fs/nilfs2/super.c

   1 /*
   2  * super.c - NILFS module and super block management.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  */
  22 /*
  23  *  linux/fs/ext2/super.c
  24  *
  25  * Copyright (C) 1992, 1993, 1994, 1995
  26  * Remy Card (card@masi.ibp.fr)
  27  * Laboratoire MASI - Institut Blaise Pascal
  28  * Universite Pierre et Marie Curie (Paris VI)
  29  *
  30  *  from
  31  *
  32  *  linux/fs/minix/inode.c
  33  *
  34  *  Copyright (C) 1991, 1992  Linus Torvalds
  35  *
  36  *  Big-endian to little-endian byte-swapping/bitmaps by
  37  *        David S. Miller (davem@caip.rutgers.edu), 1995
  38  */
  39
  40 #include <linux/module.h>
  41 #include <linux/string.h>
  42 #include <linux/slab.h>
  43 #include <linux/init.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/parser.h>
  46 #include <linux/random.h>
  47 #include <linux/crc32.h>
  48 #include <linux/smp_lock.h>
  49 #include <linux/vfs.h>
  50 #include <linux/writeback.h>
  51 #include <linux/kobject.h>
  52 #include <linux/exportfs.h>
  53 #include <linux/seq_file.h>
  54 #include <linux/mount.h>
  55 #include "nilfs.h"
  56 #include "mdt.h"
  57 #include "alloc.h"
  58 #include "btree.h"
  59 #include "btnode.h"
  60 #include "page.h"
  61 #include "cpfile.h"
  62 #include "ifile.h"
  63 #include "dat.h"
  64 #include "segment.h"
  65 #include "segbuf.h"
  66
  67 MODULE_AUTHOR("NTT Corp.");
  68 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  69                    "(NILFS)");
  70 MODULE_LICENSE("GPL");
  71
  72 struct kmem_cache *nilfs_inode_cachep;
  73 struct kmem_cache *nilfs_transaction_cachep;
  74 struct kmem_cache *nilfs_segbuf_cachep;
  75 struct kmem_cache *nilfs_btree_path_cache;
  76
  77 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  78
  79 static void nilfs_set_error(struct nilfs_sb_info *sbi)
  80 {
  81         struct the_nilfs *nilfs = sbi->s_nilfs;
  82         struct nilfs_super_block **sbp;
  83
  84         down_write(&nilfs->ns_sem);
  85         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  86                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  87                 sbp = nilfs_prepare_super(sbi, 0);
  88                 if (likely(sbp)) {
  89                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  90                         if (sbp[1])
  91                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  92                         nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
  93                 }
  94         }
  95         up_write(&nilfs->ns_sem);
  96 }
  97
  98 /**
  99  * nilfs_error() - report failure condition on a filesystem
 100  *
 101  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
 102  * reporting an error message.  It should be called when NILFS detects
 103  * incoherences or defects of meta data on disk.  As for sustainable
 104  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
 105  * function should be used instead.
 106  *
 107  * The segment constructor must not call this function because it can
 108  * kill itself.
 109  */
 110 void nilfs_error(struct super_block *sb, const char *function,
 111                  const char *fmt, ...)
 112 {
 113         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 114         va_list args;
 115
 116         va_start(args, fmt);
 117         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
 118         vprintk(fmt, args);
 119         printk("\n");
 120         va_end(args);
 121
 122         if (!(sb->s_flags & MS_RDONLY)) {
 123                 nilfs_set_error(sbi);
 124
 125                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
 126                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 127                         sb->s_flags |= MS_RDONLY;
 128                 }
 129         }
 130
 131         if (nilfs_test_opt(sbi, ERRORS_PANIC))
 132                 panic("NILFS (device %s): panic forced after error\n",
 133                       sb->s_id);
 134 }
 135
 136 void nilfs_warning(struct super_block *sb, const char *function,
 137                    const char *fmt, ...)
 138 {
 139         va_list args;
 140
 141         va_start(args, fmt);
 142         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
 143                sb->s_id, function);
 144         vprintk(fmt, args);
 145         printk("\n");
 146         va_end(args);
 147 }
 148
 149
 150 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
 151 {
 152         struct nilfs_inode_info *ii;
 153
 154         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 155         if (!ii)
 156                 return NULL;
 157         ii->i_bh = NULL;
 158         ii->i_state = 0;
 159         ii->vfs_inode.i_version = 1;
 160         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
 161         return &ii->vfs_inode;
 162 }
 163
 164 struct inode *nilfs_alloc_inode(struct super_block *sb)
 165 {
 166         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
 167 }
 168
 169 void nilfs_destroy_inode(struct inode *inode)
 170 {
 171         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 172 }
 173
 174 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
 175 {
 176         struct the_nilfs *nilfs = sbi->s_nilfs;
 177         int err;
 178         int barrier_done = 0;
 179
 180         if (nilfs_test_opt(sbi, BARRIER)) {
 181                 set_buffer_ordered(nilfs->ns_sbh[0]);
 182                 barrier_done = 1;
 183         }
 184  retry:
 185         set_buffer_dirty(nilfs->ns_sbh[0]);
 186         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 187         if (err == -EOPNOTSUPP && barrier_done) {
 188                 nilfs_warning(sbi->s_super, __func__,
 189                               "barrier-based sync failed. "
 190                               "disabling barriers\n");
 191                 nilfs_clear_opt(sbi, BARRIER);
 192                 barrier_done = 0;
 193                 clear_buffer_ordered(nilfs->ns_sbh[0]);
 194                 goto retry;
 195         }
 196         if (unlikely(err)) {
 197                 printk(KERN_ERR
 198                        "NILFS: unable to write superblock (err=%d)\n", err);
 199                 if (err == -EIO && nilfs->ns_sbh[1]) {
 200                         /*
 201                          * sbp[0] points to newer log than sbp[1],
 202                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 203                          */
 204                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 205                                nilfs->ns_sbsize);
 206                         nilfs_fall_back_super_block(nilfs);
 207                         goto retry;
 208                 }
 209         } else {
 210                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 211
 212                 nilfs->ns_sbwcount++;
 213
 214                 /*
 215                  * The latest segment becomes trailable from the position
 216                  * written in superblock.
