_mark_recovery_complete(sb, es);
1795         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1796                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1797                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1798                 "writeback");
1799
1800         lock_kernel();
1801         return 0;
1802
1803 cantfind_ext3:
1804         if (!silent)
1805                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1806                        sb->s_id);
1807         goto failed_mount;
1808
1809 failed_mount4:
1810         journal_destroy(sbi->s_journal);
1811 failed_mount3:
1812         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1813         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1814         percpu_counter_destroy(&sbi->s_dirs_counter);
1815 failed_mount2:
1816         for (i = 0; i < db_count; i++)
1817                 brelse(sbi->s_group_desc[i]);
1818         kfree(sbi->s_group_desc);
1819 failed_mount:
1820 #ifdef CONFIG_QUOTA
1821         for (i = 0; i < MAXQUOTAS; i++)
1822                 kfree(sbi->s_qf_names[i]);
1823 #endif
1824         ext3_blkdev_remove(sbi);
1825         brelse(bh);
1826 out_fail:
1827         sb->s_fs_info = NULL;
1828         kfree(sbi);
1829         lock_kernel();
1830         return -EINVAL;
1831 }
1832
1833 /*
1834  * Setup any per-fs journal parameters now.  We'll do this both on
1835  * initial mount, once the journal has been initialised but before we've
1836  * done any recovery; and again on any subsequent remount.
1837  */
1838 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1839 {
1840         struct ext3_sb_info *sbi = EXT3_SB(sb);
1841
1842         if (sbi->s_commit_interval)
1843                 journal->j_commit_interval = sbi->s_commit_interval;
1844         /* We could also set up an ext3-specific default for the commit
1845          * interval here, but for now we'll just fall back to the jbd
1846          * default. */
1847
1848         spin_lock(&journal->j_state_lock);
1849         if (test_opt(sb, BARRIER))
1850                 journal->j_flags |= JFS_BARRIER;
1851         else
1852                 journal->j_flags &= ~JFS_BARRIER;
1853         spin_unlock(&journal->j_state_lock);
1854 }
1855
1856 static journal_t *ext3_get_journal(struct super_block *sb,
1857                                    unsigned int journal_inum)
1858 {
1859         struct inode *journal_inode;
1860         journal_t *journal;
1861
1862         /* First, test for the existence of a valid inode on disk.  Bad
1863          * things happen if we iget() an unused inode, as the subsequent
1864          * iput() will try to delete it. */
1865
1866         journal_inode = iget(sb, journal_inum);
1867         if (!journal_inode) {
1868                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1869                 return NULL;
1870         }
1871         if (!journal_inode->i_nlink) {
1872                 make_bad_inode(journal_inode);
1873                 iput(journal_inode);
1874                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1875                 return NULL;
1876         }
1877
1878         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1879                   journal_inode, journal_inode->i_size);
1880         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1881                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1882                 iput(journal_inode);
1883                 return NULL;
1884         }
1885
1886         journal = journal_init_inode(journal_inode);
1887         if (!journal) {
1888                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1889                 iput(journal_inode);
1890                 return NULL;
1891         }
1892         journal->j_private = sb;
1893         ext3_init_journal_params(sb, journal);
1894         return journal;
1895 }
1896
1897 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1898                                        dev_t j_dev)
1899 {
1900         struct buffer_head * bh;
1901         journal_t *journal;
1902         ext3_fsblk_t start;
1903         ext3_fsblk_t len;
1904         int hblock, blocksize;
1905         ext3_fsblk_t sb_block;
1906         unsigned long offset;
1907         struct ext3_super_block * es;
1908         struct block_device *bdev;
1909
1910         bdev = ext3_blkdev_get(j_dev);
1911         if (bdev == NULL)
1912                 return NULL;
1913
1914         if (bd_claim(bdev, sb)) {
1915                 printk(KERN_ERR
1916                         "EXT3: failed to claim external journal device.\n");
1917                 blkdev_put(bdev);
1918                 return NULL;
1919         }
1920
1921         blocksize = sb->s_blocksize;
1922         hblock = bdev_hardsect_size(bdev);
1923         if (blocksize < hblock) {
1924                 printk(KERN_ERR
1925                         "EXT3-fs: blocksize too small for journal device.\n");
1926                 goto out_bdev;
1927         }
1928
1929         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1930         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1931         set_blocksize(bdev, blocksize);
1932         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1933                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1934                        "external journal\n");
1935                 goto out_bdev;
1936         }
1937
1938         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1939         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1940             !(le32_to_cpu(es->s_feature_incompat) &
