fs/ext3/super.c

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