 217                  */
 218                 clear_nilfs_discontinued(nilfs);
 219
 220                 /* update GC protection for recent segments */
 221                 if (nilfs->ns_sbh[1]) {
 222                         if (flag == NILFS_SB_COMMIT_ALL) {
 223                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 224                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 225                                         goto out;
 226                         }
 227                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 228                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 229                                 sbp = nilfs->ns_sbp[1];
 230                 }
 231
 232                 spin_lock(&nilfs->ns_last_segment_lock);
 233                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 234                 spin_unlock(&nilfs->ns_last_segment_lock);
 235         }
 236  out:
 237         return err;
 238 }
 239
 240 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 241                           struct the_nilfs *nilfs)
 242 {
 243         sector_t nfreeblocks;
 244
 245         /* nilfs->ns_sem must be locked by the caller. */
 246         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 247         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 248
 249         spin_lock(&nilfs->ns_last_segment_lock);
 250         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 251         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 252         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 253         spin_unlock(&nilfs->ns_last_segment_lock);
 254 }
 255
 256 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
 257                                                int flip)
 258 {
 259         struct the_nilfs *nilfs = sbi->s_nilfs;
 260         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 261
 262         /* nilfs->ns_sem must be locked by the caller. */
 263         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 264                 if (sbp[1] &&
 265                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 266                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 267                 } else {
 268                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
 269                                sbi->s_super->s_id);
 270                         return NULL;
 271                 }
 272         } else if (sbp[1] &&
 273                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 274                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 275         }
 276
 277         if (flip && sbp[1])
 278                 nilfs_swap_super_block(nilfs);
 279
 280         return sbp;
 281 }
 282
 283 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
 284 {
 285         struct the_nilfs *nilfs = sbi->s_nilfs;
 286         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 287         time_t t;
 288
 289         /* nilfs->ns_sem must be locked by the caller. */
 290         t = get_seconds();
 291         nilfs->ns_sbwtime = t;
 292         sbp[0]->s_wtime = cpu_to_le64(t);
 293         sbp[0]->s_sum = 0;
 294         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 295                                              (unsigned char *)sbp[0],
 296                                              nilfs->ns_sbsize));
 297         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 298                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 299                 sbp[1]->s_sum = 0;
 300                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 301                                             (unsigned char *)sbp[1],
 302                                             nilfs->ns_sbsize));
 303         }
 304         clear_nilfs_sb_dirty(nilfs);
 305         return nilfs_sync_super(sbi, flag);
 306 }
 307
 308 /**
 309  * nilfs_cleanup_super() - write filesystem state for cleanup
 310  * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
 311  *
 312  * This function restores state flags in the on-disk super block.
 313  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 314  * filesystem was not clean previously.
 315  */
 316 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
 317 {
 318         struct nilfs_super_block **sbp;
 319         int flag = NILFS_SB_COMMIT;
 320         int ret = -EIO;
 321
 322         sbp = nilfs_prepare_super(sbi, 0);
 323         if (sbp) {
 324                 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
 325                 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
 326                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 327                         /*
 328                          * make the "clean" flag also to the opposite
 329                          * super block if both super blocks point to
 330                          * the same checkpoint.