1941               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1942                 printk(KERN_ERR "EXT3-fs: external journal has "
1943                                         "bad superblock\n");
1944                 brelse(bh);
1945                 goto out_bdev;
1946         }
1947
1948         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1949                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1950                 brelse(bh);
1951                 goto out_bdev;
1952         }
1953
1954         len = le32_to_cpu(es->s_blocks_count);
1955         start = sb_block + 1;
1956         brelse(bh);     /* we're done with the superblock */
1957
1958         journal = journal_init_dev(bdev, sb->s_bdev,
1959                                         start, len, blocksize);
1960         if (!journal) {
1961                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1962                 goto out_bdev;
1963         }
1964         journal->j_private = sb;
1965         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1966         wait_on_buffer(journal->j_sb_buffer);
1967         if (!buffer_uptodate(journal->j_sb_buffer)) {
1968                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1969                 goto out_journal;
1970         }
1971         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1972                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1973                                         "user (unsupported) - %d\n",
1974                         be32_to_cpu(journal->j_superblock->s_nr_users));
1975                 goto out_journal;
1976         }
1977         EXT3_SB(sb)->journal_bdev = bdev;
1978         ext3_init_journal_params(sb, journal);
1979         return journal;
1980 out_journal:
1981         journal_destroy(journal);
1982 out_bdev:
1983         ext3_blkdev_put(bdev);
1984         return NULL;
1985 }
1986
1987 static int ext3_load_journal(struct super_block *sb,
1988                              struct ext3_super_block *es,
1989                              unsigned long journal_devnum)
1990 {
1991         journal_t *journal;
1992         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
1993         dev_t journal_dev;
1994         int err = 0;
1995         int really_read_only;
1996
1997         if (journal_devnum &&
1998             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1999                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2000                         "numbers have changed\n");
2001                 journal_dev = new_decode_dev(journal_devnum);
2002         } else
2003                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2004
2005         really_read_only = bdev_read_only(sb->s_bdev);
2006
2007         /*
2008          * Are we loading a blank journal or performing recovery after a
2009          * crash?  For recovery, we need to check in advance whether we
2010          * can get read-write access to the device.
2011          */
2012
2013         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2014                 if (sb->s_flags & MS_RDONLY) {
2015                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2016                                         "required on readonly filesystem.\n");
2017                         if (really_read_only) {
2018                                 printk(KERN_ERR "EXT3-fs: write access "
2019                                         "unavailable, cannot proceed.\n");
2020                                 return -EROFS;
2021                         }
2022                         printk (KERN_INFO "EXT3-fs: write access will "
2023                                         "be enabled during recovery.\n");
2024                 }
2025         }
2026
2027         if (journal_inum && journal_dev) {
2028                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2029                        "and inode journals!\n");
2030                 return -EINVAL;
2031         }
2032
2033         if (journal_inum) {
2034                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2035                         return -EINVAL;
2036         } else {
2037                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2038                         return -EINVAL;
2039         }
2040
2041         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2042                 err = journal_update_format(journal);
2043                 if (err)  {
2044                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2045                         journal_destroy(journal);
2046                         return err;
2047                 }
2048         }
2049
2050         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2051                 err = journal_wipe(journal, !really_read_only);
2052         if (!err)
2053                 err = journal_load(journal);
2054
2055         if (err) {
2056                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2057                 journal_destroy(journal);
2058                 return err;
2059         }
2060
2061         EXT3_SB(sb)->s_journal = journal;
2062         ext3_clear_journal_err(sb, es);
2063
2064         if (journal_devnum &&
2065             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2066                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2067                 sb->s_dirt = 1;
2068
2069                 /* Make sure we flush the recovery flag to disk. */
2070                 ext3_commit_super(sb, es, 1);
2071         }
2072
2073         return 0;
2074 }
2075
2076 static int ext3_create_journal(struct super_block * sb,
2077                                struct ext3_super_block * es,
