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