 331                          */
 332                         sbp[1]->s_state = sbp[0]->s_state;
 333                         flag = NILFS_SB_COMMIT_ALL;
 334                 }
 335                 ret = nilfs_commit_super(sbi, flag);
 336         }
 337         return ret;
 338 }
 339
 340 static void nilfs_put_super(struct super_block *sb)
 341 {
 342         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 343         struct the_nilfs *nilfs = sbi->s_nilfs;
 344
 345         lock_kernel();
 346
 347         nilfs_detach_segment_constructor(sbi);
 348
 349         if (!(sb->s_flags & MS_RDONLY)) {
 350                 down_write(&nilfs->ns_sem);
 351                 nilfs_cleanup_super(sbi);
 352                 up_write(&nilfs->ns_sem);
 353         }
 354         down_write(&nilfs->ns_super_sem);
 355         if (nilfs->ns_current == sbi)
 356                 nilfs->ns_current = NULL;
 357         up_write(&nilfs->ns_super_sem);
 358
 359         nilfs_detach_checkpoint(sbi);
 360         put_nilfs(sbi->s_nilfs);
 361         sbi->s_super = NULL;
 362         sb->s_fs_info = NULL;
 363         nilfs_put_sbinfo(sbi);
 364
 365         unlock_kernel();
 366 }
 367
 368 static int nilfs_sync_fs(struct super_block *sb, int wait)
 369 {
 370         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 371         struct the_nilfs *nilfs = sbi->s_nilfs;
 372         struct nilfs_super_block **sbp;
 373         int err = 0;
 374
 375         /* This function is called when super block should be written back */
 376         if (wait)
 377                 err = nilfs_construct_segment(sb);
 378
 379         down_write(&nilfs->ns_sem);
 380         if (nilfs_sb_dirty(nilfs)) {
 381                 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
 382                 if (likely(sbp)) {
 383                         nilfs_set_log_cursor(sbp[0], nilfs);
 384                         nilfs_commit_super(sbi, NILFS_SB_COMMIT);
 385                 }
 386         }
 387         up_write(&nilfs->ns_sem);
 388
 389         return err;
 390 }
 391
 392 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
 393 {
 394         struct the_nilfs *nilfs = sbi->s_nilfs;
 395         struct nilfs_checkpoint *raw_cp;
 396         struct buffer_head *bh_cp;
 397         int err;
 398
 399         down_write(&nilfs->ns_super_sem);
 400         list_add(&sbi->s_list, &nilfs->ns_supers);
 401         up_write(&nilfs->ns_super_sem);
 402
 403         err = -ENOMEM;
 404         sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
 405         if (!sbi->s_ifile)
 406                 goto delist;
 407
 408         down_read(&nilfs->ns_segctor_sem);
 409         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 410                                           &bh_cp);
 411         up_read(&nilfs->ns_segctor_sem);
 412         if (unlikely(err)) {
 413                 if (err == -ENOENT || err == -EINVAL) {
 414                         printk(KERN_ERR
 415                                "NILFS: Invalid checkpoint "
 416                                "(checkpoint number=%llu)\n",
 417                                (unsigned long long)cno);
 418                         err = -EINVAL;
 419                 }
 420                 goto failed;
 421         }
 422         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
 423         if (unlikely(err))
 424                 goto failed_bh;
 425         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
 426         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
 427
 428         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 429         return 0;
 430
 431  failed_bh:
 432         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 433  failed:
 434         nilfs_mdt_destroy(sbi->s_ifile);
 435         sbi->s_ifile = NULL;
 436
 437  delist:
 438         down_write(&nilfs->ns_super_sem);
 439         list_del_init(&sbi->s_list);
 440         up_write(&nilfs->ns_super_sem);
 441
 442         return err;
 443 }
 444
 445 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
 446 {
 447         struct the_nilfs *nilfs = sbi->s_nilfs;
 448
 449         nilfs_mdt_destroy(sbi->s_ifile);
 450         sbi->s_ifile = NULL;
 451         down_write(&nilfs->ns_super_sem);
 452         list_del_init(&sbi->s_list);
 453         up_write(&nilfs->ns_super_sem);
 454 }
 455
 456 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 457 {
 458         struct super_block *sb = dentry->d_sb;
 459         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 460         struct the_nilfs *nilfs = sbi->s_nilfs;
 461         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 462         unsigned long long blocks;
 463         unsigned long overhead;
 464         unsigned long nrsvblocks;
 465         sector_t nfreeblocks;
 466         int err;
 467
 468         /*
 469          * Compute all of the segment blocks
 470          *
 471          * The blocks before first segment and after last segment
 472          * are excluded.
 473          */
 474         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 475                 - nilfs->ns_first_data_block;
 476         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 477
 478         /*
 479          * Compute the overhead
 480          *
 481          * When distributing meta data blocks outside segment structure,
 482          * We must count them as the overhead.
 483          */
 484         overhead = 0;
 485
 486         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 487         if (unlikely(err))
 488                 return err;
 489
 490         buf->f_type = NILFS_SUPER_MAGIC;
 491         buf->f_bsize = sb->s_blocksize;
 492         buf->f_blocks = blocks - overhead;
 493         buf->f_bfree = nfreeblocks;
 494         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 495                 (buf->f_bfree - nrsvblocks) : 0;
 496         buf->f_files = atomic_read(&sbi->s_inodes_count);
 497         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
 498         buf->f_namelen = NILFS_NAME_LEN;
 499         buf->f_fsid.val[0] = (u32)id;
 500         buf->f_fsid.val[1] = (u32)(id >> 32);
 501
 502         return 0;
 503 }
 504
 505 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 506 {
 507         struct super_block *sb = vfs->mnt_sb;
 508         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 509
 510         if (!nilfs_test_opt(sbi, BARRIER))
 511                 seq_puts(seq, ",nobarrier");
 512         if (nilfs_test_opt(sbi, SNAPSHOT))
 513                 seq_printf(seq, ",cp=%llu",
 514                            (unsigned long long int)sbi->s_snapshot_cno);