2078                                unsigned int journal_inum)
2079 {
2080         journal_t *journal;
2081
2082         if (sb->s_flags & MS_RDONLY) {
2083                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2084                                 "create journal.\n");
2085                 return -EROFS;
2086         }
2087
2088         if (!(journal = ext3_get_journal(sb, journal_inum)))
2089                 return -EINVAL;
2090
2091         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2092                journal_inum);
2093
2094         if (journal_create(journal)) {
2095                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2096                 journal_destroy(journal);
2097                 return -EIO;
2098         }
2099
2100         EXT3_SB(sb)->s_journal = journal;
2101
2102         ext3_update_dynamic_rev(sb);
2103         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2104         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2105
2106         es->s_journal_inum = cpu_to_le32(journal_inum);
2107         sb->s_dirt = 1;
2108
2109         /* Make sure we flush the recovery flag to disk. */
2110         ext3_commit_super(sb, es, 1);
2111
2112         return 0;
2113 }
2114
2115 static void ext3_commit_super (struct super_block * sb,
2116                                struct ext3_super_block * es,
2117                                int sync)
2118 {
2119         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2120
2121         if (!sbh)
2122                 return;
2123         es->s_wtime = cpu_to_le32(get_seconds());
2124         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2125         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2126         BUFFER_TRACE(sbh, "marking dirty");
2127         mark_buffer_dirty(sbh);
2128         if (sync)
2129                 sync_dirty_buffer(sbh);
2130 }
2131
2132
2133 /*
2134  * Have we just finished recovery?  If so, and if we are mounting (or
2135  * remounting) the filesystem readonly, then we will end up with a
2136  * consistent fs on disk.  Record that fact.
2137  */
2138 static void ext3_mark_recovery_complete(struct super_block * sb,
2139                                         struct ext3_super_block * es)
2140 {
2141         journal_t *journal = EXT3_SB(sb)->s_journal;
2142
2143         journal_lock_updates(journal);
2144         journal_flush(journal);
2145         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2146             sb->s_flags & MS_RDONLY) {
2147                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2148                 sb->s_dirt = 0;
2149                 ext3_commit_super(sb, es, 1);
2150         }
2151         journal_unlock_updates(journal);
2152 }
2153
2154 /*
2155  * If we are mounting (or read-write remounting) a filesystem whose journal
2156  * has recorded an error from a previous lifetime, move that error to the
2157  * main filesystem now.
2158  */
2159 static void ext3_clear_journal_err(struct super_block * sb,
2160                                    struct ext3_super_block * es)
2161 {
2162         journal_t *journal;
2163         int j_errno;
2164         const char *errstr;
2165
2166         journal = EXT3_SB(sb)->s_journal;
2167
2168         /*
2169          * Now check for any error status which may have been recorded in the
2170          * journal by a prior ext3_error() or ext3_abort()
2171          */
2172
2173         j_errno = journal_errno(journal);
2174         if (j_errno) {
2175                 char nbuf[16];
2176
2177                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2178                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2179                              "from previous mount: %s", errstr);
2180                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2181                              "filesystem check.");
2182
2183                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2184                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2185                 ext3_commit_super (sb, es, 1);
2186
2187                 journal_clear_err(journal);
2188         }
2189 }
2190
2191 /*
2192  * Force the running and committing transactions to commit,
2193  * and wait on the commit.
2194  */
2195 int ext3_force_commit(struct super_block *sb)
2196 {
2197         journal_t *journal;
2198         int ret;
2199
2200         if (sb->s_flags & MS_RDONLY)
2201                 return 0;
2202
2203         journal = EXT3_SB(sb)->s_journal;
2204         sb->s_dirt = 0;
2205         ret = ext3_journal_force_commit(journal);
2206         return ret;
2207 }
2208
2209 /*
2210  * Ext3 always journals updates to the superblock itself, so we don't
2211  * have to propagate any other updates to the superblock on disk at this
2212  * point.  Just start an async writeback to get the buffers on their way
2213  * to the disk.
2214  *
2215  * This implicitly triggers the writebehind on sync().
2216  */
2217
2218 static void ext3_write_super (struct super_block * sb)
2219 {
2220         if (mutex_trylock(&sb->s_lock) != 0)
2221                 BUG();
2222         sb->s_dirt = 0;
2223 }
2224
2225 static int ext3_sync_fs(struct super_block *sb, int wait)
2226 {
2227         tid_t target;
2228
2229         sb->s_dirt = 0;
2230         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2231                 if (wait)
2232                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2233         }
2234         return 0;
2235 }
2236
2237 /*
2238  * LVM calls this function before a (read-only) snapshot is created.  This
2239  * gives us a chance to flush the journal completely and mark the fs clean.