 515         if (nilfs_test_opt(sbi, ERRORS_PANIC))
 516                 seq_puts(seq, ",errors=panic");
 517         if (nilfs_test_opt(sbi, ERRORS_CONT))
 518                 seq_puts(seq, ",errors=continue");
 519         if (nilfs_test_opt(sbi, STRICT_ORDER))
 520                 seq_puts(seq, ",order=strict");
 521         if (nilfs_test_opt(sbi, NORECOVERY))
 522                 seq_puts(seq, ",norecovery");
 523         if (nilfs_test_opt(sbi, DISCARD))
 524                 seq_puts(seq, ",discard");
 525
 526         return 0;
 527 }
 528
 529 static const struct super_operations nilfs_sops = {
 530         .alloc_inode    = nilfs_alloc_inode,
 531         .destroy_inode  = nilfs_destroy_inode,
 532         .dirty_inode    = nilfs_dirty_inode,
 533         /* .write_inode    = nilfs_write_inode, */
 534         /* .put_inode      = nilfs_put_inode, */
 535         /* .drop_inode    = nilfs_drop_inode, */
 536         .evict_inode    = nilfs_evict_inode,
 537         .put_super      = nilfs_put_super,
 538         /* .write_super    = nilfs_write_super, */
 539         .sync_fs        = nilfs_sync_fs,
 540         /* .write_super_lockfs */
 541         /* .unlockfs */
 542         .statfs         = nilfs_statfs,
 543         .remount_fs     = nilfs_remount,
 544         /* .umount_begin */
 545         .show_options = nilfs_show_options
 546 };
 547
 548 static struct inode *
 549 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
 550 {
 551         struct inode *inode;
 552
 553         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
 554             ino != NILFS_SKETCH_INO)
 555                 return ERR_PTR(-ESTALE);
 556
 557         inode = nilfs_iget(sb, ino);
 558         if (IS_ERR(inode))
 559                 return ERR_CAST(inode);
 560         if (generation && inode->i_generation != generation) {
 561                 iput(inode);
 562                 return ERR_PTR(-ESTALE);
 563         }
 564
 565         return inode;
 566 }
 567
 568 static struct dentry *
 569 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
 570                    int fh_type)
 571 {
 572         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 573                                     nilfs_nfs_get_inode);
 574 }
 575
 576 static struct dentry *
 577 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
 578                    int fh_type)
 579 {
 580         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 581                                     nilfs_nfs_get_inode);
 582 }
 583
 584 static const struct export_operations nilfs_export_ops = {
 585         .fh_to_dentry = nilfs_fh_to_dentry,
 586         .fh_to_parent = nilfs_fh_to_parent,
 587         .get_parent = nilfs_get_parent,
 588 };
 589
 590 enum {
 591         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 592         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 593         Opt_discard, Opt_nodiscard, Opt_err,
 594 };
 595
 596 static match_table_t tokens = {
 597         {Opt_err_cont, "errors=continue"},
 598         {Opt_err_panic, "errors=panic"},
 599         {Opt_err_ro, "errors=remount-ro"},
 600         {Opt_barrier, "barrier"},
 601         {Opt_nobarrier, "nobarrier"},
 602         {Opt_snapshot, "cp=%u"},
 603         {Opt_order, "order=%s"},
 604         {Opt_norecovery, "norecovery"},
 605         {Opt_discard, "discard"},
 606         {Opt_nodiscard, "nodiscard"},
 607         {Opt_err, NULL}
 608 };
 609
 610 static int parse_options(char *options, struct super_block *sb, int is_remount)
 611 {
 612         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 613         char *p;
 614         substring_t args[MAX_OPT_ARGS];
 615         int option;
 616
 617         if (!options)
 618                 return 1;
 619
 620         while ((p = strsep(&options, ",")) != NULL) {
 621                 int token;
 622                 if (!*p)
 623                         continue;
 624
 625                 token = match_token(p, tokens, args);
 626                 switch (token) {
 627                 case Opt_barrier:
 628                         nilfs_set_opt(sbi, BARRIER);
 629                         break;
 630                 case Opt_nobarrier:
 631                         nilfs_clear_opt(sbi, BARRIER);
 632                         break;
 633                 case Opt_order:
 634                         if (strcmp(args[0].from, "relaxed") == 0)
 635                                 /* Ordered data semantics */
 636                                 nilfs_clear_opt(sbi, STRICT_ORDER);
 637                         else if (strcmp(args[0].from, "strict") == 0)
 638                                 /* Strict in-order semantics */
 639                                 nilfs_set_opt(sbi, STRICT_ORDER);
 640                         else
 641                                 return 0;
 642                         break;
 643                 case Opt_err_panic:
 644                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
 645                         break;
 646                 case Opt_err_ro:
 647                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
 648                         break;
 649                 case Opt_err_cont:
 650                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
 651                         break;
 652                 case Opt_snapshot:
 653                         if (match_int(&args[0], &option) || option <= 0)
 654                                 return 0;
 655                         if (is_remount) {
 656                                 if (!nilfs_test_opt(sbi, SNAPSHOT)) {
 657                                         printk(KERN_ERR
 658                                                "NILFS: cannot change regular "
 659                                                "mount to snapshot.\n");
 660                                         return 0;
 661                                 } else if (option != sbi->s_snapshot_cno) {
 662                                         printk(KERN_ERR
 663                                                "NILFS: cannot remount to a "
 664                                                "different snapshot.\n");
 665                                         return 0;
 666                                 }
 667                                 break;
 668                         }
 669                         if (!(sb->s_flags & MS_RDONLY)) {
 670                                 printk(KERN_ERR "NILFS: cannot mount snapshot "
 671                                        "read/write.  A read-only option is "
 672                                        "required.\n");
 673                                 return 0;
 674                         }
 675                         sbi->s_snapshot_cno = option;
 676                         nilfs_set_opt(sbi, SNAPSHOT);
 677                         break;
 678                 case Opt_norecovery:
 679                         nilfs_set_opt(sbi, NORECOVERY);
 680                         break;
 681                 case Opt_discard:
 682                         nilfs_set_opt(sbi, DISCARD);
 683                         break;
 684                 case Opt_nodiscard:
 685                         nilfs_clear_opt(sbi, DISCARD);
 686                         break;
 687                 default:
 688                         printk(KERN_ERR
 689                                "NILFS: Unrecognized mount option \"%s\"\n", p);
 690                         return 0;
 691                 }
 692         }
 693         return 1;
 694 }
 695
 696 static inline void
 697 nilfs_set_default_options(struct nilfs_sb_info *sbi,
 698                           struct nilfs_super_block *sbp)
 699 {
 700         sbi->s_mount_opt =
 701                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 702 }
 703
 704 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
 705 {
 706         struct the_nilfs *nilfs = sbi->s_nilfs;
 707         struct nilfs_super_block **sbp;
 708         int max_mnt_count;
 709         int mnt_count;
 710
 711         /* nilfs->ns_sem must be locked by the caller. */
 712         sbp = nilfs_prepare_super(sbi, 0);
 713         if (!sbp)
 714                 return -EIO;
 715
 716         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 717         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 718
 719         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 720                 printk(KERN_WARNING
 721                        "NILFS warning: mounting fs with errors\n");
 722 #if 0
 723         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 724                 printk(KERN_WARNING
 725                        "NILFS warning: maximal mount count reached\n");
 726 #endif
 727         }
 728         if (!max_mnt_count)
 729                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 730
 731         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 732         sbp[0]->s_state =
 733                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 734         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
 735         /* synchronize sbp[1] with sbp[0] */
 736         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 737         return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
 738 }
 739
 740 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 741                                                  u64 pos, int blocksize,
 742                                                  struct buffer_head **pbh)
 743 {
 744         unsigned long long sb_index = pos;
 745         unsigned long offset;
 746
 747         offset = do_div(sb_index, blocksize);
 748         *pbh = sb_bread(sb, sb_index);
 749         if (!*pbh)
 750                 return NULL;
 751         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 752 }
 753
 754 int nilfs_store_magic_and_option(struct super_block *sb,
 755                                  struct nilfs_super_block *sbp,
 756                                  char *data)
 757 {
 758         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 759
 760         sb->s_magic = le16_to_cpu(sbp->s_magic);
 761
 762         /* FS independent flags */
 763 #ifdef NILFS_ATIME_DISABLE
 764         sb->s_flags |= MS_NOATIME;
 765 #endif
 766
 767         nilfs_set_default_options(sbi, sbp);
 768
 769         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
 770         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
 771         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
 772         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
 773
 774         return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
 775 }
 776
 777 int nilfs_check_feature_compatibility(struct super_block *sb,
 778                                       struct nilfs_super_block *sbp)
 779 {
 780         __u64 features;
 781
 782         features = le64_to_cpu(sbp->s_feature_incompat) &
 783                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 784         if (features) {
 785                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
 786                        "optional features (%llx)\n",
 787                        (unsigned long long)features);
 788                 return -EINVAL;
 789         }
 790         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 791                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 792         if (!(sb->s_flags & MS_RDONLY) && features) {
 793                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
 794                        "unsupported optional features (%llx)\n",
 795                        (unsigned long long)features);
 796                 return -EINVAL;
 797         }
 798         return 0;
 799 }
 800
 801 /**
 802  * nilfs_fill_super() - initialize a super block instance
 803  * @sb: super_block
 804  * @data: mount options
 805  * @silent: silent mode flag
 806  * @nilfs: the_nilfs struct
 807  *
 808  * This function is called exclusively by nilfs->ns_mount_mutex.
 809  * So, the recovery process is protected from other simultaneous mounts.
 810  */
 811 static int
 812 nilfs_fill_super(struct super_block *sb, void *data, int silent,
 813                  struct the_nilfs *nilfs)
 814 {
 815         struct nilfs_sb_info *sbi;
 816         struct inode *root;
 817         __u64 cno;
 818         int err;
 819
 820         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
 821         if (!sbi)
 822                 return -ENOMEM;
 823
 824         sb->s_fs_info = sbi;
 825
 826         get_nilfs(nilfs);
 827         sbi->s_nilfs = nilfs;
 828         sbi->s_super = sb;
 829         atomic_set(&sbi->s_count, 1);
 830
 831         err = init_nilfs(nilfs, sbi, (char *)data);
 832         if (err)
 833                 goto failed_sbi;
 834
 835         spin_lock_init(&sbi->s_inode_lock);
 836         INIT_LIST_HEAD(&sbi->s_dirty_files);
 837         INIT_LIST_HEAD(&sbi->s_list);
 838
 839         /*
 840          * Following initialization is overlapped because
 841          * nilfs_sb_info structure has been cleared at the beginning.
 842          * But we reserve them to keep our interest and make ready
 843          * for the future change.