2240  */
2241 static void ext3_write_super_lockfs(struct super_block *sb)
2242 {
2243         sb->s_dirt = 0;
2244
2245         if (!(sb->s_flags & MS_RDONLY)) {
2246                 journal_t *journal = EXT3_SB(sb)->s_journal;
2247
2248                 /* Now we set up the journal barrier. */
2249                 journal_lock_updates(journal);
2250                 journal_flush(journal);
2251
2252                 /* Journal blocked and flushed, clear needs_recovery flag. */
2253                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2254                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2255         }
2256 }
2257
2258 /*
2259  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2260  * flag here, even though the filesystem is not technically dirty yet.
2261  */
2262 static void ext3_unlockfs(struct super_block *sb)
2263 {
2264         if (!(sb->s_flags & MS_RDONLY)) {
2265                 lock_super(sb);
2266                 /* Reser the needs_recovery flag before the fs is unlocked. */
2267                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2268                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2269                 unlock_super(sb);
2270                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2271         }
2272 }
2273
2274 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2275 {
2276         struct ext3_super_block * es;
2277         struct ext3_sb_info *sbi = EXT3_SB(sb);
2278         ext3_fsblk_t n_blocks_count = 0;
2279         unsigned long old_sb_flags;
2280         struct ext3_mount_options old_opts;
2281         int err;
2282 #ifdef CONFIG_QUOTA
2283         int i;
2284 #endif
2285
2286         /* Store the original options */
2287         old_sb_flags = sb->s_flags;
2288         old_opts.s_mount_opt = sbi->s_mount_opt;
2289         old_opts.s_resuid = sbi->s_resuid;
2290         old_opts.s_resgid = sbi->s_resgid;
2291         old_opts.s_commit_interval = sbi->s_commit_interval;
2292 #ifdef CONFIG_QUOTA
2293         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2294         for (i = 0; i < MAXQUOTAS; i++)
2295                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2296 #endif
2297
2298         /*
2299          * Allow the "check" option to be passed as a remount option.
2300          */
2301         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2302                 err = -EINVAL;
2303                 goto restore_opts;
2304         }
2305
2306         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2307                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2308
2309         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2310                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2311
2312         es = sbi->s_es;
2313
2314         ext3_init_journal_params(sb, sbi->s_journal);
2315
2316         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2317                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2318                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2319                         err = -EROFS;
2320                         goto restore_opts;
2321                 }
2322
2323                 if (*flags & MS_RDONLY) {
2324                         /*
2325                          * First of all, the unconditional stuff we have to do
2326                          * to disable replay of the journal when we next remount
2327                          */
2328                         sb->s_flags |= MS_RDONLY;
2329
2330                         /*
2331                          * OK, test if we are remounting a valid rw partition
2332                          * readonly, and if so set the rdonly flag and then
2333                          * mark the partition as valid again.
2334                          */
2335                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2336                             (sbi->s_mount_state & EXT3_VALID_FS))
2337                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2338
2339                         ext3_mark_recovery_complete(sb, es);
2340                 } else {
2341                         __le32 ret;
2342                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2343                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2344                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2345                                        "remount RDWR because of unsupported "
2346                                        "optional features (%x).\n",
2347                                        sb->s_id, le32_to_cpu(ret));
2348                                 err = -EROFS;
2349                                 goto restore_opts;
2350                         }
2351
2352                         /*
2353                          * If we have an unprocessed orphan list hanging
2354                          * around from a previously readonly bdev mount,
2355                          * require a full umount/remount for now.
2356                          */
2357                         if (es->s_last_orphan) {
2358                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2359                                        "remount RDWR because of unprocessed "
2360                                        "orphan inode list.  Please "
2361                                        "umount/remount instead.\n",
2362                                        sb->s_id);
2363                                 err = -EINVAL;
2364                                 goto restore_opts;
2365                         }
2366
2367                         /*
2368                          * Mounting a RDONLY partition read-write, so reread
2369                          * and store the current valid flag.  (It may have
2370                          * been changed by e2fsck since we originally mounted
2371                          * the partition.)