 844          */
 845         get_random_bytes(&sbi->s_next_generation,
 846                          sizeof(sbi->s_next_generation));
 847         spin_lock_init(&sbi->s_next_gen_lock);
 848
 849         sb->s_op = &nilfs_sops;
 850         sb->s_export_op = &nilfs_export_ops;
 851         sb->s_root = NULL;
 852         sb->s_time_gran = 1;
 853         sb->s_bdi = nilfs->ns_bdi;
 854
 855         err = load_nilfs(nilfs, sbi);
 856         if (err)
 857                 goto failed_sbi;
 858
 859         cno = nilfs_last_cno(nilfs);
 860
 861         if (sb->s_flags & MS_RDONLY) {
 862                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
 863                         down_read(&nilfs->ns_segctor_sem);
 864                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
 865                                                        sbi->s_snapshot_cno);
 866                         up_read(&nilfs->ns_segctor_sem);
 867                         if (err < 0) {
 868                                 if (err == -ENOENT)
 869                                         err = -EINVAL;
 870                                 goto failed_sbi;
 871                         }
 872                         if (!err) {
 873                                 printk(KERN_ERR
 874                                        "NILFS: The specified checkpoint is "
 875                                        "not a snapshot "
 876                                        "(checkpoint number=%llu).\n",
 877                                        (unsigned long long)sbi->s_snapshot_cno);
 878                                 err = -EINVAL;
 879                                 goto failed_sbi;
 880                         }
 881                         cno = sbi->s_snapshot_cno;
 882                 }
 883         }
 884
 885         err = nilfs_attach_checkpoint(sbi, cno);
 886         if (err) {
 887                 printk(KERN_ERR "NILFS: error loading a checkpoint"
 888                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
 889                 goto failed_sbi;
 890         }
 891
 892         if (!(sb->s_flags & MS_RDONLY)) {
 893                 err = nilfs_attach_segment_constructor(sbi);
 894                 if (err)
 895                         goto failed_checkpoint;
 896         }
 897
 898         root = nilfs_iget(sb, NILFS_ROOT_INO);
 899         if (IS_ERR(root)) {
 900                 printk(KERN_ERR "NILFS: get root inode failed\n");
 901                 err = PTR_ERR(root);
 902                 goto failed_segctor;
 903         }
 904         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
 905                 iput(root);
 906                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
 907                 err = -EINVAL;
 908                 goto failed_segctor;
 909         }
 910         sb->s_root = d_alloc_root(root);
 911         if (!sb->s_root) {
 912                 iput(root);
 913                 printk(KERN_ERR "NILFS: get root dentry failed\n");
 914                 err = -ENOMEM;
 915                 goto failed_segctor;
 916         }
 917
 918         if (!(sb->s_flags & MS_RDONLY)) {
 919                 down_write(&nilfs->ns_sem);
 920                 nilfs_setup_super(sbi);
 921                 up_write(&nilfs->ns_sem);
 922         }
 923
 924         down_write(&nilfs->ns_super_sem);
 925         if (!nilfs_test_opt(sbi, SNAPSHOT))
 926                 nilfs->ns_current = sbi;
 927         up_write(&nilfs->ns_super_sem);
 928
 929         return 0;
 930
 931  failed_segctor:
 932         nilfs_detach_segment_constructor(sbi);
 933
 934  failed_checkpoint:
 935         nilfs_detach_checkpoint(sbi);
 936
 937  failed_sbi:
 938         put_nilfs(nilfs);
 939         sb->s_fs_info = NULL;
 940         nilfs_put_sbinfo(sbi);
 941         return err;
 942 }
 943
 944 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
 945 {
 946         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 947         struct the_nilfs *nilfs = sbi->s_nilfs;
 948         unsigned long old_sb_flags;
 949         struct nilfs_mount_options old_opts;
 950         int was_snapshot, err;
 951
 952         lock_kernel();
 953
 954         down_write(&nilfs->ns_super_sem);
 955         old_sb_flags = sb->s_flags;
 956         old_opts.mount_opt = sbi->s_mount_opt;
 957         old_opts.snapshot_cno = sbi->s_snapshot_cno;
 958         was_snapshot = nilfs_test_opt(sbi, SNAPSHOT);
 959
 960         if (!parse_options(data, sb, 1)) {
 961                 err = -EINVAL;
 962                 goto restore_opts;
 963         }
 964         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
 965
 966         err = -EINVAL;
 967         if (was_snapshot && !(*flags & MS_RDONLY)) {
 968                 printk(KERN_ERR "NILFS (device %s): cannot remount snapshot "
 969                        "read/write.\n", sb->s_id);
 970                 goto restore_opts;
 971         }
 972
 973         if (!nilfs_valid_fs(nilfs)) {
 974                 printk(KERN_WARNING "NILFS (device %s): couldn't "
 975                        "remount because the filesystem is in an "
 976                        "incomplete recovery state.\n", sb->s_id);
 977                 goto restore_opts;
 978         }
 979
 980         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 981                 goto out;
 982         if (*flags & MS_RDONLY) {
 983                 /* Shutting down the segment constructor */
 984                 nilfs_detach_segment_constructor(sbi);
 985                 sb->s_flags |= MS_RDONLY;
 986
 987                 /*
 988                  * Remounting a valid RW partition RDONLY, so set
 989                  * the RDONLY flag and then mark the partition as valid again.
 990                  */
 991                 down_write(&nilfs->ns_sem);
 992                 nilfs_cleanup_super(sbi);
 993                 up_write(&nilfs->ns_sem);
 994         } else {
 995                 __u64 features;
 996
 997                 /*
 998                  * Mounting a RDONLY partition read-write, so reread and
 999                  * store the current valid flag.  (It may have been changed
1000                  * by fsck since we originally mounted the partition.)
1001                  */
1002                 down_read(&nilfs->ns_sem);
1003                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1004                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1005                 up_read(&nilfs->ns_sem);
1006                 if (features) {
1007                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1008                                "remount RDWR because of unsupported optional "
1009                                "features (%llx)\n",
1010                                sb->s_id, (unsigned long long)features);
1011                         err = -EROFS;
1012                         goto restore_opts;
1013                 }
1014
1015                 sb->s_flags &= ~MS_RDONLY;
1016
1017                 err = nilfs_attach_segment_constructor(sbi);
1018                 if (err)
1019                         goto restore_opts;
1020
1021                 down_write(&nilfs->ns_sem);
1022                 nilfs_setup_super(sbi);
1023                 up_write(&nilfs->ns_sem);
1024         }
1025  out:
1026         up_write(&nilfs->ns_super_sem);
1027         unlock_kernel();
1028         return 0;
1029
1030  restore_opts:
1031         sb->s_flags = old_sb_flags;
1032         sbi->s_mount_opt = old_opts.mount_opt;
1033         sbi->s_snapshot_cno = old_opts.snapshot_cno;
1034         up_write(&nilfs->ns_super_sem);
1035         unlock_kernel();
1036         return err;
1037 }
1038
1039 struct nilfs_super_data {
1040         struct block_device *bdev;
1041         struct nilfs_sb_info *sbi;
1042         __u64 cno;
1043         int flags;
1044 };
1045
1046 /**
1047  * nilfs_identify - pre-read mount options needed to identify mount instance
1048  * @data: mount options
1049  * @sd: nilfs_super_data
1050  */
1051 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1052 {
1053         char *p, *options = data;
1054         substring_t args[MAX_OPT_ARGS];
1055         int option, token;
1056         int ret = 0;
1057
1058         do {
1059                 p = strsep(&options, ",");
1060                 if (p != NULL && *p) {
1061                         token = match_token(p, tokens, args);
1062                         if (token == Opt_snapshot) {
1063                                 if (!(sd->flags & MS_RDONLY))
1064                                         ret++;
1065                                 else {
1066                                         ret = match_int(&args[0], &option);
1067                                         if (!ret) {
1068                                                 if (option > 0)
1069                                                         sd->cno = option;
1070                                                 else
1071                                                         ret++;
1072                                         }
1073                                 }
1074                         }
1075                         if (ret)
1076                                 printk(KERN_ERR
1077                                        "NILFS: invalid mount option: %s\n", p);
1078                 }
1079                 if (!options)
1080                         break;
1081                 BUG_ON(options == data);
1082                 *(options - 1) = ',';
1083         } while (!ret);
1084         return ret;
1085 }
1086
1087 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1088 {
1089         struct nilfs_super_data *sd = data;
1090
1091         s->s_bdev = sd->bdev;
1092         s->s_dev = s->s_bdev->bd_dev;
1093         return 0;
1094 }
1095
1096 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1097 {
1098         struct nilfs_super_data *sd = data;
1099
1100         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1101 }
1102
1103 static int
1104 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1105              const char *dev_name, void *data, struct vfsmount *mnt)
1106 {
1107         struct nilfs_super_data sd;
1108         struct super_block *s;
1109         fmode_t mode = FMODE_READ;
1110         struct the_nilfs *nilfs;
1111         int err, need_to_close = 1;
1112
1113         if (!(flags & MS_RDONLY))
1114                 mode |= FMODE_WRITE;
1115
1116         sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1117         if (IS_ERR(sd.bdev))
1118                 return PTR_ERR(sd.bdev);
1119
1120         /*
1121          * To get mount instance using sget() vfs-routine, NILFS needs
1122          * much more information than normal filesystems to identify mount
1123          * instance.  For snapshot mounts, not only a mount type (ro-mount
1124          * or rw-mount) but also a checkpoint number is required.
1125          */
1126         sd.cno = 0;
1127         sd.flags = flags;
1128         if (nilfs_identify((char *)data, &sd)) {
1129                 err = -EINVAL;
1130                 goto failed;
1131         }
1132
1133         nilfs = find_or_create_nilfs(sd.bdev);
1134         if (!nilfs) {
1135                 err = -ENOMEM;
1136                 goto failed;
1137         }
1138
1139         mutex_lock(&nilfs->ns_mount_mutex);
1140
1141         if (!sd.cno) {
1142                 /*
1143                  * Check if an exclusive mount exists or not.
1144                  * Snapshot mounts coexist with a current mount
1145                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1146                  * ro-mount are mutually exclusive.
1147                  */
1148                 down_read(&nilfs->ns_super_sem);
1149                 if (nilfs->ns_current &&
1150                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1151                      & MS_RDONLY)) {
1152                         up_read(&nilfs->ns_super_sem);
1153                         err = -EBUSY;
1154                         goto failed_unlock;
1155                 }
1156                 up_read(&nilfs->ns_super_sem);
1157         }
1158
1159         /*
1160          * Find existing nilfs_sb_info struct
1161          */
1162         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1163
1164         /*
1165          * Get super block instance holding the nilfs_sb_info struct.