2372                          */
2373                         ext3_clear_journal_err(sb, es);
2374                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2375                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2376                                 goto restore_opts;
2377                         if (!ext3_setup_super (sb, es, 0))
2378                                 sb->s_flags &= ~MS_RDONLY;
2379                 }
2380         }
2381 #ifdef CONFIG_QUOTA
2382         /* Release old quota file names */
2383         for (i = 0; i < MAXQUOTAS; i++)
2384                 if (old_opts.s_qf_names[i] &&
2385                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2386                         kfree(old_opts.s_qf_names[i]);
2387 #endif
2388         return 0;
2389 restore_opts:
2390         sb->s_flags = old_sb_flags;
2391         sbi->s_mount_opt = old_opts.s_mount_opt;
2392         sbi->s_resuid = old_opts.s_resuid;
2393         sbi->s_resgid = old_opts.s_resgid;
2394         sbi->s_commit_interval = old_opts.s_commit_interval;
2395 #ifdef CONFIG_QUOTA
2396         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2397         for (i = 0; i < MAXQUOTAS; i++) {
2398                 if (sbi->s_qf_names[i] &&
2399                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2400                         kfree(sbi->s_qf_names[i]);
2401                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2402         }
2403 #endif
2404         return err;
2405 }
2406
2407 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2408 {
2409         struct super_block *sb = dentry->d_sb;
2410         struct ext3_sb_info *sbi = EXT3_SB(sb);
2411         struct ext3_super_block *es = sbi->s_es;
2412         ext3_fsblk_t overhead;
2413         int i;
2414         u64 fsid;
2415
2416         if (test_opt (sb, MINIX_DF))
2417                 overhead = 0;
2418         else {
2419                 unsigned long ngroups;
2420                 ngroups = EXT3_SB(sb)->s_groups_count;
2421                 smp_rmb();
2422
2423                 /*
2424                  * Compute the overhead (FS structures)
2425                  */
2426
2427                 /*
2428                  * All of the blocks before first_data_block are
2429                  * overhead
2430                  */
2431                 overhead = le32_to_cpu(es->s_first_data_block);
2432
2433                 /*
2434                  * Add the overhead attributed to the superblock and
2435                  * block group descriptors.  If the sparse superblocks
2436                  * feature is turned on, then not all groups have this.
2437                  */
2438                 for (i = 0; i < ngroups; i++) {
2439                         overhead += ext3_bg_has_super(sb, i) +
2440                                 ext3_bg_num_gdb(sb, i);
2441                         cond_resched();
2442                 }
2443
2444                 /*
2445                  * Every block group has an inode bitmap, a block
2446                  * bitmap, and an inode table.
2447                  */
2448                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2449         }
2450
2451         buf->f_type = EXT3_SUPER_MAGIC;
2452         buf->f_bsize = sb->s_blocksize;
2453         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2454         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2455         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2456         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2457                 buf->f_bavail = 0;
2458         buf->f_files = le32_to_cpu(es->s_inodes_count);
2459         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2460         buf->f_namelen = EXT3_NAME_LEN;
2461         fsid = le64_to_cpup((void *)es->s_uuid) ^
2462                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2463         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2464         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2465         return 0;
2466 }
2467
2468 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2469  * is locked for write. Otherwise the are possible deadlocks:
2470  * Process 1                         Process 2
2471  * ext3_create()                     quota_sync()
2472  *   journal_start()                   write_dquot()
2473  *   DQUOT_INIT()                        down(dqio_mutex)
2474  *     down(dqio_mutex)                    journal_start()
2475  *
2476  */
2477
2478 #ifdef CONFIG_QUOTA
2479
2480 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2481 {
2482         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2483 }
2484
2485 static int ext3_dquot_initialize(struct inode *inode, int type)
2486 {
2487         handle_t *handle;
2488         int ret, err;
2489
2490         /* We may create quota structure so we need to reserve enough blocks */
2491         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2492         if (IS_ERR(handle))
2493                 return PTR_ERR(handle);
2494         ret = dquot_initialize(inode, type);
2495         err = ext3_journal_stop(handle);
2496         if (!ret)
2497                 ret = err;
2498         return ret;
2499 }
2500
2501 static int ext3_dquot_drop(struct inode *inode)
2502 {
2503         handle_t *handle;
2504         int ret, err;
2505
2506         /* We may delete quota structure so we need to reserve enough blocks */