1166          * A new instance is allocated if no existing mount is present or
1167          * existing instance has been unmounted.
1168          */
1169         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1170         if (sd.sbi)
1171                 nilfs_put_sbinfo(sd.sbi);
1172
1173         if (IS_ERR(s)) {
1174                 err = PTR_ERR(s);
1175                 goto failed_unlock;
1176         }
1177
1178         if (!s->s_root) {
1179                 char b[BDEVNAME_SIZE];
1180
1181                 /* New superblock instance created */
1182                 s->s_flags = flags;
1183                 s->s_mode = mode;
1184                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1185                 sb_set_blocksize(s, block_size(sd.bdev));
1186
1187                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0,
1188                                        nilfs);
1189                 if (err)
1190                         goto cancel_new;
1191
1192                 s->s_flags |= MS_ACTIVE;
1193                 need_to_close = 0;
1194         }
1195
1196         mutex_unlock(&nilfs->ns_mount_mutex);
1197         put_nilfs(nilfs);
1198         if (need_to_close)
1199                 close_bdev_exclusive(sd.bdev, mode);
1200         simple_set_mnt(mnt, s);
1201         return 0;
1202
1203  failed_unlock:
1204         mutex_unlock(&nilfs->ns_mount_mutex);
1205         put_nilfs(nilfs);
1206  failed:
1207         close_bdev_exclusive(sd.bdev, mode);
1208
1209         return err;
1210
1211  cancel_new:
1212         /* Abandoning the newly allocated superblock */
1213         mutex_unlock(&nilfs->ns_mount_mutex);
1214         put_nilfs(nilfs);
1215         deactivate_locked_super(s);
1216         /*
1217          * deactivate_locked_super() invokes close_bdev_exclusive().
1218          * We must finish all post-cleaning before this call;
1219          * put_nilfs() needs the block device.
1220          */
1221         return err;
1222 }
1223
1224 struct file_system_type nilfs_fs_type = {
1225         .owner    = THIS_MODULE,
1226         .name     = "nilfs2",
1227         .get_sb   = nilfs_get_sb,
1228         .kill_sb  = kill_block_super,
1229         .fs_flags = FS_REQUIRES_DEV,
1230 };
1231
1232 static void nilfs_inode_init_once(void *obj)
1233 {
1234         struct nilfs_inode_info *ii = obj;
1235
1236         INIT_LIST_HEAD(&ii->i_dirty);
1237 #ifdef CONFIG_NILFS_XATTR
1238         init_rwsem(&ii->xattr_sem);
1239 #endif
1240         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1241         ii->i_bmap = &ii->i_bmap_data;
1242         inode_init_once(&ii->vfs_inode);
1243 }
1244
1245 static void nilfs_segbuf_init_once(void *obj)
1246 {
1247         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1248 }
1249
1250 static void nilfs_destroy_cachep(void)
1251 {
1252         if (nilfs_inode_cachep)
1253                 kmem_cache_destroy(nilfs_inode_cachep);
1254         if (nilfs_transaction_cachep)
1255                 kmem_cache_destroy(nilfs_transaction_cachep);
1256         if (nilfs_segbuf_cachep)
1257                 kmem_cache_destroy(nilfs_segbuf_cachep);
1258         if (nilfs_btree_path_cache)
1259                 kmem_cache_destroy(nilfs_btree_path_cache);
1260 }
1261
1262 static int __init nilfs_init_cachep(void)
1263 {
1264         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1265                         sizeof(struct nilfs_inode_info), 0,
1266                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1267         if (!nilfs_inode_cachep)
1268                 goto fail;
1269
1270         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1271                         sizeof(struct nilfs_transaction_info), 0,
1272                         SLAB_RECLAIM_ACCOUNT, NULL);
1273         if (!nilfs_transaction_cachep)
1274                 goto fail;
1275
1276         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1277                         sizeof(struct nilfs_segment_buffer), 0,
1278                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1279         if (!nilfs_segbuf_cachep)
1280                 goto fail;
1281
1282         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1283                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1284                         0, 0, NULL);
1285         if (!nilfs_btree_path_cache)
1286                 goto fail;
1287
1288         return 0;
1289
1290 fail:
1291         nilfs_destroy_cachep();
1292         return -ENOMEM;
1293 }
1294
1295 static int __init init_nilfs_fs(void)
1296 {
1297         int err;
1298
1299         err = nilfs_init_cachep();
1300         if (err)
1301                 goto fail;
1302
1303         err = register_filesystem(&nilfs_fs_type);
1304         if (err)
1305                 goto free_cachep;
1306
1307         printk(KERN_INFO "NILFS version 2 loaded\n");
1308         return 0;
1309
1310 free_cachep:
1311         nilfs_destroy_cachep();
1312 fail:
1313         return err;
1314 }
1315
1316 static void __exit exit_nilfs_fs(void)
1317 {
1318         nilfs_destroy_cachep();
1319         unregister_filesystem(&nilfs_fs_type);
1320 }
1321
1322 module_init(init_nilfs_fs)
1323 module_exit(exit_nilfs_fs)