2507         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2508         if (IS_ERR(handle))
2509                 return PTR_ERR(handle);
2510         ret = dquot_drop(inode);
2511         err = ext3_journal_stop(handle);
2512         if (!ret)
2513                 ret = err;
2514         return ret;
2515 }
2516
2517 static int ext3_write_dquot(struct dquot *dquot)
2518 {
2519         int ret, err;
2520         handle_t *handle;
2521         struct inode *inode;
2522
2523         inode = dquot_to_inode(dquot);
2524         handle = ext3_journal_start(inode,
2525                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2526         if (IS_ERR(handle))
2527                 return PTR_ERR(handle);
2528         ret = dquot_commit(dquot);
2529         err = ext3_journal_stop(handle);
2530         if (!ret)
2531                 ret = err;
2532         return ret;
2533 }
2534
2535 static int ext3_acquire_dquot(struct dquot *dquot)
2536 {
2537         int ret, err;
2538         handle_t *handle;
2539
2540         handle = ext3_journal_start(dquot_to_inode(dquot),
2541                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2542         if (IS_ERR(handle))
2543                 return PTR_ERR(handle);
2544         ret = dquot_acquire(dquot);
2545         err = ext3_journal_stop(handle);
2546         if (!ret)
2547                 ret = err;
2548         return ret;
2549 }
2550
2551 static int ext3_release_dquot(struct dquot *dquot)
2552 {
2553         int ret, err;
2554         handle_t *handle;
2555
2556         handle = ext3_journal_start(dquot_to_inode(dquot),
2557                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2558         if (IS_ERR(handle))
2559                 return PTR_ERR(handle);
2560         ret = dquot_release(dquot);
2561         err = ext3_journal_stop(handle);
2562         if (!ret)
2563                 ret = err;
2564         return ret;
2565 }
2566
2567 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2568 {
2569         /* Are we journalling quotas? */
2570         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2571             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2572                 dquot_mark_dquot_dirty(dquot);
2573                 return ext3_write_dquot(dquot);
2574         } else {
2575                 return dquot_mark_dquot_dirty(dquot);
2576         }
2577 }
2578
2579 static int ext3_write_info(struct super_block *sb, int type)
2580 {
2581         int ret, err;
2582         handle_t *handle;
2583
2584         /* Data block + inode block */
2585         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2586         if (IS_ERR(handle))
2587                 return PTR_ERR(handle);
2588         ret = dquot_commit_info(sb, type);
2589         err = ext3_journal_stop(handle);
2590         if (!ret)
2591                 ret = err;
2592         return ret;
2593 }
2594
2595 /*
2596  * Turn on quotas during mount time - we need to find
2597  * the quota file and such...
2598  */
2599 static int ext3_quota_on_mount(struct super_block *sb, int type)
2600 {
2601         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2602                         EXT3_SB(sb)->s_jquota_fmt, type);
2603 }
2604
2605 /*
2606  * Standard function to be called on quota_on
2607  */
2608 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2609                          char *path)
2610 {
2611         int err;
2612         struct nameidata nd;
2613
2614         if (!test_opt(sb, QUOTA))
2615                 return -EINVAL;
2616         /* Not journalling quota? */
2617         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2618             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2619                 return vfs_quota_on(sb, type, format_id, path);
2620         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2621         if (err)
2622                 return err;
2623         /* Quotafile not on the same filesystem? */
2624         if (nd.mnt->mnt_sb != sb) {
2625                 path_release(&nd);
2626                 return -EXDEV;
2627         }
2628         /* Quotafile not of fs root? */
2629         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2630                 printk(KERN_WARNING
2631                         "EXT3-fs: Quota file not on filesystem root. "
2632                         "Journalled quota will not work.\n");
2633         path_release(&nd);
2634         return vfs_quota_on(sb, type, format_id, path);
2635 }
2636
2637 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2638  * acquiring the locks... As quota files are never truncated and quota code
2639  * itself serializes the operations (and noone else should touch the files)
2640  * we don't have to be afraid of races */
2641 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2642                                size_t len, loff_t off)
2643 {
2644         struct inode *inode = sb_dqopt(sb)->files[type];
2645         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2646         int err = 0;
2647         int offset = off & (sb->s_blocksize - 1);
2648         int tocopy;
2649         size_t toread;
2650         struct buffer_head *bh;
2651         loff_t i_size = i_size_read(inode);
2652
2653         if (off > i_size)
2654                 return 0;
2655         if (off+len > i_size)
2656                 len = i_size-off;
2657         toread = len;
2658         while (toread > 0) {
2659                 tocopy = sb->s_blocksize - offset < toread ?
2660                                 sb->s_blocksize - offset : toread;
2661                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2662                 if (err)
2663                         return err;
2664                 if (!bh)        /* A hole? */
2665                         memset(data, 0, tocopy);
2666                 else
2667                         memcpy(data, bh->b_data+offset, tocopy);
2668                 brelse(bh);
2669                 offset = 0;
2670                 toread -= tocopy;
2671                 data += tocopy;
2672                 blk++;
2673         }
2674         return len;
2675 }
2676
2677 /* Write to quotafile (we know the transaction is already started and has
2678  * enough credits) */
2679 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2680                                 const char *data, size_t len, loff_t off)
2681 {
2682         struct inode *inode = sb_dqopt(sb)->files[type];
2683         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2684         int err = 0;
2685         int offset = off & (sb->s_blocksize - 1);
2686         int tocopy;
2687         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2688         size_t towrite = len;
2689         struct buffer_head *bh;
2690         handle_t *handle = journal_current_handle();
2691
2692         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2693         while (towrite > 0) {
2694                 tocopy = sb->s_blocksize - offset < towrite ?
2695                                 sb->s_blocksize - offset : towrite;
2696                 bh = ext3_bread(handle, inode, blk, 1, &err);
2697                 if (!bh)
2698                         goto out;
2699                 if (journal_quota) {
2700                         err = ext3_journal_get_write_access(handle, bh);
2701                         if (err) {
2702                                 brelse(bh);
2703                                 goto out;
2704                         }
2705                 }
2706                 lock_buffer(bh);
2707                 memcpy(bh->b_data+offset, data, tocopy);
2708                 flush_dcache_page(bh->b_page);
2709                 unlock_buffer(bh);
2710                 if (journal_quota)
2711                         err = ext3_journal_dirty_metadata(handle, bh);
2712                 else {
2713                         /* Always do at least ordered writes for quotas */
2714                         err = ext3_journal_dirty_data(handle, bh);
2715                         mark_buffer_dirty(bh);
2716                 }
2717                 brelse(bh);
2718                 if (err)
2719                         goto out;
2720                 offset = 0;
2721                 towrite -= tocopy;
2722                 data += tocopy;
2723                 blk++;
2724         }
2725 out:
2726         if (len == towrite)
2727                 return err;
2728         if (inode->i_size < off+len-towrite) {
2729                 i_size_write(inode, off+len-towrite);
2730                 EXT3_I(inode)->i_disksize = inode->i_size;
2731         }
2732         inode->i_version++;
2733         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2734         ext3_mark_inode_dirty(handle, inode);
2735         mutex_unlock(&inode->i_mutex);
2736         return len - towrite;
2737 }
2738
2739 #endif
2740
2741 static int ext3_get_sb(struct file_system_type *fs_type,
2742         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2743 {
2744         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2745 }
2746
2747 static struct file_system_type ext3_fs_type = {
2748         .owner          = THIS_MODULE,
2749         .name           = "ext3",
2750         .get_sb         = ext3_get_sb,
2751         .kill_sb        = kill_block_super,
2752         .fs_flags       = FS_REQUIRES_DEV,
2753 };
2754
2755 static int __init init_ext3_fs(void)
2756 {
2757         int err = init_ext3_xattr();
2758         if (err)
2759                 return err;
2760         err = init_inodecache();
2761         if (err)
2762                 goto out1;
2763         err = register_filesystem(&ext3_fs_type);
2764         if (err)
2765                 goto out;
2766         return 0;
2767 out:
2768         destroy_inodecache();
2769 out1:
2770         exit_ext3_xattr();
2771         return err;
2772 }
2773
2774 static void __exit exit_ext3_fs(void)
2775 {
2776         unregister_filesystem(&ext3_fs_type);
2777         destroy_inodecache();
2778         exit_ext3_xattr();
2779 }
2780
2781 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2782 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2783 MODULE_LICENSE("GPL");
2784 module_init(init_ext3_fs)
2785 module_exit(exit_ext3_fs)